1 views

Uploaded by shafinur04

- ON APPROXIMATE JACOBIAN MATRICES IN SIMULATION OF MULTIBODY SYSTEMS
- 012277281AI
- Validation of FACTS Models
- paper1.pdf
- pbis lesson- overview of pride1
- EXP_7.doc
- kubf8owsd6zdx0oj056w-signature-9d21828b8995bb45daf6bd98ae827dae2f0010efa3f03f445d719c9fcbe7ef93-poli-150624011216-lva1-app6892 (1)
- EE Control System 2001-2011(Solutions)
- Yichao Chen, Jonathan L. Gross and Toufik Mansour- Genus Distributions of Mobius Ladders: A Website Supplement
- magdaleno2017.pdf
- Matlab Manual
- FAXieeetac04
- Capitulo 9 Control
- Determination of State Space Matrices for Active Vibration
- 117CG - DIGITAL CONTROL SYSTEMS.pdf
- Three-Dimensional Iso Parametric Elements
- Array Formula
- Assignment Buck Design n Analysis 2012 Final
- FYBMS Sem 2 Syllabus
- Observer Pole Placement

You are on page 1of 266

New Developments in

Multibody System Dynamics

and Its Applications

Guest Editors: Xiaoting Rui, Chuanzeng Zhang, and Caishan Liu

New Developments in Multibody System

Dynamics and Its Applications

Advances in Mechanical Engineering

Dynamics and Its Applications

and Caishan Liu

Copyright © 2014 Hindawi Publishing Corporation. All rights reserved.

This is a special issue published in “Advances in Mechanical Engineering.” All articles are open access articles distributed under the

Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the

original work is properly cited.

Editorial Board

Koshi Adachi, Japan Luı́s Godinho, Portugal David R. Salgado, Spain

Mehdi Ahmadian, USA Tian Han, China Mohammad R. Salimpour, Iran

Rehan Ahmed, UK Francisco J. Huera-Huarte, Spain Sunetra Sarkar, India

Muhammad T. Akhtar, Japan Davood Jalali-Vahid, Iran Pietro Scandura, Italy

Nacim Alilat, France Jiin Y. Jang, Taiwan A. S. Sekhar, India

M. Affan Badar, USA Zhongmin Jin, UK Liyuan Sheng, China

Luis Baeza, Spain Xiaodong Jing, China Xi Shi, China

R. Balachandran, UK S.-W. Kang, Republic of Korea Seiichi Shiga, Japan

Claude Bathias, France Xianwen Kong, UK Chow-Shing Shin, Taiwan

Adib Becker, UK Michal Kuciej, Poland Ray W. Snidle, UK

Leonardo Bertini, Italy Yaguo Lei, China Margaret M. Stack, UK

Liam A. Blunt, UK Zili Li, The Netherlands Neil Stephen, UK

Noël Brunetière, France Yangmin Li, Macau Kumar K. Tamma, USA

Marco Ceccarelli, Italy Jaw-Ren Lin, Taiwan Yaya Tan, China

Fakher Chaari, Tunisia Cheng-Xian Lin, USA Anand Thite, UK

Chin-Lung Chen, Taiwan Seyed N. Mahmoodi, USA Cho W. Solomon To, USA

Hyung H. Cho, Republic of Korea Oronzio Manca, Italy Yoshihiro Tomita, Japan

Seung-Bok Choi, Korea Ramiro Martins, Portugal Shan-Tung Tu, China

Ahmet S. Dalkilic, Turkey Aristide F. Massardo, Italy Sandra Velarde-Suárez, Spain

Kangyao Deng, China Francesco Massi, Italy Moran Wang, China

Francisco D. Denia, Spain T. H. New, Singapore Junwu Wang, China

T. S. Dhanasekaran, USA Kim C. Ng, Singapore Jia-Jang Wu, Taiwan

Nihad Dukhan, USA Cong T. Nguyen, Canada Fengfeng Xi, Canada

Farzad Ebrahimi, Iran Hirosi Noguchi, Japan Gongnan Xie, China

Ali El Wahed, UK Hakan F. Oztop, Turkey Wei Mon Yan, Taiwan

Bogdan I. Epureanu, USA Duc T. Pham, UK Jianqiao Ye, UK

Mohammad R. Eslami, Iran Jurij Prezelj, Slovenia Byeng D. Youn, USA

Ali Fatemi, USA Xiaotun Qiu, USA Bo Yu, China

Mario L. Ferrari, Italy Pascal Ray, France Jianbo Yu, China

Siegfried Fouvry, France Robert L. Reuben, UK Zhongrong Zhou, China

Ian Frigaard, Canada Pedro A. R. Rosa, Portugal

Mergen H. Ghayesh, Canada Elsa de Sá Caetano, Portugal

Contents

New Developments in Multibody System Dynamics and Its Applications, Xiaoting Rui,

Chuanzeng Zhang, and Caishan Liu

Volume 2014, Article ID 671604, 1 page

Automatic Deduction Theorem of Overall Transfer Equation of Multibody System, Xiaoting Rui,

Jianshu Zhang, and Qinbo Zhou

Volume 2014, Article ID 378047, 12 pages

Optimal Design of One-Folded Leaf Spring with High Fatigue Life Applied to Horizontally Vibrating

Linear Actuator in Smart Phone, Ki Bum Lee, Chang Hyun Park, and Jin Ho Kim

Volume 2014, Article ID 545126, 7 pages

Near-Earth Orbits, Jingrui Zhang, Changqin Yuan, Dongmei Jiang, and Dawei Jin

Volume 2014, Article ID 512583, 9 pages

Application of Krylov Reduction Technique for a Machine Tool Multibody Modelling, M. Sulitka,

J. Šindler, J. Sušeň, and J. Smolı́k

Volume 2014, Article ID 592628, 9 pages

Study on Launch Dynamics of Self-Propelled Artillery Based on Transfer Matrix Method of Multibody

System, Hailong Yu and Xiaoting Rui

Volume 2014, Article ID 308049, 9 pages

Higher Order Variational Integrators for Multibody System Dynamics with Constraints, Jieyu Ding and

Zhenkuan Pan

Volume 2014, Article ID 383680, 8 pages

Free Vibration Characteristic of Multilevel Beam Based on Transfer Matrix Method of Linear

Multibody Systems, Laith K. Abbas and Xiaoting Rui

Volume 2014, Article ID 792478, 16 pages

Case Study on Human Walking during Wearing a Powered Prosthetic Device: Effectiveness of the System

“Human-Robot”, Svetlana Grosu, Pierre Cherelle, Chris Verheul, Bram Vanderborght, and Dirk Lefeber

Volume 2014, Article ID 365265, 9 pages

Controller Parameters Tuning Based on Transfer Matrix Method for Multibody Systems,

Hossam Hendy, Xiaoting Rui, Qinbo Zhou, and Mostafa Khalil

Volume 2014, Article ID 957684, 8 pages

All Terrain Vehicle Flexible Multibody Dynamic Simulation for Fatigue Prediction, Jia-Shiun Chen and

Hsiu-Ying Hwang

Volume 2013, Article ID 725315, 6 pages

Projectile Impact Point Prediction Based on Self-Propelled Artillery Dynamics and Doppler Radar

Measurements, Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy

Volume 2013, Article ID 153913, 12 pages

Study on the Dynamics of Laser Gyro Strapdown Inertial Measurement Unit System Based on Transfer

Matrix Method for Multibody System, Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang,

and Qinbo Zhou

Volume 2013, Article ID 854583, 9 pages

Recursive Formulation for Dynamic Modeling and Simulation of Multilink Spatial Flexible Robotic

Manipulators, Zhenjie Qian, Dingguo Zhang, and Jun Liu

Volume 2013, Article ID 216014, 15 pages

Volume 2013, Article ID 139498, 8 pages

Dynamic Control and Disturbance Estimation of 3D Path Following for the Observation Class

Underwater Remotely Operated Vehicle, Hai Huang, Qirong Tang, Yueming Li, Lei Wan, and Yongjie Pang

Volume 2013, Article ID 604393, 16 pages

Application of Finite Volume Method to Structural Stochastic Dynamics, Weidong Chen, Yanchun Yu,

Ping Jia, Xiande Wu, and Fengchao Zhang

Volume 2013, Article ID 391704, 10 pages

Numerical Simulation on Dense Packing of Granular Materials by Container Oscillation, Jun Liu and

Dongxu You

Volume 2013, Article ID 284693, 7 pages

Research on Flexible Joint Stiffness Test and Trajectory Planning of Space Manipulator, Yongtai Yang,

Jili Rong, Jian Li, and Ling Tang

Volume 2013, Article ID 280453, 7 pages

Case Study on Influence of Step Blast-Excavation on Support Systems of Existing Service Tunnel with

Small Interval, Shaorui Sun, Ling Yue, Jimin Wu, Jin Liu, and Jihong Wei

Volume 2013, Article ID 257457, 13 pages

The Effect of First-Order Bending Resonance of Wheelset at High Speed on Wheel-Rail Contact

Behavior, Shuoqiao Zhong, Xinbiao Xiao, Zefeng Wen, and Xuesong Jin

Volume 2013, Article ID 296106, 19 pages

Dynamic Modeling and Analysis of Hub-Tapered Beam System, Si-jia Chen and Ding-guo Zhang

Volume 2013, Article ID 313279, 12 pages

Based on the Transfer Matrix Method of Multibody System, Jianguo Ding, Yuwei Dai, and Zhi Qiao

Volume 2013, Article ID 784525, 10 pages

Cooperative Search by Combining Simulated and Real Robots in a Swarm under the View of Multibody

System Dynamics, Qirong Tang and Peter Eberhard

Volume 2013, Article ID 284782, 11 pages

Global Impact Dynamic Modeling and Verification of a Flexible Beam with Large Overall Motion,

Yuechen Duan, Dingguo Zhang, and Jiazhen Hong

Volume 2013, Article ID 362317, 8 pages

Evaluating Energy Flux in Vibrofluidized Granular Bed, N. A. Sheikh, S. Manzoor, and U. Ghani

Volume 2013, Article ID 327379, 7 pages

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 671604, 1 page

http://dx.doi.org/10.1155/2014/671604

Editorial

New Developments in Multibody System Dynamics and

Its Applications

1

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

2

Department of Civil Engineering, University of Siegen, 57068 Siegen, Germany

3

College of Engineering, Peking University, Beijing 100871, China

Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Multibody system dynamics is nowadays a widely used tool- and precision, the model and modal reduction techniques

box in the development of modern engineering science and for multibody system dynamics, the optimization and control

technology. It provides powerful dynamic design theories, of a swarm of robots, the optimal design of smart systems,

computational and analytical methods, and test technology the dynamics and control of underwater vehicle and space-

for innovative engineering applications. In order to sys- craft formation, the perturbation-finite volume method for

tematically show the important effects of multibody system stochastic multibody system dynamics, the blast-excavation

dynamics on the development of modern science and tech- dynamics, the granular system dynamics, the vehicle-track

nology and to strengthen the academic exchanges in the coupling dynamics, and the human-robot dynamics. Thus,

research field of multibody system dynamics and its applica- the general fundamental theories and several key engineering

tions, we organized this special issue containing twenty-five applications are covered representing the state of the art of

papers selected from the contributions coming from Belgium, recent intensive and extensive research activities in the field

Czech Republic, Egypt, Germany, Iraq, Netherlands, Pak- of multibody system dynamics to a certain extent.

istan, China, Republic of Korea, United Kingdom, Taiwan, The guest editors hope that this special issue will pro-

and so forth. vide some valuable information and guidance for scientists

The twenty-five papers selected for this Special Issue are working on multibody system dynamics and its applications

devoted to the following topics: an automatic deduction the- in modern engineering science and technology.

orem of the overall transfer equations of the transfer matrix

method for multibody systems (MSTMM), the application

of MSTMM in dynamics of a multilevel beam system, the Acknowledgments

application of MSTMM in dynamics of an elastic-plastic Many thanks are due to Professor Edwin Kreuzer from

system subjected to earthquakes, the application of MSTMM Hamburg University of Technology and Professor Dieter

in dynamics of controlled cars, the application of MSTMM Bestle from Brandenburg University of Technology Cottbus-

in dynamics of a laser gyro strapdown inertial measurement Senftenberg for their great unselfish help in organizing the

unit system, the application of MSTMM in launch dynamics special issue, to reviewers and authors, and to the reviewers

of a self-propelled artillery system and flight dynamics of for their voluntary works and constructive comments. All

projectile, the modeling of trivariate isogeometrics for flexible of these promoted greatly the projected publication of this

multibody dynamics, the modeling of hub-tapered beam special issue.

system dynamics, the formulations and modeling approaches

of spatial flexible robotic manipulators, the impact in flexible Xiaoting Rui

systems, the modeling and test methods for flexible joint Chuanzeng Zhang

stiffness of a space manipulator, the computational stability Caishan Liu

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 378047, 12 pages

http://dx.doi.org/10.1155/2014/378047

Research Article

Automatic Deduction Theorem of Overall Transfer

Equation of Multibody System

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Transfer matrix method for multibody System (MSTMM) is a new multibody dynamics method developed in recent 20 years. It

has been widely used in both science research and engineering for its special features as follows: without global dynamics equations

of the system, high programming, low order of system matrix, and high computational speed. Based on MSTMM and its above

features, a theorem to deduce automatically the overall transfer equations of multibody systems by handwriting or by computer

is proposed in this paper. The theorem is effective for multibody systems with various topological structures, including chain

systems, closed-loop systems, tree systems, general systems composed of one tree subsystem, and some closed-loop subsystems.

This theorem makes it possible to program large scale software of multibody system dynamics with much higher programming,

and much higher computational speed because of the above features of MSTMM. Formulations of the proposed method as well as

two examples are given to verify this method.

equations of the system, keep low order of the system

Lots of methods dealing with multibody system dynamics matrix so very high computational speed, and avoid the

(MSD) have been studied by many authors since 1960s [1– difficulties in computation caused by high-order matrices and

18]. They are widely used in many engineering fields such as high programming. It has been proved by lots of theories

aeronautics, astronautics, spacecraft, vehicle, robot, precision and experiments that MSTMM is effective for linear time-

machinery, and biomechanics. It is well known that almost invariant multibody systems [22], nonlinear time-variant

all the previous ordinary methods for MSD have the same multibody systems, multi-rigid-body systems [19, 21], multi-

characteristics as follows: it is necessary and very complicated rigid-flexible-body systems [22–24], and controlled multi-

to develop the global dynamics equations of the system; the body systems [19, 29].

order of system matrix depends on the number of degrees of Generally speaking, various multibody systems may be

freedom of the system and hence it is rather high for complex considered as one of the following four cases in topology [24]:

multibody system. (1) chain system, (2) closed-loop system, (3) tree system, (4)

To avoid establishing the global dynamics equations of general systems composed of one tree subsystem, and some

the system, simplify the study procedure and especially keep closed-loop subsystems. A chain system can be considered

high computational efficiency independent of the number as a special example of a tree system at the case with only

of degree of freedom of system in studying MSD, new two boundary ends. By “cutting” at one connection point

analytical method for MSD, namely, transfer matrix method of a system, a closed-loop system can be considered as

for multibody system (MSTMM), is presented by Rui and a chain system, and a general system composed of one

his co-workers [19–21] and constantly developed in recent tree subsystem and some closed-loop subsystems can be

20 years [22–25]. Nowadays, MSTMM is widely applied in dealt with a tree system [24]. Based on MSTMM and its

science research and engineering [26–29] for the features of above features, a theorem to deduce automatically the overall

2 Advances in Mechanical Engineering

transfer equations of various multibody systems mentioned time derivatives of the generalized coordinates describing

above by handwriting and by computer is proposed in this the deformation of flexible bodies using modal method; the

paper. superscript 𝑛 is the highest order of the modal considered, r,̈

𝜃,̈m, and q are the column matrices of accelerations, second

2. General Theorems and Steps of MSTMM time derivatives of the space-three angles, internal torques,

and internal forces in 𝑜𝑥𝑦𝑧, respectively.

2.1. Basic Idea of MSTMM. The basic idea of MSTMM [19] is For body and hinge elements moving in a plane, a similar

to break up a multibody system into the elements contain- definition of the state vector can be introduced, which is a

ing bodies (including rigid bodies, flexible bodies, lumped special example of spatial motion.

masses, etc.) and hinges (including joints, ball-and-sockets, The transfer equations of the 𝑗th element can be obtained

pins, springs, rotary springs, dampers and rotary dampers, easily by rewriting its dynamics equations as follows [19, 21,

etc.) whose dynamics properties can be readily expressed 30]

in matrix forms. These matrices of elements are considered

as building blocks that provide the dynamics properties of z𝑗,𝑗+1 = U𝑗 (𝑡𝑖 ) z𝑗,𝑗−1 , (5)

the entire system when assembling them together according

to the topology of the system. Particularly, the positions where U𝑗 (𝑡𝑖 ) is the transfer matrix of the 𝑗th element, which

of bodies and hinges are considered equivalent in transfer expresses the relationship between the state vectors of its

equations and transfer matrices, which is totally different output end z𝑗,𝑗+1 and input end z𝑗,𝑗−1 and is already known

from ordinary methods for MSD [1–18] and results in the at time instant 𝑡𝑖 .

very low order of system matrix and very high computational It should be pointed out that there are general linear

speed in MSTMM. relations among accelerations, angular accelerations, forces,

and torques of any mechanics system in an inertial coordinate

2.2. State Vector, Transfer Equation, and Transfer Matrix of system 𝑜𝑥𝑦𝑧, according to Newton motion law and Euler

Element. The same coordinate systems and sign conventions theorem of moment of momentum. It is to say that there are

as are described in [19, 21, 30] will be used. The state vector strict linear relations between the state vectors of output end

of the connection point between any rigid body and hinge and input end of any element and among all state vectors

moving in space is defined as of a multibody system. Thus, the transfer equation (5) is a

general equation and effective for any mechanics element in

𝑇 the inertial coordinate system.

z = [𝑥̈𝑦̈𝑧̈𝜃𝑥̈ 𝜃𝑦̈ 𝜃𝑧̈ 𝑚𝑥 𝑚𝑦 𝑚𝑧 𝑞𝑥 𝑞𝑦 𝑞𝑧 1] , (1)

𝑇 the System. The overall transfer equations of a chain system

𝑇

z = [r𝑇̈ 𝜃̈ m𝑇 q𝑇 1] , (2) can be deduced automatically as [19, 24]

[ ]

r̈= [𝑦]̈, 𝜃̈= [𝜃𝑦̈] ,

𝑧̈ ̈ U𝑛−1 = U1 U2 U3 ⋅ ⋅ ⋅ U𝑛 .

[ ] [𝜃𝑧 ] (3)

(7)

m = [𝑚𝑦 ] , q = [𝑞𝑦 ] . transfer equation for a chain system can be clearly seen;

[𝑚𝑧 ] [𝑞𝑧 ] overall transfer matrix of a chain system can be deduced auto-

matically by successive premultiplication of transfer matrices

The state vector of connection point between flexible of every element of the system along the transfer path from

body and hinge moving in space is defined as the one end to another end.

2.4. Solutions of the System Motion. Applying the boundary

(4) conditions of the system, z1,0 and z𝑛,0 , to the overall system

𝑇

𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 𝑞1̈, 𝑞2̈, . . . , 𝑞𝑛̈, 1] , transfer equation (6), the unknown state variables in the

boundary state vectors can be computed. Then, the state

where 𝑥,̈𝑦,̈and 𝑧̈are the accelerations of the connection point vectors of each element at time 𝑡𝑖 can be computed by the

with respect to the global inertial coordinate system 𝑜𝑥𝑦𝑧; 𝜃𝑥̈, repeated use of corresponding transfer equations of element

𝜃𝑦̈, and 𝜃𝑧̈are the second time derivatives of the orientation (5). Then, using the computed values of the last step as the

angles of the body using space-three-angles 1-2-3 𝜃𝑥 , 𝜃𝑦 , and initial conditions, the entire procedure can be repeated from

𝜃𝑧 defined in [18]; 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , and 𝑞𝑧 are the cor- time 𝑡𝑖+1 until the time required for complete analysis. It can

responding internal torques and internal forces in the same be seen clearly from equations (5)–(7) that MSTMM belongs

reference system, respectively; 𝑞1̈, 𝑞2̈, . . . , 𝑞𝑛̈are the second to one of strict analytical methods in principle.

Advances in Mechanical Engineering 3

𝒛17,0 0 𝒛18,0

Sign Conventions 15 16 17 18

3.1. Topology Figure of a Multibody System. A dynamics

model of any complex multibody system can be constructed 11 12 13 14

with dynamics elements including bodies and hinges. In

order to describe the transfer relationship among the state 9 10

vectors of elements and the transfer directions in a system, the

topology figure of the model will be very useful for deduction 8

of overall transfer equation in MSTMM. And the topology 7 6

figure of the dynamics model of a system, for example, a tree 5

multibody system, as shown in Figure 1, can be got very easily 4

and directly from its dynamics model if using the following

sign conventions. 3

2

3.2. The Sign Conventions. Besides the sign conventions

1

introduced in [19, 21], the sign conventions as follows are used Root

in the paper. 0

𝒛1,0

(1) A circle I denotes a body element and the number Figure 1: Topology figure of a dynamics model of tree multibody

inside this circle is the sequence number of the body system.

element.

(2) An arrow → denotes a hinge element and the transfer 4. Automatic Deduction of Overall Transfer

direction of state vectors; the number beside the

arrow is the sequence number of the hinge element. Equations of System

4.1. Automatic Deduction of the Overall Transfer Equation of

(3) Each body element is dealt with single output end and

a Chain System. The topology figure of any chain system is

single input end if the body has two connection ends

shown in Figure 2.

with other elements; otherwise, it is dealt with single

It is clear that we can rewrite the overall transfer equation

output end and multiple input ends if the body has

(6) of the chain system as

more two connection ends.

𝑇

(4) For a nonboundary end, the first and second sub- Uall [z𝑇1,0 z𝑇𝑛,0 ] = 0, Uall = [−I U𝑛−1 ] . (8)

scripts, 𝑖 and 𝑗 (𝑖, 𝑗 ≠0), in a state vector z𝑖,𝑗 of the

end denote the sequence numbers of the adjacent From equations (6) or (7), it can be seen clearly that

body element and hinge element, respectively. For the overall transfer matrix of any chain system Uall can be

a boundary end, the second subscript 𝑗 = 0 in deduced automatically by handwriting and by computer. In

the state vector z𝑖,𝑗 ; that is, the second subscript 0 the overall transfer matrix, the coefficient matrix of the state

means boundary end; then the first subscript 𝑖 in the vector of root is a minus unit matrix I, while the coefficient

state vector z𝑖,0 of the boundary end stands for the matrix of the state vector of tip U𝑛−1 is the successive

sequence number of the element involved. premultiplication of the transfer matrices of all elements in

the transfer path from this tip to the root as shown in equation

(5) In a multibody system, only one boundary end is

(7).

considered as the root; the state vector of root is noted

The highest order of the overall transfer matrix is 13 for

as z𝑖,0 , where 𝑖 is the sequence number of the root

spatial chain multi-rigid-body system or (13 + 𝑛) for chain

element; all of other boundary ends are considered as

multi-rigid-flexible-body system, where 𝑛 is the highest order

the tips; the state vectors of tips are denoted as z𝑗,0 ,

of the modal considered.

where 𝑗 is the sequence number of the tip element.

The transfer directions of a system are always from its

tips to the root. 4.2. Automatic Deduction of the Overall Transfer Equations

of a Closed-Loop System. For any closed-loop system, whose

(6) The subscript 𝑖 in transfer matrix U𝑖 denotes the topology figure is shown in Figure 3, after “cutting” at the

sequence number of element 𝑖. The subscript (𝑖 − 𝑘) junction of any two adjacent elements such as body 1 and

in the transfer matrix U𝑖−𝑘 and the partitioned matrix hinge 𝑛 as shown in Figure 4, consider the couple of “cutting

u𝑖−𝑘 means from element 𝑖 to element 𝑘. U𝑖−𝑘 and u𝑖−𝑘 point” as the “boundary ends” with the same state vectors

mean the successive premultiplication of the transfer noted as z𝑛,0 and z1,0 . Then the original closed-loop system

matrices of all elements in the transfer path from the becomes a chain system with the same “boundary ends” state

element 𝑖 to element 𝑘 of the system. vectors, z𝑛,0 and z1,0 , as shown in Figure 4.

4 Advances in Mechanical Engineering

0 n − 1 ··· 7 5 3 1 0 Root from the state vector when it premultiplies z𝑗,𝐼𝑘 , acted on the

𝒛n,0 n 6 4 2 same point.

𝒛1,0

However, in equation(13), the number of unknown vari-

Figure 2: Topology figure of a chain system. ables is more than that of algebraic equations. Therefore,

geometrical equations of the body, which describes the

geometrical relationship between the first input end and

kth (𝑘 = 2, 3, 4, . . . , 𝐿) input end of the body, should be

7 ··· n−1 n 1 introduced for body elements with single output end and

multiple input ends. It is verified later that the geometrical

6 equation can be written in the form of

2

5

4

where H𝑗 is a constant matrix extracting displacement vari-

Figure 3: Topology figure of a closed-loop system. ables (including position coordinates and orientation angles)

from a state vector when premultiplying z𝑗,𝐼1 ; H𝑗,𝐼𝑘 is related

to the relative position between the first input end and kth

Thus, the transfer equation of the closed-loop system can input end of body. The number of the geometrical equations

be deduced automatically as is achieved for the chain system is (𝐿 − 1).

A spatial motion rigid body with more than two ends is

z1,0 = U𝑛−1 z𝑛,0 (9) shown in Figure 5; the concrete form of the transfer equations

and geometrical equations of body element will be exhibited.

according to the proposed sign conventions, where The state vectors of inboard ends and outboard ends of

the rigid body, as defined in Section 2.2, are

U𝑛−1 = U1 U2 U3 ⋅ ⋅ ⋅ U𝑛 . (10)

𝑇

Attention should be paid that the state vectors of a couple z𝐼𝑘 = [𝑥,̈𝑦,̈𝑧,̈𝜃𝑥̈, 𝜃𝑦̈, 𝜃𝑧̈, 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 1]𝐼

𝑘

of “cutting points” are the same, namely

(𝑘 = 1, 2, . . . , 𝐿) (15)

z𝑛,0 = z1,0 . (11)

𝑇

z𝑂 = [𝑥,̈𝑦,̈𝑧,̈𝜃𝑥̈, 𝜃𝑦̈, 𝜃𝑧̈, 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 1]𝑂.

Then, the transfer equation (9) of the closed-loop system

can be deduced automatically by handwriting and by com- As shown in Figure 5, 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿), 𝑂, and 𝐶 denote

puter the inboard ends, outboard end, and mass center of the rigid

body, respectively; the coordinate system with subscript 2

Uall z1,0 = 0, Uall = (I − U𝑛−1 ) . (12) denotes the body-fixed coordinate system whose initial point

𝑜2 is on the first inboard end 𝐼1 of the rigid body. According

4.3. Automatic Deduction of the Overall Transfer Equations to the properties of a rigid body, the geometrical relationship

of a Tree System. For each body element with more than between the first inboard end and the outboard end of the

two ends, as is mentioned in the convention, only one of rigid body can be obtained as

the ends is considered as output end and all the other ends

are input ends. Moreover, its transfer equations should cover 𝜃𝑂 = 𝜃𝐼1 ,

the geometrical relationship between its first input end and (16)

output end and describe the mechanical principle for the r𝑂 = r𝐼1 + Al𝐼𝐼 𝑂,

forces and moments acting on the element. Thus, it can be

verified later that the transfer equations of a rigid body 𝑗 with where 𝜃𝐼1 and 𝜃𝑂 are the column matrices of the orientation

𝐿 input ends can be written in the following form coordinates of the first inboard end 𝐼1 and the outboard end

𝑂; r𝑂 and l𝐼𝐼 𝑂 are the column matrices of the position coordi-

z𝑗,𝑂 = U𝑗 z𝑗,𝐼1 + U𝑗,𝐼2 z𝑗,𝐼2 + ⋅ ⋅ ⋅ + U𝑗,𝐼𝐿 z𝑗,𝐼𝐿 , (13) nates of point 𝑂 with respect to the global inertial coordinate

system 𝑜𝑥𝑦𝑧 and the body-fixed coordinate system 𝑜2 𝑥2 𝑦2 𝑧2

where the subscript 𝑗 is the sequence number of the body; respectively; A is the direction cosine matrix that defines the

𝑂 and 𝐼1 , 𝐼2 , 𝐼3 ⋅ ⋅ ⋅ 𝐼𝐿 denote the output end and input ends orientation of the body-fixed coordinate system 𝑜2 𝑥2 𝑦2 𝑧2 in

respectively, and the first input end 𝐼1 is considered as the the global inertial coordinate system 𝑜𝑥𝑦𝑧.

dominant input end; z𝑗,𝑂 and z𝑗,𝐼𝑘 (𝑘 = 2, 3, 4, . . . , 𝐿) are Similarly, the geometrical equations between the first

the state vectors of the output end and the kth input end of inboard end 𝐼1 and the other inboard ends 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿)

the body element respectively; U𝑗 is just the transfer matrix can be obtained easily

of element 𝑗 when point 𝐼1 and point 𝑂 are considered as

the only input end and output end of this element; U𝑗,𝐼𝑘 is 𝜃𝐼𝑘 = 𝜃𝐼1 (𝑘 = 2, 3, . . . , 𝐿)

the corresponding extraction matrix that extracts the force (17)

variables (including internal force and internal moment) r𝐼𝑘 = r𝐼1 + Al𝐼𝐼 𝐼𝑘 (𝑘 = 2, 3, . . . , 𝐿) .

Advances in Mechanical Engineering 5

𝒛n,0 𝒛1,0 By equations (16), (18), and rewriting them in the form of

7 ··· n−1 0 0 1

equation (13), transfer matrices can be obtained as

n

I3 E1 O3×3 O3×3 E2

6

2 [ O3×3 I3 O3×3 O3×3 O3×1 ]

[ ]

U𝑗 = [

[E6 E3 + E7 E6 E4 + E8 I3 E6 E6 E5 + E9 ]

],

5 3 [ E3 E4 O3×3 I3 E5 ]

4

[ O1×3 O1×3 O1×3 O1×3 1 ]

Figure 4: Topology figure of a closed-loop system after “cutting” the (19)

hinge 𝑛.

O3×3 O3×3 O3×3 O3×3 O3×1

[O3×3 O3×3 O3×3 O3×3 O3×1 ]

[ ]

U𝑗,𝐼𝑘 =[

[O3×3 O3×3 I3 ̃r𝐼1 𝐼𝑘 O3×1 ]

]

[O3×3 O3×3 O3×3 I3 O3×1 ] (20)

y2

O [O1×3 O1×3 O1×3 O1×3 1 ]

y

x2 (𝑘 = 2, 3, 4, . . . , 𝐿) ,

C

o2 IL E1 = −Ãl𝐼𝑂H, E2 = Ã

𝜔𝜔̃l𝐼𝑂 − Ãl𝐼𝑂Ḣ𝜃̇

𝐼,

I1 I2

E3 = −𝑚I3 , E4 = 𝑚Ãl𝐼𝐶H,

o z2

x

E5 = f𝐶 − 𝑚Ã

𝜔𝜔̃l𝐼𝐶 + 𝑚Ãl𝐼𝐶Ḣ𝜃̇

𝐼,

z

E6 = ̃r𝐼𝑂, E7 = 𝑚̃r𝐼𝐶, E8 = AJ𝐼 H,

Figure 5: A spatial motion rigid body with more than two ends.

𝜔J𝜔 + AJḢ𝜃̇

E9 = −m𝐶 − ̃r𝐼𝐶f𝐶 + Ã 𝐼,

(21)

̃ = A𝑇 A,̇

𝜔 𝜔 = H𝜃,̇

conventions, the dynamics equations of the rigid body can be A = [𝑐𝑦 𝑠𝑧 𝑠𝑥 𝑠𝑦 𝑠𝑧 + 𝑐𝑥 𝑐𝑧 𝑐𝑥 𝑠𝑦 𝑠𝑧 − 𝑠𝑥 𝑐𝑧 ] ,

obtained in the global inertial reference frame as [30] [ −𝑠𝑦 𝑠𝑥 𝑐𝑦 𝑐𝑥 𝑐𝑦 ]

1 0 −𝑠𝑦

H = [0 𝑐𝑥 𝑠𝑥 𝑐𝑦 ] ,

𝐿

𝑚r𝐶̈= ∑ q𝐼𝑘 − q𝑂 + f𝐶, [0 −𝑠𝑥 𝑐𝑥 𝑐𝑦 ]

𝑘=1

where l𝐴𝐵 is the column matrix of the position vector from

𝐿 𝐿 (18) point 𝐴 to point 𝐵 in the body-fixed coordinate whose

Ġ𝐼1 = − ∑ m𝐼𝑘 + ∑̃r𝐼1 𝐼𝑘 q𝐼𝑘 + m𝑂 − ̃r𝐼1 𝑂q𝑂 original point is the first input end 𝐼1 ; ̃l𝐴𝐵 is the cross product

𝑘=1 𝑘=2

matrix of l𝐴𝐵 . The meanings of other signs are the same with

− m̃r𝐼1 𝐶r𝐼̈

1

+ m𝐶 + ̃r𝐼1 𝐶f𝐶, those in [19, 30].

By geometrical equations (17) and rewriting them in the

form of equation (14), the concrete form of matrices H𝑗 and

H𝑗,𝐼𝑘 can be acquired as

where 𝑚 and r𝐶̈are the mass and the column matrix of mass

center acceleration of the rigid body respectively; q𝐼𝑘 (𝑘 = H𝑗 = [I6 O6×7 ]

1, 2, . . . , 𝐿) and q𝑂 are the column matrices of internal forces

acting on the point 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) and 𝑂, respectively; H𝑗,𝐼𝑘

f𝐶 and m𝐶 are the column matrices of external force and the

external torque acting on the mass center of the rigid body; I3 Ãl𝐼1 𝐼𝑘 H O3×3 O3×3 Ãl𝐼1 𝐼𝑘 Ḣ𝜃̇− Ã

𝜔𝜔̃l𝐼1 𝐼𝑘

G𝐼1 is the column matrix of relative moment of momentum =[ ]

O3×3 I3 O3×3 O3×3 O3×1

with respect to the first input end 𝐼1 of the body element;

m𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) and m𝑂 are the column matrices (𝑘 = 2, 3, 4, . . . , 𝐿) .

of internal torques acting on points 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) (22)

and 𝑂, respectively; r𝐼1 𝑃 represents the column matrix of

position vectors from 𝐼1 to point 𝑃 with respect to the inertial It can be clearly seen from equation (19) that the matrix is

coordinate system, where 𝑃 represents 𝐼𝑘 (𝑘 = 2, 3, . . . , 𝐿), 𝐶, exactly the same with the transfer matrix of rigid body with

or 𝑂. single input end and single output end, which in fact can be

6 Advances in Mechanical Engineering

regarded as a special case of rigid body with more than two Similarly, the geometrical equations corresponding to

ends (multiple input ends and single output end). body elements 10 and 4 can be deduced as

Based on the transfer equations and geometrical equa-

tions of elements, it is then easy to get the overall transfer G17−10 z17,0 + G18−10 z18,0 = 0,

equation of the system automatically.

G15−4 z15,0 + G16−4 z16,0 + G17−4 z17,0 (30)

According to the topology figure of the system shown in

Figure 1, the relations among the state vectors and transfer + G18−4 z18,0 = 0,

equations of elements can been described more intuitively

and directly using topology described by state vectors and where

transfer equations as shown in Figure 6. G17−10 = −H10 U13 U17 ,

Then the main transfer equations of the system in

Figure 6 can be easily deduced, that is, G18−10 = H10,14 U14 U18 ,

(31)

+ U1 U2 U3 U4,6 z4,6 G16−4 = −H4 U5 U7 U8 U9,12 U12 U16 ,

(23)

+ U1 U2 U3 U4 U5 U7 U8 U9,12 z9,12 The overall transfer equation of the system can be

obtained by combining the main transfer equation (23) and

+ U1 U2 U3 U4,6 U6 U10 z10,13 all the geometrical equations (28) and (30) of the system:

+ U1 U2 U3 U4,6 U6 U10,14 z10,14 Uall zall = 0, (32)

= T15−1 z15,0 + T16−1 z16,0 + T17−1 z17,0 + T18−1 z18,0 , where

−I T15−1 T16−1 T17−1 T18−1

where [ O G15−9 G16−9

[ O O ] ],

Uall = [

T15−1 = U1 U2 U3 U4 U5 U7 U8 U9 U11 U15 , O O O G17−10 G18−10 ]

[ O G15−4 G16−4 G17−4 G18−4 ]

T16−1 = U1 U2 U3 U4 U5 U7 U8 U9,12 U12 U16 ,

(24) z1,0 (33)

T17−1 = U1 U2 U3 U4,6 U6 U10 U13 U17 , [z15,0 ]

[ ]

zall = [ ]

[z16,0 ] .

T18−1 = U1 U2 U3 U4,6 U6 U10,14 U14 U18 . [z17,0 ]

Further, the geometrical equation of the body element 9 is [z18,0 ]

According to the sign conventions and equations (32)

H9 z9,11 = H9,12 z9,12 . (25) and (33), it can be seen clearly that the overall transfer

equation of the tree system, such as Figure 1, can be deduced

Applying system topology described by state vectors and automatically by handwriting and by computer. For more

transfer equations in Figure 6, the state vectors z9,11 and z9,12 details, see Section 5 please.

can be expressed using state vectors of system boundaries as As a short conclusion, in the overall transfer equation, zall

consists of all the state vectors at the boundary ends of system.

z9,11 = U11 U15 z15,0 ,

(26) Besides the main transfer equation, there exists a geometrical

z9,12 = U12 U16 z16,0 . equation at multiple input body elements that finally leads to a

branch in a tree system. Moreover, the number of geometrical

Substituting equation (26) into equation (25), one obtains equations in the overall transfer equation of a system is equal

to the number of tips of a tree minus one. For the system

−H9 U11 U15 z15,0 + H9,12 U12 U16 z16,0 = 0 (27) shown in Figure 1, the number of the tips is 4; thus the

number of geometric equations of the system is 3.

which can be written as It can be seen clearly that overall transfer equation (8) of a

chain system is a special example of overall transfer equations

G15−9 z15,0 + G16−9 z16,0 = 0, (28) (32) and (33) at the case of the tree system with only two

boundary ends.

where

G15−9 = −H9 U11 U15 , 4.4. Automatic Deduction of the Overall Transfer Equations of

(29) a General System. As is shown in Figure 7, a general multi-

G16−9 = H9,12 U12 U16 . body system can be considered as a system which consists

Advances in Mechanical Engineering 7

𝒛15,11 = 𝑼15 𝒛15,0 𝒛16,12 = 𝑼16 𝒛16,0 𝒛17,13 = 𝑼17 𝒛17,0 𝒛18,14 = 𝑼18 𝒛18,0

𝒛9,11 = 𝑼11 𝒛15,11 𝒛9,12 = 𝑼12 𝒛16,12 𝒛10,13 = 𝑼13 𝒛17,13 𝒛10,14 = 𝑼14 𝒛18,14

𝒛7,8 = 𝑼8 𝒛8,9

𝒛7,5 = 𝑼7 𝒛7,8

𝒛2,3 = 𝑼3 𝒛4,3

𝒛2,1 = 𝑼2 𝒛2,3

𝒛1,0 = 𝑼1 𝒛2,1

After “cutting” at the junction of any two adjacent elements

body 16 and hinge 19 in a closed-loop subsystem, a couple

of “new boundaries” noted as z16,0 and z19,0 , which are the −I T15−1 T16−1 T19−1 T18−1

same in nature, will emerge at the “cutting point.” Then, as [ O G15−9 G16−9 O O ]

[ ]

shown in Figure 8, the original nontree system shown in Uall = [

[ O O O G 19−10 G ]

18−10 ] ,

Figure 7 will become a tree system with the same two “new [ O G15−4 G16−4 G19−4 G18−4 ]

boundaries” state vectors z16,0 and z19,0 at the “cutting points.”

[O O C −I O ]

And the overall transfer equation of the general system (37)

with closed-loop subsystems can be deduced automatically z1,0

by handwriting and by computer according to the method [z15,0 ]

[ ]

proposed in Section 4.3. zall = [ ]

[z16,0 ] .

It should be pointed out that either z16,0 or z19,0 is the [z19,0 ]

input end state vector of elements 16 or 19 respectively. Thus, [z18,0 ]

the internal force and internal moment in z16,0 and z19,0 are

of the same quantities but with opposite direction due to the

sign conventions, that is,

Considering that z19,0 can be explicitly expressed by z16,0

z19,0 = Cz16,0 , (34) due to equation (34), the overall transfer equation, the overall

transfer matrix, and the overall state vectors equation (37) of

where the general system can be rewritten in the following form

𝐼𝑛 𝑂𝑛 𝑂𝑛×1

C = [ 𝑂𝑛 −𝐼𝑛 𝑂𝑛×1 ] (35)

[𝑂1×𝑛 0 1 ] Uall zall = 0

𝑛 is equal to 3 for planar motion or 6 for spatial motion. −I T15−1 T16−1 + T19−1 C T18−1

Considering the method proposed in Section 4.3 for a [ O G15−9 G16−9 O ]

[

Uall = [ ],

tree system, and regarding equation (34) concerned with the O O G19−10 C G18−10 ]

relationship between z16,0 and z19,0 , the overall transfer equa- [ O G15−4 G16−4 + G19−4 C G18−4 ] (38)

tions of a general system with a closed-loop subsystem shown

in Figure 8 can be obtained automatically by handwriting and z1,0

[z15,0 ]

by computer: zall = [ ]

[z16,0 ] ,

Uall zall = 0, (36) [z18,0 ]

8 Advances in Mechanical Engineering

0

𝒛15,0

0 𝒛18,0 0 𝒛15,0 0 𝒛 𝒛19,0 0 𝒛18,0

16,0

15 16 17 18 15 16 19 17 18

19

11 12 13 14 11 12 13 14

9 10 9 10

8 8

7 6 6

7

5 5

4 4

3 3

2 2

Root 1 1

0 Root 0

𝒛1,0 𝒛1,0

Figure 7: Topology figure of a nontree system. Figure 8: Topology figure of a nontree system becoming a tree

system after “cutting” the hinge 19.

where

(2) For a tree system, in the first line of the overall

T15−1 = U1 U2 U3 U4 U5 U7 U8 U9 U11 U15 , transfer matrix, the coefficient matrix of the state

vector of root is a minus unit matrix, while each

T16−1 = U1 U2 U3 U4 U5 U7 U8 U9,12 U12 U16 , coefficient matrix of the state vector of a tip is the

successive premultiplication of the transfer matrices

T19−1 = U1 U2 U3 U4,6 U6 U10 U13 U17 U19 ,

of all elements in the transfer path from this tip to

T18−1 = U1 U2 U3 U4,6 U6 U10,14 U14 U18 , the root; besides the first line in the overall transfer

matrix, all coefficient matrices of state vectors in

G15−9 = −H9 U11 U15 , the first column are zero matrices. Except the first

line, in each row, each nonzero partitioned matrix

G16−9 = H9,12 U12 U16 , corresponds to the coefficient matrix of the tip state

(39)

G19−10 = −H10 U13 U17 U19 , vector, from which there is a transfer path to the input

end of the element with multiple input ends. Each

G18−10 = H10,14 U14 U18 , nonzero coefficient matrix of the state vector of a

tip is the successive premultiplication of all transfer

G15−4 = −H4 U5 U7 U8 U9 U11 U15 , matrices of elements in the transfer path from this tip

to the 𝑘th input end 𝐼𝑘 of jth body element, which has

G16−4 = −H4 U5 U7 U8 U9,12 U12 U16 ,

more than two ends, then premultiplied by −H𝑗 for

G19−4 = H4,6 U6 U10 U13 U17 U19 , 𝑘 = 1 or premultiplied by H𝑗,𝐼𝑘 for 𝑘 = 2, 3, 4 ⋅ ⋅ ⋅ 𝐿.

G18−4 = H4,6 U6 U10,14 U14 U18 . (3) For a chain system, its overall transfer matrix is

deduced automatically by successive premultiplica-

tion of the transfer matrices of all elements in the

5. Automatic Deduction Theorem of Overall transfer path from the tip to the root of system. In

Transfer Equation fact, any chain system can be considered as a special

example of the tree system in the case with only two

The following features of the overall transfer equation of a boundary ends.

multibody system can be clearly concluded from equations

(4) For a closed-loop system, its overall transfer equation

(23), (24), (28), and (30). These features make up the theorem

is deduced automatically as the chain system, after

to deduce automatically the overall transfer equation as the

treating the original system as the chain system by

following for tree system (1, 2), for chain system (1, 3), for

“cutting” a junction of any two adjacent elements and

closed-loop system (1, 3, 4), and for general system (1, 2, 5).

letting the couple of “cutting points” as the tip and

(1) The state vectors involved in an overall transfer root with the same state vectors of the chain system.

equation are the column matrix comprising the state (5) For a general system composed of one tree subsystem

vectors of all boundary ends of the system. and some closed-loop subsystems, its overall transfer

Advances in Mechanical Engineering 9

0 𝐾𝑥,7 = 1000 N/m, 𝐾𝑦,6 = 𝐾𝑦,7 = 500 N/m, and 𝐾6 = 𝐾7 =

8 150 N ⋅ m/rad.

3 2 1 0 overall transfer equation, the topology figure of the system

4

5 x3 can be got as in Figure 10, and the overall transfer equation of

y above system is deduced automatically by handwriting and by

6

o x computer as follows

z

0

Uall zall = 0, (40)

Figure 9: Tree multi-rigid-flexible-body system.

where the overall transfer matrix is

−I T6−1 T8−1

0 𝒛 Uall = [ ]. (41)

8,0 O G6−5 G8−5

𝒛6,0 8

0 State vectors of all boundary ends are

7

6

5

z1,0

4 zall = [z6,0 ] . (42)

3

[z8,0 ]

2 According to the proposed sign conventions, we know that

1 T6−1 = U1 U2 U3 U4 U5 U6 ,

𝒛1,0

Root 0 T8−1 = U1 U2 U3 U4 U5,7 U7 U8 ,

(43)

Figure 10: Topology figure of the system. G6−5 = −H5 U6 ,

G8−5 = H5,7 U7 U8 ,

equation is deduced automatically as the tree system, where z1,0 is state vector of root, z6,0 and z8,0 are the state

after treating the original system as the tree system by vectors of tips, and U1 , U2 , U3 , U4 , U5 , U5,7 , U6 , U7 , and U8

“cutting” the junctions of any two adjacent elements are the transfer matrices of body 1, fixed hinge 2, uniform

in every closed-loop subsystem and letting every beam 3, fixed hinge 4, body 5, elastic hinge 6, elastic hinge 7,

couple of “cutting points” as new “boundary ends” and body 8. All of the transfer matrices can be found directly

with the same state vectors of the tree system. from the library of transfer matrices in [19, 30].

There are boundary conditions of the system

The theorem above is effective for various multi-rigid-body

𝑇

systems, chain multi-rigid-flexible-body systems, and any z1,0 = [𝑥̈𝑦̈ 𝜃𝑧̈ 0 0 0 1]1,0 ,

closed-loop multi-rigid-flexible-body systems and is effective

for various tree multi-rigid-flexible-body systems and general 𝑇

z6,0 = [0 0 0 𝑚𝑧 𝑞𝑥 𝑞𝑦 1]6,0 , (44)

multi-rigid-flexible-body systems if the bodies with more

than two ends are rigid bodies. For more general systems 𝑇

z8,0 = [𝑥̈𝑦̈𝜃𝑧̈ 0 0 0 1]8,0 .

including the flexible bodies with more than two ends, the

theorem to deduce automatically overall transfer equation is

The system experiences a step upward force at the mass

undergoing study and will be discussed in another paper.

center of body element 1 at time instant zero, while the

initial displacement and velocity of the whole system are zero.

6. Numerical Examples The computational results of the system dynamics obtained

by the proposal method and by Newton-Euler method are

By comparison with Newton-Euler method, the numerical shown in Figure 11. It can be seen from Figure 11 that the

examples here are carried out to validate the proposed computational results obtained by the above two methods

method. have good agreements.

Example 1. A tree multi-rigid-flexible-body system moving Example 2. A multi-rigid-body system containing a close

in plane, as shown in Figure 9, consists of two fixed hinges, 2 loop moving in plane is shown in Figure 12. The planar rigid

and 4, two elastic hinges, 6 and 7, three rigid bodies, 1, 5, and 8, bodies (numbered 1, 3, 5, 7, 9, 11, and 13) are connected by pin

and one uniform beam element, 3, with three boundary ends. hinges (numbered 2, 4, 6, 8, 10, 12, and 14), and rigid body

The simulation parameters are given as follows: 𝑚1 = 𝑚5 = 1 is connected with the ground by a smooth pin. Each rigid

𝑚8 = 7.8 kg, 𝐽𝐶,1 = 𝐽𝐶,5 = 𝐽𝐶,8 = 0.013 kg ⋅ m2 , 𝑙3 = 3 m, body has the identical dynamics parameter as 𝑚 = 1 kg and

𝐸𝐴 3 = 1000 N, 𝑚3 = 0.78 kg/m, 𝐸𝐼3 = 166.67 N ⋅ m2 , 𝐾𝑥,6 = 𝐽𝐶 = (1/6) kg ⋅ m2 .

10 Advances in Mechanical Engineering

0

0.2

1

0.15

Rotation angle (rad)

3

0.1

4 10

5 11

0.05

6 12

7 13

0

0 1 2 3 4 5 8 14

9

t (s)

Figure 12: A general system moving in plane.

MSTMM

Newton-Euler method

0 𝒛9,0

Figure 11: Computational results of the angle of the right end of the 𝒛14,0

beam. 9 0

8 14

and following the proposed theorem, topology figure of the 6 12

system can be drawn in Figure 13. And the overall transfer

equation of the system can be derived automatically as 5 11

Uall zall = 0, 4 10

(45)

I1 3 I2

where the overall transfer matrix takes the form: 2

−I T9−1 T14−1 1

Uall = [ O G9−3 G14−3 ] . (52) 0 𝒛1,0

[O C −I ]

Figure 13: Topology figure of the system.

State vectors of all boundary ends are

z1,0

zall = [ z9,0 ] . (46) The initial angle of rigid body 1 is (𝜋/6) rad and the

relative angles of pin hinges (numbered 2, 4, 6, 8, 10,

[z14,0 ] 12, and 14) are all zero. The system moves from the rest

According to the proposed sign conventions, one can acquire under the effect of gravity. The computational results of the

system dynamics are obtained by the proposal method and

T9−1 = U1 U2 U3 U4 U5 U6 U7 U8 U9 , by Newton-Euler method. The time history of rigid body

1’s angle is exhibited in Figure 14, which shows that the

T14−1 = U1 U2 U3,10 U10 U11 U12 U13 U14 , computational results obtained by the above two methods

(47)

G9−3 = −H3 U4 U5 U6 U7 U8 U9 , have good agreements.

z1,0 is the state vector of the root; z9,0 and z14,0 are the state Based on MSTMM and the features of the overall transfer

vectors of tips emerging after the “cutting.” U1 , U3 , U3,10 , equations of multibody systems with various topological

U5 , U7 , U9 , U11 , and U13 are the transfer matrices of the structures, including chain systems, closed-loop systems, tree

corresponding planar rigid bodies. U2 , U4 , U6 , U8 , U10 , U12 , systems, general systems composed of one tree subsystem,

and U14 are transfer matrices of the corresponding planar pin and some closed-loop subsystems, the theorem to deduce

hinges. automatically the overall transfer equation by handwriting

There are boundary conditions of the system: and by computer is presented. Formulations of the proposed

𝑇 theorem as well as two numerical examples are given to verify

z1,0 = [0 0 𝜃𝑧̈ 0 𝑞𝑥 𝑞𝑦 1]1,0 . (48) the theorem. This makes it possible to program large-scale

Advances in Mechanical Engineering 11

ulator arms,” The International Journal of Robotics Research, vol.

0.6

3, no. 3, pp. 87–101, 1984.

[9] R. L. Huston, “Multi-body dynamics including the effects of

0.4

Rotation angle (rad)

no. 5-6, pp. 443–451, 1981.

0.2

[10] Y. A. Khulief and A. A. Shabana, “Dynamic analysis of con-

strained system of rigid and flexible bodies with intermittent

0

motion,” ASME Journal of Mechanisms, Transmissions and

Automation in Design, vol. 108, no. 1, pp. 38–45, 1986.

−0.2

[11] R. A. Laskin, P. W. Likins, and R. W. Longman, “Dynamical

equations of a free-free beam subject to large overall motions,”

−0.4 Journal of the Astronautical Sciences, vol. 31, no. 4, pp. 507–527,

1983.

0 2 4 6 8 10 [12] M. Pascal, “Dynamical analysis of a flexible manipulator arm,”

t (s)

Acta Astronautica, vol. 21, no. 3, pp. 161–169, 1990.

[13] J. Rauh and W. Schiehlen, “A unified approach for the modeling

MSTMM of flexible robot arms,” in Proceedings of the 6th CISM-IFToMM

Newton-Euler method Symposium on Theory and Practice of Robots and Manipulators,

Cracow, Poland, September 1986.

Figure 14: Computational results of the angle of rigid body 1. [14] R. E. Roberson, “A form of the translational dynamical equa-

tions for relative motion in systems of many non-rigid bodies,”

Acta Mechanica, vol. 14, no. 4, pp. 297–308, 1972.

software for MSD with much higher computational speed and [15] O. A. Bauchau, “Computational schemes for flexible, nonlinear

much higher programming. multi-body systems,” Multibody System Dynamics, vol. 2, no. 2,

pp. 169–225, 1998.

[16] S. C. Chu and K. C. Pan, “Dynamic response of a high speed

Conflict of Interests slider crank mechanism with an elastic connecting rod,” ASME

Journal of Engineering for Industry, vol. 97, no. 2, pp. 542–550,

The authors declare that there is no conflict of interests 1975.

regarding the publication of this paper. [17] E. Kreuzer and O. Kust, “Analysis of long torsional strings by

proper orthogonal decomposition,” Archive of Applied Mechan-

Acknowledgments ics, vol. 67, no. 1-2, pp. 68–80, 1996.

[18] T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft

The research was supported by Research Fund for the Dynamics, McGraw-Hill, New York, NY, USA, 1983.

Doctoral Program of Higher Education of China (no. [19] X. Rui, L. Yun, Y. Lu et al., Transfer Matrix Method for Multibody

20113219110025). The authors owe special thanks to Professor System and Its Application, Science Press, Beijing, China, 2008.

Dieter Bestle for his important discussion with them during [20] X. Rui, K. Huang, and Y. Qin, “Transfer matrix method of

his five months visit in Nanjing University of Science and multibody system and its application in spacecraft dynamics,”

Technology. in Proceeding of the 3rd China-Russia-Ukraine Symposium on

Astronautical Science and Technology, pp. 933–935, Northwest-

ern Polytechnical University Press, Xi’an, China, 1994.

References [21] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer

matrix method for multibody system dynamics,” Multibody

[1] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,

System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.

Germany, 1990.

[22] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for

[2] A. A. Shabana, Dynamics of Multibody Systems, Cambridge linear multibody system,” Multibody System Dynamics, vol. 19,

University Press, New York, NY, USA, 2009. no. 3, pp. 179–207, 2008.

[3] D. Bestle and P. Eberhard, “Analyzing and optimizing multibody [23] X. Rui, B. He, B. Rong, G. Wang, and Y. Lu, “Discrete time

systems,” Mechanics of Structures and Machines, no. 20, pp. 67– transfer matrix method for dynamics of a multi-rigid- flexible-

92, 1992. body system moving in plane,” Proceedings of the Institution of

[4] K. Thoma, F. Schäfer, S. Hiermaier, and E. Schneider, “An Mechanical Engineers K, vol. 223, no. 1, pp. 23–42, 2009.

approach to achieve progress in spacecraft shielding,” Advances [24] X. Rui and J. Zhang, “Automatical transfer matrix method of

in Space Research, vol. 34, no. 5, pp. 1063–1075, 2004. multibody system,” in Proceedings of the 2nd Joint International

[5] J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G. Conference on Multibody System Dynamics, Stuttgart, Germany,

Teubner, Stuttgart, Germany, 1977. May-June 2012.

[6] O. A. Bauchau, Flexible Multibody Dynamics, Springer, [25] X. Rui and B. Rong, “Advance in transfer matrix method for

Dodrecht, The Netherlands, 2011. multibody system dynamics,” Advance in Mechanics, vol. 42, no.

[7] J. A. C. Ambrósio, M. A. Neto, and R. P. Leal, “Optimization of a 1, pp. 4–17, 2012.

complex flexible multibody systems with composite materials,” [26] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer

Multibody System Dynamics, vol. 18, no. 2, pp. 117–144, 2007. matrix method for dynamics analysis of complex weapon

12 Advances in Mechanical Engineering

1061–1071, 2011.

[27] X. Rui, “Launch dynamics of multibody system and its applica-

tion,” Engineering Science, vol. 13, no. 10, pp. 76–81, 2011.

[28] G. Wang, X. Rui, F. Yang, and H. Xu, “Dynamics analysis

for wheeled and tracked multiple launch rocket system,” Acta

Armamentarii, vol. 33, no. 11, pp. 1286–1290, 2012.

[29] B. Rong, X. Rui, G. Wang, and F. Yang, “Discrete time transfer

matrix method for dynamics of multibody system with real-

time control,” Journal of Sound and Vibration, vol. 329, no. 6,

pp. 627–643, 2010.

[30] X. Rui, D. Bestle, and J. Zhang, “A new form of the trans-

fer matrix method for multibody systems,” in Proceedings of

the ECCOMAS Thematic Conference on Multibody Dynamics,

Zagreb, Croatia, 2013.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 545126, 7 pages

http://dx.doi.org/10.1155/2014/545126

Research Article

Optimal Design of One-Folded Leaf Spring with

High Fatigue Life Applied to Horizontally Vibrating

Linear Actuator in Smart Phone

1

Department of Mechanical Engineering, Yeungnam University, Dae-dong, Gyeongsan-si,

Gyeongsangbuk-do 712-160, Republic of Korea

2

PIDOTECH Inc., Republic of Korea

Received 4 July 2013; Revised 12 November 2013; Accepted 25 November 2013; Published 24 February 2014

Copyright © 2014 Ki Bum Lee et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Horizontally vibrating linear actuator (HVLA) instead of VVLA has been under study in a few past years and recently HVLA

with thickness of 2.5 mm was developed. The one-folded leaf spring to guide the moving part is newly designed and applied in

HVLA, but unfortunately it makes HVLA be wider. Accordingly, this paper presents the optimal design of one-folded leaf spring,

which results in reduction of HLVA width. The commercial design optimization tool “PIAnO” was utilized based on design of

experiments (DOE), approximation techniques, and optimization algorithm. In addition, for the vibration modal analysis and

harmonic response analysis to generate metamodeling, the software “ANSYS” is utilized. The optimal width of leaf spring was

reduced by 46% compared to the initial one, while all the design constraints were satisfied, which clearly showed the validity of the

proposed design approach.

1. Introduction part of the actuator and enable the actuator to vibrate elas-

tically. Accordingly, the various designs of guide spring had

The slimming of a smart phone has become a competitive been presented to achieve the high fatigue life [3, 4] and

issue in the smart phone industry because the display panel is among them the one-folded leaf spring introduced by Lee and

enlarged to show more information to the user. A smart Kim has the highest fatigue life [5]. Figures 2(a) and 2(b) show

phone is comprised of numerous components: the battery, the the one-folded spring and the schematic diagrams of the

camera module, the vibrating actuator, and so forth. One of HVLA which has the one-folded spring, respectively. Table 1

the thickest components is vibrating actuator which imposes shows the performance characteristics of the HVLA.

the restrictions on the slimming of smart phones. In currently This one-folded leaf spring enables the HVLA to improve

manufactured smart phone in the market, a vertical vibrating the vibration force and product life. Instead, the HVLA has

linear actuator (VVLA) is being widely used as a vibration been enlarged in width due to the wide structure of one-

motor, as shown in Figure 1. folded spring. Figure 3 shows the top view of the HVLA with

However, it imposes the thickness problem on a smart one-folded leaf spring.

phone, because it needs vibration space in the vertical direc- This one-folded leaf spring has occupied 45% of total

tion [1]. Therefore, a horizontally vibrating linear actuator width of HVLA. Therefore, the design of one-folded leaf

(HVLA) has been developed in recent years, but it has not spring needs be optimized to reduce its width. Several

been commercialized in the smart phone industry so far [2]. researches about the optimal design of the mechanical springs

The primary reason is that the fatigue fracture of guide spring such as helical, coil, and leaf spring had been reported in

due to the cyclic compression and tension loads has not been past decades. These springs are usually designed to avoid

resolved. The guide spring has the role to support the moving resonance, but one-folded leaf spring of HVLA needs to

2 Advances in Mechanical Engineering

the following equation:

Case

2.1.2. Design Constraint on the Maximum Stress. A one-

Spring folded spring that is affected by cyclic compression and ten-

sion loads should have a yield stress of approximately 215 MPa

Yoke

in order to maintain its fatigue life [7]. The one-folded spring

is represented as follows:

Magnet

𝜎max < 𝜎yield . (2)

Coil

2.1.3. Design Constraint on the Mechanical Natural Frequency.

The mechanical natural frequency of HVLA ranges from

175 Hz to 185.5 Hz:

Figure 1: Schematic diagram of VVLA.

175 ≤ 𝑓 ≤ 185.5. (3)

Table 1: Specification of the HVLA. ness, height, and radius as design variables for the leaf spring.

Figure 4 shows the shape of the leaf spring, and Table 2

Requirements Values

shows the initial, lower, and upper bound values of the design

Natural frequency 183 Hz

variables.

Thickness of actuator 2.5 mm

Vibration acceleration 2.1 to 2.3 G

2.3. Design Problem Formulation. The design problem for

Fatigue life Over 1,000,000 cycles

determining the design variables that satisfy all the design

requirements can be mathematically formulated as

no studies reported about the optimal design of one-folded

leaf spring. Therefore, this paper presents an optimization of to minimize Width of spring

the one-folded leaf spring. For optimal design process, we subject to 𝜎max ≤ 𝜎yield

used the commercial optimization software PIAnO (Process

Integration, Automation, and Optimization) which provides 175 ≤ 𝑓 ≤ 185.5

the users with various tools for the efficient optimization

design [6]. (Ver.3.5, PIDOTECH) The optimal design objec- 1.79 ≤ 𝑊 ≤ 3.6 (4)

tive is to minimize the width of one-folded spring to satisfy

the stiffness and maximum stress at cyclic loading condition. 6 ≤ 𝐿 ≤ 15.8

Five design variables were selected to derive the optimal 0.15 ≤ 𝑇 ≤ 0.35

design of one-folded leaf spring and an orthogonal array used.

Modal and harmonic response analyses were performed 0.5 ≤ 𝐻 ≤ 1.5

according to the design of experiments to obtain the natural

frequency of the first vibration mode due to stiffness and the 0.5 ≤ 𝑅 ≤ 0.75.

maximum stress at resonance using the commercial struc-

tural analysis program ANSYS. (Version 14.5, ANSYS) And 3. Analysis Procedures and Optimal Design

then, analyses results were saved as a text file and reimported

into the PIAnO for generating metamodel. Metamodels were 3.1. Analysis Procedures. Figure 5 shows our analysis proce-

generated based on the results of structural analysis using dures. First, we generated a design of experiments (DOE) and

a Kriging model with PIAnO. Finally, we derived a global then the modal and harmonic analyses were response per-

optimal point using evolution algorithms. formed according to sampling points. After, metamodels

were generated using the Kriging algorithm provided by the

PIAnO software. We used an optimization technique using an

2. Design Problem evolution algorithm (EA) to find the optimum solution.

2.1. Design Requirements

3.2. Design of Experiments. After determining the experi-

2.1.1. Minimization of the Width of Leaf Spring. The width of mental design using an orthogonal array, which is one of the

one-folded spring should be minimized in order to reduce the tools provided by 𝐿 98 (715 ), the leaf spring was designed

Advances in Mechanical Engineering 3

15 mm

m

m

.5

16

Case

Leaf spring

Housing

Weight

Magnets Yoke

Coil

PCB

Bracket

(a) (b)

Figure 2: (a) The one-folded spring of HVLA and (b) schematic diagram of HVLA.

Table 2: Initial, lower, and upper bound values of the selected design variables.

𝑥1 𝑊 (mm) 1.79 3.6 3.6

𝑥2 𝐿 (mm) 6 11.6 15.8

𝑥3 𝑇 (mm) 0.15 0.25 0.35

𝑥4 𝐻 (mm) 0.5 1 1.5

𝑥5 𝑅 (mm) 0.5 0.64 0.75

Width Width W

of spring of spring

Length L

of weight

R

H

Figure 3: Top view of HVLA with the one-folded leaf spring. saturated points is the minimum required number of sam-

pling points to generate a full quadratic polynomial model.

An orthogonal array with the main effect and interaction of

according to each design variable. 𝐿 98 (715 ) was selected each factor was represented by a table, which allows the

by considering the number of saturated points and levels design of experiments to be easily established [8]. An orthog-

according to the number of design variables. The number of onal array, regardless of the lack of theoretical background,

4 Advances in Mechanical Engineering

(1) Design of

Extract experiments

experiments

(2) ANSYS

PIAnO

(modal and harmonic response)

(3) Approximation Generate meta models

(4) Design

optimization Derive the optimal

Figure 7: Fatigue fracture of the one-folded spring at folded area.

has the advantage of being able to apply easily fractional repli- flux density in the airgap, 𝑖 is the input electric current, 𝐿 eff is

cation, decomposition method, and so on. Also, the experi- the effective coil length, and 𝑓 is the input frequency of the

ment can use many factors without enlarging the size of the electric current.

experiment, and the interpretation of the experiment data is Accordingly, the harmonic response analysis was per-

simple. formed to calculate the maximum stress near the folded area

at resonance frequency. The maximum stress must be less

3.3. Modal Analysis and Harmonic Response Analysis. 98 leaf than the yield stress. The harmonic responses were performed

spring FE models are created according to sampling points in with 30 intervals between ±10 natural frequencies [12].

the orthogonal array. And then the modal analysis was per- Figure 8 shows the result of harmonic response analysis.

formed to obtain the vibration natural frequency and the

harmonic response analysis was performed to calculate the 3.4. Metamodeling. Metamodeling builds a metamodel that

maximum mechanical stress at the resonance using ANSYS. approximates the relationship between performance indices

and design variables of a real model by using the analysis

(1) Modal Analysis. Figure 6 shows a 3-dimensional finite ele- results at the sampling points specified by a DOE. Figure 9

ment (FE) model of moving part with the one-folded spring shows the result of a parameter study of the nonlinearity of

and constraint. This moving part consists of permanent mag- 𝑥1 (𝑊). The simulation was performed such that 𝑥1 (𝑊) for

nets, yoke, weight, housing, and one-folded spring. Table 3 the upper and lower bounds was divided into five sections.

lists the material properties used in FE model. The modal Then, we modeled the one-folded spring according to a

analysis was conducted to calculate the natural vibration fre- divided 𝑥1 (𝑊) after a fixed initial value of 𝑥2 (𝐿), 𝑥3 (𝑇),

quency according to DOE [9]. 𝑥4 (𝐻), and 𝑥5 (𝑅).

We found that 𝑥3 (𝑇) was a sensitive value. In a previous

(2) Harmonic Response Analysis. Figure 7 shows the fatigue optimization of the leaf spring, we found a sensitive value of

fracture of leaf spring at folded area due to the cyclic loading. 𝑥3 (𝑇) because the first frequency and stress level were very

Cyclic loading is created by the magnetic force which is responsive, depending on the thickness of the spring. For

energized by sinusoidal electric current. It is expressed by example, the stiffness of cantilever beams is given by

[10, 11]

3𝐸𝐼 𝐸𝐻𝑇3 𝐻𝑇3

𝑘= = , (𝐼 = ). (6)

𝐹magnetic = 𝑁𝐵𝑔 𝑖𝐿 eff , 𝑖 = 𝑖0 sin (2𝜋𝑓𝑡) , (5) 𝐿3 4𝐿3 12

Advances in Mechanical Engineering 5

Leaf spring SUS304 8.00 193–200 0.29

Weight Tungsten 19.3 400 0.28

Housing SUS304 8.00 193–200 0.29

Magnet NdFe35 7.50–7.80 150–160 0.24

Steel yoke Steel 1010 7.84–7.87 200–205 0.29

148

Amplitude (MPa)

117

93.1

58.8

46.7

37.1

23.4

174 175 178 180 183 185 188 190 193

Frequency (Hz)

189 200

188

First frequency (Hz)

150

187

Stress (MPa)

186 100

185

50

184

183 0

1.79 2.2425 2.695 3.1475 3.6 1.79 2.2425 2.695 3.1475 3.6

x1 width x1 width

The stiffness of a cantilever beam is directly proportional response analyses. Metamodels were regenerated with respect

to the cube of the thickness. Therefore, the thickness of the to the width of the one-folded spring, maximum stress, and

leaf spring has a nonlinearity, so we selected the Kriging natural frequency of the first mode. At this point, we added

model, which is one of the metamodels provided in the only 39 types of sampling points to satisfy the 15 multiples

PIAnO software that was selected. The Kriging model, which of the number of design variables corresponding to the

a type of interpolation model, was mathematically estab- sampling points (75). We used augmented Latin hypercube

lished by Metheron in 1963 based on research conducted design (ALHD) that is not overlapped with the existing

by mining researcher Krige in 1951 [13, 14]. And it shows sampling points [16]. Also, it is widely known for excellent

superb predictive performance under many design variables performance of the space filling. Nevertheless, the accuracy of

2

and is in strongly nonlinear systems, provides a statistical 𝑅pred was only about 80%. Therefore, 98 sampling points were

estimation [15]. Also, there are no parameters that depend on reextracted in order to obtain higher accuracy over 90% using

the experience and intuition of customers when choosing the orthogonal array 𝐿 98 (715 ) at a time. Finally, in the evaluation

design parameters because the Kriging model can optimize results of the metamodel, the accuracy of 𝑅pred 2

is over 90%.

design parameters through maximum likelihood estimation

(MLE). Therefore, a recent trend is an increase in the use of

Kriging models in the field of engineering. Our first Kriging 3.5. Optimization Technique. We selected an evolution algo-

model was generated by using an orthogonal array 𝐿 36 (313 ), rithm that is provided in PIAnO. The evolution algorithm

which was based on the analytical results of the width of (EA) was proposed by Holland in 1975 and is a global

the one-folded spring, the maximum stress, and the natural optimization technique [17]. This algorithm describes the

frequency. However, an accuracy evaluation of the Kriging evolution of the biological genetic trait, so it searches for a

2

model using 𝑅pred , which corresponds to the maximum stress, global optimum through a process of selection, recombina-

was underestimated 55% as accuracy. Consequently, we tion, and mutation. On the downside, it requires consider-

added 39 types of sampling points to improve the accuracy of able computation time, depending on the analytical model,

the metamodel and then performed the modal and harmonic because more calculation functions are required compared to

6 Advances in Mechanical Engineering

210

Violation area

200

200

Violation area

Maximum stress (MPa)

180

190 187.32

160

148.75

141.12 180 178.17 176.67

142.06

140

170

120 Violation area

100 160

Initial Opt meta Opt exact Initial Opt meta Opt exact

(a) Maximum stress at bending area on leaf spring (b) Natural frequency of first mode

𝑥1 𝑊 (mm) 1.79 3.60 1.9 3.60

𝑥2 𝐿 (mm) 6.00 11.64 12.45 15.80

𝑥3 𝑇 (mm) 0.15 0.25 0.26 0.25

𝑥4 𝐻 (mm) 0.50 1.00 1.04 1.24

𝑥5 𝑅 (mm) 0.28 0.64 0.63 0.88

disadvantage of the evolution algorithm because we used a 3.6

metamodel with a short analysis time. 3.5

3

Width (mm)

3.6. Results. In our results for the optimal design using the

2.5

metamodel (Opt meta), the optimal design satisfied the con-

straints, including the maximum stress and natural frequency 1.9

2 1.9

of first mode, and the width of one-folded leaf spring

decreased by 47% compared to the initial width of 1.9 mm. 1.5

However, the optimal design results can be changed based on

using the metamodel instead of the actual analytical model in 1

this research. The accuracy of the optimization results should Initial Opt meta Opt exact

be verified by actual analysis using ANSYS. To do this, the Figure 11: Comparison of the objective function.

Kriging model results (Opt meta) of the optimal design vari-

ables and the analysis results from ANSYS (Opt exact) were

compared as shown in Figures 10(a) and 10(b). 15 mm 12.2 mm

The Kriging model results (Opt meta) and the ANSYS

model results (Opt exact) were very similar; therefore, we

confirmed the high accuracy of the Kriging model’s predic-

tion. The initial and optimal values of the design variables

16.5 mm

pared widths of the leaf spring. Figure 12 shows a comparison

of the existing actuator and the actuator with the optimal

width of one-folded spring.

4. Conclusion

We minimized the existing size of a HVLA (16.5 × 15 × 2.5) by

decreasing the width of the leaf spring. The following conclu- Figure 12: Comparison of HLVA according to the initial design and

sions were drawn from our results. optimal design.

Advances in Mechanical Engineering 7

mize the maximum stress in the bending area and to [8] A. S. Hedayat, N. J. A. Sloane, and J. Stufken, Orthogonal Arrays:

satisfy the required first mode natural frequency, Theory and Applications, Springer, New York, NY, USA, 1999.

based on the design requirements. [9] Y. Kang, G.-J. Sheen, M.-H. Tseng, S.-H. Tu, and H.-W. Chiang,

(2) We generated an orthogonal array using the com- “Modal analyses and experiments for engine crankshafts,” Jour-

mercial optimization software PIAnO. Then, each leaf nal of Sound and Vibration, vol. 214, no. 3, pp. 413–430, 1998.

spring designed by sampling points was analyzed by [10] K. I. Hwang and J. H. Kim, “Single-axis flat electro-magnetic

using the commercial structural analysis software actuator using shorted turn for fast initial response,” Journal of

Korean Medical Science, vol. 19, no. 6, pp. 222–226, 2009.

ANSYS.

[11] S.-M. Yoon, J.-H. Kim, and J.-H. Kim, “Mathematical modeling

(3) The Kriging model, provided PIAnO, was generated and analysis of vibration characteristics of smart-phone,” Inter-

based on simulation results according to the design of national Journal of Precision Engineering and Manufacturing,

experiments 𝐿 98 (715 ). Then, the optimal design was vol. 14, no. 3, pp. 505–508, 2013.

determined using the evolution algorithm of the [12] Z. Wu, C. Xu, J. Zhang, D. Yu, and P. Feng, “Modal and harmonic

global optimization technique. As a result, all con- reponse analysis and evaluation of machine tools,” in Proceed-

straints were satisfied, and the derived the optimal ings of the International Conference on Digital Manufacturing

width of one-folded spring was decreased by 47% and Automation (ICDMA ’10), vol. 1, pp. 929–933, Changsha,

compared to the initial values. China, December 2010.

[13] D. G. Krige, “A statistical approach to some basic mine valuation

problems on the witwatersrand,” Journal of the Chemical, Met-

Abbreviations allurgical and Mining Society of South Africa, vol. 52, no. 6, pp.

𝑇: Thickness of leaf spring 119–139, 1951.

𝐻: Height of leaf spring [14] G. Metheron, “Principles of geostatistics, economic geology,”

𝐿: Length of leaf spring Economic Geology, vol. 58, no. 8, pp. 1246–1266, 1963.

𝑊: Width of leaf spring [15] T. W. Simpson, J. J. Korte, T. M. Mauery, and F. Mistree, “Com-

𝑅: Radius of leaf spring. parisons of response surface and kriging models for multidis-

ciplinary design optimization,” in Proceedings of the 7th AIAA/

USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis

Conflict of Interests & Oprimization, vol. 1, pp. 381–3391, 1998.

[16] M. Stein, “Large sample properties of simulations using latin

The authors declare that there is no conflict of interests

hypercube sampling,” Technometrics, vol. 29, no. 2, pp. 143–151,

regarding the publication of this paper. 1987.

[17] J. H. Holland, Adaptation in Natural and Artificial Systems, The

Acknowledgment University of Michinan Press, 1975.

[18] R. L. Haupt and S. E. Haupt, Practical Evolutionary Algorithms,

The authors express their gratitude to PIDOTECH, Inc., for Wiley, 1998.

providing their PIAnO software as a PIDO tool.

References

[1] J.-H. Kim, H.-K. Il, J.-W. Lee, and J.-H. Yoo, Horizontally Vibra-

tion Motor for Slimming Mobile Smart Phone, vol. 6, Korean

Society for Precision Engineering, 2011.

[2] J.-H. Kim and J.-H. Kim, “A horizontally vibration linear actua-

tor for slimming of smart phone,” in Proceedings of the Interna-

tional Conference on Computer, Electrical and Systems Confer-

ence (ICCESSE ’11), Phuket, Thailand, December 2011.

[3] JA HWA ELECTRONICS CO., LTD., Linear type vibration

motor of horizontal vibration capable of generating vibration to

horizontal direction by linearly reciprocating mover, KOEAR

Patent, 1007201970000 (20070514).

[4] SAMSUNG ELECTRO-MECHANICS CO., LTD, Linear vibra-

tion motor capable of minimizing a friction area with adja-

cent components when a motor is driven, KOREA Patent,

1010595990000 (20110819).

[5] K.-B. Lee and J.-H. Kim, “Design of leaf spring with high fatigue

life applied to horizontal linear vibrating actuator,” The Korea

Academia-Industrial Cooperation Society, vol. 13, no. 12, pp.

5684–5688, 1975.

[6] PIDOTECH Inc, Process Intergration, Automation, and Opti-

mization, PIAnO User’s Manual, 2010.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 512583, 9 pages

http://dx.doi.org/10.1155/2014/512583

Research Article

Adaptive Terminal Sliding Mode Control of Electromagnetic

Spacecraft Formation Flying in Near-Earth Orbits

1

School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, China

2

Department of Aircraft and Dynamics, Aviation University of Air Force, Changchun 130022, China

Received 25 August 2013; Revised 17 November 2013; Accepted 9 December 2013; Published 11 February 2014

Copyright © 2014 Jingrui Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An adaptive terminal sliding mode control for six-degree-of-freedom electromagnetic spacecraft formation flying (EMFF) in near-

Earth orbits is presented. By using terminal sliding mode (TSM) technique, the output tracking error can converge to zero in

finite time, and strong robustness with respect to disturbance forces can be guaranteed. Based on a rotated frame 𝐹𝑟 and the

adaptive TSM controller, the special magnetic moment of the steerable magnetic dipole is computed. The angular momentum

management strategy (AMM) is implemented in a periodically switching fashion, by which the angular momentum buildup was

limited. Illustrative simulations of EMFF are conducted to verify the effectiveness of the proposed controller.

1. Introduction forces on the order of 10-9 N/W may be generated using elec-

tromagnetic gradient forces or scattering forces; microwave

Spacecraft formation flying (SFF) represents the concept beam powers of 10 kW can thus produce restoring forces of

of distributing the functionality of large spacecraft among approximately 10 𝜇N, which are sufficient to correct a number

smaller, less-expensive, cooperative spacecraft [1, 2]. Specif- of orbital perturbations. Miller et al. [10, 11] address the

ically, NASA and the U.S. Air Force have identified space- novel concept of electromagnetic formation flying (EMFF) in

craft formation flying as an enabling technology for future which high temperature superconducting (HTS) wire tech-

missions. The practical implementation of the SFF concept nology is used to create magnetic dipoles on each spacecraft

relies on the accurate control of the relative positions and ori- that can be used to maintain and reconfigure the spacecraft

entations between the participating spacecraft for formation formation. However, Since magnetic force on each spacecraft

configuration. The conventional thruster-based schemes may in the formation can be applied in any arbitrary direction

require continuous expenditure of fuel to maintain formation which can be easily created by steerable magnetic dipoles,

geometry that can contaminate the sensitive sensors on board EMFF has advantages in terms of controllability [4].

and mission lifetime also becomes dependent on the fuel A critical component of EMFF is an effective formation

available [3, 4]. flying control. The dynamics and control problem associated

To alleviate these concerns, several propellant-free for- with EMFF become highly challenging, due to the nonlinear

mation flying methods have been proposed in the literature. nature of the magnetic forces. Ahsun and Miller [3] have

The propulsive conducting tethers and spin-stabilized tether presented a hybrid adaptive control scheme in which trans-

systems have been proposed in place of on-board propulsion lation control is implemented in a centralized fashion with

systems to form and maintain satellite formations [5, 6]. a decentralized attitude control. Elias et al. [12] designed a

King et al. [7] have presented Coulomb force approaches linear optimal controller based on the linearized dynamics.

to maintain a formation. The flux pinning technology has Kong et al. [13] addressed the use of electromagnetic dipoles

been applied to achieve passively stable configurations by for relative position and orientation maintenance as needed

HTS electromagnetics [8]. LaPointe [9] has presented the for the terrestrial planet finder (TPF). Reference [14] derived

microwave scattering formation flight method. Radiation the dynamics of an N-spacecraft EMFF (in 2D) for deep

2 Advances in Mechanical Engineering

(follower) axis aligned in the radial (zenith) direction, the 𝑧-axis is

Z rij perpendicular to the orbital plane and points in the direction

Rj r0j of the angular momentum vector, and the 𝑦-axis completes

r0i the right-hand system. (3) The orbital frame 𝐹𝑂𝑖 is used in

R0 ith spacecraft the attitude control. The reference frame is rotating about the

Leader

(follower) 𝑦-axis with respect to ECI frame at orbital rate. The roll axis

Ri 𝑥 is in the fight direction, the pitch axis 𝑦 is perpendicular

to the orbital plane, and the yaw axis 𝑧 points toward the

O

Earth

X Earth. (4) A body frame 𝐹𝐵𝑖 attached to the body of 𝑖th

spacecraft with the center of mass of the spacecraft is also

defined to describe the orientation of each spacecraft in the

Y inertial space. (5) R𝑖 ∈ R3 denotes the position vector from

the origin of ECI frame to 𝑖th spacecraft. r𝑖𝑗 ∈ R3 denotes

Reference orbit

Figure 1: Schematic representation of the EMFF system. 𝑇

𝑗th spacecraft. 𝜌𝑖 ≜ [𝑥𝑖 𝑦𝑖 𝑧𝑖 ] ∈ R3 is the expression of

the relative position vector r0𝑖 from the origin of the leader

spacecraft coordinates system to the follower spacecraft 𝑖 in

space missions and discussed a nonlinear control law using the Hill frame 𝐹𝐻. (6) The rotated frame 𝐹𝑟𝑖 is used in a

potential functions. simplified algebraic form of the magnetic force equation. The

The orientation of a dipole obviously depends on the ori- 𝑥-axis is aligned with vector r𝑖𝑗 [3].

entation of the body axes in the inertial space, and changing

the dipole on one spacecraft affects actuation on all other 2.1. Translational Dynamics. Designating spacecraft 0 as the

spacecraft in EMFF. Therefore, general asymptotical stability leader of the formation, the relative dynamics of 𝑖th spacecraft

may not deliver fast enough convergence to meet EMFF in the Hill reference frame 𝐹𝐻 can be represented as

control for high-precision situation. The recently developed

𝜌̈ ̂ ̂

terminal sliding mode (TSM) control enables convergence 𝑖 + C𝑖𝑇 (𝜔𝑂, 𝜌̇

𝑖 ) + N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 ) = F𝑑𝑖 + F𝑐𝑖 , (1)

to the desired state in finite time [15–18]. This technique where 𝜔𝑂 is the orbital angular velocity of the leader. C𝑖𝑇 (⋅) ∈

has been used successfully in some control designs, such as R3 is a nonlinear term defined as

robotic manipulators [17] and mobile target tracking [18]. The

𝑇

physical interpretation of finite time convergence lies in the C𝑖𝑇 (𝜔𝑂, 𝜌̇𝑖 ) ≜ 2𝜔𝑂[−𝑦𝑖̇ 𝑥𝑖̇ 0] . (2)

fact that the convergence rate of TSM grows exponentially

when the state is near equilibrium. N𝑖𝑇 (⋅) ∈ R3 is nonlinear term defined as

In this Paper, we confine our attention to adaptive TSMC N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 )

design for EMFF in near-Earth orbits. The remainder of

this paper is organized as follows. Section 2 presents a 𝑥𝑖 + R0 1 2

detailed 6-DOF dynamic modeling for EMFF in near-Earth [ 𝜇 ( 3

− 2 ) − 𝜔𝑂 𝑥𝑖 − 𝜔̇𝑂𝑦𝑖 ]

[

R + 𝜌

R

]

orbits. In Section 3, a position/attitude tracking controller [ 0 𝑖 0 ] (3)

[ 𝑦 ]

based on TSM is developed for EMFF, which ensures error ≜[ [ 𝜇 𝑖

− 𝜔 2

𝑦 + 𝜔̇ 𝑥 ],

]

convergence in finite time and strong robustness of the [ R0 + 𝜌 3 𝑂 𝑖 𝑂 𝑖

]

bounded disturbances. Based on a rotated frame 𝐹𝑟 and the [ 𝑖 ]

[ 𝑧𝑖 ]

adaptive TSM controller, the special magnetic moments of 𝜇 3

the Steerable Magnetic Dipole (SMD) are computed. The [ R0 + 𝜌𝑖 ]

AMM strategy is implemented in a periodically switching where 𝜇 is gravitational constant of earth.

fashion. Simulation results are presented in Section 4. Finally, Note that F̂ 𝑑𝑖 denotes the parametric uncertainty and

Section 5 concludes this paper. external disturbances, such as differential J2 and higher-order

terms, differential solar pressure, and differential drag. The

2. System Model specific relative magnetic force F ̂𝑐𝑖 provided by the coils needs

to be strong enough to cancel these disturbances and at the

Spacecraft formation constitutes an 𝑁-body mechanics prob- same time provide additional force for trajectory following.

lem. Figure 1 shows a typical formation system including 𝑁 Consider the following:

spacecrafts orbiting a central body (Earth). The spacecraft 0’s

̂ 1 1

designated as the “leader” spacecraft. We make the following F𝑐𝑖 = F − F , (4)

considerations. (1) The ECI frame 𝐹𝐼 is defined with its 𝑚𝑖 𝑐𝑚𝑖 𝑚0 𝑐𝑚0

origin at the center of Earth, its 𝑋-axis points toward vernal where

equinox, 𝑍-axis points toward celestial north pole, and 𝑌- 𝑁−1

axis completes a right-handed axis system. (2) The Hill frame F𝑐𝑚𝑖 = ∑ F𝑚

𝑖𝑗 . (5)

𝐹𝐻 is used to visualize the relative motion of each follower 𝑗=0, 𝑗 ≠

𝑖

Advances in Mechanical Engineering 3

By approximating the coils on each spacecraft to SMD, the where 𝜏𝑚𝑖𝑗 is the torque exerted on the SMD of 𝑖th spacecraft

magnetic field due to the 𝑗th spacecraft can be written as [3, due to that of 𝑗th spacecraft [3, 10],

10]

𝜇0 𝜇𝑖 3r𝑖𝑗 (𝜇𝑗 ⋅ r𝑖𝑗 ) 𝜇𝑗

𝜇 3𝜇𝑗 ⋅ r𝑖𝑗 𝜇𝑗 𝜏𝑚

𝑖𝑗 = ×( − 3) (13)

B𝑗 (r𝑖𝑗 ) = 0 ( 5

r𝑖𝑗 − 3 ) (6) 4𝜋 𝑟𝑖𝑗5 𝑟𝑖𝑗

4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗

𝜏𝑐𝑤𝑖 and 𝜏𝑑𝑖 are the control and disturbance torques, respec-

and the force between spacecrafts can be written as

tively. 𝜏𝑑𝑖 includes the gravity gradient torques 𝜏𝑔𝑖 and the

F𝑚 Earth magnetic torques 𝜏𝑒𝑖 . Consider the following:

𝑖𝑗 ≜ 𝑓 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 )

𝜏𝑑𝑖 = 𝜏𝑔𝑖 + 𝜏𝑒𝑖 . (14)

3𝜇 𝜇𝑖 ⋅ 𝜇𝑗 𝜇𝑖 ⋅ r𝑖𝑗 𝜇𝑖 ⋅ r𝑖𝑗

= 0 (− 5

r𝑖𝑗 − 5

𝜇𝑗 − 𝜇𝑖 A very important factor presented in the disturbance

4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗 𝑟𝑖𝑗5 (7) torque term is the torque that acts on the spacecraft due to

the Earth’s magnetic field. This torque can be written as [4]

(𝜇𝑖 ⋅ r𝑖𝑗 ) (𝜇𝑗 ⋅ r𝑖𝑗 )

+5 r𝑖𝑗 ) , 𝜏𝑒𝑖 = 𝜇𝑖 × B𝑒 (R𝑖 ) , (15)

𝑟𝑖𝑗7

where B𝑒 is Earth’s magnetic field strength at the location of

where 𝜇0 = 4𝜋 × 10−7 T⋅m/A is the permeability constant. 𝜇𝑖 the spacecraft.

and 𝜇𝑗 are the dipole strength of the 𝑖th satellite and the 𝑗th Substituting (10) into (11), the attitude dynamics can be

satellite, respectively. Note that (7) gives the force on dipole written compactly as

𝑖 (located on 𝑖th spacecraft) due to dipole 𝑗 (located on 𝑗th

spacecraft), which depends on the distance between the two M𝑖𝐴 (J𝑖 , q𝑖 ) 𝜀̈

𝑖 + C𝑖𝐴 (J𝑖 , Ω𝑖 , h) + N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 )

dipoles and the orientation of both dipoles in the inertial = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,

space. It is the dependence on the orientation of the dipoles (16)

that gives rise to the complexity of the expression for the force,

since the orientation of a dipole obviously depends on the where

orientation of the body axes in the inertial space [3].

M𝑖𝐴 (J𝑖 , q𝑖 ) = 2J𝑖 T𝑖 −1 ,

2.2. Attitude Dynamics. To avoid singular points, the Euler 𝑇

C𝑖𝐴 (J𝑖 , Ω𝑖 , h) = Ω×𝑖 (J𝑖 Ω𝑖 + h) , X𝑖 = [𝜌𝑇𝑖 , 𝜀𝑇𝑖 ] ,

parameter is chosen to describe the attitude of the spacecraft.

Let q𝑖 represent the Euler parameters corresponding to the

N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 ) = 2J𝑖 T𝑖 −1 𝑓𝑖 , 𝑓𝑖 = −𝜔×𝑏𝑖 𝜔𝑂𝑖 + 𝜔̇𝑂𝑖 .

attitude of the body frame relative to the orbital frame 𝐹𝑂𝑖 (17)

[19],

𝑇

q𝑖 = [𝜀𝑇𝑖 𝜂𝑖 ] ∈ R4 , 2.3. Combined Attitude and Translational Dynamics. Com-

(8) bining the attitude dynamics in (1) and the translational

𝑇

𝜀𝑖 ≡ [𝜀𝑖1 𝜀𝑖2 𝜀𝑖3 ] ∈ R3×1 , 𝜂𝑖 ∈ R1×1 . dynamics in (16), the following 6-DOF dynamics equation for

formation flying is obtained:

The Euler parameters, which are equivalent to the coeffi-

cients of unit quaternion, have unit norm by definition; hence M𝑖 Ẍ

𝑖 + C𝑖 + N𝑖 = u𝑖 + D𝑖 , (18)

where

𝜀𝑇𝑖 𝜀𝑖 + 𝜂𝑖2 = 1. (9)

I3×3 0 C

Attitude kinematics and dynamics of the spacecraft are M𝑖 = [ ] ∈ R6×6 , C𝑖 = [ 𝑖𝑇 ] ∈ R6×1 ,

0 M𝑖𝐴 C𝑖𝐴

governed by [19]

N𝑖𝑇 F̂

Ω𝑖 = 2T𝑖 −1 𝜀̇𝑖 + 𝜔𝑂𝑖 , (10) N𝑖 = [ ] ∈ R6×1 , u𝑖 = [ 𝑐𝑖 ] ,

(19)

N𝑖𝐴 𝜏𝑐𝑤𝑖

J𝑖 Ω̇𝑖 + Ω×𝑖 (J𝑖 Ω𝑖 + h) = 𝜏𝑚𝑖 + 𝜏𝑐𝑤𝑖 + 𝜏𝑑𝑖 , (11)

𝑇 F̂

D𝑖 ≜ [𝑑1𝑖 𝑑2𝑖 𝑑3𝑖 𝑑4𝑖 𝑑5𝑖 𝑑6𝑖 ] = [ 𝑑𝑖 ].

where T𝑖 ≡ 𝜂𝑖 I + 𝜀×𝑖 3×3

∈ R , the superscript “×” denotes 𝜏𝑚𝑖 + 𝜏𝑑𝑖

skew-symmetric matrix of a vector, and Ω𝑖 = 𝜔𝑂𝑖 + 𝜔𝑏𝑖 , 𝜔𝑏𝑖 is

angular velocity of the spacecraft relative to the orbital frame 3. Adaptive Terminal Sliding

𝐹𝑂𝑖 . 𝜔𝑂𝑖 is the orbital angular velocity of 𝑖th spacecraft. J𝑖 Mode Control Design

is the moment of inertia of the spacecraft. h is the angular

momentum of the flywheel. The magnetic torque term in (11) In this section, an adaptive terminal sliding mode controller

can be written as is designed for the follower spacecraft based on the dynamic

𝑁−1 model in (18) and the TSM technique. With this controller,

𝜏𝑚𝑖 = ∑ 𝜏𝑚

𝑖𝑗 , (12) the follower spacecraft can track the desired attitude and

𝑗=0, 𝑖 ≠

𝑗 relative position trajectories simultaneously.

4 Advances in Mechanical Engineering

3.1. Error Dynamics Equation. The trajectory tracking errors Substituting (24) and (25) into (11), the error attitude

of the follower spacecraft are defined as dynamics are derived as follows [19]:

𝑖 + C𝑖𝐴 (J𝑖 , Ω𝑖 , h) + N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑏𝑖 , 𝜔̇𝑏𝑖 )

where 𝜌𝑑𝑖 , 𝜌̇𝑑𝑖 ∈ R3×1 are the relative position/velocity of the = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,

desired trajectory with respect to the leader. Differentiating (27)

(20) and substituting it into (1), one can obtain that −1 −1

where N𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 f𝑖𝑒 , M𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 .

𝑇

𝑒

𝜌̈ 𝑒 𝑑 ̂ ̂ 𝑇 𝑇

𝑖 + C𝑖𝑇 (𝜔𝑂, 𝜌̇

𝑖 ) + N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 , 𝜌̈

𝑖 ) = F𝑑𝑖 + F𝑐𝑖 , Let e𝑖 = [𝜌𝑒𝑖 𝜀𝑒𝑖 ] , and utilizing (18), (21), and (27), the

(21) error dynamics of the whole system can be obtained

𝑑 𝑑 M𝑒𝑖 ë 𝑒

𝑖 + C𝑖 + N𝑖 = u𝑖 + D𝑖 , (28)

𝑖 ) = N𝑖𝑇 + 𝜌̈

𝑖.

The error Euler parameter is chosen to describe the atti-

where

tude error of the spacecraft with respect to the reference atti-

tude. The error Euler parameter is determined as follow [19]: I3×3 0 N𝑒𝑖𝑇

M𝑒𝑖 = [ ] ∈ R6×6 , N𝑒𝑖 = [ ] ∈ R6×1 .

0 M𝑒𝑖𝐴 N𝑒𝑖𝐴

𝑇

𝜀𝑒𝑖 = U𝑑𝑖 q𝑖 , (22) (29)

3.2. Controller Design. The terminal switching planes can be

𝑇 describe as [16–18]

𝜂𝑖𝑒 = q𝑑𝑖 q𝑖 , (23)

𝛾

S𝑖 = ė𝑖 + 𝛼𝑖 e𝑖 + 𝛽𝑖 e𝑖 𝑖 , (30)

𝑇 𝑇

where U𝑑𝑖 ≡ [T𝑑𝑖 −𝜀𝑖 ] ∈ R

4×3

, T𝑑𝑖 ≡ 𝜂𝑖𝑑 I + 𝜀×𝑑

𝑖 ∈R 3×3

, and

𝑇

𝑇 𝑇 where S𝑖 ≜ [𝑠1𝑖 𝑠2𝑖 𝑠3𝑖 𝑠4𝑖 𝑠5𝑖 𝑠6𝑖 ] , 𝛼𝑖 = diag(𝑎1𝑖 , 𝑎2𝑖 , . . . ,

q𝑑𝑖 ≡ [𝜀𝑑𝑖 R4×1 is the desired attitude.

𝜂𝑖𝑑 ] ∈ 𝑎6𝑖 ), 𝛽𝑖 = diag(𝑏1𝑖 , 𝑏2𝑖 , . . . , 𝑏6𝑖 ), 𝑎𝑛 > 0, 𝑏𝑛 > 0 (𝑛 = 1, 2, . . . , 6),

Differentiating (22) with respect to time and utilizing (10), and 0 < 𝛾𝑖 < 1. The unknown disturbances D𝑖 in (18) are

the attitude kinematics equations with error Euler parameter ∗

bounded but unknown. 𝜎𝑛𝑖 is the upper bound of 𝑑𝑛𝑖 . Here, it

are given by can be estimated by

1

Ω𝑏𝑖 = 2T𝑖 𝑒−1 𝜀̇𝑒𝑖 + A𝑒𝑖 𝜔𝑑𝑏𝑖 + A𝑏𝐼𝑖 𝜔𝑂𝑖 , (24) ̂̇𝑛𝑖 =

𝜎 𝑠 , (𝑛 = 1, 2, . . . , 6) , (31)

𝜗𝑛𝑖 𝑛𝑖

where A𝑒𝑖 is the transformation matrix from the reference

where 𝜗𝑛𝑖 is the adaptive gain.

attitude frame to the body frame.

Utilizing (28), (30), and (31), an adaptive terminal sliding

Differentiating (24), one can obtain that

mode control law is designed as follows:

−1

Ω̇𝑏𝑖 = 2T𝑒𝑖 𝜀̈

𝑒 𝑒

𝑖 + f𝑖 ,

(25) u𝑖𝑇

u𝑖 ≜ [ ] = C𝑖 + N𝑒𝑖

u𝑖𝐴

where

𝜎𝑖 ) sign (S𝑖 )) ,

f𝑖𝑒 = f𝑖 + 𝑖𝑒2 (𝜔𝑏𝑖 − A𝑒𝑖 𝜔𝑑𝑏𝑖 ) (𝜔𝑏𝑖 − A𝑒𝑖 𝜔𝑑𝑏𝑖 ) (32)

2𝜂𝑖

(26)

𝑇

̂ 𝑖 ≜ [𝜎̂1𝑖 𝜎̂2𝑖 𝜎̂3𝑖 𝜎̂4𝑖 𝜎̂5𝑖 𝜎̂6𝑖 ] and the vector

where 𝜎

+ A𝑒𝑖 𝜔̇𝑑𝑏𝑖 − 𝜔×𝑏𝑖 A𝑒𝑖 𝜔𝑑𝑏𝑖 signum function sign(S𝑖 ) is a column of signum functions

𝑇

sign (S𝑖 ) ≜ [sign (𝑠1𝑖 ) sign (𝑠2𝑖 ) sign (𝑠3𝑖 ) sign (𝑠4𝑖 ) sign (𝑠5𝑖 ) sign (𝑠6𝑖 )] . (33)

Substituting (32) into (18) produces the closed-loop dynamics Consider a Lyapunov function as follows:

𝛾−1

𝑖 = −𝛼𝑖 ė

ë 𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 − diag (̂

𝜎𝑖 ) sign (S𝑖 ) + D𝑖 . 1 1 𝑇

𝑉 = S𝑇𝑖 S𝑖 + 𝜎

̃ diag (𝜗𝑖 ) 𝜎

̃𝑖, (35)

(34) 2 2 𝑖

Advances in Mechanical Engineering 5

𝑇

where 𝜗𝑖 ≜ [𝜗1𝑖 𝜗2𝑖 𝜗3𝑖 𝜗4𝑖 𝜗5𝑖 𝜗6𝑖 ] . The estima- It means that the switching planes 𝑠𝑛𝑖 (𝑛 = 1, 2, . . . , 6)

tion error is defined as 𝜎 ̃𝑖 = 𝜎 ̂ 𝑖 − 𝜎∗𝑖 , 𝜎∗𝑖 ≜ converge to zero [20]. On the other hand, in the TSM 𝑠𝑛𝑖 = 0

∗ ∗ ∗ ∗ ∗ ∗ 𝑇

[𝜎1𝑖 , 𝜎2𝑖 , 𝜎3𝑖 , 𝜎4𝑖 , 𝜎5𝑖 , 𝜎6𝑖 ] . (𝑛 = 1, 2, . . . , 6), the system state will reach zero in finite time

[17].

The first derivative of (30) can be expressed as

Remark 1. For the purpose of eliminating chattering, a

Ṡ𝑖 = ë

𝑖 + 𝛼𝑖 ė

𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖

𝛾−1

) ė𝑖 . (36) common practice is to replace the signum function of (32)

with a continuous saturation function

Differentiating (35) and utilizing (36), one can obtain that sat (𝑠𝑛𝑖 , 𝜉𝑛𝑖 )

𝑠𝑛𝑖

{

{ if 𝑠𝑛𝑖 < 𝜉𝑛𝑖 (39)

𝑉̇= S𝑇𝑖 Ṡ𝑖 + ̃̇𝑖

̃ 𝑇𝑖 diag (𝜗𝑖 ) 𝜎

𝜎 𝜉𝑛𝑖

={ 𝑛 = 1, 2, . . . , 6,

(37) {

= S𝑇𝑖 (ë

𝑖 + 𝛼𝑖 ė

𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖

𝛾−1

̃ 𝑇𝑖 S𝑖 ,

) ė𝑖 ) + 𝜎 { sign (𝑠𝑛𝑖 ) if 𝑠

𝑛𝑖 ≥ 𝜉𝑛𝑖 ,

𝑇 (32) can be written as

where ‖S𝑖 ‖ ≜ [|𝑠1𝑖 | |𝑠2𝑖 | |𝑠3𝑖 | |𝑠4𝑖 | |𝑠5𝑖 | |𝑠6𝑖 |] . Substituting

(31) and (34) into (37) produces that u𝑖 = C𝑖 + N𝑒𝑖

𝛾 −1

𝑉̇= −S𝑇𝑖 (diag (̂ 𝜎𝑇𝑖 − 𝜎∗𝑇

𝜎𝑖 ) sign (S𝑖 ) − D𝑖 ) + (̂

+ M𝑒𝑖 (−𝛼𝑖 ė𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 𝑖 ) ė𝑖 (40)

𝑖 )S𝑖

6 (38) − diag (̂

𝜎𝑖 ) sat (S𝑖 , 𝜉𝑖 ) ) ,

∗

≤ − ∑ 𝑠𝑛𝑖 (𝜎𝑛𝑖 − 𝑑𝑛𝑖 ) < 0 for 𝑠𝑛𝑖 ≠

0.

𝑛=1 where

𝑇

sat (S𝑖 , 𝜉𝑖 ) = [sat(𝑠1𝑖 , 𝜉1𝑖 ) sat(𝑠2𝑖 , 𝜉2𝑖 ) sat(𝑠3𝑖 , 𝜉3𝑖 ) sat(𝑠4𝑖 , 𝜉4𝑖 ) sat(𝑠5𝑖 , 𝜉5𝑖 ) sat (𝑠6𝑖 , 𝜉6𝑖 )] . (41)

Remark 2. There exists a possible singularity in sliding mode Substituting (5) and (7) into (44) produces

controller as e𝑖 → 0. Since e𝑖 = 0 only approaches along a

sliding mode, we observe that for a general choice of 𝛾𝑖 𝑁−1 𝑚𝑖 𝑚𝑗

∑ 𝑓𝑖𝑗 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = u𝑖𝑇 . (45)

𝑗=0, 𝑗 ≠

𝑖 𝑚𝑖 + 𝑚𝑗

𝛾

ė𝑖 = −𝛼𝑖 e𝑖 − 𝛽𝑖 diag (e𝑖 ) (42)

Note that (7) gives the force on the 𝑖th SMD due to the 𝑗th

while sliding and that the component in (32) SMD and it depends on the distance between the two SMDs

and the orientation of both SMDs in the inertial space. It rises

𝛾−1 𝛾 2𝛾𝑖 −1 to the complexity of the expression for the magnetic. Here, we

𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 → −𝛾𝑖 𝛽𝑖 (𝛼𝑖 diag (e𝑖 𝑖 ) + 𝛽𝑖 diag (e𝑖 )) .

consider a two-spacecraft electromagnetic formation flying

(43) (EMFF) array. By defining a rotated frame 𝐹𝑟 (see Section 2),

a simplified algebraic form of (7) is obtained

Consequently there will be a singularity in (32) unless 𝛾𝑖 is

chosen so that 2𝛾𝑖 > 1. To satisfy this requirement we set 𝛾𝑖 = 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧

3/5 for the examples to follow. 3𝜇0 [ ]

𝑓𝑖𝑗𝑟 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] . (46)

4𝜋𝑟𝑖𝑗4

[ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]

3.3. Compute the Special Magnetic Moments of SMD. The

adaptive sliding mode controller u𝑖𝑇 gives the desired special Utilizing (45) and (46) produces that

forces that can be used as input for a thruster-based system.

However, for EMFF, the desired special forces are produced 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧 4

by SMD. The SMD of each spacecraft is a complicated func- [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] 4𝜋𝑟𝑖𝑗 𝑚𝑖 𝑚𝑗

[ ]= A A u ,

tion of current position and orientation [3, 10]. For EMFF, 3𝜇0 𝑚𝑖 + 𝑚𝑗 𝑟𝐼𝑖 𝐼𝑏𝑖 𝑖𝑇

the control variables consist of the dipoles of individual [ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]

spacecraft. Utilizing (32) and (4), one can obtain that (47)

F𝑐𝑚1 = u , F𝑐𝑚0 = − u . (44)

𝑚0 + 𝑚1 𝑖𝑇 𝑚0 + 𝑚1 𝑖𝑇 ECI reference frame 𝐹𝐼 to the rotated frame 𝐹𝑟 . A𝐼𝑏𝑖 is

6 Advances in Mechanical Engineering

Spacecraft Semimajor axis Inclination Right ascension of Argument of Eccentricity Mean anomaly

ascending node perigee

LS 6878 km 1.106538745764 rad 0.523598775598 rad 0 rad 0.003 0 rad

FS 6878 km 1.106538745764 rad 0.523598071512 rad 0.000000315261 rad 0.002999636522 0 rad

the transformation matrix from the body frame 𝐹𝐵𝑖 of 𝑖th on a unique nonlinearity of the magnetic torques [3, 4]. The

spacecraft to ECI frame. force acting between a pair of dipoles depends on the product

𝑟 𝑟 𝑟 𝑟 𝑇 4

Let ŭ𝑖𝑇 ≜ [𝑢̆ 𝑖𝑇𝑥 , 𝑢̆

𝑖𝑇𝑦 , 𝑢̆

𝑖𝑇𝑧 ] = (4𝜋𝑟12 /3𝜇0 )(𝑚1 𝑚2 /(𝑚1 +

of the individual dipole values. By switching the polarity of

𝑚2 ))A𝑟𝐼𝑖 A𝐼𝑏𝑖 u𝑖𝑇 , 𝜇𝑖 = −𝜇𝑗 , and utilizing (47), one can obtain all dipoles in the EMFF, the torque acting on the spacecraft

that due to Earth’s magnetic field changes sign, but the torques

and forces due to the other spacecraft in the system do not.

1 As can be seen from (15), it results in a net cancellation

𝜇𝑖𝑥 = √ √𝑢̆

𝑟2 𝑟2

𝑖𝑇𝑥 + 2𝑢̆

𝑟2

𝑖𝑇𝑦 + 2𝑢̆

𝑟2

𝑖𝑇𝑧 − 𝑢̆

𝑖𝑇𝑥 ,

2 of the effect of the Earth’s magnetic field on the average

𝑟2

sense.

𝑢̆

𝑖𝑇𝑦

𝜇𝑖𝑦 = ,

𝑟2

√ √𝑢̆ 𝑟2 𝑟2 𝑟2

𝑖𝑇𝑥 + 2𝑢̆

𝑖𝑇𝑦 + 2𝑢̆

𝑖𝑇𝑧 − 𝑢̆

𝑖𝑇𝑥

(48) 4. Numerical Simulation

𝑟2 The adaptive terminal sliding mode controller equation (40)

𝑢̆

𝑖𝑇𝑧

𝜇𝑖𝑧 = . was simulated for a two-spacecraft formation flying control.

𝑟2

√ √𝑢̆ 𝑟2 𝑟2 𝑟2 Considering the nonlinear dynamics with disturbance D1 ,

𝑖𝑇𝑥 + 2𝑢̆

𝑖𝑇𝑦 + 2𝑢̆

𝑖𝑇𝑧 − 𝑢̆

𝑖𝑇𝑥

the effectiveness of the proposed controller was verified.

The orbital parameters of leader and follower spacecraft

The special magnetic moments of SMD can be computed in the simulations are listed in Table 1. The parameters of

using (48), by which the follower spacecraft can track the spacecraft are summarized in Table 2. The dynamics includes

desired relative position trajectories. the second harmonic of the gravitational field as well as

the Earth’s magnetic field, of which the vector is calculated

3.4. Angular Momentum Management. For EMFF in near- using the World Magnetic (WMM2005)Model block in the

Earth orbits, a constant disturbance torque may act on each paper.

spacecraft due to the Earth’s magnetic field that causes the The follower was commanded to move around the leader

reaction wheels on each spacecraft to quickly become satu- in an elliptic orbit. The desired trajectory was generated by

rated. In order to avoid this situation, the angular momentum solving nonlinear equation (1) numerically (set F̂ ̂

𝑑1 and F𝑐1

management strategy herein can be utilized, which is based equal to zero) with the following initial condition:

(49)

𝜌̇𝑑1 (0) = [0.000000000000, −0.000005559083863, −0.000000000758] [m/s] .

(50)

𝜌̇1 (0) = [−0.000000000000, −0.005557083863, −0.000000000758] [m/s] .

3

The parameters of the adaptive sliding mode controller are 𝛾1 = , 𝜉 = [5, 5, 5, 1, 1, 1] × 10−3 ,

then 5

𝛼1 = diag (2, 2, 3, 250, 80, 80) × 10−4 , 𝜗1 = [15, 22, 22, 80, 125, 125] .

(51)

𝛽1 = diag (10, 15, 15, 1200, 800, 800) × 10−5 ,

Advances in Mechanical Engineering 7

0.2

Spacecraft Mass Moment of inertia

(deg)

0

Ls 100 kg Diag (19, 19, 32) kg⋅m2

FS 100 kg Diag (19, 19, 32) kg⋅m2 −0.2

0 1000 2000 3000 4000 5000 6000

(a)

Relative trajectory rou

Error (𝜃2 )

5 0.2

(deg)

0

z (m)

0 −0.2

∗ 0 1000 2000 3000 4000 5000 6000

(b)

−5

10 Error (𝜃3 )

5 4 0.5

0 2

0

(deg)

y (m −5 −2 ) 0

) −10 −4 x (m

−0.5

Figure 2: Actual trajectory 𝜌1 (𝑡) of the follower spacecraft (∗ the 0 1000 2000 3000 4000 5000 6000

leader spacecraft.) Time (s)

(c)

Error (x-direction)

0.1 Figure 4: Attitude tracking error.

(m)

0.05

0

0 1000 2000 3000 4000 5000

(a) 𝜇x (Am2 )

1000

Error (y-direction) 0

0.8 −1000

0.6 −2000

(m)

0.2

0 (a)

0 1000 2000 3000 4000 5000

𝜇y (Am2 )

(b)

0

Error (z-direction)

−2000

0.2

−4000

(m)

0

(b)

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500

Time (s) 𝜇z (Am2 )

1000

(c) 0

−1000

−2000

Figure 3: Position tracking error.

0 1000 2000 3000 4000 5000

Time (s)

(c)

In this simulation, a switching time of 100 second was

Figure 5: Magnetic dipole strength in 𝐹𝐵1 frame.

used for AMM. Figures 2–6 show the simulation results,

which are obtained by simulating the adaptive terminal

sliding mode controller. The phase portrait of the trajectory

𝜌1 (𝑡) of the follower spacecraft relative to the leader spacecraft

is illustrated in Figure 3, where ∗ represents the leader

spacecraft at the origin. Figure 3 depicts the position tracking and it is clear from this figure that the angular momentum

error of the follower spacecraft. Figure 4 gives attitude error buildup was limited by AMM. In a stable tracking situation,

of the follower spacecraft. Figure 5 shows the magnetic dipole simulation results show that ‖𝜌𝑒1 ‖ is smaller than 6 × 10−3 m

strength of the follower spacecraft. Figure 6 gives the reaction and ‖𝜀𝑒1 ‖ is smaller than 0.075∘ which meets the required

wheel angular momentum buildup of the follower spacecraft, tracking accuracy.

8 Advances in Mechanical Engineering

hx Conflict of Interests

5

(Nms)

regarding the publication of this paper.

−5

0 1000 2000 3000 4000 5000 6000

(a) Acknowledgment

hy This work was supported by the Natural Science Foundation

2

of China under Grant Nos. 11372353 and 10902125.

(Nms)

−2 References

0 1000 2000 3000 4000 5000 6000

[1] J. Bristow, D. Folta, and K. Hartman, “A formation flying

(b)

technology vision,” in Proceedings of the AIAA Space Conference

hz and Exposition, Long Beach, Calif, USA, September 2000,

5

AIAA-2000-5194.

(Nms)

formation flying experiment: earth observing-1 (EO-1),” in Pro-

−5

0 1000 2000 3000 4000 5000 6000 ceedings of the AIAA/AAS Astrodynamics Specialist Conference

Time (s) and Exhibit, Monterey, Calif, USA, August 2002, AIAA-2002-

4743.

(c)

[3] U. Ahsun and D. W. Miller, Dynamics and control of electromag-

Figure 6: Reaction wheel angular momentum. netic satellite formations [Ph.D. thesis], Massachusetts Institute

of Technology, 2007.

[4] U. Ahsun, D. W. Miller, and J. L. Ramirez, “Control of electro-

magnetic satellite formations in near-Earth orbits,” Journal of

Guidance, Control, and Dynamics, vol. 33, no. 6, pp. 1883–1891,

5. Conclusions 2010.

[5] L. Johnson, B. Gilchrist, R. D. Estes, and E. Lorenzini, “Over-

In this paper, we consider the control problem for six- view of future NASA tether applications,” Tech. Rep. 1998-

degree-of-freedom electromagnetic spacecraft formation fly- 0237034, NASA, 1998.

ing (EMFF) in near-Earth orbits. Firstly, the combined atti- [6] D. Quinn and D. Folta, “A tethered formation flying concept for

tude and translational dynamics are presented. Secondly, we the SPECS missions,” Tech. Rep. 2000-0032752, NASA, 2000.

confine our attention mainly to the position/attitude tracking [7] L. B. King, G. G. Parker, S. Deshmukh, and J. Chong, “Spacecraft

control using the terminal sliding mode and adaptive control formation-flying using inter-vehicle coulomb forces,” NIAC

technique. Based on a rotated frame 𝐹𝑟 and the adaptive Phase I Final Report, 2002, http://www.niac.usra.edu/.

TSM controller, the special magnetic moments of SMD are [8] J. P. Shoer and M. A. Peck, “A flux-pinned magnet-supe-

computed. The AMM strategy is implemented in a peri- rconductor pair for close-proximity station keeping and self-

odically switching fashion. Through numerical simulation, assembly of spacecraft,” in Proceedings of the AIAA Guidance,

it is clear that the output tracking error can converge to Navigation, and Control Conference and Exhibit, pp. 611–628,

zero in finite time by using the proposed controller and that Hilton Head, SC, USA, August 2007, AIAA Paper 2007-6352.

the angular momentum buildup was limited by AMM. At [9] M. R. LaPointe, “Formation flight method,” NIAC Phase I Final

the same time, strong robustness with respect to bounded Report, 2001, http://www.niac.usra.edu/.

disturbances can be guaranteed (in the simulation, the 𝐽2 dis- [10] D. W. Miller, R. J. Sedwick, D. Jiang, D. Jin et al., “Electro-

turbance, the gravity gradient torques 𝜏𝑔1 , the Earth magnetic magnetic formation flight,” NIAC Phase I Final Report, Mas-

torques 𝜏𝑒1 , and the magnetic torque term 𝜏𝑚1 have been sachusetts Institute of Technology, 2002.

considered). [11] D. W. Miller, R. J. Sedwick, E. M. C. Kong, and S. Schweighart,

“Electromagnetic formation flight for sparse aperture tele-

scopes,” in Proceedings of the IEEE Aerospace Conference, Big

Nomenclature Sky, Mont, USA, March 2002.

[12] L. M. Elias, D. W. Kwon, R. J. Sedwick, and D. W. Miller, “Elec-

EMFF: Electromagnetic spacecraft formation flying tromagnetic formation flight dynamics including reaction

TSM: Terminal sliding mode wheel gyroscopic stiffening effects,” Journal of Guidance, Con-

AMM: Angular momentum management trol, and Dynamics, vol. 30, no. 2, pp. 499–511, 2007.

SFF: Spacecraft formation flying [13] E. M. C. Kong, D. W. Kwon, S. A. Schweighart, L. M. Elias, R.

HTS: High-temperature superconducting J. Sedwick, and D. W. Miller, “Electromagnetic formation flight

TSMC: Terminal sliding mode control for multisatellite arrays,” Journal of Spacecraft and Rockets, vol.

SMD: Steerable magnetic dipole 41, no. 4, pp. 659–666, 2004.

ECI: earth center inertial frame [14] U. Ahsun and D. W. Miller, “Dynamics and control of electro-

TPF: terrestrial planet finder. magnetic satellite formations,” in Proceedings of the American

Advances in Mechanical Engineering 9

June 2006.

[15] Y. Feng, X. Yu, and Z. Man, “Non-singular terminal sliding

mode control of rigid manipulators,” Automatica, vol. 38, no.

12, pp. 2159–2167, 2002.

[16] X. Yu and Z. Man, “Variable structure systems with terminal

sliding modes,” in Variable Structure Systems: Towards the 21st

Century, vol. 274 of Lecture Notes in Control and Information

Sciences, pp. 109–128, Springer, Berlin, Germany, 2002.

[17] X. Yu, Y. Wu, and Z. Man, “On global stabilization of nonlinear

dynamical systems,” in Variable Structure Systems, Sliding Mode

and Nonlinear Control, vol. 274 of Lecture Notes in Control and

Information Sciences, pp. 109–122, Springer, Berlin, Germany,

1999.

[18] D. J. Stonier and R. J. Stonier, “Obstacle avoidance and finite-

time tracking of mobile targets,” in Proceedings of the 2nd

International Conference on Autonomous Robots and Agents, pp.

58–63, Palmerston North, New Zealand, December 2004.

[19] C.-Q. Yuan, J.-F. Li, T.-S. Wang, and H.-X. Baoyin, “Robust

attitude control for rapid multi-target tracking in spacecraft

formation flying,” Applied Mathematics and Mechanics, vol. 29,

no. 2, pp. 185–198, 2008.

[20] D. R. Merkin, Introduction to the Theory of Stability, Spring, New

York, NY, USA, 1996.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 592628, 9 pages

http://dx.doi.org/10.1155/2014/592628

Research Article

Application of Krylov Reduction Technique for a Machine Tool

Multibody Modelling

Research Center of Manufacturing Technology, Czech Technical University in Prague, Horská 3, 128 00 Prague, Czech Republic

Copyright © 2014 M. Sulitka et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Quick calculation of machine tool dynamic response represents one of the major requirements for machine tool virtual modelling

and virtual machining, aiming at simulating the machining process performance, quality, and precision of a workpiece. Enhanced

time effectiveness in machine tool dynamic simulations may be achieved by employing model order reduction (MOR) techniques

of the full finite element (FE) models. The paper provides a case study aimed at comparison of Krylov subspace base and mode

truncation technique. Application of both of the reduction techniques for creating a machine tool multibody model is evaluated.

The Krylov subspace reduction technique shows high quality in terms of both dynamic properties of the reduced multibody model

and very low time demands at the same time.

In the case of ball screw driven motion axes, the above

In recent years, with increasing computational power of mentioned issue relates predominantly to active length of the

computers, machine tool virtual modeling has become one ball screw. Next to it, structural modal properties of large

of the intensively developed tools for simulating the machine machine tools may vary significantly also depending on the

tool behavior in interaction with machining process, aiming kinematical configuration and actual position of the motion

especially at predicting the machining performance, process axes.

quality, and stability [1–4]. A key element in machine tool

dynamic behavior represents the compliant machine tool Relevant capturing of the machine tool structural prop-

mechanical system, including both the structural parts and erties represents an important tasks in virtual modeling.

the components of feed drive systems as well. Excitation Coupled modeling techniques employing the reduced FE

of natural eigenfrequencies of the machine tool mechanics models of the machine tool structure have been developed for

stands in the foreground of problems and represents one of describing the ball screw feed drive and machine tool inter-

the limiting factors related to enhancing the machining per- action. Commonly the coupled models rely on one kinematic

formance, since it can lead to either deteriorated machined configuration and application of modal or CMS reduction

surface quality, workpiece precision, or even generation of techniques [6–9]. Importance of proper prediction of higher

instable machining process. frequency range dynamics of feed drive mechanical structure

The modern machine tool, as a typical mechatronic with respect to feed drive control parameters shows [10].

system, features a number of interactions between the control Using the coupled models, machining process stability and

and mechanics [5]. Achievable parameters of both the CNC performance may be predicted by the oriented directional

control and feed drive control as well relate directly to frequency response functions (FRFs) at tool center point

mechanical eigenfrequencies and damping of the mechanical (TCP) [11].

system. Commonly, the setting of the control parameters is For updating of machine tool structural modal prop-

defined by the most compliant configuration of the machine erties, a technique based on continuous redistribution of

tool mechanics, and thus the potential of the structural contact force along the motion axis linear guideway has been

2 Advances in Mechanical Engineering

developed in [12]. Another approach employing an addition- The Krylov subspace methods [26, 29–31] are very

ally created FE node moving between the end positions of the interesting because of their iterative nature which allows

motion axes stroke has been described in [13]. the reduction of large-scale problems. The computational

The finite element method (FEM) is used as a common efficiency of Krylov subspace based MOR has encouraged

way to simulate the dynamic behavior of a machine tool. wide interest in the method, and therefore wide knowledge

The solution of FE problems usually involves the solution in different fields is available. We will mention the most

of a large set of sparse algebraic equations. The number important observations to date. The passivity and stability

of unknowns in ordinary engineering FE model nowadays preservation has been achieved using the Krylov MOR

is commonly 106 -107 . The computational time needed to methods in [26, 32]. A Krylov algorithm preserving structure

solve static analysis of such problems on current hardware is of second order ordinary differential equations has been pre-

acceptably low (∼hours). The issue arises when the harmonic sented in [33]. Handling of nonlinear convection coefficient

or transient analysis is required. The solution then requires was studied in [34]. Reduction of coupled physics problems

10 s or 100 s of iterations and therefore requires days or even was studied in [35] for the case of a thermomechanical model

weeks to solve. Such long solution time effectively hinders of packages and in [36] for the case of structural-acoustic

this kind of simulations in machine tool virtual model coupled models. Krylov subspace MOR was also successfully

simulations. The model order reduction (MOR) methods are used in optimization of MEMS devices [37] and sensitivity

one of possible ways to speed up the solution and make analysis of structural frequency response [38]. One of the

it feasible in practice. The idea behind MOR is to reduce most important directions in development on the Krylov base

the number of unknowns while producing sufficiently good reductions is a parametric model order reduction (PMOR).

approximation to the input/output behavior. The PMOR allows preservation of parameters which the

The following text gives a basic review of model order system depends on [39, 40]. The dependence of parameters

reduction techniques. The idea of reduction of the number may be either linear or nonlinear.

of unknowns during the solution of the FE model is almost The comparison of different model order reduction meth-

as old as the FE method itself. The first MOR method was ods has been discussed in [17], where the Krylov subspace

MOR method was found to be one of the best methods. The

static reduction proposed by Guyan [14] and Irons [15]. This

comparison of Krylov, CMS, and balanced truncation can be

method was introduced for structural mechanics problems

found in [41].

but it is also valid for thermal analysis and other analyses

Based on the findings presented in previous works the

regardless of the underlying physics. However, this method

Krylov subspace based MOR is very robust and computation-

is of questionable quality when using it for dynamic analyses

ally efficient. The goal of this study is to show a new strategy

as was shown in [16, 17].

of employing the Krylov MOR technique for creating the

To remedy the insufficiencies in the static condensation coupled models of compliant systems. The idea is to reduce

method, the component mode synthesis (CMS, [18]) was each part of the machine tool structure separately and to

proposed by Bampton and Craig. The CMS has become produce the coupled model of the whole assembly. This way

widely used by the engineering community. The CMS was enables coupling of reduced components in any kinematic

used to efficiently conduct not only large-scale structural configuration. To assess the quality of Krylov MOR and to

eigenanalysis [19] but also transient heat conduction analysis show its soundness the following comparison properties will

[20] and heat conduction/convection analysis [21]. Another be evaluated:

field of application of the CMS is coupled physics simulations.

The weakly coupled thermomechanical models were studied (1) low error in approximation,

in [22]. There is still active research regarding improvement (2) fast computation.

of the CMS [23]. A low error in the approximation of a full FE model

Another method enhanced to approximate dynamic sys- is required, as the objective is to replace the full FE model

tems well is the improved reduced system (IRS) proposed by in simulations. Fast computations are required to meet the

O’Callahan in [24]. Later, Friswell et al. developed an iterated requirements in virtual machine tools simulations.

version of IRS in [25]. The static condensation, CMS, and The proposed procedure will be tested on industrial scale

IRS can be viewed as engineering approaches to reduce the (106 -107 DOFs) FE model of machine tool. The harmonic

number of equations. and transient simulations on such large models take a lot of

The global error bounds and the preservation of passivity time (days, weeks), and therefore some kind of acceleration is

and stability are important questions posed on the MOR usually used. Mode truncation [42] is frequently used for fast

methods in a more mathematical point of view. Two of simulations and will be compared with Krylov based MOR.

the MOR techniques proposed in accordance with these Both methods will be employed in reduction of individual

questions are Krylov subspace reduction [26] and Balanced parts of machine tool. The reduced part will be then coupled

truncation [26]. Balanced truncation methods [27] have a in any kinematic position. The purpose is to simulate flexible

great advantage because there exists a priori global error multibody systems accurately and very fast (with possible

bound. But it also has a great disadvantage in that the applications in real-time CNC control simulations). The

Lyapunov equation [28] needs to be solved in order to proposed method of coupling the Krylov based reduced

reduce the system. Thus the usage of balanced truncation in substructures has not yet been reported in the literature, to

reduction of large-scale systems is limited. the best the authors’ knowledge.

Advances in Mechanical Engineering 3

The paper is organized in the following way: Section 1 Here 𝑥𝑢 is a vector of unique DOFs. Because (5) indicates that

contains the introduction and the motivation of work; the substructures DOF are obtained from the unique set 𝑥𝑢 ,

approach to substructuring is considered in Section 2; the compatibility equation (3) is satisfied for any set 𝑥𝑢 :

Section 3 contains the description of Krylov subspace reduc-

tion; Section 4 introduces the used machine tool model; and B𝑥 = BL𝑥𝑢 = 0 ∀𝑥𝑢 . (6)

Section 5 presents the comparison of the results obtained This means that L is null space of B and vice versa. And it is

by the methods. Conclusion suggestions on future work are only necessary to define one of the matrices L and B:

given in Section 6.

L = null (B) ,

2. Substructuring (7)

B𝑇 = null (L𝑇 ) .

One way to couple dynamic systems is to connect their mass

M𝑠 , damping C𝑠 , and stiffness K𝑠 matrices; external force Substituting (5) into (1), one gets equation of motion

vectors 𝑓𝑠 , and internal force vectors 𝑔𝑠 by coupling equations ML𝑥𝑢̈ + CL𝑥𝑢̇ + KL𝑥𝑢 = 𝑓 + 𝑔. (8)

[43]; the subscript 𝑠 denotes the systems being coupled. This

approach is similar to the global matrix assembly in FE Further multiplication of (8) from the left-hand side by L𝑇

software. The equations of motion of 𝑛 coupled subsystems yields the final form of the equation of motion of the coupled

can be written as system

M𝑥̈+ C𝑥̇+ K𝑥 = 𝑓 + 𝑔. (1) M ̃𝑥𝑢̇ + K𝑥

̃𝑥𝑢̈ + C ̃

̃ 𝑢 = 𝑓, (9)

where matrices M, ̃ and vector 𝑓̃ are expressed as

̃ and K

the subsystem matrices. Vectors 𝑓 and 𝑔 are column vectors

containing the subsystem internal and external force vectors, ̃ ≜ L𝑇 ML,

M

respectively.

Consider ̃ ≜ L𝑇 CL,

C

(10)

M = diag (M(1) , . . . , M(𝑛) ) , ̃ ≜ L𝑇 KL,

K

Coupling of Reduced Components. Let us consider the case of

K = diag (K(1) , . . . , K(𝑛) ) , (2) 𝑛 independent substructures, each reduced using projection

𝑥(1) 𝑓(1) 𝑔(1) matrix R(𝑖) :

[ .. ] [ .. ] [ .. ]

𝑥 = [ . ], 𝑓 = [ . ], 𝑔 = [ . ]. 𝑥 ≅ R𝜂, R ≜ diag (𝑅(1) ⋅ ⋅ ⋅ 𝑅(𝑛) ) , (11)

(𝑛) (𝑛) (𝑛)

[𝑥 ] [𝑓 ] [𝑔 ] where 𝑥 are physical coordinates of substructures and 𝜂 are

generalized coordinates of substructures. Substituting (11)

The compatibility condition (coupling equations) is ex-

into (1) we get

pressed as

M(𝑠) R𝜂(𝑠)

̈ + CR𝜂(𝑠)

̇ + K(𝑠) R𝜂(𝑠) = 𝑓(𝑠) + 𝑔(𝑠) + 𝑟(𝑠) , (12)

B𝑥 = 0, (3)

where 𝑟(𝑠) are residual forces due to model order reduction.

B is a Boolean if the interface degrees of freedom match 𝑇

Following Galerkin method, we enforce R(𝑠) 𝑟(𝑠) = 0. The

(interfaces are conforming). In this case, the coupling equa- equations of motion in generalized coordinates 𝜂 have the

tions are very simple 𝑥(𝑘) − 𝑥(𝑙) = 0. If the connected degrees form

of freedom do not coincide, the relations among them are

more complex and the matrix B is real. M𝑚 𝜂̈+ C𝑚 𝜂̇+ K𝜂𝑚 = 𝑓𝑚 + 𝑔𝑚 , (13)

Equilibrium condition is given by

where

𝑇

L 𝑔 = 0, (4) M𝑚 ≜ R𝑇 MR,

where L is a Boolean matrix describing the relations among C𝑚 ≜ R𝑇 CR,

the interface forces.

Equations (1), (2), (3), and (4) can be now used to couple K𝑚 ≜ R𝑇 KR, (14)

any number of subsystems. Before it can be done, however,

the unknown interface forces have to be eliminated: 𝑓𝑚 ≜ R𝑇 𝑓,

𝑥 = L𝑥𝑢 . (5) 𝑔𝑚 ≜ R𝑇 𝑔.

4 Advances in Mechanical Engineering

Then following, procedure for coupling full matrices (3)– Full model with n physical DOFs

(10), we get coupled reduced system. This procedure allows

reducing substructures independently and then coupling of

reduced systems. The coupling interface is defined using

matrix B (3), and therefore the specification of the coupling

interface is arbitrary.

Mode superposition

Full model with n model DOFs

Order Systems

n

The underlying physics of the problem is described by a m1 m2 m3 mn

system of linear second order ordinary differential equations

with constant coefficients k1 k2 k3 ··· kn

c1 c2 c3 cn

M𝑥̈(𝑡) + C𝑥̇(𝑡) + K𝑥 (𝑡) = F𝑢 (𝑡) ,

m

where 𝑥(𝑡) ∈ 𝑅𝑁 is displacement vector of state variables,

𝑢(𝑡) ∈ 𝑅𝑁 is input function, and 𝑦(𝑡) ∈ 𝑅𝑚 is output function. m1 m2 m3 mm

The matrices M, C, and K ∈ 𝑅𝑁×𝑁 are mass matrix, damping

k1 k2 k3 ··· km

matrix and stiffness matrix. F ∈ 𝑅𝑁×𝑙 , and L ∈ 𝑅𝑁×𝑚 are c1 c2 c3 cm

input and output matrices. 𝑥0 , 𝑥1 ∈ 𝑅𝑁 are initial conditions.

The system matrices M, C, and K are constant therefore the

Figure 1: Diagram of modal truncation [42].

system (15) is a linear time invariant system (LTI).

model order reduction is modal truncation method [42]. independent because the matrices E, C𝑞 , and Λ are diagonal.

The projection matrix T is in this case obtained from modal This means that the whole system of equations can be looked

analysis performed on the full model. The diagram of model at as a sum of many one-degree-of-freedom systems (Mode

reduction by modal truncation can be seen in Figure 1. superposition).

The equation of motion of the full model is The model order is reduced by including only a small

number of mode shapes and degrees of freedom in (19). The

M𝑥̈+ BV 𝑥̇+ K𝑥 = 𝑓, (16)

number of mode shapes included depends on the required

where M is the mass matrix, B is damping matrix, K is accuracy of the reduced model. It is recommended that the

stiffness matrix, 𝑓 is force vector and 𝑥 is the displacement natural frequency of the last included mode shape is at least

vector. twice as high as the highest working frequency of the resulting

The projection matrix T consists of the system’s mode reduced model. The number of included DOFs does not affect

shapes; each column of T corresponds to one mode shape 𝑉𝑖 the model accuracy so only the important ones are included.

and each row represents a degree of freedom: One disadvantage of this method is the fact that the

reduced model’s FRF never fits the full model’s FRF near 0 Hz.

T = [𝑉1 ⋅ ⋅ ⋅ 𝑉𝑛 ] . (17) Another drawback is the relatively long time needed to create

the reduced model (modal analysis with extraction of lots of

The vector 𝑥 can be transformed to vector of modal coordi- modes has to be performed). The method is on the other hand

nates using simple and accurate in chosen frequency range.

𝑥 = T𝑞. (18)

3.2. Krylov MOR. In this section only the basics behind

This equation can be then substituted to (16) and after Krylov reductions will be described. The reader is encouraged

multiplying the result by T𝑇 from the left-hand side, one to read an excellent mathematical description of Krylov based

obtains the equation of motion in the full system in the modal reductions in [26]. An overview of the reduction methods

coordinates of is given in [27]. Although optimal Krylov based reduction

algorithms are available [44], a simpler and a possibly more

E𝑞 ̈+ C𝑞 𝑞 ̇+ Λ𝑞 = T𝑇 𝑓, (19) computationally efficient method will be used in this work—

a block rational Krylov method [45]. The structure of second

where E is identity matrix, C𝑞 is damping ratio matrix, order ODEs in (15) will be preserved using Bai’s algorithm

and Λ is matrix of eigenvalues. The equations in (19) are [33].

Advances in Mechanical Engineering 5

using the following coordinate transformation in case of (15):

e

x

𝑥 (𝑡) = 𝑥̃ (𝑡) + 𝑥0 ,

̃̇(𝑡) + 𝑥1 ,

𝑥̇(𝑡) = 𝑥 (20)

q

̃̈(𝑡) .

𝑥̈(𝑡) = 𝑥

V

Substituting (20) into (15),

̃̈(𝑡) + C𝑥

M𝑥 ̃̇(𝑡) + K𝑥̃ (𝑡) = F𝑢 (𝑡) − C𝑥1 − K𝑥0 . (21)

Figure 2: Projection onto K𝑛 .

Right-hand side of (21) is enriched with the constant term

−C𝑥1 − K𝑥0 , and we may consider this term as additional

force. Then we may assume initial conditions (15) without loss where

of generality in form of

M𝑛 = V𝑇𝑛 MV𝑛 ,

𝑥 (0) = 0, 𝑥̇(0) = 0. (22)

This kind of treatment of a nonzero IC was introduced in [46]. K𝑛 = V𝑇𝑛 KV𝑛 ,

(30)

C𝑛 = V𝑇𝑛 CV𝑛 ,

3.2.2. Krylov Subspace Based MOR. The Laplace transform of

(15) has the form of F𝑛 = V𝑇𝑛 F,

−1

H (𝑠) = L𝑇 (𝑠2 M + 𝑠D + K) F. (23) where M𝑛 , Cn , and K𝑛 ∈ 𝑅𝑛×𝑛 , F ∈ 𝑅𝑛 , and L𝑛 ∈ 𝑅𝑛×𝑚 .

The transfer function of the reduced system (30) has the

And the McLaurin series of transfer function (23) has the

form

form

−1

∞ H𝑛 (𝑠) = L𝑇𝑛 (𝑠C𝑛 + K𝑛 ) Q𝑛 . (31)

H (𝑠) = ∑𝑚𝑙 𝑠𝑙 , (24)

𝑖=0 The above procedure assures that the first 𝑛 moments of the

where 𝑚𝑙 are the so-called moments of the transfer function transfer function (23) of the full system equal the first 𝑛

moments of the transfer function (31) of the reduced system

𝑚𝑙 = L𝑇 𝑟𝑙 , [26].

The error induced by the projection (28) in the output

𝑟0 = K−1 F, function 𝑦(𝑡) has the form

(25)

𝑟1 = −K−1 D𝑟0 , 𝜖 = max 𝑦 (𝑡) − 𝑦̃ (𝑡) . (32)

𝑡>0

𝑟𝑙 = −K−1 (D𝑟𝑙−1 − M𝑟𝑙−2 ) . An a priori expression for error norm (32) is not known

although there exist algorithms minimizing the error [44, 47].

The first 𝑛 vectors 𝑟𝑙 span Krylov space

The algorithm used in this paper to produce the reduced

K𝑛 = span (𝑟0 , . . . , 𝑟𝑛−1 ) . (26) order systems is the block Arnoldi algorithm [45].

There exist wide possibilities to improve the computa-

Letting V𝑛 be the orthonormal basis of K𝑛 tional performance of Krylov methods. One of the most

obvious options is parallelization [48]. Another is to use an

K𝑛 = span (V𝑛 ) , V𝑇𝑛 V𝑛 = I, V𝑛 ∈ 𝑅𝑁×𝑛 . (27)

iterative algorithm to solve the system [49]. The presented

The projection of state coordinates 𝑥 onto K𝑛 using V𝑛 is case is of medium size, and it is therefore suitable to use the

called generalized state coordinates 𝑞 ∈ 𝑅𝑛 direct sparse solver [50].

The procedure is easily extended to a multi-input/multi-

𝑥 = V𝑞 + 𝜖. (28) output case where Q, F ∈ 𝑅𝑁×𝑙 and L ∈ 𝑅𝑁×𝑚 . The size of

𝑛

The error 𝜖 ∈ 𝑅 in the projection rises while performing the reduced system is determined by the size of Q and L.

projection of 𝑥 onto K𝑛 (Figure 2). However, it is possible to use the superposition property [51]

We obtain reduced system of (29) by substituting gener- to keep the matrices small.

alized coordinates 𝑞 into (15) and using the Galerkin method.

The reduced equations have the form of 4. Machine Tool Model for the Case Study

M𝑛 𝑞 ̈(𝑡) + C𝑛 𝑞 ̇(𝑡) + K𝑛 𝑞 (𝑡) = F𝑛 𝑢 (𝑡) , The dynamic performance of a machine tool may be evalu-

(29) ated at the TCP using the oriented directional FRFs, which

𝑦̃ (𝑡) = L𝑇𝑛 𝑞 (𝑡) , directly relate according to stability theory to achievable

6 Advances in Mechanical Engineering

Full harmonic — 111 hours 111 hours

Mode truncation 40 min <1 s 40 min

Krylov MOR 6,9 min <1 s 6,9 min

Tool tip the X-slide (face coupling)—or by force interaction in the case

of linear guides (symbols of springs in Figure 4).

Z Z

X Y Y

5. Results

The following case studies compare computational efficiency

Figure 3: Machine tool FE model. of full harmonic in ANSYS software package and model

order reduction using mode truncation and Krylov MOR. To

assess the quality of results obtained using MOR the error

Face coupling norm is evaluated. The study is performed in one kinematic

Face coupling

configuration, and it is assumed that properties of MOR

techniques will be similar in other kinematic configurations.

The model used in the study is a multibody system.

response function at the tip of the tool has been computed

Face coupling using a full FEA model and MOR using mode truncation

Z Face coupling

Y with first 100 modes for each reduced component and Krylov

subspace MOR with first 100 moments. Additionally FRF

computed using modal reduction is provided for comparison.

The comparison is shown in Table 1. The error norm is

Figure 4: Multibody system of the machine tool structure.

concluded to be

𝜖Mode Truncation = max 𝑦 (𝑓) − 𝑦̃ (𝑓) > 4𝑒 − 5,

𝑓∈(0,400)

chip thickness [52–54]. FRF can be obtained by harmonic (33)

𝜖KRYLOV = max 𝑦 (𝑓) − 𝑦̃ (𝑓) < 4𝑒 − 6.

simulation of either a full FE model or reduced models. 𝑓∈(0,400)

The purpose of the study is to perform mathematical

verification of the FRFs generated at the TCP using a full FE The Table 1 shows comparison of methods in terms of time

model, or reduced order models. Modal reduction technique needed to produce the reduced order model, time needed to

and Krylov subspace method are considered. At the same computed FRF, and total time. The time needed to produce

time, the study aims at testing the suitability of reduction coupled reduced components is added to the time required

methods for creating a machine tool multibody model, to compute FRF. The reduced models are computed only

assembled from reduced FE models of separate structural once and then it is possible to couple them in any kinematic

parts. In this way, a model allowing for quick update of configuration. The coupling is computationally very cheap.

machine tool dynamic properties according to actual kine- Figures 5 and 6 show FRFs in axes 𝑥 and 𝑦. The results

matic configuration is created. clearly show the quality of approximation obtained using

A model of a large portal milling machine tool is con- Krylov MOR as well as computational efficiency compared

sidered. The FE model (Figure 3) consists of the volume, to the full solution and modal reduction. The Krylov MOR

shell, spring, matrix, and mass elements. The mesh consists of is about 1000x more computationally efficient than the full

about 1,5⋅106 nodes; the total number of degrees of freedom solution, and the approximation error is almost negligible.

number is almost 6⋅106 . The model was built in ANSYS v 14.5. The mode truncation is computationally less efficient and also

Linear guideways are represented by linear spring elements; the approximation error on frequencies higher than 50 Hz is

similarly the rack and pinion feed drives are modeled using very high rendering this method almost unusable.

linear spring elements, which reflect the closed position

control. 6. Conclusions

Figure 4 shows the decomposition of the FE model into

separate structural parts with a description of the type of In the paper a novel application of Krylov subspace based

coupling chosen according to the character of connections model order reduction method for coupling of compliant

between the parts. The parts are connected either through the components, modelled as FE bodies, has been proposed and

faces—typically, for example, connection of the column with successfully tested. To the best of the authors’ knowledge,

Advances in Mechanical Engineering 7

1.2

will improve the quality of tool oscillation simulation and

1.0 will help increasing the accuracy of machined surface quality

Compliance (mm/N)

0.8 requirements of the Krylov subspace reduction technique

set also a good prerequisite for possible future real-time

0.6 applications of machine tool dynamic virtual model as an

0.4

observer in a real CNC control.

0 50 100 150 200 250 300 350 400 The authors declare that there is no conflict of interests

Frequency (Hz) regarding the publication of this paper.

Full FE model

Modal reduction Acknowledgment

Krylov MOR

This research has been supported by the Competence Center-

Figure 5: Comparison of FRF, 𝑋-axis. Manufacturing Technology Project TE01020075 funded by

the Technology Agency of the Czech Republic.

×10−5

8 References

7

[1] Y. Altintas, C. Brecher, M. Weck, and S. Witt, “Virtual machine

Compliance (mm/N)

5 pp. 115–138, 2005.

4 [2] Y. Altintas, “Virtual machining system,” in Proceedings of the

4th International Conference and Exhibition on Design and

3 Production of Machines and Dies/Molds, Cesme, Turkey, 2007.

2 [3] M. F. Zaeh, M. Hennauer, and A. Poehler, “The machine tool

1 of tomorrow: an optimised, complete system,” in Proceedings of

the 6th International Conference on High Speed Machining, San

0 50 100 150 200 250 300 350 400 Sebastian, Spain, 2007.

Frequency (Hz) [4] G. Pritschow and N. Croon, “Wege zur virtuellen werkzeugmas-

chine,” Wt Werkstattstechnik, vol. 92, no. 5, pp. 194–199, 2002.

Full FE model

[5] J. Veselý, M. Sulitka, and J. Hornych, “Simulated impact of

Modal reduction

Krylov MOR

machine tool CNC system and feed drive control on workpiece

quality,” in Proceedings of the 2nd International conference

Figure 6: Comparison of FRF, 𝑌-axis. Process Machine Interactions, Vancouver, Canada, 2010.

[6] M. Zatarain, E. Lejardi, and F. Egaña, “Modular synthesis of

machine tools,” CIRP Annals—Manufacturing Technology, vol.

47, no. 1, pp. 333–336, 1998.

coupling of arbitrary substructures reduced using Krylov

[7] P. De Fonseca, D. Vandepitte, H. Van Brussel, and P. Sas,

MOR has not yet been reported in the literature. The “Dynamic model reduction of a flexible three-axis milling

proposed method has been tested on a dynamic simulation machine,” in Proceedings of the International Seminar on Modal

of multibody system of the machine tool structure. Analysis (ISMA ’98), pp. 1197–1206, Leuven, Belgium, Septem-

The case study has shown superior quality of the Krylov ber 1998.

subspace based MOR in comparison to mode truncation [8] R. Maj and G. Bianchi, “Mechatronic analysis of machine tools,”

technique for creating the machine tool multibody model, in Proceedings of the 9th SAMTECH Users Conference, Paris,

in which each of the structural parts is reduced separately France, February 2005.

and coupled to each other using either faced, or force cou- [9] M. Sulitka and P. Strakoš, “Complex model of a real machine

pling. The multibody model employing the Krylov subspace tool feed drive axis with ball screw,” in Proceedings of the

base MOR features almost identical match of the dynamic Conference VIDA, Poznan, Poland, 2007.

properties with the full FE model. At the same time, Krylov [10] J. Vesely and M. Sulitka, “Machine tool virtual model,” in

subspace method is very time efficient and proves to provide Proceedings of the International Conference on Machine Tools,

a very effective tool for quick model updates related to vary- Automation, Robotics and Technology (MATAR ’08), Prague,

ing structural dynamics with respect to various kinematic Czech Republic, 2008.

configurations or testing of different design variants in the [11] J. J. Zulaika, F. J. Campa, and L. N. Lopez de Lacalle, “An

machine tool development. The Krylov subspace will be next integrated processmachine approach for designing productive

used for creating a multibody model with movable motion and lightweight milling machines,” International Journal of

8 Advances in Mechanical Engineering

Machine Tools and Manufacture, vol. 51, no. 7-8, pp. 591–604, [29] E. Grimme, Krylov projection methods for model reduction

2011. [Ph.D. thesis], ECE Department, University of Illinois, Urbana-

[12] M. Zaeh and D. Siedl, “A new method for simulation of machin- Champaign, Ill, USA, 1997.

ing performance by integrating finite element and multi-body [30] E. Grimme and K. Gallivan, “Krylov projection methods

simulation for machine tools,” CIRP Annals—Manufacturing for rational interpolation,” 1997 http://citeseerx.ist.psu.edu/

Technology, vol. 56, no. 1, pp. 383–386, 2007. viewdoc/download?doi=10.1.1.54.5011&rep=rep1&type=pdf.

[13] R. Neugebauer, M. Kolouch, M. Richter, V. Wittstock, and S. [31] Z. Bai, “Krylov subspace techniques for reduced-order mod-

N. Saha, “Model updating of machine tools,” in Proceedings of eling of large-scale dynamical systems,” Applied Numerical

the 9th International Conference on Machine Tools, Automation, Mathematics, vol. 43, no. 1-2, pp. 9–44, 2002.

Robotics and Technology (MATAR ’12), Prague, Czech Republic,

[32] A. Odabasioglu, M. Celik, and L. T. Pileggi, “PRIMA: pas-

September 2012, (MM Science Journal 2012, Special Issue).

sive reduced-order interconnect macromodeling algorithm,”

[14] R. J. Guyan, “Reduction of stiffness and mass matrices,” AIAA in Proceedings of the IEEE/ACM International Conference on

Journal, vol. 3, p. 380, 1965. Computer-Aided Design (ICCAD ’97), pp. 58–65, Washington,

[15] B. M. Irons, “Structural eigenvalue problems: elimination of DC, USA, November 1997.

unwanted variables,” AIAA Journal, vol. 3, no. 5, pp. 961–962,

[33] Z. Bai, K. Meerbergen, and Y. Su, “Arnoldi methods for

1965.

structure-preserving dimension reduction of second-order

[16] L. B. Bushard, “On the value of Guyan reduction in dynamic dynamical systems,” in Dimension Reduction of Large-Scale

thermal problems,” Computers & Structures, vol. 13, no. 4, pp. Systems, P. Benner, D. Sorensen, and V. Mehrmann, Eds., vol.

525–531, 1981. 45 of Lecture Notes in Computational Science and Engineering,

[17] P. Koutsovasilis and M. Beitelschmidt, “Comparison of model pp. 173–189, Springer, Berlin, Germany, 2005.

reduction techniques for large mechanical systems,” Multibody

[34] L. H. Feng, E. B. Rudnyi, J. G. Korvink, C. Bohm, and T. Hauck,

System Dynamics, vol. 20, no. 2, pp. 111–128, 2008.

“Compact electro-thermal model of semiconductor device with

[18] M. Bampton and J. R. Craig, “Coupling of substructures for nonlinear convection coefficient,” in Proceedings of the 6th

dynamic analyses,” AIAA Journal, vol. 6, no. 7, pp. 1313–1319, International Conference on Thermal, Mechanical and Multi-

1968. Physics Simulation and Experiments in Micro-Electronics and

[19] Y. Aoyama and G. Yagawa, “Component mode synthesis for Micro-Systems (EuroSimE ’05), pp. 372–375, Berlin, Germany,

large-scale structural eigenanalysis,” Computers & Structures, April 2005.

vol. 79, no. 6, pp. 605–615, 2001.

[35] E. Zukowski, J. Wilde, E.-B. Rudnyi, and J.-G. Korvink, “Model

[20] D. Botto, S. Zucca, M. Gola, E. Troncarelli, and G. Pasquero, reduction for thermo-mechanical simulation of packages,” in

“Component modes synthesis applied to a thermal transient Proceedings of the 11th International Workshop on Thermal

analysis of a turbine disc,” in Proceedings of the 3rd Worldwide Investigations of ICs and Systems (THERMINIC ’05), pp. 134–

Aerospace Conference & Technology Showcase, Toulouse, France, 138, Belgirate, Italy, 2006.

2002.

[36] R. S. Puri, D. Morrey, A. J. Bell, J. F. Durodola, E. B. Rudnyi, and

[21] D. Botto, S. Zucca, and M. M. Gola, “Reduced-order models

J. G. Korvink, “Reduced order fully coupled structural-acoustic

for the calculation of thermal transients of heat conduc-

analysis via implicit moment matching,” Applied Mathematical

tion/convection FE models,” Journal of Thermal Stresses, vol. 30,

Modelling, vol. 33, no. 11, pp. 4097–4119, 2009.

no. 8, pp. 819–839, 2007.

[37] J. S. Han, E. B. Rudnyi, and J. G. Korvink, “Efficient optimization

[22] P. Nachtergaele, D. J. Rixen, and A. M. Steenhoek, “Efficient

of transient dynamic problems in MEMS devices using model

weakly coupled projection basis for the reduction of thermo-

order reduction,” Journal of Micromechanics and Microengineer-

mechanical models,” Journal of Computational and Applied

ing, vol. 15, no. 4, pp. 822–832, 2005.

Mathematics, vol. 234, no. 7, pp. 2272–2278, 2010.

[23] Z. Bai and B.-S. Liao, “Towards an optimal substructuring [38] J. S. Han, “Efficient frequency response and its direct sensitivity

method for model reduction,” in Applied Parallel Computing. analyses for large-size finite element models using Krylov

State of the Art in Scientific Computing, J. Dongarra, K. Madsen, subspace-based model order reduction,” Journal of Mechanical

and J. Waśniewski, Eds., vol. 3732 of Lecture Notes in Computer Science and Technology, vol. 26, no. 4, pp. 1115–1126, 2012.

Science, pp. 276–285, Springer, Berlin, Germany, 2006. [39] U. Baur, C. Beattie, P. Benner, and S. Gugercin, “Interpolatory

[24] J. O’Callahan, “A procedure for an improved reduced system projection methods for parameterized model reduction,” SIAM

(IRS) model,” in Proceedings of the 7th International Modal Journal on Scientific Computing, vol. 33, no. 5, pp. 2489–2518,

Analysis Conference (IMAC ’89), Las Vegas, Nev, USA, 1989. 2011.

[25] M. I. Friswell, S. D. Garvey, and J. E. T. Penny, “Model reduction [40] J. Šindler, A. Suleng, T. Jelstad Olsen, and P. Bárta, “Krylov

using dynamic and iterated IRS techniques,” Journal of Sound model order reduction of a thermal Subsea model,” World

and Vibration, vol. 186, no. 2, pp. 311–323, 1995. Academy of Science, Engineering and Technology, vol. 77, pp. 217–

[26] A. C. Antoulas, Approximation of Large-Scale Dynamical Sys- 224, 2013.

tems, Advances in Design and Control, SIAM, Philadelphia, Pa, [41] W. Witteveen, “Comparison of CMS, Krylov and balanced trun-

USA, 2005. cation based model reduction from a mechanical application

[27] A. C. Antoulas and D. C. Sorensen, “Approximation of large- engineer’s perspective,” in Topics in Experimental Dynamics

scale dynamical systems: an overview,” International Journal of Substructuring and Wind Turbine Dynamics, R. Mayes, D.

Applied Mathematics and Computer Science, vol. 11, no. 5, pp. Rixen, D. Griffith et al., Eds., vol. 2 of Conference Proceedings

1093–1121, 2001. of the Society for Experimental Mechanics Series, pp. 319–331,

[28] T. Stykel, Analysis and numerical solution of generalized Lya- Springer, New York, NY, USA, 2012.

punov equations [Ph.D. thesis], Technical University of Berlin, [42] Z. Qu, Model Order Reduction Techniques: With Applications in

Berlin, Germany, 2002. Finite Element Analysis, Springer, New York, NY, USA, 2004.

Advances in Mechanical Engineering 9

framework for dynamic substructuring: history, review, and

classification of techniques,” AIAA Journal, vol. 46, no. 5, pp.

1169–1181, 2008.

[44] S. Gugercin, A. C. Antoulas, and C. Beattie, “H2 model reduc-

tion for large-scale linear dynamical systems,” SIAM Journal on

Matrix Analysis and Applications, vol. 30, no. 2, pp. 609–638,

2008.

[45] I. M. Elfadel and D. D. Ling, “A block rational Arnoldi algorithm

for multipoint passive model-order reduction of multiport

RLC networks,” in Proceedings of the IEEE/ACM International

Conference on Computer-Aided Design (ICCAD ’97), pp. 66–71,

Washington, DC, USA, November 1997.

[46] L. F. L. Feng, D. Koziol, E. Rudnyi, and J. Korvink, “Model

order reduction for scanning electrochemical microscope: the

treatment of nonzero initial condition,” in Proceedings of IEEE

Sensors, vol. 3, pp. 1236–1239, 2004.

[47] G. Flagg, C. Beattie, and S. Gugercin, “Convergence of the

iterative rational Krylov algorithm,” Systems & Control Letters,

vol. 61, no. 6, pp. 688–691, 2012.

[48] E. F. Yetkin and H. Daǧ, “Parallel implementation of iterative

rational Krylov methods for model order reduction,” in Pro-

ceedings of the 5th International Conference on Soft Computing,

Computing with Words and Perceptions in System Analysis, Deci-

sion and Control (ICSCCW ’09), Famagusta, Cyprus, September

2009.

[49] C. A. Beattie and S. Gugercin, “Inexact solves in Krylov-based

model reduction,” in Proceedings of the 45th IEEE Conference

on Decision and Control (CDC ’06), pp. 3405–3411, San Diego,

Calif, USA, December 2006.

[50] T. Davis, Direct Methods for Sparse Linear Systems, Fundamen-

tals of Algorithms, SIAM, Philadelphia, Pa, USA, 2006.

[51] P. Benner, L. Feng, and E. B. Rudnyi, “Using the superposition

property for model reduction of linear systems with a large

number of inputs,” in Proceedings of the 18th International

Symposium on Mathematical Theory of Networks and Systems

(MTNS ’08), pp. 1–12, Blacksburg, Va, USA, 2008.

[52] Y. Altintas, Manufacturing Automation: Metal Cutting Mechan-

ics, Machine Tool Vibrations, and CNC Design, Cambridge

University Press, New York, NY, USA, 2000.

[53] Y. Altintas and E. Budak, “Analytical prediction of stability lobes

in milling,” CIRP Annals—Manufacturing Technology, vol. 44,

no. 1, pp. 357–362, 1995.

[54] S. D. Merdol and Y. Altintas, “Multi frequency solution of

chatter stability for low immersion milling,” Journal of Manu-

facturing Science and Engineering, vol. 126, no. 3, pp. 459–466,

2004.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 308049, 9 pages

http://dx.doi.org/10.1155/2014/308049

Research Article

Study on Launch Dynamics of Self-Propelled Artillery Based on

Transfer Matrix Method of Multibody System

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Copyright © 2014 H. Yu and X. Rui. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Due to the importance of increasing the firing precision of self-propelled artillery system corresponding to the projectile impact

point, the launch dynamics of self-propelled artillery based on the transfer matrix method for multibody system (MSTMM) is

studied in this paper. By using MSTMM to study launch dynamics of self-propelled artillery, there is no need of global dynamical

equation of the system, the order of the system matrix is low, the computational speed is high, and it meets the requirements

of dynamics of self-propelled artillery multibody system. The dynamical model of multibody system and its topological figure,

the overall transfer equation, overall transfer matrix, characteristics equation and generalized coordinates equation, and launch

dynamical equation of self-propelled artillery are developed. The vibration characteristics, initial disturbance of projectile,

dynamical response, and firing dispersion for self-propelled artillery are simulated. The results of eigenfrequency and the time

history of system dynamics got by simulations and experiments have good agreements. The study results can be used to improve

the firing precision and design of self-propelled artillery.

of mechanical system are mainly finite element method,

The research results of launch dynamics provided the the- modal analysis method, and the structural modal syn-

oretical basis and technical means for design and test of thesis method. The multibody system dynamics methods,

various weapon systems [1–6]. The movements of projectile such as Wittenburg method, Schiehlen method, and Kane

and artillery are very complicated because of the complex method have been widely accepted and applied by engi-

mechanical structure of self-propelled artillery and the severe neer of mechanical system dynamics [7–9]. Finite element

mechanical environments, such as high temperature, high method and multibody system dynamics method have been

pressure, high speed, instantaneous state, multibody, and an important basis of weapon dynamics. However, when

mutation in launch process. Because of their great influence calculating natural vibration characteristics of weapons for

on firing precision of self-propelled artillery, the vibration complex multirigid-flexible body system containing rigid and

characteristics and dynamical response of the system are paid flexible bodies using these methods, it faces the difficulties

great attention in the study of launch dynamics. To increase that computational scale is large and it is easy to result in com-

firing precision of self-propelled artillery, it is necessary to putation singularity caused by computation ill-condition. To

compute accurately the frequencies distribution and establish meet the requirement of accurate dynamics modeling of

the quantitative relationship among the global structure launch dynamics and fast calculation, Rui et al. presented

parameters of the system and its vibration characteristics, transfer matrix method for multibody systems [2, 10–20].

dynamical response, and firing precision. By adjusting the Due to the important feature that order of the overall transfer

structure parameters to design system vibration frequency matrix is very low, MSTMM avoids eigenvalue computation

distribution, it is then able to improve the firing precision for ill-condition for complex multi-rigid-flexible body system,

self-propelled artillery system. which significantly improves the computational speed of

2 Advances in Mechanical Engineering

vibration characteristics and is successfully applied to many Taking self-propelled artillery as an example, according

engineering design and test of various types of vehicles, to its system structure, from bottom to top in sequence, it can

airborne, and ship-borne weapons [1, 2, 4]. be divided into road wheels, hull, and revolving part which

In this paper, by using MSTMM and its automatic does not contain elevating part, elevating part which does

deduction method [2, 20], the dynamical model of multibody not contain recoil part, gun breech, and muzzle brake. Each

system for self-propelled artillery is developed; the overall component can be regarded as rigid body and elastic beam

transfer equation, overall transfer matrix, and characteristics according to its natural attribute. Rigid body and elastic beam

equation are deduced. Generalized coordinates equation of are called “body,” the connection between “body” and “body”

self-propelled artillery is obtained by using the orthogonality is called the “hinge,” which can be seen in [2].

of augmented eigenvector and body dynamical equation. The ground that supports the self-propelled artillery is

Combining with launch dynamical equation of projectile, regarded as an infinity rigid body, whose sequence number

launch dynamical equation of self-propelled artillery system is 0. The elastic and damping effect of each road wheel

is developed. The vibration characteristics, initial disturbance and the interaction between ground and each road wheel

of projectile, dynamical response, and firing dispersion are are, respectively, modeled as springs, rotary springs, and

simulated, and the factors of different ground conditions the accompanying dampers connected in parallel, which can

and connection stiffness among various components, which represent relative linear motion and relative angular motion

influence the firing precision and vibration characteristics, in 3 directions at the same time; their sequence numbers are

are analyzed. It provides a theoretical basis and simulation 42, 43, . . . , 53. Each road wheel can be regarded as a rigid

tool for improving the firing precision by adjusting structural body, whose sequence number is 30, 31, . . . , 41. The inter-

parameters of the system to change the vibration characteris- action between each road wheel and hull are, respectively,

tics of the self-propelled artillery. modeled as springs, rotary springs, and the accompanying

dampers connected in parallel, which can represent relative

2. Dynamical Model of Self-Propelled Artillery linear motion and relative angular motion in 3 directions

Multibody System at the same time; their sequence numbers are 18, 19, . . . , 29.

Hull, revolving part, elevating part, gun breech, and muzzle

Main components of self-propelled artillery are muzzle brake, brake can be regarded as a rigid body, respectively, whose

barrel, gun breech, recoil and counter-recoil mechanisms, sequence numbers are 17, 15, 13, 10, and 1. The effect of

cradle, elevating mechanism, equilibrator, turret, travers- traversing mechanism associating the elastic, and damping

ing mechanism, chassis, torsion bar, balance elbow, shock effects of hull, the effect of elevating mechanism and equi-

absorber, track chain, road wheel, and so on. According to the librator associating the elastic and damping effects between

motion state of each component, the firepower system of self- revolving part and elevating part are, respectively, modeled as

propelled artillery can be divided into recoil part, elevating springs and rotary springs accompanying dampers which can

part, revolving part, suspension part, walking part, and so represent relative linear motion and relative angular motion

on. The recoil part contains muzzle brake, barrel, gun breech, in 3 directions at the same time, whose sequence numbers

and recoil and counter-recoil mechanisms. The elevating are 16, 14. The interaction between barrel and elevating

part contains total recoil part, cradle, and the components par, is, respectively, modeled as springs and rotary springs

moving with cradle, which includes elevating mechanism, accompanying dampers which can represent relative linear

equilibrator, and so on. The revolving part contains total motion and relative angular motion in 3 directions at the

elevating part, turret, and the elements moving with turret same time, whose sequence numbers are 11, 12. According

that include traversing mechanism and so on. The walking to its structure characteristics, the barrel is divided into 6

part is used to support the weight of the self-propelled segments, each segment can be regarded as a beam with

artillery and drive self-propelled artillery to run placidly, equal sectional area, whose sequence numbers are 2, 3, 4, 5, 7,

which contains track chain and road wheels. The suspension and 9. The connection points among joints and beam can be

part is used to connect chassis to walking part, which contains regarded as massless rigid body, whose sequence numbers are

torsion bar, balance elbow, and shock absorber. The self- 6, 8. The dynamical model of self-propelled artillery multi-

propelled artillery is shown in Figure 1. rigid-flexible system is composed of 19 rigid bodies, 6 beams,

Advances in Mechanical Engineering 3

and 28 joints, as shown in Figure 2. The topology figure of where, 1 is the sequence number of body and 0 denotes the

dynamical model of the self-propelled system is shown in boundary.

Figure 3. The form of Z10,0 , Z11,13 , Z13,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 ,

Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , and Z53,0 is similar

3. Overall Transfer Equation of to Z1,0 .

Self-Propelled Artillery

3.2. Overall Transfer Equation of Self-Propelled Artillery Sys-

3.1. The State Vector of Self-Propelled Artillery. According to tem. According to MSTMM and the topology figure of

the dynamical model and its topology figure, the state vectors the dynamical model, the overall transfer equation of self-

at boundary points for self-propelled artillery are defined as propelled artillery system is automatically deduced as follows:

follows

𝑇 Uall Zall = 0, (2)

Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]1,0 ,

(1) where overall transfer matrix

−I12 T10−1 T42−1 T43−1 T44−1 T45−1 T46−1 T47−1 T48−1 T49−1 T50−1 T51−1 T52−1 T53−1 T13−1 + T11−1

[006×12 06×12 G42−17 G43−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]

[ 6×12 06×12 G42−17 06×12 G44−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]

[06×12 06×12 G42−17 06×12 06×12 G45−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]

[06×12 06×12 G42−17 06×12 06×12 06×12 G46−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]

[ ]

[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 G47−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]

[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 G48−17 06×12 06×12 06×12 06×12 06×12 06×12 ]

[ ]

Uall = [06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 G49−17 06×12 06×12 06×12 06×12 06×12 ].

[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G50−17 06×12 06×12 06×12 06×12 ]

[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G51−17 06×12 06×12 06×12 ]

[0 ]

[ 6×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G52−17 06×12 06×12 ]

[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G53−17 06×12 ]

[0 06×12 G42−13 G43−13 G44−13 G45−13 G46−13 G47−13 G48−13 G49−13 G50−13 G51−13 G52−13 G53−13 G13,𝐼2 ]

[ 6×12 ]

06×12 G10−8 G42−8 G43−8 G44−8 G45−8 G46−8 G47−8 G48−8 G49−8 G50−8 G51−8 G8−52,𝑏12 G53−8 G13−8

[06×12 G10−6 G42−6 G43−6 G44−6 G45−6 G46−6 G47−6 G48−6 G49−6 G50−6 G51−6 G52−6 G53−6 G13−6 + G11−6 ]

(3)

The overall state vector of all boundary points is given by Z𝑖,0 (𝑖 = 42, 43, . . . , 53), which contains 6 variables. Z10,0 and

Z1,0 are the state vectors at the boundary points of gun breech

and muzzle. Each of them contains 12 variables, including

displacements, angle displacements, forces, and moments in 3

Zall = [Z1,0 , Z10,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , directions, respectively. Both gun breech and muzzle are free

boundary, therefore, the 6 variables representing the forces

Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , Z53,0 , Z13,0 ] .

and moments are always equal to 0. In the same way, let the

(4) symbol Z10,0 and Z1,0 represent the remnant after these zeros

are removed from Z10,0 and Z1,0 , respectively, and each of

The elements of the overall transfer matrix Uall are deter- them has 6 variables. In this case, there are 84 variables always

mined in appendix. equal to 0 in Zall ; eliminating them from Zall , a new state

vector Zall can be obtained, which has 96 variables, given by

4. Characteristics Equation of Self-Propelled

Artillery System Zall = [Z1,0 , Z10,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 ,

(5)

In (4), Zall is composed of the state vectors Z1,0 , Z10,0 , Z13,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , Z53,0 , Z13,0 ] .

Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 ,

Z52,0 , and Z53,0 at the system boundary points. For the state Deleting the columns 7∼12, 19∼24, 25∼30, 37∼42, 49∼54, 61∼

vectors at the boundary points, half of its elements have 66, 73∼78, 85∼90, 97∼102, 109∼114, 121∼126, 133∼138, 145∼150,

been identified using boundary conditions. For self-propelled 157∼162 in Uall , a new matrix Uall with 96 orders can be

artillery system, each state vector Z𝑖,0 (𝑖 = 42, 43, . . . , 53) obtained; (2) can be written as

between 12 road wheels and the ground contains 12 variables,

including displacements, angle displacements, forces, and Uall Zall = 0. (6)

96×96 96×1

moments in 3 directions respectively, in which the 6 vari-

ables representing the displacements and angle displacements The components of Uall or Uall are only decided by parameters

are always equal to 0; therefore, note symbol Z𝑖,0 (𝑖 = and eigenfrequency of the system. When the system parame-

42, 43, . . . , 53) as the state vector that only includes the ters are determined, (6) should have nonzero solution for any

4 Advances in Mechanical Engineering

Elevating part 13

Projectile

Gun breech 10 14 11

12 Barrel 2∼5 Muzzle brake 1

0

16

18 19 20 21 22 23

30 31 32 33 34 35

42 43 44 45 46 47 Road wheel 30∼41

0 0

48 49 50 51 52 53

36 37 38 39 40 41

9 7 5-2

Z10.0 10 I1 8 I1 6 1 Z1.0

24 25 26 27 28 29 I2 I2

I7 I8 I9 I10 I11 I12 12 11

Z11.13

17 15 I1 13 Z13.0

16 14 I2

I1 I2 I3 I4 I5 I6

18 19 20 21 22 23

30 31 32 33 34 35

42 43 44 45 46 47

eigenvalue of the system. Hence, the system characteristics an arbitrary point on beam (barrel) can be obtained. Thus,

equation is obtained by the vibration characteristics of self-propelled artillery are

obtained, which include the eigenfrequency 𝜔𝑘 (𝑘 = 1, 2, . . .)

det (Uall ) = 0. (7) and the mode shape corresponding to each 𝜔𝑘 , and the datum

of mode shape is included in the variables of all state vec-

By solving (7), the eigenfrequency of self-propelled artillery, tors of system, so that the mode shape can be got after the

𝜔𝑘 (𝑘 = 1, 2, . . .), can be obtained. Solving (6), the state vector corresponding variables selected them from the state vec-

𝑘

Zall corresponding to 𝜔𝑘 can be obtained, in other words, Z𝑘all tors.

is obtained. That is to say, the state vectors Z1,0 , Z10,0 , Z13,0 , It can be seen from (6) that the order of the overall

Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , transfer matrix of the self-propelled artillery system is only

Z52,0 , and Z53,0 corresponding to 𝜔𝑘 are obtained. Based on 96 and is much lower than that for other multibody system

these state vectors at the system boundaries, through using dynamics methods so computational speed is high and the

the transfer equation of each element one by one, all the computational ill-condition caused by high matrix order and

state vectors in system including each connection points and large stiffness gradient is avoided.

Advances in Mechanical Engineering 5

5. Launch Dynamical Equation of 5.3. Launch Dynamical Equation of the Projectile. Launch

Self-Propelled Artillery dynamical equation of a projectile is given by [1, 5]

Body dynamical equation of self-propelled artillery can be 𝑎𝑝 = − 2

𝑚𝜑3 𝜕𝑡

written as

M𝑗 k𝑗,𝑡𝑡 + C𝑗 k𝑗,𝑡 + K𝑗 k𝑗 = f𝑗 ̈ = − 𝑔 cos 𝜃1

𝑦𝑜𝑐

(8)

(𝑗 = 1, 2, 3, 4, 5, 7, 9, 10, 13, 15, 17, 30, 31, . . . , 41) , 𝐾

𝐹𝑦𝑠𝑓

− [𝑦𝑜𝑜 − 𝜇𝑧𝑜𝑜 sin 𝛼] + − 𝑦𝑜̈,

𝑚 𝑚

where 𝑗 is the sequence number of all body elements of

the self-propelled artillery and M𝑗 , C𝑗 , K𝑗 , k𝑗 , and f𝑗 are,

respectively, named as parameter matrix of mass, parameter −𝐾 𝐹𝑠𝑓

matrix of damper, parameter matrix of stiffness, displacement 𝑧̈

𝑜𝑐 = [𝑧𝑜𝑜 + 𝜇𝑦𝑜𝑜 sin 𝛼] + 𝑧 − 𝑧̈

𝑜,

𝑚 𝑚

matrix, and external force matrix, which can be seen in [2].

𝐶 𝐶

𝛿1𝐼̈ = − 𝐼

𝛾̇(𝜓̇ 𝐼̇

2 + 𝛿2 ) + (1 − ) (𝛾2̇𝛽𝐷𝜂 + 𝛾𝛽

̈𝐷𝜉 )

5.2. Dynamic Response of Self-Propelled Artillery System. The 𝐴 𝐴

dynamic response in the physical coordinate system may be 𝐶𝛾̈ 𝐼 𝑃𝑏 𝑆𝑏

− 𝛿 + 𝑦

expanded using augmented eigenvectors V𝑘1 up to a selected 𝐴 2 𝐴𝜑3 𝑜𝑐

modal order 𝑛,

𝐾ℎ2 𝐼 𝐾 (𝑙𝑅 + 𝑟𝑏 𝜇)

+∞

𝑘 − (𝛿 − 𝛿2𝐼 𝜇 sin 𝛼) + 𝑦𝑜𝑜

k𝑗 = ∑ V𝑗 1 𝑞𝑘1 (𝑡) . (9) 12𝐴 1 𝐴

𝑘1 =1

𝐾𝑙𝑅 𝜇 sin 𝛼 𝑙

− 𝑧𝑜𝑜 + 1 𝐹𝑦𝑠𝑓 − 𝜓̈

𝐼

1,

Substituting (9) into (8), thus we obtain 𝐴 𝐴

+∞ +∞ 𝐶 𝐶

𝑘

∑ (M𝑗 V𝑗 1 ) 𝑞𝑘̈1 (𝑡) + ∑ (C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡)

𝑘 𝛿2𝐼̈ = 𝐼

𝛾̇(𝜓̇ 𝐼̇

1 + 𝛿1 ) + (1 − ) (𝛾2̇𝛽𝐷𝜉 − 𝛾𝛽

̈𝐷𝜂 )

𝐴 𝐴

𝑘1 =1 𝑘1 =1

(10) 𝐶𝛾̈ 𝐼 𝑃𝑏 𝑆𝑏

+∞ + 𝛿 + 𝑧

𝑘 𝐴 1 𝐴𝜑3 𝑜𝑐

+ ∑ (K𝑗 V𝑗 1 ) 𝑞𝑘1 (𝑡) = f𝑗 .

𝑘1 =1

𝐾ℎ2 𝐼 𝐾 (𝑙𝑅 + 𝑟𝑏 𝜇)

− (𝛿2 + 𝛿1𝐼 𝜇 sin 𝛼) + 𝑧𝑜𝑜

Taking the inner product of both sides of (10) with V𝑘𝑗 (𝑘 = 12𝐴 𝐴

1, 2, . . . , 𝑛), thus we obtain 𝐾 𝑙

+ 𝑙 𝜇𝑦 sin 𝛼 + 1 𝐹𝑧𝑠𝑓 − 𝜓̈

𝐼

2,

∑𝑗 ⟨∑+∞

𝑘

(C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡) , V𝑘𝑗 ⟩ 𝐴 𝑅 𝑜𝑜 𝐴

𝑘1 =1

𝑞𝑘̈(𝑡) + + 𝜔𝑘2 𝑞𝑘 (𝑡) 2𝑡𝑔𝛼0 𝑘𝛼 2

𝑑𝑘 {

(11) { 𝑑 𝑥𝑞 + 𝑑 𝑥𝑞

{ (𝑥𝑞 < 𝑙𝛼 ) ,

0

𝛾 = { 2𝑡𝑔𝛼 0

∑𝑗 ⟨f𝑗 , V𝑘𝑗 ⟩ {

{ 𝑔 𝑘𝛼 2

= . 𝑥𝑞 − 𝑙𝛼 (𝑥𝑞 ≥ 𝑙𝛼 ) ,

𝑑𝑘 { 𝑑0 𝑑0

For proportional damping, (16)

C𝑗 = 𝛼M𝑗 + 𝛽K𝑗 . (12) where the specific meaning of all symbols is no longer given,

which can be seen in [1, 5].

There are

+∞

𝑘

∑ ⟨ ∑ (C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡) , V𝑘𝑗 ⟩ = (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) 𝑑𝑘 . (13) 6. Numerical Simulation and

𝑗 𝑘1 =1 Experimental Validation

Generalized coordinate equation of self-propelled artillery is Launch dynamics of self-propelled artillery can be computed

obtained, by combining (14) and (16). Eigenfrequency, vertical target

dispersion, the time history of barrel recoil displacement,

𝑞𝑘̈(𝑡) + (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) + 𝜔𝑘2 𝑞𝑘 (𝑡) = 𝑝𝑘 (𝑡) , (14)

and the time history of muzzle displacement in the plumb

where direction in cement ground for a self-propelled artillery got

by simulation and test are shown, respectively, in Tables 1 and

𝑘

∑𝑗 ⟨f𝑗 , V𝑘𝑗 ⟩ 2 and Figures 4 and 5. The simulation parameters can be seen

𝑝 (𝑡) =

𝑑𝑘 (15) in [1].

It can be seen from Tables 1 and 2 and Figures 4 and 5

(𝑗 = 1, 2, 3, 4, 5, 7, 9, 10, 13, 15, 17, 30, 31, . . . , 41) . that results got by simulation and test have good agreements,

6 Advances in Mechanical Engineering

Mode order 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Simulation 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8 87.5 94.2 101.5 — 141.0 214.2 225.9 243.7

Test — 15.5 18.7 29.9 42.2 49.8 — — 82.1 — 105.5 128.4 136.3 202.9 229.2 244.4

Error (%) −1.3 −1.0 −2.7 4.3 6.0 6.2 −3.8 3.4 5.6 −1.4 −0.3

0 0.6

−0.2

Barrel recoil displacement (m)

0.4

−0.4

0.2

−0.6

0

−0.8

−1 −0.2

0 0.5 1 1.0 2 0 1 2 3

t (s) t (s)

Test Test

Simulation Simulation

Figure 4: The time history of barrel recoil displacement got by Figure 5: The time history of muzzle displacement in plumb

simulation and test. direction got by simulation and test.

simulation and test.

is significantly great and then affects the firing precision of

Vertical target dispersion self-propelled artillery. The simulation results explained this

Vertical dispersion Direction dispersion phenomenon that firing precision of self-propelled artillery is

Test 0.400 0.380 different due to different ground conditions.

Simulation 0.391 0.404 For self-propelled artillery, the vibration characteristics

Error (%) −2.3 6.3 and firing dispersion on the cement ground are simulated by

changing the connection stiffness between hull and revolving

part; eigenfrequencies and firing dispersion got by simulation

which shows that the study on launch dynamics of self- are shown, respectively, in Tables 5 and 6.

propelled artillery by using MSTMM is very effective. For a self-propelled artillery, the vibration characteristic

and firing dispersion on the cement ground are simulated by

changing the connection stiffness between revolving part and

7. Analysis of Factors Influenced Firing elevating part; eigenfrequencies and firing dispersion got by

Precision of Self-Propelled Artillery simulation are shown, respectively, in Tables 7 and 8.

It can be seen from Tables 5–8 that the connection

Many shooting tests have shown that the difference of firing

stiffness between hull and revolving part, revolving part, and

dispersion is big due to different ground conditions and

elevating part has great effect on vibration characteristic and

different connection stiffness among the various components.

firing dispersion of self-propelled artillery. The factors, such

In fact, different ground conditions mean that connection

as different ground hardness, connection stiffness among

stiffness between road wheel and ground is different; the

various components, have a great influence on vibration

different ground hardness can be simulated by changing

characteristics, and then greatly influence firing precision of

connection stiffness between road wheel and ground. The

self-propelled artillery.

vibration characteristics and firing dispersion on the different

ground are simulated for self-propelled artillery. Eigenfre-

quencies and firing dispersion of self-propelled artillery got 8. Conclusion

by simulation are shown, respectively, in Tables 3 and 4.

It can be seen from Tables 3 and 4 that the effect of differ- In this paper, the launch dynamics of self-propelled artillery is

ent ground conditions on the natural vibration characteristics studied using MSTMM and the automatic deduction method

Advances in Mechanical Engineering 7

Mode order 1 2 3 4 5 6 7 8

Cement

Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8

Mode order 1 2 3 4 5 6 7 8

Soil

Eigenfrequencies 3.0 7.7 9.2 12.0 15.2 21.1 30.2 31.2

ground got by simulation.

× U8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,

Vertical target dispersion

Different ground conditions Vertical Direction T45−1 = U1 U2 U3 U4 U5 U6 U7

dispersion/m dispersion/m

× U8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 ,

Cement 0.411 0.367

Soil 0.283 0.393 T46−1 = U1 U2 U3 U4 U5 U6 U7

× U8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,

propelled artillery multibody system and its topology figure × U8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,

are developed. According to MSTMM, dynamical model of

self-propelled artillery and its topology figure, overall transfer T48−1 = U1 U2 U3 U4 U5 U6 U7

equation, characteristics equation, and launch dynamical

equations of self-propelled artillery are developed. By solving × U8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,

launch dynamical equations of self-propelled artillery, eigen- T49−1 = U1 U2 U3 U4 U5 U6 U7 U8,12

frequency, firing dispersion, and dynamical response of self-

propelled artillery are obtained. Results got by simulation × U8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,

and experiment have good agreements, which show that

MSTMM and the automatic deduction method of overall T50−1 = U1 U2 U3 U4 U5 U6 U7

transfer equation are useful to study the launch dynamics of × U8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,

self-propelled artillery.

Different ground conditions and connection stiffness T51−1 = U1 U2 U3 U4 U5 U6 U7

among various components greatly influences vibration char-

acteristics and firing precision of self-propelled artillery. × U8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,

Decreasing connection stiffness between hull and revolving T52−1 = U1 U2 U3 U4 U5 U6 U7

part, revolving part, and elevating part makes firing disper-

sion of self-propelled change significantly. Firing precision × U8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,

of self-propelled artillery can be significantly improved by

reasonable adjustment of the system stiffness or connection T53−1 = U1 U2 U3 U4 U5 U6 U7

stiffness among various components.

× U8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,

T11−1 = U1 U2 U3 U4 U5 U6,11 U11 C,

Overall Transfer Matrix Elements for

Self-Propelled Artillery G42−17 = −H17 U18 U30 U42 ,

All elements for overall transfer matrix mentioned in (3) are G43−17 = H17,19 U19 U31 U43 ,

expressed as

G44−17 = H17,20 U20 U32 U44 ,

× U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , G47−17 = H17,23 U23 U35 U47 ,

× U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , G49−17 = H17,25 U25 U37 U49 ,

8 Advances in Mechanical Engineering

(2.2 × 107 , 2.2 × 107 , 1.1 × 108 ) Mode order 1 2 3 4 5 6 7 8

(5.5 × 107 , 1.3 × 107 , 1.1 × 107 ) Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8

(2.2 × 106 , 2.2 × 106 , 1.1 × 107 ) Mode order 1 2 3 4 5 6 7 8

(5.5 × 106 , 1.3 × 106 , 1.1 × 106 ) Eigenfrequencies 1.6 3.0 10.9 19.7 25.6 39.1 45.1 52.0

Table 6: Firing dispersion of a self-propelled artillery got by G46−8 = H8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,

simulation on the cement ground.

G47−8 = H8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,

Vertical target dispersion

Different connection stiffness Vertical Direction G48−8 = H8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,

dispersion/m dispersion/m

(2.2 × 107 , 2.2 × 107 , 1.1 × 108 ) G49−8 = H8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,

0.411 0.367

(5.5 × 107 , 1.3 × 107 , 1.1 × 107 )

G50−8 = H8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,

(2.2 × 106 , 2.2 × 106 , 1.1 × 107 )

0.543 0.380

(5.5 × 106 , 1.3 × 106 , 1.1 × 106 ) G51−8 = H8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,

G52−8 = H8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,

G53−8 = H8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,

G50−17 = H17,26 U26 U38 U50 ,

G51−17 = H17,27 U27 U39 U51 , G13−8 = H8,12 U12 U13,0 ,

G52−17 = H17,28 U28 U40 U52 , G42−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 ,

G53−17 = H17,29 U29 U41 U53 , G43−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 ,

G42−13 = H13 U14 U15 U16 U17 U18 U30 U42 , G44−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,

G43−13 = H13 U14 U15 U16 U17,19 U19 U31 U43 , G45−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 ,

G44−13 = H13 U14 U15 U16 U17,20 U20 U32 U44 , G46−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,

G45−13 = H13 U14 U15 U16 U17,21 U21 U33 U45 , G47−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,

G46−13 = H13 U14 U15 U16 U17,22 U22 U34 U46 , G48−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,

G47−13 = H13 U14 U15 U16 U17,23 U23 U35 U47 ,

G49−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,

G48−13 = H13 U14 U15 U16 U17,24 U24 U36 U48 ,

G50−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,

G49−13 = H13 U14 U15 U16 U17,25 U25 U37 U49 ,

G51−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,

G50−13 = H13 U14 U15 U16 U17,26 U26 U38 U50 ,

G52−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,

G51−13 = H13 U14 U15 U16 U17,27 U27 U39 U51 ,

G53−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,

G52−13 = H13 U14 U15 U16 U17,28 U28 U40 U52 ,

G13−6 = H6 U7 U8,12 U12 U13,0 ,

G53−13 = H13 U14 U15 U16 U17,29 U29 U41 U53 ,

G11−6 = −H6,11 U11 C,

G13,I2 = −H13,0 ,

G10−6 = H6 U7 U8 U9 U10 ,

G10−8 = −H8 U9 U10 ,

I O6×6

G42−8 = H8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , C=[ 6 ].

O6×6 −I6

G43−8 = H8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , (A.1)

G44−8 = H8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,

All matrices is no longer given in detail, which can be seen in

G45−8 = H8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 , [1, 2, 20].

Advances in Mechanical Engineering 9

(2.2 × 107 , 2.1 × 107 , 5.3 × 107 ) Mode order 1 2 3 4 5 6 7 8

(5.5 × 107 , 1.1 × 107 , 1.1 × 108 ) Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8

(2.2 × 106 , 2.1 × 106 , 5.3 × 106 ) Mode order 1 2 3 4 5 6 7 8

(5.5 × 106 , 1.1 × 106 , 1.1 × 107 ) Eigenfrequencies 2.1 2.9 7.9 16.6 27.9 36.5 42.9 50.1

Table 8: Firing dispersion of self-propelled artillery got by simula- [11] X. Rui and W. Schiehlen, “Multibody system dyanamics: pref-

tion on the cement ground. ace,” Multibody System Dynamics, vol. 18, no. 4, pp. 485–492,

2007.

Vertical target dispersion

[12] X. Rui, L. Yun, G. Wang et al., “Augmented eigenvector and its

Different connection stiffness Vertical Direction orthogonality of linear multi-rigid-flexible-body system,” Jour-

dispersion/m dispersion/m nal of China Ordnance, vol. 4, no. 2, pp. 100–105, 2008.

(2.2 × 107 , 2.1 × 107 , 5.3 × 107 ) [13] X. Rui, B. He, B. Rong et al., “Discrete time transfer matrix

0.411 0.367

(5.5 × 107 , 1.1 × 107 , 1.1 × 108 ) method for dynamics of multi-rigid-flexible-body system mov-

(2.2 × 106 , 2.1 × 106 , 5.3 × 106 ) ing in plane,” Journal of Multi-Body Dynamics, vol. 23, no. 1, pp.

0.458 0.371

(5.5 × 106 , 1.1 × 106 , 1.1 × 107 ) 23–42, 2009.

[14] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for

linear multibody system,” Multibody System Dynamics, vol. 19,

Conflict of Interests no. 3, pp. 179–207, 2008.

[15] X. Rui, E. Kreuzer, B. Rong, and B. He, “Discrete time transfer

The authors declare that there is no conflict of interests matrix method for dynamics of multibody system with flexible

regarding the publication of this paper. beams moving in space,” Acta Mechanica Sinica, vol. 28, pp.

490–504, 2012.

Acknowledgments [16] X. Rui, B. Rong et al., “Some new developments in discrete time

transfer matrix method of multibody systems,” in Proceedings

The research was supported by the Research Fund for of the International Conference on Dynamics Vibration and

the Doctoral Program of Higher Education of China Control, Hangzhou, China, 2010.

(20113219110025), the Natural Science Foundation of China [17] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer

Government (11102089), and the Program for New Century matrix method for dynamics analysis of complex weapon sys-

Excellent Talents in University (NCET-10-0075). tems,” Science China Technological Sciences, vol. 54, no. 5, pp.

1061–1071, 2011.

[18] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer

References matrix method for multibody system dynamics,” Multibody

System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.

[1] X. Rui, Y. Liu, and H. Yu, Launch Dynamics of Tank and Self-

Propelled Artillery, Science Press, Beijing, China, 2008. [19] B. He, X. Rui, and G. Wang, “Riccati discrete time transfer

matrix method for elastic beam undergoing large overall mo-

[2] X. Rui, L. Yun, Y. Lu et al., Transfer Matrix Method for Multibody

tion,” Multibody System Dynamics, vol. 18, no. 4, pp. 579–598,

System and its Application, Science Press, Beijing, China, 2008.

2007.

[3] X. Rui, L. Yun, G. Wang et al., Direction to Launch Safety of

[20] X. Rui and J. Zhang, “Automatical transfer matrix method of

Ammunition, National Defense Industry Press, Beijing, China,

multibody system,” in Proceedings of the 2nd Joint International

2009.

Conference on Multibody System Dynamics, Stuttgart, Germany,

[4] X. Rui, Y. Lu, G. Wang et al., Simulation and Test Methods of May-June 2012.

Launch Dynamics of Multiple Launch Rocket System, National

Defense Industry Press, Beijing, China, 2003.

[5] X. Rui, Launch Dynamics of Multibody System, National Defense

Industry Press, Beijing, China, 1995.

[6] X. Rui and Q. Yang, Launch Process of Projectile, South East

University Press, Nanjing, China, 1991.

[7] J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G.

Teubner, Stuttgart, Germany, 1977.

[8] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,

Germany, 1990.

[9] T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft Dynam-

ics, McGraw-Hill, New York, NY, USA, 1983.

[10] X. Rui, D. Bestle, and J. Zhang, “A new form of the trans-

fer matrix method for multibody systems,” in Proceedings of

the ECCOMAS Thematic Conference on Multibody Dynamics,

Zagreb, Croatia, 2013.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 383680, 8 pages

http://dx.doi.org/10.1155/2014/383680

Research Article

Higher Order Variational Integrators for Multibody System

Dynamics with Constraints

College of Information Engineering, Qingdao University, Qingdao 266071, China

Received 25 August 2013; Revised 16 November 2013; Accepted 25 November 2013; Published 23 January 2014

Copyright © 2014 J. Ding and Z. Pan. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The continuous and discrete Euler-Lagrangian equations with holonomic constraints are presented based on continuous and

discrete Hamiltonian Principle. Using Lagrangian polynomial to interpolate state variables and Gauss quadrature formula to

approximate Hamiltonian action integral, the higher order variational Galerkin integrators for multibody system dynamics with

constraints and the computation procedure are given. Numerical examples are provided to show the long-time behavior of the

methods proposed in this paper via comparisons with traditional Runge-Kutta methods.

the same time [6]. The variational integrators [7, 8] based

Dynamics of multibody systems [1] are usually described on discrete variational principle [9] provides an excellent

by a set of differential-algebraic equations (DAEs) [2, 3], framework for design of geometric numerical integrators

which are solved traditionally using numerical methods with many merits: it preserves momentum naturally with

of ordinary differential equations (ODEs) and combining good behavior of approximate energy conservation, even

some constraint stabilization techniques. In the community preserves strictly symplectic-momentum-energy [10]; it can

of numerical analysis, numerical methods for ODEs are be easily extended to a large class of problems [11, 12],

always designed in a small area with local stability analysis, such as the construction of geometric structure-preserving

which lead to a lot of difficulties for long-time simulation. numerical integrators for PDEs [13], nonsmooth collisions

Geometric numerical integration methods [3] originated [14, 15], stochastic systems [16, 17], nonholonomic systems

from long-term simulation of molecular dynamics and planet [18, 19], and constrained systems [20], dissipative systems

dynamics in solar system with global numerical stability due [21], optimal control [22, 23], parameter optimization [24]; it

to the property of structure-preserving provide alternatives can generate easily a large class of higher order methods based

to overcome these problems. These numerical methods are on polynomial interpolation and numerical quadrature for-

referred to as the ones preserving the invariants in their mulas systematically [25–29] and some Lie group variational

corresponding continuous systems, such as preserving sym- integrators [30–32] combining Lie group methods [33].

plectic, energy, momentum, symmetry, and orthogonality, For the general conservative systems without constraints,

in the original systems, thus revolutionizing the traditional [25, 26] propose the same framework of design higher order

concepts of numerical stability. It has been widely accepted variational Galerkin integrators using numerical interpola-

that the more invariants conserved in a numerical method, tion and quadrature techniques for state variables and action

the more stable it is. integral, respectively, during one integration step based on

Symplectic algorithms [4] and energy conservation meth- discrete variational principle. Motivated by these works, we

ods [5] are two frequently used methods for numerical will investigate higher order variational Galerkin integrators

integration of constrained Hamiltonian systems, but they are for multibody system dynamics with constraints for the long-

designed for conservative systems and it is difficult to design term simulation purpose. Due to the space limit, we will

2 Advances in Mechanical Engineering

focus on the cases with holonomic constraints, but the results where 𝐿 𝑑 (q𝑖 , q𝑖+1 , ℎ) and 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) are defined as the

can be easily extended to the cases with nonholonomic discrete approximation of the integration of 𝐿(q, q,̇ 𝑡) and

constraints and the cases with nonconservative forces. 𝜆𝑇 Φ(q, 𝑡) on the interval 𝑡 ∈ [𝑡𝑖 , 𝑡𝑖+1 ],

The paper is organized as follows. In Section 2, the 𝑡𝑖+1

continuous and discrete Euler-Lagrangian equations with 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) = ∫ 𝐿 (q, q,̇ 𝑡) 𝑑𝑡,

𝑡𝑖

holonomic constraints are presented based on continuous (5)

𝑡𝑖+1

and discrete Hamiltonian Principle, respectively; in Section 3, 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) = ∫ 𝜆𝑇 Φ (q, 𝑡) 𝑑𝑡.

we derive the higher order variational Galerkin integra- 𝑡𝑖

tors for multibody system dynamics with constraints using

Lagrangian polynomial to interpolate state variables and Using standard variational method, the discrete Euler-

using Gauss quadrature formula to approximate Hamilto- Lagrangian (DEL) equations are derived as

nian action integral; the computation procedure is given in 𝐷1 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) + 𝐷2 𝐿 𝑑 (q𝑖−1 , q𝑖 , 𝑡𝑖−1 )

Section 4; numerical examples are provided in Section 5 to

show the long-time behavior of the methods proposed in this − ℎΦ𝑇q (q𝑖 , 𝑡𝑖 ) 𝜆𝑖 = 0, (6)

paper via comparisons with traditional methods. The last one Φ (q𝑖+1 , 𝑡𝑖+1 ) = 0,

is concluding remarks including summary and future works.

where 𝐷𝑗 𝐿 𝑑 (𝑗 = 1, 2) is the partial derivative of 𝐿 𝑑 with

2. Discrete Euler-Lagrangian respect to the 𝑗th variable.

Equations with Constraints

3. Higher Order Variational Galerkin

The Hamilton Principle of a multibody system with holo- Integrators of Multibody

nomic constraints and conservative forces can be stated as

Systems with Constraints

𝑡𝑓

𝛿𝑆 = 𝛿 ∫ (𝐿 (q, q,̇ 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡 = 0, (1) The key technique to improve the accuracy of variational

0

integrators is the approximation of the discrete Hamilto-

where 𝑆 is Hamilton action integral, 𝐿 is Lagrangian, q, q̇are nian action integral, which can be realized through general

generalized displacement and velocity, respectively, and 𝜆 is Galerkin methods [24, 25]. Using the same methods in this

Lagrange multiplier vector corresponding to the constraint section, Hamiltonian action integral is approximated in a

subinterval [𝑡𝑖 , 𝑡𝑖+1 ] with an introduced local time parameter

Φ (q, 𝑡) = 0. (2) 𝜏 ∈ [0, 1], where

Via variational method, the following Euler-Lagrange 𝑡 − 𝑡𝑖

𝜏= . (7)

equations of a constrained mechanical system can be got: 𝑡𝑖+1 − 𝑡𝑖

𝑑 𝜕𝐿 (q, q,̇ 𝑡) 𝜕𝐿 (q, q,̇ 𝑡) Given 𝑠 + 1 control points 0 = 𝜏0 < 𝜏1 < ⋅ ⋅ ⋅ < 𝜏𝑠 = 1,

( )− + Φ𝑇 (q, 𝑡) 𝜆 = 0, for 𝑘 = 0, 1, . . . , 𝑠, Lagrange interpolation polynomial 𝑙𝑘 (𝜏) :

𝑑𝑡 𝜕q̇ 𝜕q (3) [0, 1] → 𝑅 is defined as

Φ (q, 𝑡) = 0. 𝜏 − 𝜏𝑗

𝑙𝑘 (𝜏) = ∏ . (8)

It is a typical differential-algebraic equation which can be 𝜏 −𝜏 𝑘 𝑘

0≤𝑗≤𝑠,𝑗 ≠ 𝑗

solved using different traditional numerical methods of DAEs

[2, 3]. The state variables q(𝑡) on [𝑡𝑖 , 𝑡𝑖+1 ] can be interpolated as

Following the variational integrators based on discrete follows:

𝑠

variational principle, the time interval 𝑡 ∈ [0, 𝑡𝑓 ] is divided q𝑑 (𝜏) = ∑ 𝑙𝑘 (𝜏) q𝑖,𝑘 , (9)

into 𝑁 subintervals with time step ℎ = 𝑡𝑓 /𝑁 equally along 𝑘=0

with 𝑁 + 1 grids, 0 = 𝑡0 , 𝑡1 , 𝑡2 , . . . , 𝑡𝑁−1 , 𝑡𝑁 = 𝑡𝑓 , 𝑡𝑖 = 𝑖ℎ,

𝑖 = 0, 1, 2, . . . , 𝑁. The discrete generalized displacement and where q𝑖,𝑘 = q(𝑡𝑖 + 𝑘 ⋅ (ℎ/𝑠)). The derivative of q𝑑 (𝜏) with

velocity and Lagrangian multiplier can be denoted as q𝑖 = ̇ on [𝑡𝑖 , 𝑡𝑖+1 ] as

respect to 𝑡 provides an approximation of q(𝑡)

q(𝑡𝑖 ), q̇𝑖 = q(𝑡

̇ 𝑖 ), 𝜆𝑖 = 𝜆(𝑡𝑖 ), 𝑖 = 0, 1, 2, . . . , 𝑁. Equation (1) can 1 𝑠

be rewritten as the following discrete counterpart: q̇𝑑 (𝜏) = ∑ 𝑙 (𝜏) q𝑖,𝑘 , (10)

ℎ 𝑘=0 𝑘

𝛿𝑆𝑑 (q0 , q1 , . . . , q𝑁−1 , q𝑁, 𝜆0 , . . . , 𝜆𝑁−2 , 𝜆𝑁−1 )

where 𝑙𝑘 (𝜏) = (𝜕/𝜕𝜏)𝑙𝑘 (𝜏). Then, Hamiltonian action integral

𝑁−1 can be approximated on [𝑡𝑖 , 𝑡𝑖+1 ] as follows:

= 𝛿 ∑ 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) 𝑡𝑖+1

𝑖=0

(4) ∫ (𝐿 (q (𝑡) , q̇(𝑡) , 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡

𝑁−1 𝑡𝑖

1

− 𝛿 ∑ 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) (11)

𝑖=0

= ∫ ℎ (𝐿 (q𝑑 (𝜏) , q̇𝑑 (𝜏) , 𝑡𝑖 + 𝜏ℎ)

0

Advances in Mechanical Engineering 3

derivatives with respect to 𝜏. x1

𝑠 𝑙𝑘 (𝜏) 𝑙𝑘 (𝜏)

𝑙0 (𝜏) = 1 − 𝜏

1 𝑙0 (𝜏) = −1

𝑙1 (𝜏) = 𝜏 𝑙1 (𝜏) = 1

𝑙0 (𝜏) = (2𝜏 − 1) (𝜏 − 1) 𝑙0 (𝜏) = 4𝜏 − 3 𝜃1

y1 l1

2 𝑙1 (𝜏) = −4𝜏 (𝜏 − 1) 𝑙1 (𝜏) = −4 (2𝜏 − 1)

𝑙2 (𝜏) = 𝜏 (2𝜏 − 1) 𝑙2 (𝜏) = 4𝜏 − 1

y2

Table 2: Quadrature points and weights of the Gauss quadrature m1

formula with respect to the interval [−1, 1].

𝜃2 l2

𝑛 𝜏̃𝑟 𝐴𝑟

2 ±√3/3 1 m2

±√15/5 5/9

3

0 8/9

Lagrange interpolation polynomials in (9) and comparable

derivatives with respect to 𝜏 in (10) are listed in Table 1.

Using Gauss quadrature formula, which has maximal Step 1. Given the order of Lagrange interpolation polynomial

degree of accuracy 2𝑛 − 1 for a fixed number 𝑛 of quadrature 𝑠, the number of Gauss quadrature points 𝑛, fixed time step ℎ,

points, we have initial state variables q0 , and velocity variables q̇0 .

∫ (𝐿 (q (𝑡) , q̇(𝑡) , 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡 method. Use q0,0 = q0 , and q0,𝑠 = q1 to calculate q0,𝑘 , 𝑘 =

𝑡𝑖

1, . . . , 𝑠 − 1.

𝑛

𝜏𝑟 ) , q̇𝑑 (̃

= ∑𝐴 𝑟 ℎ (𝐿 (q𝑑 (̃ 𝜏𝑟 ) , 𝑡𝑖 + 𝜏̃𝑟 ℎ) Step 3. For 𝑖 = 2 to 𝑁 − 1, solve (13) by Newton iteration

𝑟=1

method to get q𝑖,1 , . . . , q𝑖,𝑠 and 𝜆𝑖 , thus q𝑖+1 = q𝑖,𝑠 , q̇𝑖+1 =

−𝜆𝑇𝑖 Φ (q𝑑 (̃

𝜏𝑟 ) , 𝑡𝑖 + 𝜏̃𝑟 ℎ)) q̇𝑑 (1) = (1/ℎ) ∑𝑠𝑘=0 𝑙𝑘 (1)q𝑖,𝑘 .

= 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 )

Step 4. Use the solution q𝑖 , q̇𝑖+1 , 𝑖 = 0, 1, . . . , 𝑁 − 1 to get

− 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 ) ,

the potential energy 𝑇, kinetic energy 𝑉, total energy of the

(12) system 𝐻 = 𝑇 + 𝑉, and constraints Φ.

Quadrature points and weights of the Gauss quadrature 5. Numerical Example

formula with respect to the interval [−1, 1] are listed in Consider a simple double pendulum system shown in

Table 2. Figure 1 the state variables are chosen as q = [𝑥1 𝑦1 𝑥2

Apply discrete Hamilton’s principle, the derivatives of

𝑦2 ]𝑇 , then the potential energy, kinetic energy of the system

the action with respect to q𝑖,𝑘 have to vanish for all 𝑖 =

are

0, 1, . . . , 𝑁 − 1 and 𝑘 = 0, 1, . . . , 𝑠. That leads to the discrete

Euler-Lagrange equations 1 1

𝑇 = 𝑚1 (𝑥1̇2 + 𝑦12̇) + 𝑚2 (𝑥2̇2 + 𝑦22̇) ,

2 2 (14)

𝐷1 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 ) − 𝐷1 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 )

𝑉 = 𝑚1 𝑔𝑦1 + 𝑚2 𝑔𝑦2 .

+ 𝐷𝑠+1 𝐿 𝑑 (q𝑖−1,0 , q𝑖−1,1 , . . . , q𝑖−1,𝑠 , 𝑡𝑖−1 )

− 𝐷𝑠+1 𝑓𝑑 (q𝑖−1,0 , q𝑖−1,1 , . . . , q𝑖−1,𝑠 , 𝜆𝑖−1 , 𝑡𝑖−1 ) = 0 The mass matrix is a constant matrix as follows:

𝐷𝑘 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 ) M = diag [𝑚1 𝑚1 𝑚2 𝑚2 ] . (15)

− 𝐷𝑘 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 ) = 0, 𝑘 = 1, . . . , 𝑠

The constraint equations are

Φ (q𝑖+1 , 𝑡𝑖+1 ) = 0.

(13) 𝑥12 + 𝑦12 − 𝑙12

Φ (q) = ( 2 2 ) = 0. (16)

(𝑥2 − 𝑥1 ) + (𝑦2 − 𝑦1 ) − 𝑙22

4. Computation Procedure

The following computation procedure can be used to get the Given 𝑚1 = 𝑚2 = 1 kg, 𝑙1 = 𝑙2 = 1 m, with time step ℎ =

solution of the discrete Euler-Lagrange equations (13). 0.01 s, the results of the method presented in this paper are

4 Advances in Mechanical Engineering

Runge-Kutta 15.0385 3.0045 0.2345

P1L-Q2G 16.0993 0.4049 1.9984 × 10−15

P2L-Q2G 24.8510 0.2270 8.8818 × 10−16

P2L-Q3G 31.2938 0.2273 1.3323 × 10−15

−4

Runge-Kutta 99.7938 4.7378 × 10 3.6633 × 10−5

P1L-Q2G 121.4468 0.0405 9.6040 × 10−11

P2L-Q2G 212.9882 0.0023 1.2912 × 10−9

P2L-Q3G 265.9505 0.0021 1.2912 × 10−9

Runge-Kutta P1L-Q2G

3 3

Total energies (N·m)

2 2

1 1

0 0

0 50 100 0 50 100

Time (s) Time (s)

(a) (b)

P2L-Q2G P2L-Q3G

3 3

Total energies (N·m)

2 2

1 1

0 0

0 50 100 0 50 100

Time (s) Time (s)

(c) (d)

Advances in Mechanical Engineering 5

y

xj

yj A

xi i

x

B

yi

Oj

Oi

𝜃

O x

P1L-Q2G P1L-Q2G

30 3

20 2

10 1

Total energies (N·m)

Energies (N·m)

0 0

−10 −1

−20 −2

−30 −3

0 20 40 60 80 100 0 20 40 60 80 100

Time (s) Time (s)

Kinetic energy

Potential energy

Total energy

(a) (b)

compared with the Runge-Kutta But using the discrete variational methods presented in this

method in Table 3 for the ter- paper, the time step can be chosen longer as 0.01 s. Table 3 sh-

minal time 100 s. Here, P1L-Q2G means that the order of Lag- ows that, with this longer time step, the discrete variational

range interpolation polynomial is 1 and the number of Gauss methods can keep the total energy and constraints in higher

quadrature points is 2, and the same as P2L-Q2G and P2L- accuracy than Runge-Kutta method, the cost instead is the

Q3G. 𝜀(Η), 𝜀(Φ) are the maximum errors of the total energy longer computation time because of the iterations during the

𝐻 and constraints Φ. nonlinear equations solving process, but it is still shorter than

Using the traditional method such as Runge-Kutta met- the computation time of Runge-Kutta method with smaller

hod, to keep the errors of the total energy 𝐻 and constraints Φ time step to reach the same accuracy of the results.

are all smaller, the time step should be chosen as about 0.001 s It is also shown in Tables 3 and 4 that when the order of

in this example, and the total computation time is 99.7938 s. Lagrange interpolation polynomial is higher, the total energy

6 Advances in Mechanical Engineering

×10−16 ×10−16

4 4

2 2

Constraints 1

Constraints 2

0 0

−2 −2

−4 −4

0 50 100 0 50 100

Time (s) Time (s)

(a) (b)

−16 −16

×10 ×10

4 4

2 2

Constraints 3

Constraints 4

0 0

−2 −2

−4 −4

0 50 100 0 50 100

Time (s) Time (s)

(c) (d)

Figure 5: Constraints of rotary rod slider system by method P1L-Q2G, ℎ = 0.01.

and constraints can be kept better, but there are no big In the case of nonconstant mass matrix, the methods of

differences between 2 or 3 points in Gauss quadrature higher order variational Galerkin integrators are also applica-

formula. ble. For another example, Figure 3 shows a rotary rod slider

Figure 2 shows the energies of the previously mentioned system. 𝑂𝐴 is a rigid rod with uniform mass 𝑚𝐴 and length 𝑙,

methods. With the time step ℎ = 0.01 s, the total energy is which rotates round 𝑂 in the plane 𝑂𝑋𝑌. 𝐵 is a slider with the

up and down around zero with little errors during the long-ti- mass 𝑚𝐵 , and the stiffness of the spring on it is 𝑘; the mass of

me simulation by the methods presented in this paper, wh- the spring is ignored. Suppose only gravity in the plane 𝑂𝑋𝑌

ile increases quickly during the simulation by Runge-Kutta is considered for the system.

method. The results of high order variational Galerkin inte- The state variables are chosen as q = [𝑅𝑥𝑖 𝑅𝑦𝑖 𝑅𝑥𝑗

grators are better than low order integrators. 𝑅𝑦𝑗 𝜃 𝑥𝑖 ]𝑇 . Then, the mass matrix is a nonconstant matrix

[ 0 𝑚 0 0 𝑚𝐵 𝑥𝑖 cos 𝜃 𝑚𝐵 sin 𝜃 ]

[ 𝐵 ]

[ 0 0 𝑚𝐴 0 0 0 ]

[ ]

M=[ 0 0 0 𝑚𝐴 0 0 ]. (17)

[ 1 ]

[ ]

[−𝑚𝐵 𝑥𝑖 sin 𝜃 𝑚𝐵 𝑥𝑖 cos 𝜃 0 0 𝑚𝐴 𝑙2 + 𝑚𝐵 𝑥𝑖2 0 ]

3

[ 𝑚𝐵 cos 𝜃 𝑚𝐵 sin 𝜃 0 0 0 𝑚𝐵 ]

Advances in Mechanical Engineering 7

The constraint equations are [6] Z. Ge and J. E. Marsden, “Lie-Poisson Hamilton-Jacobi theory

and Lie-Poisson integrators,” Physics Letters A, vol. 133, no. 3, pp.

𝑙 134–139, 1988.

𝑅𝑥𝑗 − cos 𝜃

2 [7] J. M. Wendlandt and J. E. Marsden, “Mechanical integrators der-

( 𝑙 ) ived from a discrete variational principle,” Physica D, vol. 106,

𝑅𝑦𝑗 − sin 𝜃

Φ=(

( 2 ) = 0.

)

no. 3-4, pp. 223–246, 1997.

[8] J. E. Marsden and M. West, “Discrete Mechanics and Variational

(𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) cos 𝜃 + (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) sin 𝜃 − 𝑥𝑖 Integrators,” Acta Numerica, vol. 10, pp. 357–514, 2001.

( (𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) sin 𝜃 − (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) cos 𝜃 ) [9] J. Moser and A. P. Veselov, “Discrete versions of some classical

(18) integrable systems and factorization of matrix polynomials,”

Communications in Mathematical Physics, vol. 139, no. 2, pp.

Given 𝑙1 = 2𝑚, 𝑚𝐴 =2 kg, 𝑚𝐵 = 5 kg, 𝑘 = 1000 N/m. 217–243, 1991.

With the initial step length ℎ = 0.01 s and the terminal [10] C. Kane, J. E. Marsden, and M. Ortiz, “Symplectic-energy-mo-

time 100 s, using the method P1L-Q2G, where the order of mentum preserving variational integrators,” Journal of Mathe-

matical Physics, vol. 40, no. 7, pp. 3353–3371, 1999.

Lagrange interpolation polynomial is 1 and the number of

Gauss quadrature points is 2, the results are also satisfying. [11] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Variational time

integrators,” International Journal for Numerical Methods in

The total computer time is 86.0034 s; the maximum errors of

Engineering, vol. 60, no. 1, pp. 153–212, 2004.

the total energy 𝐻 and constraints Φ are 0.2611 and 3.8858 ×

[12] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “An overview of

10−16 , which is illustrated in Figures 4 and 5. variational integrators,” in Finite Element Methods: 1970s and

Beyond, pp. 85–146, CIMNE, Barcelona, Spain, 2003.

6. Conclusions [13] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Asynchronous

variational integrators,” Archive for Rational Mechanics and

Based on Lagrange interpolation and Gauss quadrature, hig- Analysis, vol. 167, no. 2, pp. 85–146, 2003.

her order variational Galerkin integrators have been devel- [14] A. Pandolfi, C. Kane, J. E. Marsden, and M. Ortiz, “Time-dis-

oped for multibody systems with constraints. Numerical res- cretized variational formulation of non-smooth frictional con-

ults validate the accuracy of the method for longer time step tact,” International Journal for Numerical Methods in Engineer-

and long-time simulation. Nevertheless, as the computation ing, vol. 53, no. 8, pp. 1801–1829, 2002.

is more complex, the computation cost is higher. Further [15] R. C. Fetecau, J. E. Marsden, M. Ortiz, and M. West, “Nonsmo-

study is needed to have more efficient integrators. oth lagrangian mechanics and variational collision integrators,”

SIAM Journal on Applied Dynamical Systems, vol. 2, no. 3, pp.

381–416, 2003.

Conflict of Interests [16] N. Bou-Rabee and H. Owhadi, “Stochastic variational integra-

tors,” IMA Journal of Numerical Analysis, vol. 29, no. 2, pp. 421–

The authors declare that there is no conflict of interests regar-

443, 2009.

ding the publication of this paper.

[17] W. Hu, Z. Deng et al., “Generalized multi-symplectic integrators

for a class of Hamiltonian nonlinear wave PDEs,” Journal of

Acknowledgments Computational Physics, vol. 235, pp. 394–406, 2013.

[18] J. Cortés and S. Martı́nez, “Non-holonomic integrators,” Nonlin-

This research was supported by the National Natural Sci- earity, vol. 14, no. 5, pp. 1365–1392, 2001.

ence Foundation of China (Grants nos. 11272166, 11002075, [19] R. McLachlan and M. Perlmutter, “Integrators for nonholonom-

10972110). ic mechanical systems,” Journal of Nonlinear Science, vol. 16, no.

4, pp. 283–328, 2006.

References [20] S. Leyendecker, J. E. Marsden, and M. Ortiz, “Variational integr-

ators for constrained dynamical systems,” ZAMM Zeitschrift für

[1] A. A. Shabana, Dynamics of Multibody Systems, Cambridge Uni- Angewandte Mathematik und Mechanik, vol. 88, no. 9, pp. 677–

versity Press, 3rd edition, 2005. 708, 2008.

[2] A. Laulusa and O. A. Bauchau, “Review of classical approaches [21] C. Kane, J. E. Marsden, M. Ortiz, and M. West, “Variational

for constraint enforcement in multibody systems,” Journal of integrators and the newmark algorithm for conservative and

Computational and Nonlinear Dynamics, vol. 3, no. 1, Article ID dissipative mechanical systems,” International Journal for Num-

011004, 2008. erical Methods in Engineering, vol. 49, no. 10, pp. 1295–1325,

[3] O. A. Bauchau and A. Laulusa, “Review of contemporary appr- 2000.

oaches for constraint enforcement in multibody systems,” Jour- [22] S. Leyendecker, S. Ober-Blöbaum, J. E. Marsden, and M. Ortiz,

nal of Computational and Nonlinear Dynamics, vol. 3, no. 1, “Discrete mechanics and optimal control for constrained sys-

Article ID 011005, 2008. tems,” Optimal Control Applications and Methods, vol. 31, no. 6,

[4] L. Jay, “Symplectic partitioned Runge-Kutta methods for con- pp. 505–528, 2010.

strained Hamiltonian systems,” SIAM Journal on Numerical [23] S. Ober-Blöbaum, O. Junge, and J. E. Marsden, “Discrete me-

Analysis, vol. 33, no. 1, pp. 368–387, 1996. chanics and optimal control: an analysis,” Control, Optimisation

[5] O. Gonzalez, “Mechanical systems subject to holonomic con- and Calculus of Variations, vol. 17, no. 2, pp. 322–352, 2011.

straints: differential-algebraic formulations and conservative [24] K. Witting, S. Ober-Blöbaum, and M. Dellnitz, “A variational

integration,” Physica D, vol. 132, no. 1-2, pp. 165–174, 1999. approach to define robustness for parametric multi-objective

8 Advances in Mechanical Engineering

no. 2, pp. 331–345, 2013.

[25] M. Leok and J. Zhang, “Discrete Hamiltonian variational int-

egrators,” IMA Journal of Numerical Analysis, vol. 31, no. 4, pp.

1497–1532, 2011.

[26] S. Ober-Blöbaum and N. Saake, “Construction and analysis of

higher order Galerkin variational integrators,” Numerical Anal-

ysis. In press. http://arxiv.org/abs/1304.1398 .

[27] M. Leok, T. Shingel, and T. P, “rolongation-collocation variatio-

nal integrators,” IMA Journal of Numerical Analysis, vol. 32, no.

3, pp. 1194–1216, 2012.

[28] S. Leyendecker and S. Ober-Blöbaum, “A variational approach

to multirate integration for constrained systems,” in Multibody

Dynamics, J.-C. Samin and P. Fisette, Eds., vol. 28 of Computa-

tional Methods in Applied Sciences, pp. 97–121, Springer, 2013.

[29] C. M. Campos, O. Junge, and S. Ober-Blöbaum, “Higher order

variational time discretization of optimal control problems,” in

Proceedings of the 20th International Symposium on Mathemat-

ical Theory of Networks and Systems,, pp. 9–13, Melbourne,

Australia, July 2012.

[30] M. Leok and T. Shingel, “General techniques for constructing

variational integrators,” Frontiers of Mathematics in China, vol.

7, no. 2, pp. 273–303, 2012.

[31] N. Bou-Rabee and J. E. Marsden, “Hamilton-Pontryagin inte-

grators on lie groups part I: introduction and structure-

preserving properties,” Foundations of Computational Mathe-

matics, vol. 9, no. 2, pp. 197–219, 2009.

[32] T. Lee, M. Leok, and N. H. McClamroch, “Lie group variational

integrators for the full body problem,” Computer Methods in

Applied Mechanics and Engineering, vol. 196, no. 29-30, pp.

2907–2924, 2007.

[33] A. Iserles, H. Z. Munthe-Kaas, S. P. Nørsett, and A. Zanna, “Lie-

Group methods,” Acta Numerica, vol. 9, pp. 215–365, 2000.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 792478, 16 pages

http://dx.doi.org/10.1155/2014/792478

Research Article

Free Vibration Characteristic of Multilevel Beam Based on

Transfer Matrix Method of Linear Multibody Systems

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Copyright © 2014 L. K. Abbas and X. Rui. This is an open access article distributed under the Creative Commons Attribution

License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly

cited.

In this paper, an approach based on transfer matrix method of linear multibody systems (MS-TMM) is developed to analyze the free

vibration of a multilevel beam, coupled by spring/dashpot systems attached to them in-span. The Euler-Bernoulli model is used for

the transverse vibration of the beams, and the spring/dashpot system represents a simplified model of a viscoelastic material. MS-

TMM reduces the dynamic problem to an overall transfer equation which only involves boundary state vectors. The state vectors at

the boundaries are composed of displacements, rotation angles, bending moments, and shear forces, which are partly known and

partly unknown, and end up with reduced overall transfer matrix. Nontrivial solution requires the coefficient matrix to be singular

to yield the required natural frequencies. This paper implements two novel algorithms based on the methodology by reducing the

zero search of the reduced overall transfer matrix’s determinate to a minimization problem and demonstrates a simple and robust

algorithm being much more efficient than direct enumeration. The proposal method is easy to formulate, systematic to apply, and

simple to code and can be extended to complex structures with any boundary conditions. Numerical results are presented to show

the validity of the proposal method against the published literature.

coupled by translational springs or elastic layers have been

The vibration problem of beam-type structures is of partic- studied extensively in the literature. Inceoğlu and Gürgöze

ular urgent issue in many branches of modern aerospace, [1] studied the bending vibrations of a combined system

mechanical, and civil engineering. Natural vibration frequen- consisting of two clamped-free beams carrying tip masses to

cies and modes are one of the most important dynamic which several double spring-mass systems are attached across

characteristics of these kinds of systems. For example, the the span. Using Green’s function method, the frequency

precision in manufacturing can be highly influenced by equation of the system is established. Kukla [2] solved the

vibrations. If the vibration characteristics cannot be solved or problem of free vibration of two axially loaded beams which

preestimated exactly when designing a mechanism system, are connected by many translational springs. The solution

it is often hard to obtain a good dynamic performance of contains possible combinations of the classical boundary

the mechanism system and consequently hard to control its conditions. The technique of the solution consists of devel-

vibration. oping a Green function. In [3], the vibrations of uniform

There are different types of beam models. One of the well- beams connected by homogeneous elastic layer are devoted.

known models is the Euler-Bernoulli beam theory that works Oniszczuk [4] discussed the free transverse vibrations of

well for slender beams. According to the Euler-Bernoulli two parallel simply supported beams continuously joined

beam theory, the length of each beam section is much greater by a Winkler elastic layer. The motion of the system was

than the height of each section and the shear and rotary described by a homogeneous set of two partial differential

inertia effects are ignored. equations, which was solved by using the classical Bernoulli-

The vibration theory of single-beam systems is well Fourier method. Oniszczuk [5] is devoted to analyze the

developed and studied in detail in hundreds of contributions. undamped forced transverse vibrations of an elastically

2 Advances in Mechanical Engineering

connected complex double-beam system in the case of simply and 𝐼 is the area moment of inertia. Beam mass per unit

supported beams. The classical modal expansion method was length 𝑚 = 𝜌𝐴, where 𝜌 is the material density and 𝐴 is the

applied to determine the dynamic responses of the beams beam cross-section area. Of course, any different materials,

due to arbitrarily distributed continuous loads. Vu et al. [6] different dimensions, or different boundary conditions could

presented an exact method for solving the vibration problem be considered in MS-TMM. The beams have the same length

of a damped double-beam system subjected to harmonic 𝐿 and are joined by the 𝑗-system of spring/dashpot located

excitation. The double-beam system consists of two identical at different positions (e.g., 𝑥1 and 𝑥𝑗 , where 𝑗 is the number

beams with the same boundary conditions on both sides. of spring/dashpot systems between two levels). 𝑘𝑦 and 𝑐 are

The beams are connected by a viscoelastic layer. Natural linear translational spring constant and damping coefficient,

frequencies and mode shapes of vibration of the system respectively, as shown in Figure 1.

are determined and the forced vibrations are investigated.

Gürgöze and Erol [7] determined the natural frequencies

of a clamped-free double-beam system carrying tip masses 3. MS-TMM Strategy in the Context of

to which several spring mass systems are attached across Free Vibration Characteristics

the span. However, there are only few contributions dealing

with the vibration of multibeam systems. That is, probably, 3.1. MS Topologies. According to the natural attribute of

the general vibration analyses of an elastically connected bodies, a complicated MS can be represented by various

multibeam system are complicated and laborious in view bodies (e.g., rigid bodies, elastic bodies, lumped masses,

of a large variety of possible combinations of boundary etc.) interconnected by hinges (e.g., spherical joints, sliding

conditions, and thus, the solution of the governing coupled joints, cylindrical joints, dampers, springs, etc.). In MS-

partial differential equations is difficult [8]. TMM, there are different topologies based on a certain set

Multibody system dynamics (MSD) has become an of modeling variables to formulate the dynamic equations of

important theoretical tool for wide engineering problems MS. Such topologies are chain, tree, closed loop (as illustrated

analysis in the world. Lots of methods of MSD have been in Figure 2 for reference), network, and so forth.

studied by many authors on theory and computational

method [9–15]. Professor Rui Xiaoting and his students have

been enlightened by the method of letting state vectors (SVs) 3.2. State Vector, State Variables, and Transfer Direction.

be transferred into classical transfer matrix method and The state vector (SV) at a connection point of MS is a

built up a new multibody dynamics method called “Transfer column vector denoting the mechanics state of this point.

Matrix Method of Linear Multibody Systems-MS-TMM” It includes the displacements of the point (including angu-

[16]. Using MS-TMM, the eigenvalue of linear multi-rigid- lar displacements) and the corresponding internal forces

flexible-body system is computed easily, the computational (including internal moments). Therefore, the SV is given

ill-condition is overcome, and the computational efficiency is by kinematics (displacements) and kinetics (internal forces)

increased. Over 20 years, MS-TMM has been developed and quantities, called state variables. For convention in this paper,

used widely in engineering applications. z with bold lowercase represents the SV in the physical

Motivated by the interesting study by Kukla [2] which coordinates and Z with bold capital represents the SV in

was published on the problem of the natural longitudinal the modal coordinates. Vibrations in space are described

vibrations of two rods coupled by many translational springs by displacement coordinates 𝑥, 𝑦, 𝑧 along the Cartesian axes

and by the two novel algorithms based on the new recursive and angular rotations 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 about Cartesian axes. Cutting

scanning approach presented by Bestle et al. [17], this paper forces and moments are given by 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 and 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 ,

presents a unique yet simple scenario of obtaining the respectively. Positive directions at input points are shown

exact free vibration characteristics of undamped/damped in Figure 3(a). Positive directions of forces and moments at

multilevel beam coupled elastically. The scenario developed output points (Figure 3(b)) are opposite due to the principle

in this paper is based on MS-TMM and Euler-Bernoulli beam of action equals reaction. In 3D case with 𝑛𝑠 = 12 (𝑛𝑠 is the

theory. number of state variables in the SV), the SVs in physical and

The text is organized as follows. The problem statement modal coordinates at the connection point 𝑝𝑖,𝑘 (where the

is presented in Section 2. In Sections 3 and 4, the general first subscript 𝑖 is the serial number of element for boundary

theorem brief of MS-TMM and problem solution scenario are end and the second subscript 𝑘 is the serial number of the

shown. In Section 5, some results calculated by MS-TMM and hinge element and 𝑘 = 0 for boundary end) are summarized

the other method are given which can validate the proposed in a vector, receptively:

method. The conclusions are presented in Section 6.

z𝑖,𝑘 physical coordinates

2. Problem Statement 𝑇

= [𝑥, 𝑦, 𝑧, 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 , 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 ]𝑖,𝑘 ,

The transversely vibrating system in a plane under considera- (1)

tion consists of multi-level parallel, elastic, and homogeneous Z𝑖,𝑘 modal coordinates

Euler-Bernoulli beam with general boundary conditions. 𝑇

Beam bending stiffness is 𝐸𝐼, where 𝐸 is the elastic modulus = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]𝑖,𝑘 .

Advances in Mechanical Engineering 3

y

General

boundary x Beam 1 m1 , (EI)1

conditions

ky c ··· ky c

Beam 2 m2 , (EI)2

ky c ··· ky c

..

x1 .

xj General

boundary

conditions

ky c ··· ky c

Beam m mm , (EI)m

Hinges

Loop

Bodies

Loop

Base

(a) (b) (c)

Figure 2: Multibody systems (a) chain, (b) tree, and (c) closed loop topology.

In case of 1D or 2D applications, the SV will be reduced direction shown in Figure 3, the transfer equation between the

as shown later. Defining a boundary point of the MS as component input and output is

the transfer end, the direction from all other boundary points

to the transfer end is called transfer direction. Along the z𝑛,0 = U𝑛 z𝑛−1,𝑛

transfer direction, the nodes entering into elements are called

inputs denoted by 𝐼 and the nodes leaving from elements are z𝑛−1,𝑛 = U𝑛−1 z𝑛−2,𝑛−1

called outputs 𝑂. z𝑛−2,𝑛−1 = U𝑛−2 z𝑛−3,𝑛−2 (2)

..

3.3. Transfer Equation, Transfer Matrix, Overall System Trans- .

fer Matrix, and Overall System State Vector. A vibrating MS

comprised of 𝑛-components, see Figure 4, is used as an z1,2 = U1 z1,0 .

example to show how to deduce the overall transfer equation

The constant matrix U𝑘 is the transfer matrix of the

and overall transfer matrix of the system. In order to describe

𝑘th component. Transfer matrices of basic components are

conveniently the idea, the chain topology is considered

considered as building blocks, which can be assembled

in the following. This vibrating system is comprised of 𝑛

together to provide the transfer matrix of the whole system

components and 𝑛 + 1 connection points. The SVs of the

according to the chain MS-TMM topology as follows:

boundary right extremity and other boundary left extremity

of the system are expressed as z𝑛,0 and z1,0 , respectively. 𝑛−1

Transfer direction of the system is always from another z𝑛,0 = Tz1,0 , where T = ∏U𝑛−𝑘 . (3)

boundary end to may call it the root. Following the transfer 𝑘=0

4 Advances in Mechanical Engineering

yI , qyI yO , −qyO

−𝜃xO , mxO

𝜃xI , −mxI

I xI , qxI O xO , −qxO

zI , qzI zO , −qzO

(a) (b)

Transfer direction

2 4 ··· n − 1 n 0 Root

0 1 3

Left end Right end

z1,0 zn,0

I 3 O

{

z2,3 z3,4

Figure 4: TMM-MS in the sense of chain topology. A vibrating system comprised of 𝑛-components with 𝑛 + 1 connections.

Rewrite (2) as from elimination of all columns of Uall (4c) associated zeros

in zall . For harmonic vibrations, solutions maybe written as

Uall 𝑛𝑠 ×(2×𝑛𝑠 ) zall (2×𝑛𝑠 )×1 = 0, (4a) z𝑖,𝑗 = Z𝑖,𝑗 𝑒𝜆𝑡 where 𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 , 𝜆𝑟 , 𝜆𝑖 ∈ R are the

eigenvalues. The real part (−𝜆𝑟 ) is related to the magnitude

where

of damping, where the imaginary part (𝜆𝑖 ) is related to the

𝑇 vibration frequency of the damped system. For undamped

z𝑇all = [z𝑇1,0 z𝑇𝑛,0 ] , (4b)

systems, 𝜆𝑟 = 0 and 𝜆𝑖 = 𝜔. Finally, Uall is only a function of

Uall = [T −I𝑛𝑠 ] . (4c) the unknown 𝜆 𝑖 of the system. For nontrivial solutions, the

Eigenfrequency equation

Herein, Uall is the overall system transfer matrix and zall is the

!

overall system state vector. Δ (𝜆) = det Uall = 0 (5)

3.4. Eigenfrequency Equation of the Whole MS. The overall has to be fulfilled. The natural frequencies of the system can

transfer equation (4a) only involves the boundary SVs, and now be computed.

the SVs at all other connection points do not appear. The

SVs at the boundary are composed of displacements, rotation 3.5. Beam Transfer Matrix. The full derivation of transfer

angles, moments, and shears, which are partly known and matrices for the Timoshenko and Euler-Bernoulli beams

partly unknown. For common boundary conditions, half of vibrating in a plane (with kinematics and kinetics’ SV defined

state variables of zall (4b) are zeros due to known constraints. as z = [𝑦, 𝜃𝑧 , 𝑚𝑧 , 𝑞𝑦 ]𝑇 ) may be found in [18], which is an

Thus, (4a) reduces to Uall zall = 0, where zall is composed of the open access article and the reader may download it from the

unknown state variables and Uall is a square matrix resulting Internet. However, for completeness, only the transfer matrix

Advances in Mechanical Engineering 5

for the Euler-Bernoulli beam will be presented, Figure 5. The where 𝑌 (𝑥) = 𝐴 1 cosh 𝛽𝑥 + 𝐴 2 sinh 𝛽𝑥

differential equation of a Euler-Bernoulli beam is

+ 𝐴 3 cos 𝛽𝑥 + 𝐴 4 sin 𝛽𝑥,

(6)

𝜕4 𝑦 𝜕2 𝑦

𝐸𝐼 4 + 𝑚 2

𝜕𝑥 𝜕𝑡 𝐴 1 , 𝐴 2 , 𝐴 3 , and 𝐴 4 are arbitrary constants, and 𝛽 =

√4 −𝑚𝜆2 /(𝐸𝐼). For the Euler-Bernoulli beam, the linearized

𝜕4 𝑌 (𝑥) 𝑚𝜆2 relations in modal coordinates Θ𝑧 = 𝑌 , 𝑀𝑧 = 𝐸𝐼𝑌 , and

= 0 𝑦 (𝑥, 𝑡) = 𝑌 (𝑥) 𝑒𝜆𝑡 4

+ 𝑌 (𝑥) = 0,

→ 𝜕𝑥 𝐸𝐼 𝑄𝑦 = 𝑀𝑧 maybe added to end up with the transfer relation:

[ ] [ ][ ]

[ Θ𝑧 ] [ 𝛽 sinh 𝛽𝑥 𝜆 cos 𝛽𝑥 ]

[ ] [ 𝛽 cosh 𝛽𝑥 −𝛽 sin 𝛽𝑥 ][ 𝐴 2]

[ ] [ ][[ ]

]

[ ]

Z (𝑥) = B (𝑥) a or [

[𝑀 ]

] =[ ] [ ]. (7)

[ 𝑧] [𝐸𝐼𝛽2 cosh 𝛽𝑥 𝐸𝐼𝛽2 sinh 𝛽𝑥 −𝐸𝐼𝛽2 cos 𝛽𝑥 −𝐸𝐼𝛽2 sin 𝛽𝑥 ] [ ]

[𝐴 3 ]

[ ] [ ]

[ ] [ ][ ]

[

]

[ 3 3 3 3

]

𝑄𝑦 𝐸𝐼𝛽 sinh 𝛽𝑥 𝐸𝐼𝛽 cosh 𝛽𝑥 𝐸𝐼𝛽 sin 𝛽𝑥 −𝐸𝐼𝛽 cos 𝛽𝑥 𝐴4

[ ]𝑥 [ ][ ]

The coefficient vector a = [𝐴 1 , 𝐴 2 , 𝐴 3 , 𝐴 4 ]𝑇 summarizes (1) Break up the complicated MS into components with

the unknown constants to be adopted to boundary condi- simple dynamic properties, which can be expressed in matrix

tions. At input end Z𝐼 (𝑥 = 0), we get Z𝐼 = [B(0)]a. Thus, form and SVs (for each component, it is possible to obtain

the coefficient vector can be expressed as a = [B(0)]−1 Z𝑂 and the close form expression of the transfer matrix giving the

substituting it into (7) for the beam output end at 𝑥 = 𝑙, one displacements and the forces applied to one extremity to

gets the displacements and forces applied to the other extremity).

In other words, on the component level, the governing

Z𝑂 = [B (𝑙)] a = [B (𝑙)] [B (0)]−1 Z𝐼 = UZ𝐼 , (8a) partial and ordinary differential or algebraic equations are

transformed to algebraic transfer equations, where the output

where state results from a product of the input state and an

element specific transfer matrix. These component matrices

U = B (𝑙) B−1 (0) are considered as building blocks. In fact, the transfer matrix

of such components needs not to be rededuced but may be

𝑇 𝑈 𝑉

[ 𝑆

taken directly from a transfer matrix library.

𝛽 𝐸𝐼𝛽 2 𝐸𝐼𝛽3 ]

[ ] (2) Following the transfer direction that has been

[ 𝑇 𝑈 ]

[ ] designed already by the analyst and according to the topology

[ 𝛽𝑉 𝑆 ]

=[

[

𝐸𝐼𝛽 𝐸𝐼𝛽2 ] ,

]

of the MS, these component transfer matrices are then

[ 𝑇 ] assembled and end up with a system of linear algebraic

[𝐸𝐼𝛽2 𝑈 𝐸𝐼𝛽𝑉 𝑆 ]

[ 𝛽 ]

equations called the overall transfer equation.

[ ] (3) Substitute the boundary conditions into the overall

3 2

[ 𝐸𝐼𝛽 𝑇 𝐸𝐼𝛽 𝑈 𝛽𝑉 𝑆 ] transfer equation to construct the eigenfrequency equation.

Consequently, the vibration characteristics such as frequen-

𝑐ℎ + 𝑐 𝑠ℎ + 𝑠

𝑆= , 𝑇= , cies can be deduced as the roots of a transcendental equation.

2 2 Due to narrow couples of natural frequencies, the classical

𝑐ℎ − 𝑐 𝑠ℎ − 𝑠 zero search method is likely to fail. However, a new recursive

𝑈= , 𝑉= , scanning approach for minima of the absolute values of the

2 2

determinant shows much more efficiency and reliability than

𝑐ℎ = cosh (𝛽𝑙) , 𝑠ℎ = sinh (𝛽𝑙) , direct enumeration.

𝑐 = cos (𝛽𝑙) , 𝑠 = sin (𝛽𝑙)

4. Problem Solution Scenario

(8b)

Figure 6 illustrates the suggested scenario for the solution

is the transfer matrix of the Euler-Bernoulli beam compo- of the problem statement and as follows. There are (1 : 𝑚)

nent. multi-level beam. Each beam level is divided into (1 : n)

components, which have (0 : 𝑛 + 1) connection points

3.6. Summary. In the context of free vibration characteristic, and coupled with another beam level by a system (say 1 :

the general strategy of linear MS-TMM, in summary, is as 𝑗) of a viscoelastic material (modeled as a spring/dashpot

follows. system). The massless dummy body as shown in Figure 6

6 Advances in Mechanical Engineering

qyI

𝜃zO mzO

mzI 𝜃zI m, L, EI

qyO

yI yO

i−1 i n−1 n

0 1 2

A G

U1,1 U1,2 ··· U1,i ··· U1,n−1 U1,n 1

1, 1 j, 1

B C

.. ..

. .

.. 1, m − 1

.

..

.. j, m − 1 .

.

V m

Um,1 Um,2 ··· Um,i ··· Um,n−1 Um,n

Spring/dashpot component

(e.g., 𝐴, 𝐵, 𝐶, and so on) is a connection point between Figure 7(a) sketches the first step. According to the continuity,

the two beam segment components and spring/dashpot 𝑗 has identical displacements, angles, and moments at input

system(s). The main key of MS-TMM is transferring the and output:

SV from one component to another following the general

transfer equation (2). For the beam segment component, 𝑌𝑂 = 𝑌𝐼 , Θ𝑧𝑂 = Θ𝑧𝐼 , 𝑀𝑧𝑂 = 𝑀𝑧𝐼 . (9a)

the transfer matrix is available (8b), while it is not for

spring/dashpot system at the connection point between two And from the force analysis shown in Figure 7(b), the

or multi-level beam in this paper. However, it is based on the spring/dashpot force 𝐹 changes the shear forces:

kinematics and kinetics of the spring/dashpot to formulate 𝑄𝑦𝑂 = 𝑄𝑦𝐼 + 𝐹. (9b)

the transfer matrix and it needs two steps. First, as an

example, let us consider a connection point 𝑗 between two Equations (9a) and (9b) in the view of the SV as a matrix form

beam segment components and system of spring/dashpot

may be grounded or connected to another connection point. Z𝑂 = Z𝐼 + e4 𝐹. (9c)

Advances in Mechanical Engineering 7

O

Fsys yO

Qy,I F Qy,O

ky c

YI YO

ky c

I j O yI

Maybe grounded or connected

to another connection point ZI ZO Fsys

Z1,I 1 Z1,O

G Level

1

F

Fsys

ky𝑗,1

cj,1

Fsys

Z1,I 1 F

Z1,O Level

Level

A 1 C 2

Z2,I Z2,O

F F

Fsys

Fsys

ky1,1 ky1,𝑚−1

c1,1 c1,m−1

Fsys Fsys

F F

Level Level

B 2 V 3

Z2,I Z2,O Z3,I Z3,O

(d) (e)

Figure 7: (a) Two beam segments and spring/dashpot system connected at massless dummy body, (b) state vectors (SV) at the massless

dummy body 𝑗-connection point, (c) force analysis of a spring/dashpot system, (d) a system 1,1 connected two levels 1-2 through two

connection points 𝐴 and 𝐵 shown in Figure 6, and (e) two systems 𝑗, 1 and 1, 𝑚−1 connected with multi-level (1-2-3) through three connection

points 𝐺, 𝐶, and 𝑉 shown in Figure 6.

8 Advances in Mechanical Engineering

𝑇

e4 = [0 0 0 1] is a unit vector assigning 𝐹 to the points 𝐺, 𝐶, and 𝑉 shown in Figures 6 and 7(e) can be

transfer equation of the massless connection body. The obtained similarly:

spring compression and damping forces are given as (see

Figure 7(c))

U𝑑𝑗,1 U𝑐𝑗,1 0

Z [ ] Z

𝐹spring = 𝑘𝑦 (𝑌𝐼 − 𝑌𝑂) = 𝑘𝑦 Δ𝑌 = 𝑘𝑦 e𝑇1ΔZ { }

{ 1

} [ 𝑐 𝑑 𝑐 ] { 1}

{ }

{ Z2 } = [ 𝑗,1 𝑗,1/1,𝑚−1 1,𝑚−1 ]

[ U U U

] { Z2} ,

{ } [ ] { }

) 𝑦 = 𝑌𝑒𝜆𝑡 = 𝑐𝜆 (𝑌𝐼 − 𝑌𝑂) = 𝑐𝜆e𝑇1 ΔZ [ ]

𝐹dashpot = 𝑐 (𝑦𝐼̇

− 𝑦𝑂̇ {Z3 }𝑂 0 U𝑐1,𝑚−1 U𝑑1,𝑚−1 {Z3 }𝐼

→ [ ]

→ 𝐹sys = (𝑘𝑦 + 𝑐𝜆) e𝑇1ΔZ.

where U𝑑𝑗,1 ≡ I − D𝑗,1 , U𝑐𝑗,1 ≡ D𝑗,1

(10) (12d)

Herein, e𝑇1 = [1 0 0 0] is a unit vector, which extracts 𝑌

from the transfer SV. Substitute (10) into (9c) and knowing

that 𝐹 = ± 𝐹sys , we ended the first step with U𝑑𝑗,1/1,𝑚−1 ≡ I − D𝑗,1/1,𝑚−1 ,

0 0 0 0

[ 0 0 0 0] (11)

where D = (𝑘𝑦 + 𝑐𝜆) e4 e𝑇1 =[

[

]. Now we are in the position to end the solution scenario.

0 0 0 0] From the MS-TMM topology point of view, the system shown

[ 𝑘𝑦 + 𝑐𝜆 0 0 0] in Figure 1 or equivalently Figure 6 can be modeled as a mul-

tichain, following the similar procedures mentioned above

For system 1,1 connected two levels 1-2 through two in Section 3.3 to determine the overall transfer equation and

connection points 𝐴 and 𝐵 shown in Figure 6, the second step ending up with determinant solution of the reduced overall

is how to deduce the transfer equation between the input and transfer matrix.

output related to these two levels. For the connection point 𝐴

(see Figure 7(d)) and using (11) with 𝐹 = 𝐹sys ,

5. Numerical Examples

Z1,𝑂 = Z1,𝐼 + D1,1 (Z2,𝐼 − Z1,𝐼 ) Before the free vibration analysis of a multi-level elastic

beam coupled by a spring/dashpot system(s) is performed,

(I − D1,1 )Z1,𝐼 + D

= ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ ⏟⏟Z2,𝐼

⏟⏟⏟⏟⏟

1,1 the reliability of the suggested scenario and the Matlab

U𝑑1,1 U𝑐1,1

(12a)

computer program developed for this paper are confirmed

by comparing the present results with those obtained from

Z1,𝑂 = U𝑑1,1 Z1,𝐼 + U𝑐1,1 Z2,𝐼 . the existing literature.

5.1. Reliability of the Suggested Scenario. Figure 8(a) consists

−𝐹sys ,

of two free-free and clamped-clamped undamped beams

connected to each other with two symmetrically distributed

Z2,𝑂 = Z2,𝐼 − D1,1 (Z2,𝐼 − Z1,𝐼 ) , linear translational springs. The hybrid system is vibrating

(12b) transversely in the 𝑥 − 𝑦 plane. The beams are supposed to

Z2,𝑂 = U𝑐1,1 Z1,𝐼 + U𝑑1,1 Z2,𝐼 . have the same length, material, and geometrical data, that is,

(𝐿 = 1 m, 𝑚1 = 𝑚2 = 𝑚 and (𝐸𝐼)1 = (𝐸𝐼)2 = 𝐸𝐼). It is worth

Combining (12a) and (12b) in one matrix form, we ended up mentioning here that the proposal scenario can be extended

the second step by writing the transfer equation and transfer to a system consisting of any number of uniform/nonuniform

matrix of a system located between two levels as beams coupled with any number of spring/dashpot systems.

However, the two springs 𝑘𝑦1,1 and 𝑘𝑦2,1 are located at distances

𝑥1 and 𝑥2 = 𝐿 − 𝑥1 , respectively, and are moving as a

U𝑑1,1 U𝑐1,1 function of 𝑥1 from 0 to 0.5 m as shown in Figure 8(a). This

Z1 ] {Z1 } .

{ } =[ 𝑐

Z2 𝑂 U1,1 U𝑑1,1 Z2 𝐼 (12c) example is presented by Kukla [2], which studied the dynam-

[ ]

⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ ics of the longitudinal vibrations of two rods coupled by

transfer matrix several translational springs using Green’s functional method.

Figure 8(b) shows a vibrating two-level beam comprised of

The transfer equation for two systems 𝑗, 1 and 1, 𝑚 − 1 10 components with 4 boundaries, namely, Z1,0 , Z6,0 on the

connected with multi-level (1-2-3) through three connection system left side and Z5,0 , Z10,0 on the right side. Herein,Z𝑖,𝑘 =

Advances in Mechanical Engineering 9

y Transfer direction

x m1 , (EI)1

Beam 1 1 3 5

ky1,1 ky2,1 0 0 Level 1

Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0

m2 , (EI)2 Chain 1

Beam 2 ky1,1 ky2,1

2 4

6 7 8

x1 0 Level 2

0 Z Z6,2 Z2,7 Z7,4 Z4,8 Z8,0

x2 = 1 − x1 6,0 Chain 2

x1 L − 2x1 x1

L

Beam segement

(a) (b)

Figure 8: (a) Free-free and clamped-clamped beams connected with two linear translational springs, (b) chain TMM-MS: State vectors and

transfer direction of the suggested solution scenario.

apply. Following the transfer direction from the left to right Z5,0

{ }

and applying (2) for two-level, the transfer equation can be Z8,0 𝑂

written as

U5 U4,𝑑 2,𝑑

2,1 U3 U1,1 U1 U5 U4,𝑑 2,𝑐

2,1 U3 U1,1 U6

[( ]

[ +U U4,𝑐 U U2,𝑐 U ) (+U U4,𝑐 U U2,𝑑 U )]

U5 0 [ 5 2,1 7 1,1 1 5 2,1 7 1,1 6 ]

Z5,0 Z4,5 [ ]

{ } =[ ]{ } =[ ]

Z8,0 𝑂 0 U8 Z4,8 𝐼 [ U8 U4,𝑐 U U 2,𝑑

U U U 4,𝑐

U U 2,𝑐

U ]

[ 2,1 3 1,1 1 8 2,1 3 1,1 6 ]

[( 4,𝑑 2,𝑐 ) ( 4,𝑑 2,𝑑 )]

↓ +U8 U2,1 U7 U1,1 U1 +U U U U

8 2,1 7 1,1 6 U

[ ]

U4,𝑑 U4,𝑐 Z1,0

Z4,5 [ 2,1 2,1 ] Z3,4

{ } = [ 4,𝑐 4,𝑑 ] { } ×{ }

Z4,8 𝑂 U2,1 U2,1 Z7,4 𝐼 Z6,0 𝐼

[ ] ↓

↓

T1,1 T1,2

Z5,0 ] {Z1,0 } .

Z3,4 U3 0 Z2,3 { } =[

Z8,0 𝑂 T2,1 T2,1 Z6,0 𝐼

{ } =[ ]{ } [ ]

Z7,4 𝑂 0 U7 Z2,7 𝐼

(13)

↓

U1 , U3 , U5 , U6 , U7 , and U8 represent the beam segment

U2,𝑑 U2,𝑐 transfer matrices (8b). U2,𝑑 2,𝑐 4,𝑑 4,𝑐

1,1 , U1,1 , and U2,1 , U2,1 are the

Z2,3 [ 1,1 1,1 ] Z1,2

{ } =[ 𝑐 ]{ } transfer matrices of a system between two connection points

Z2,7 𝑂 U1,1 U2,𝑑

1,1

Z6,2 𝐼

located between two beams (12c). The overall transfer equa-

[ ]

tion, overall SV, and overall transfer matrix are, respectively,

↓

Uall 8×16 Zall 16×1 = 0, (14a)

Z1,2 U1 0 Z1,0

{ } =[ ]{ } 𝑇 𝑇 𝑇 𝑇 𝑇

Z6,2 𝑂 0 U6 Z6,0 𝐼 where Z𝑇all = [Z ⏟⏟⏟ ⏟ Z

⏟⏟⏟1,0 ⏟⏟⏟ ⏟ Z

⏟⏟⏟6,0 ⏟⏟⏟ ⏟ Z

⏟⏟⏟5,0 ⏟⏟⏟8,0

⏟⏟⏟ ⏟] , (14b)

4×1 4×1 4×1 4×1

⇓

T1,1 4×4 T1,2 4×4 −I4×4 O4×4

U4,𝑑 U4,𝑐 Uall = [ ]. (14c)

Z5,0 U5 0 U 02,1 2,1

{ } = [ ][ ][ 3 ] T T

[ 2,1 4×4 2,2 4×4 O 4×4 −I4×4 ]

Z8,0 𝑂 0 U8 4,𝑐 4,𝑑 0 U7

[U2,1 U2,1 ]

Applying the boundary conditions listed in Table 1, half

U2,𝑑 2,𝑐

1,1 U1,1

of state variables of Zall (14b) are zeros due to known

×[ 𝑐 ] [U1 0 ] {Z1,0 } constraints. Thus, the overall transfer equation reduced to

U1,1 U2,𝑑

1,1

0 U6 Z6,0 𝐼

Uall |8 × 8 Zall |8 × 1 = 0 that is ready for eigenproblem (5).

[ ]

10 Advances in Mechanical Engineering

Table 1: Common boundary conditions for a beam vibrating in a grids 𝑁𝑥0 = 500, absolute precision tolerance 𝜀 = 10−6 ).

plane. Figure 9 shows log10 |Δ| obtained from the fMin1D algorithm

versus the first 10 dimensionless frequency values (𝜔 =

Support type Zero terms Nonzero terms

Fixed 𝑌, Θ𝑧 𝑄𝑦 , 𝑀𝑧 √4 (𝜆𝑖 ≡ 𝜔)2 𝑚𝐿4 /𝐸𝐼) for 𝑘𝑦 = 𝑘𝑦 = 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ) and

1,1 2,1

Free 𝑄𝑦 , 𝑀𝑧 𝑌, Θ𝑧 two springs in Figures 10(a) and 10(b) for 𝑘𝑦 = 100(𝐸𝐼/𝐿3 )

Guided 𝑄𝑦 , Θ𝑧 𝑌, 𝑀𝑧 and 𝑘𝑦 = 1000(𝐸𝐼/𝐿3 ), respectively. The first and second

springs are moving opposite to each other from 𝑥1 = 0 at

the beams support ends and coincide together at the middle

The natural frequencies of the system can now be com- of the beams (𝑥1 = 0.5 m). Reference [2] does not provide the

puted by zero search of the determinant, which is based on results as tables or specific data but only showed figures. The

sign change of Δ(𝜔) (in case of real numbers) or Δ(𝜆) (in reader may download the reference from the library or from

case of complex numbers) during a scanning of an interesting the Internet because the authors of the present paper cannot

frequency range. This procedure can be cumbersome for include the figures to maintain the journal publication rights.

several reasons. Therefore, reliable and efficient algorithms However, the MS-TMM results are in very good agreement

called recursive scanning approach are applied (see [17] for with Figures 2(a) and 2(b) [2, page 131].

more details of the proposed algorithm) by switching from

zero search for Δ to minimization of the absolute value |Δ| 5.2. Free Vibration of Damped Elastically Coupled Triple

of the determinant, which is equally well applicable to both Beams. The physical model of the transversely vibrating

the real and the complex cases. The algorithm general idea system under consideration is composed of three paral-

is as follows. In a first iteration step, it divides an interesting lel uniform rectangular Euler-Bernoulli beams of homoge-

band of frequencies into a number of sample points and nous properties. Each two-level beam is joined (connected)

searches for lower peaks. Each region having a lower peak is together by two spring/dashpot systems located at 𝑥1 and 𝑥2 ,

then divided again into small intervals to find more narrow respectively (see Figure 1). The beams have the same length

regions of lower peaks as second iteration step. The algorithm and are pinned at their ends. The small damped vibrations of

proceeds until the required precision of roots is achieved. MS- the system are considered. In the sense of the chain MS-TMM

TMM natural frequency results are obtained using fMin1D (see Figure 11), the overall transfer equation, overall transfer

[1, 3000] rad/sec, scanning

algorithm [17] (frequency ranges ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ matrix, and overall SV are, respectively,

𝜆𝑖 =𝜔

T1,1 4×4 T1,2 4×4 T1,3 4×4 −I4×4 O4×4 O4×4

[ ]

[ ]

where Uall = [ T2,1 4×4 T2,2 4×4 T2,3 4×4 O4×4 −I4×4 O4×4 ] , (15b)

[ ]

[ T3,1 4×4 T3,2 4×4 T3,3 4×4 O4×4 O4×4 −I4×4 ]12 × 24

𝑇

Z𝑇all = [Z1,0 𝑇 Z6,0 𝑇 Z11,0 𝑇 Z5,0 𝑇 Z10,0 𝑇 Z13,0 𝑇 ]24 × 1 , (15c)

2,1 U3 U1,1 U1 + U5 U2,1 U8 U1,1 U1 ,

2,1 U3 U1,1 U6 + U5 U2,1 U8 U1,1/1,2 U6 ,

2,1 U8 U1,2 U11 ,

2,1 U3 U1,1 U1 + U10 U2,1/2,2 U8 U1,1 U1 ,

2,1 U3 U1,1 U6 + U10 U2,1/2,2 U8 U1,1/1,2 U6 + U10 U2,2 U12 U1,2 U6 ,

(15d)

2,1/2,2 U8 U1,2 U11 + U10 U2,2 U12 U1,2 U11 ,

2,2 U8 U1,1 U1 ,

2,2 U8 U1,1/1,2 U6 + U13 U2,2 U12 U1,2 U6 ,

2,2 U8 U1,2 U11 + U13 U2,2 U12 U1,2 U11 .

Advances in Mechanical Engineering 11

15

10

ky1,1 ky2,1

1010

x1

x2 = 1 − x1

5

10

log10 |Δ|

100

10−5

10−10

0 2 4 6 8 10 12 14 16

𝜔

8 10

10 10

106 108

log10 |Δ|

log10 |Δ|

104 106

102

104

10.6 10.7 10.8 10.9 11 11.1 11.2 11.3 11.4 14 14.05 14.1 14.15 14.2

𝜔

𝜔

Figure 9: fMin1D function determinant of a system consisting of free-free and clamped-clamped beams coupled with two springs. 𝑘𝑦 =

100(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.3 m.

⏟⏟⏟⏟⏟ ⏟⏟Z⏟⏟⏟

all all⏟⏟ = 0. The geometric and material properties

U13 represent the beam segment transfer matrices; 12 × 1212 × 1

U2,𝑑 2,𝑐 4,𝑑 4,𝑐 7,𝑑 7,𝑐 9,𝑑

1,1 , U1,1 , U2,1 , U2,1 , U1,2 , U1,2 , U2,2 , and U9,𝑐2,2 are the (steel) of the multi-level beam system are given as follows:

transfer matrices of a system between two connection 𝐿 = 1 m, 𝑏1 = 𝑏2 = 𝑏3 = 0.02 m, ℎ1 = ℎ2 = ℎ3 = 0.01 m,

points located between two beams (12c). U2/7,𝑑 4/9,𝑑 𝐸1 = 𝐸2 = 𝐸3 = 𝐸 = 2.069 × 1011 N/m2 , and

1,1/1,2 and U2,1/2,2 𝜌1 = 𝜌2 =𝜌3 = 𝜌 = 7850 kg/m3 . Here, 𝑏𝑖 and ℎ𝑖 represent

are the transfer matrices for two systems connected with

three-level through three connection points (12d). Applying the width and height of the beam, respectively. The cross-

sectional area and the moment of inertial of the beam

the boundary conditions, that is, Z1,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇1,0 ;

cross-section are 𝐴 𝑖 = 𝑏𝑖 × ℎ𝑖 and 𝐼𝑖 = 𝑏𝑖 × ℎ𝑖3 /12, respectively.

Z6,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇6,0 ; Z11,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇11,0 , Z5,0 =

For comparison, the lowest three “exact” dimensionless

[0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇5,0 , Z10,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇10,0 , and Z13,0 = frequency values 𝜔 for the single pinned-pinned beam

[0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇13,0 , the overall transfer equation (15a) reduced that might be found in structural or vibration text books

12 Advances in Mechanical Engineering

16 16

14 14

12 12

10 10

8 8

𝜔

𝜔

6 6

4 4

2 2

0 0

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

x1 (m) x1 (m)

(a) (b)

Figure 10: Dimensionless frequency parameter 𝜔 values as a function of the spring locations (𝑥1 = 0 → 0.5 m) for a system consisting of

free-fee and clamped-clamped beams coupled by two springs. (a) 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ); (b) 𝑘𝑦 = 1000(𝐸𝐼/𝐿3 ).

Transfer direction

1 3 5

0 0 Level 1

Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0 Chain 1

ky1,1 2 ky2,1 4

c1,1 c2,1

6 8 10

0 0 Level 2

Z6,0 Z6,2 Z2,8 Z8,4 Z4,10 Z10,0 Chain 2

or or or or

Z6,7 Z7,8 Z8,9 Z9,10

ky1,2 7 ky2,2 9

c1,2 c2,2

11 12 13

0 0 Level 3

Z11,0 Z11,7 Z7,12 Z12,9 Z9,13 Z13,0 Chain 3

x1 L − 2x1 x1

Figure 11: Chain TMM-MS: state vectors and transfer direction of pinned-pinned multi-level beam connected by spring/dashpot systems.

1010

105

100

10−5

log10 |Δ|

10−10

L 10−15

10−20

10−25

10−30

0 1 2 3 4 5 6 7 8 9 10 11 12

𝜔

Figure 12: fMin1D function determinant to evaluate the lowest three 𝜔 for undamped, uncoupled pinned-pinned three-beam.

Advances in Mechanical Engineering 13

ky1,1 ky2,1

10

ky1,2 ky2,2 9

8

x1 7

x2 = 1 − x1 6

5

𝜔

L

4

3

9.54

2

9.52

1

9.5 0

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

9.48

𝜔

x1 (m)

9.46 (a) ky = 25(EI/L3 )

9.44

9.42 1010

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

x1 (m)

105

(a) ky = 25(EI/L3 )

log10 |Δ|

6.6

100

6.55

6.5

6.45 10−5

6.4

𝜔

6.35 10−10

0 2 4 6 8 10 12

6.3

𝜔

6.25

(b) ky = 25(EI/L3 ) and x1 = 0.25 m

6.2

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

x1 (m)

(a) ky = 25(EI/L3 )

102

100 100

10−2 10−2

log10 |Δ|

log10 |Δ|

10−4 10−4

10−6 10−6

6.25 6.3 6.35 6.4 6.45 6.5 6.55 6.6 𝜔

𝜔

(b) ky = 25(EI/L3 ) and x1 = 0.25 m

(b) ky = 25(EI/L3 ) and x1 = 0.25 m

Figure 13: (a) The lowest three 𝜔 values as a function of the spring systems locations (𝑥1 = 0 → 0.5 m) for undamped, coupled pinned-pinned

three-beam; (b) fMin1D function determinant to evaluate the dimensionless frequency parameter 𝜔 for 𝑘𝑦 = 25(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.25 m.

14 Advances in Mechanical Engineering

ky1,1 ky2,1

c1,1 c2,1

ky1,2 ky2,2

c1,2 c2,2

x1

x2 = 1 − x1

2000 1000

1500 800

600

1000

400

500

200

𝜆r = 𝛿

𝜆r = 𝛿

0 0

−500 −200

−400

−1000

−600

−1500

−800

−2000 −1000

−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600

𝜆i = 𝜔 (rad/s) 𝜆i = 𝜔 (rad/s)

(a) (b)

500 600

400

500

300

200 400

100

300

𝜆r = 𝛿

𝜆r = 𝛿

0

200

−100

−200 100

−300

0

−400

−500 −100

−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600

i i

𝜆 = 𝜔 (rad/s) 𝜆 = 𝜔 (rad/s)

(c) (d)

Figure 14: fMin2D first step: damping (𝜆𝑟 = 𝛿) range scanning of a specific frequency band (𝜆𝑖 = 𝜔 ≤ 1500 rad/sec) for damped, coupled

pinned-pinned three-beam, (a) [−2000, 2000], (b) [−1000, 1000], (c) [−500, 500], and (d) [−100, 600].

Advances in Mechanical Engineering 15

8 frequency vibration versus the spring systems location

resembles a half or full sine wave with insignificant variation

6 in the frequency ranges. The strong ability of fMin1D

4 algorithm is shown in Figure 13(b) by capturing the very

narrow regions of lower peaks.

log10 |Δ|

0 making the model more complicated. The two spring/dashpot

systems coupled with three-beam are located at position 𝑥1 =

−2 0.25 m. The spring/dashpot system in parallel is similar to

−4 viscoelastic material model. The given values of damping are

600 𝑐1,1 = 𝑐2,1 = 𝑐1,2 = 𝑐2,2 = 𝑐 = 5 √𝐸𝐼 𝑚/𝐿2 . Next, it is

400 1500

200 1000 the turn to implement fMin2D algorithm [17] for complex

0 500

𝜆r= −200 −500 0 eigenproblem solution. For undamped given system as it is

𝛿 (rad/ s)

𝜆 =𝜔

i

computed above, 𝜔 values do not exceed more than 10, which

is equivalent to (𝜆𝑖 = 𝜔 ≤ 1500 rad/sec). However, the

damping (𝜆𝑟 = 𝛿) is unknown in range. Therefore, within

the required system frequency band, a couple scanning

6 grids are demanded to understand and explore the damping

4 ranges of the system. Figure 14 shows different scanning

2 grids for damping ranges as follows: (a) [−2000, 2000], (b)

0 [−1000, 1000], (c) [−500, 500], and (d) [−100, 600]. It is

log10 |Δ|

−4 randomly on the surface. These regions represent the complex

−6 eigenvalues of the system. In the following searching complex

−8 roots, the damping range (Figure 14(d)) is considered. MS-

−10 TMM for damped, coupled three-beam results are shown in

600 Figure 15 and Table 2 after using fMin2D algorithm coded

400 1500 under Matlab environment with the following input data:

2000 1000

𝜆r =𝛿 0 0

500

s) ([−1,

⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ [−100, 600], 𝑁𝑥0 = 35, 𝑁𝑦0 = 20, 𝜀𝑥 = 10−8 ,

1500], ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

(rad/

𝜆 =𝜔

i

𝜆𝑖 𝜆𝑟

−8

𝜀𝑦 = 10 ).

Figure 15: fMin2D second step: function determinant to evaluate

Chain MS-TMM for undamped/damped coupled multi-

the eigenvalues (𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 ) for a damped, coupled pinned-

pinned three-beam. level beam with the suggested scenario and the two novel

recursive scanning algorithms provide a closed-form solu-

tion, not only presents the principles of the vibration problem

but also shed light on practical applications. Since the

are 3.1416, 6.2832 and 9.4248. MS-TMM results for solution is almost exact, it allows a complete understanding

undamped, uncoupled pinned-pinned three-beam are of a problem.

3.141592, 6.28318, and 9.424777 as shown in Figure 12

using fMin1D algorithm. For undamped elastically coupled

three-beam, Figure 13 shows that 𝜔 values were plotted 6. Conclusions

versus the location movement of spring systems. Systems

Starting from the principle of mechanics and the elementary

2 and 4 (𝑘𝑦1,1 = 𝑘𝑦1,2 = 𝑘𝑦 = 25 (𝐸𝐼/𝐿3 )) and systems 7 and formulations for the flexible beam, the free vibration analysis

9 (𝑘𝑦1,2 = 𝑘𝑦2,2 = 𝑘𝑦 = 25 (𝐸𝐼/𝐿3 )) are moving opposite to of laterally vibrating system made up of a multi-level Euler-

each other from (𝑥1 = 0 → 0.5 m). The system shows Bernoulli beam to which spring/dashpot systems are attached

symmetric and antisymmetric vibrations. For symmetric across the span is performed using one of the Transfer Matrix

vibration, the frequency parameters 𝜔1,1 , 𝜔1,2 , and 𝜔1,3 Method of Linear Multibody Systems (MS-TMM) scenarios.

(where 𝑙 and 𝑝 are the beam number and vibration mode, Although the number of coupling springs or spring/dashpot

resp.) are constant throughout the spring systems movement systems considered in the examples given was limited to

as seen in Figure 13(a). The antisymmetric of the first mode three, there is no inherent difficulty in extending the current

vibration 𝜔2,1 and 𝜔3,1 started with equally values as 𝜔1,1 at method to solve the problems of vibration of systems con-

𝑥1 = 0 (spring systems are rigidly mounted at the support sisting of any number of uniform/nonuniform beams with

ends) and increased significantly at 𝑥1 = 0.5 m (both spring different boundary conditions and coupled with any number

systems are in the same position). However, it is evident of spring/dashpot systems. The numerical results obtained to

that 𝜔2,1 and 𝜔3,1 are sensitive to the variation of spring reveal that the eigenfrequencies calculated by this method are

positions. The results presented of the other antisymmetric in very good agreement with those obtained by the published

vibration modes, 𝜔2,2 , 𝜔3,2 , 𝜔2,3 , and 𝜔3,3 , indicate that as literature. Moreover, MS-TMM is encouraging for further

16 Advances in Mechanical Engineering

Table 2: Chain MS-TMM eigenvalues results of damped, coupled Journal of Sound and Vibration, vol. 269, no. 1-2, pp. 431–438,

pinned-pinned three-beam. 2004.

[8] M. Abu-Hilal and N. Beithou, “Free transverse vibrations of a

𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 Chain MS-TMM results using fMin2D

triple-beam system,” Journal of Mechanical Engineering, vol. 58,

𝜆𝑖 (rad/sec)

no. 1, pp. 30–50, 2007.

𝜆1 −2.463512 × 102 + 4.010226 × 101 𝑖 [9] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,

𝜆2 −2.074369 × 10−9 + 1.462698 × 102 𝑖 Germany, 1990.

𝜆3 −7.409510 × 101 + 1.652322 × 102 𝑖 [10] W. Schiehlen, “Multibody system dynamics: roots and perspec-

𝜆4 −4.577982 × 102 + 4.615502 × 102 𝑖 tives,” Multibody System Dynamics, vol. 1, no. 2, pp. 149–188,

1997.

𝜆5 −2.074369 × 10−9 + 5.850790 × 102 𝑖

[11] J. Wittenburg, Dynamics of Systems of Rigid Bodies, Edited by:

𝜆6 −1.484833 × 102 + 5.861790 × 102 𝑖 B. G. Teubner, Stuttgart, Germany, 1977.

𝜆7 −2.159398 × 102 + 1.243120 × 103 𝑖 [12] J. Wittenburg, Dynamics of Multibody Systems, Springer, Berlin,

𝜆8 −7.448404 × 101 + 1.311750 × 103 𝑖 Germany, 2nd edition, 2008.

𝜆9 −2.074369 × 10−9 + 1.316428 × 103 𝑖 [13] A. A. Shabana, Dynamics of Multibody Systems, Cambridge

University Press, New York, NY, USA, 3rd edition, 2010.

[14] A. A. Shabana, Computational Dynamics, John Wiley & Sons,

investigations of more complex multibody systems of this New York, NY, USA, 3rd edition, 2010.

type with rigid bodies due to simplicity in the formulation [15] A. A. Shabana, “Flexible multibody dynamics: review of past

of the transfer equation, being systematic to apply, and being and recent developments,” Multibody System Dynamics, vol. 1,

easy to program. no. 2, pp. 189–222, 1997.

[16] X. Rui, L. Yun, Y. Lu, B. He, and G. Wang, Transfer Matrix

Method of Multibody System and Its Application, Science Press,

Conflict of Interests Beijing, China, 2008, (Chinese).

The authors declare that there is no conflict of interests [17] D. Bestle, L. K. Abbas, and X. Rui, “Recursive eigenvalue

search algorithm for transfer matrix method of linear flexible

regarding the publication of this paper.

multibody systems,” Multibody System Dynamics, 2013.

[18] L. K. Abbas, M. J. Li, and X. Rui, “Transfer matrix method for the

Acknowledgments determination of the natural vibration characteristics of realistic

thrusting launch vehicle—part I,” Mathematical Problems in

The research was supported by the Research Fund for Engineering, vol. 2013, Article ID 764673, 16 pages, 2013.

the Doctoral Program of Higher Education of China

(20113219110025), the Natural Science Foundation of China

Government (11102089), and the Program for New Century

Excellent Talents in University (NCET-10-0075).

References

[1] S. Inceoğlu and M. Gürgöze, “Bending vibrations of beams

coupled by several double spring-mass systems,” Journal of

Sound and Vibration, vol. 243, no. 2, pp. 370–379, 2001.

[2] S. Kukla, “Free vibration of the system of two beams connected

by many translational springs,” Journal of Sound and Vibration,

vol. 172, no. 1, pp. 130–135, 1994.

[3] S. Kukla and B. Skalmierski, “Free vibration of a system

composed of two beams separated by an elastic layer,” Journal

of Theoretical and Applied Mechanics, vol. 32, no. 3, pp. 581–590,

1994.

[4] Z. Oniszczuk, “Free transverse vibrations of elastically con-

nected simply supported double-beam complex system,” Jour-

nal of Sound and Vibration, vol. 232, no. 2, pp. 387–403, 2000.

[5] Z. Oniszczuk, “Forced transverse vibrations of an elastically

connected complex simply supported double-beam system,”

Journal of Sound and Vibration, vol. 264, no. 2, pp. 273–286,

2003.

[6] H. V. Vu, A. M. Ordóñez, and B. H. Karnopp, “Vibration of a

double-beam system,” Journal of Sound and Vibration, vol. 229,

no. 4, pp. 807–822, 2000.

[7] M. Gürgöze and H. Erol, “On laterally vibrating beams carrying

tip masses, coupled by several double spring-mass systems,”

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 365265, 9 pages

http://dx.doi.org/10.1155/2014/365265

Research Article

Case Study on Human Walking during Wearing a Powered

Prosthetic Device: Effectiveness of the System ‘‘Human-Robot’’

Svetlana Grosu,1 Pierre Cherelle,1 Chris Verheul,2 Bram Vanderborght,1 and Dirk Lefeber1

1

Department of Mechanical Engineering, Vrije Universiteit Brussel, 1050 Brussels, Belgium

2

SayField International, Broeksloot 10, 3474 HP Zegveld, The Netherlands

Copyright © 2014 Svetlana Grosu et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

It is known that force exchanges between a robotic assistive device and the end-user have a direct impact on the quality and

performance of a particular movement task. This knowledge finds a special reflective importance in prosthetic industry due to

the close human-robot collaboration. Although lower-extremity prostheses are currently better able to provide assistance as their

upper-extremity counterparts, specific locomotion problems still remain. In a framework of this contribution the authors introduce

the multibody dynamic modelling approach of the transtibial prosthesis wearing on a human body model. The obtained results are

based on multibody dynamic simulations against the real experimental data using AMP-Foot 2.0, an energy efficient powered

transtibial prosthesis for actively assisted walking of amputees.

control architecture [2].

A definition of the functionalities/duties between a human Compared with healthy persons, walking amputees

and a robotic device, also the organization of their interac- require 10–60% more metabolic energy depending on walk-

tion, basically, includes a number of different criteria that ing speed, physical individual properties, cause of amputa-

influence the effectiveness of the “human-robot” system. tion, amputation level, and prosthetic intervention charac-

The hierarchy of criteria importance depends on a general teristics. Furthermore, amputees walk at 11–40% slower self-

approach in a certain domain application. Generally, the selected gait speed than do persons with intact limbs [3, 4].

requirements in a robotic device design should assure the To date, commercially available prostheses comprise spring

maximum economical effectiveness of the system in combi- structures that store and release elastic energy throughout

nation with a personal security of the end-user. each walking stance period [5]. Due to their passive nature,

Robots for physical assistance to humans are meant such prostheses are unable to generate more mechanical

to reduce fatigue and stress, increase human capabilities energy than what is stored during each walking step. Also,

in terms of force, speed, and precision, and improve in experiences in clinical environment indicate that transtibial

general the quality of life. In other words, the crucial goal (TT) amputees suffer from a nonsymmetrical gait while

of a robot for physical human-robot interactions (pHRI) is wearing a completely passive prosthesis [6]. In distinction,

a generation of supplementary forces to overcome human the human ankle performs positive net work and has a greater

physical limits. Moreover, the human can bring experience, peak power over the stance period, especially at moderate to

global knowledge, and understanding for a correct execution fast walking speeds [7, 8].

of movements [1]. In case of assistive devices, an improved In literature several engineering challenges still slowing

analysis of the problems related to the physical interaction down the further development of a powered ankle-foot

with robots becomes mandatory. Also, in a special perspec- prosthesis [9, 10] are discussed. In the field of prosthetic leg

tive for the interaction with humans should be considered design, a critical objective is to progress a powered ankle

2 Advances in Mechanical Engineering

ankle. A study of TT prosthetics research contributions & ballscrew

shows that none of the commercially available devices are

skillful of mimicking an human ankle-foot complex. With

current actuator technology, it is challenging to build an PO spring

ankle-foot prosthesis that matches the size and weight of

the human ankle-foot structure but still provides sufficient Locking system

stance-period work and instantaneous power output to drive

an amputee. In 1998, Klute and colleagues [11] were the

first to build a powered ankle-foot prosthesis efficient in

performing net positive work. Their device employed a

pneumatic actuation strategy with an off-board power supply.

More recent work has focused on the design of energetically

autonomous powered systems [12–14]. In the growing field

of rehabilitation robotics, the use of compliant actuators is Lever arm

becoming a standard where accurate trajectory tracking is not PF spring

required. The advantages of such actuators are represented by

safely interaction with the patient and large forces absorption

Figure 1: Design representation of AMP-Foot 2.0.

during gait. In the particular case of trans-tibial prostheses,

compliance of the actuation provides even more advantages.

Besides shock absorption in case of collision with objects

during walking, energy provided by the actuator (e.g., electric

motor) can be stored into its elastic element (e.g., spring in

series). The current state-of-the-art in powered ankle-foot loading a “push-off (PO)” spring during the complete stance

prostheses with focus on devices using compliant actuators phase. The prosthesis includes a locking mechanism which

has been presented in [15]. provides the energy implementation into the PO spring and

In the next sections, the authors are focused on the can be delayed and released at push-off. This way, full torque

influences on normal human gait of the forces that are and power required for locomotion can be obtained with less

generated by a motor and forces that are stored in and then power from the actuator.

released by springs, also the reaction of the mechanism as a Structurally, the device consists of three bodies: a leg, a

whole in interaction with user. Then, the comparison between foot, and a lever arm, pivoting around the ankle axis; see

the real “human-robot” setup and virtual model is discussed. Figure 2. As mentioned before, the system comprises 2 spring

sets: a PF and a PO spring set. The PF spring is placed

between a fixed point 𝑝 on the foot and a cable that runs

2. The Ankle Mimicking Prosthetic- (AMP-) over a pulley 𝑎 to the lever arm at point 𝑏 and is attached

Foot 2.0: Background to the lever arm at point c, while the PO spring is placed

between the motor-ballscrew assembly and a fixed point 𝑑

This section is dedicated to a summary description of the

on the lever arm. A critical part of this device is the locking

prosthesis used in the study. The section includes presenta-

mechanism, that is meant to bear high forces while being as

tion of the AMP-Foot 2.0 mechanical design and validation

compact and lightweight as possible and is represented by a

part in the framework of real experiments, with an amputee

four bar linkage moving in and out of its singular position.

subject participation.

The working principle of such a system has been proposed in

[17].

2.1. Design of the AMP-Foot 2.0. The AMP-Foot 2.0, see The locking mechanism which provides a rigid connec-

Figure 1, claims a new energy concept, based on a principle tion between the leg and the lever arm when energy is injected

of optimal power distribution which is elaborated in [16]. into the system is not represented in Figure 2.

Basically, the working principle is similar to existing powered In order to illustrate the behaviour of the AMP-Foot 2.0

assistive devices, except that the actuator is working during prosthesis, the gait cycle is divided in 5 phases starting with

the complete stance phase. Gravitational potential energy is a controlled plantar flexion from heel strike (HS) to foot flat

gradually stored into a series elastic element, in time which (FF); see Figure 3. A step is initiated by touching the ground

the drive still has to provide the same torque and power with the heel. During this phase the foot rotates with respect

output. But there is approximately 3 times more time available to the leg, until 𝜃 (= 𝜙) reaches approximately −5∘ . This is

to generate the power by the electric motor. As a consequence, followed by a controlled dorsiflexion phase ending in push-

the overall power rating of the actuator can be divided by off at heel off (HO), during which a generation of propulsive

approximately the same amount, reducing the weight and size forces by Soleus and Gastrocnemius muscle groups will have

of the drive power considerably. place. In the late stance phase, the torque produced by the

The functional principle of the AMP-Foot 2.0 uses a ankle decreases until the leg enters the swing phase at toe off

“plantar flexion (PF)” spring, which stores energy from the (TO). Once the leg is engaged in the swing phase, the foot

controlled dorsiflexion phase of stance. An electric actuator is resets the locking mechanism.

Advances in Mechanical Engineering 3

and ballscrew the need to develop simulations and analysis for mechanical

𝜃

systems that consist of interconnected bodies. Simulation

Leg is meant to describe and analyze the behaviour of various

system configurations, ask what-if questions about the real

system, and optimize the structural design.

Due to the high complexity of modern robotic systems,

PO spring

almost any research conducted in the area of robotics

k2

can benefit from a simulation of the system behaviour

L1 before experiments on a real platform take place [20]. The

Lever arm computational modelling including anatomic, physiologic,

L2 L3 A d

and engineering analyses serve to study various activities

c L4 in a normal and pathological condition of humans. Com-

puter simulations represent an effective, faster, and cheap-

PF spring a

k1

est approach than experiments, which necessarily consume

b

physical resources. Computer modelling is considered the

p 𝜙 most effective, when employed in combination with real

experiments.

Foot

3.1. Modelling Approach for AMP-Foot 2.0. A 3D model was

developed in conformity with the real mechanical properties

of AMP-Foot 2.0 device. In other words, all material and

The more detailed description of AMP-Foot 2.0 dynamic structural characteristics are preserved relative to the existing

behaviour during a complete gait cycle can be found under prototype. In order to model and simulate the AMP-Foot

contribution [16]. 2.0 prosthesis, MSC Adams 3D multibody dynamics software

was used [21]. The geometries of the AMP-Foot 2.0 compo-

2.2. Validation of the AMP-Foot 2.0 Device. The experiments nent parts were exported from Autodesk Inventor software

with participation of one disabled patient were effectuated in MSC Adams environment and were converted into a set

by the research group of Vrije Universiteit Brussel [18] of Adams/View geometry elements. This importing approach

and, in present paper, the authors are referring to results reduces the need to recreate geometry primitives within

obtained from those experiments. Briefly, in the performed Adams software and enhances the ability to realistically

experiments one transfemoral amputee subject of 75 kg was view the behavior of complicated mechanical systems. After

considered and the experiment was divided in three trials: (1) importing, the co-rrelated constraints between all geometry

walking at a self-selected speed on a treadmill; (2) walking parts were defined and applied to the model. Two spring

speed raised to a faster cadence; (3) an overground walking sets (PF and PO), which play a critical role in AMP-Foot

at self-selected speed; see Figure 4 [19]. 2.0 dynamic behaviour, have been modelled according to

Conform analysed data acquired during the experiments, the real design stiffness and damping characteristics. The

and compared to existing powered prosthetic devices, it was stiffness of the PF spring is about 300 N/mm and for every PO

observed that the AMP-Foot 2.0 prosthesis can improve an spring the value of stiffness is modeled as 60 N/mm. Damping

amputee gait, in conditions when little power is required values were determined experimentally, and were found to

for the actuation. Since the present contribution is focused be 10 Ns/mm for PF spring set and 1.2 Ns/mm for each PO

on modelling and simulation approach of human-robot spring respectively. The FF phase is considered as the initial

system, all additional details regarding the real experiment position of the model. In the beginning stage of this study the

design, prosthesis behaviour, power consumption, and torque model dynamic behaviour, see Figure 5, was analysed without

characteristics are given in [18]. human body model and was elaborated in [22]. The AMP-

Foot 2.0 model complexity can be appreciated and visualised

in Figure 5, where moving parts and constraints of the model

3. Modelling and Simulations in are presented. There are 11 degrees of freedom (DoF) in the

Virtual Environment AMP-Foot 2.0 system.

The model includes 2 actuation forces: one represents

In this section, the authors present modelling methods of the motor actuator, which is constantly pulling up the PO

the AMP-Foot 2.0 device. Then, modelling and simulation springs from HS to TO period of the gait cycle; the second

results of an amputee walking using the assistive device force is responsible for triggering the locking mechanism.

are described. Further, the comparison of a normal walking The control of the force actuation period is based on timing

(healthy person) and the same person wearing the AMP-Foot approach, implemented in the model in form of a STEP

2.0 is effectuated. The section is concluded with discussion on function.

obtained results from simulations and the real experiments, The STEP function approximates the Heaviside step func-

referred in Section 2.2. tion with a cubic polynomial and has the following format:

4 Advances in Mechanical Engineering

150

Torque (Nm)

100 HO

50

FF

0 TO HC

−25 −20 −15 −10 −5 0 5 10 15

Angle (deg)

AMP-Foot 2 ankle angle data

AMP-Foot 2 ankle angle average

D. winter data

(a) (b)

axial stiffness, and strain value, are defined with respect to

the ones used in real mechanical system and were determined

experimentally. These properties are presented in Figure 6(c)

and can be easily modified for further experimental purposes

within the framework of Graphical User Interface (GUI),

developed in MSC Adams software; see Figure 6(c); the mass

of the cable elements is assumed to be negligible.

Figure 5: The AMP-Foot 2.0 model composition and DoF. 3.2. Modelling and Simulation of Human Walking during

Wearing the AMP-Foot 2.0 Prosthesis. There are specialized

commercially available simulation tools that can be used for

analysis of human walking, such as AnyBody [23], OpenSim

STEP (𝑥, 𝑥0 , ℎ0 , 𝑥1 , ℎ1 ). It has continuous first derivatives. Its

[24], SIMM [25], and LifeMod [26]. However, these tools

second derivatives are discontinuous at 𝑥 = 𝑥0 and 𝑥 = 𝑥1 ,

cannot be used for analyzing human-robot interactions,

where 𝑥 is the independent variable (time, in present model),

which is becoming an essential requirement for modeling and

𝑥0 is a real variable that specifies the 𝑥 value at which the

simulation of robotic systems as their physical interaction

STEP function begins, 𝑥1 is a real variable that specifies the 𝑥

with humans becomes more complex. In general, the model-

value at which the STEP function ends, ℎ0 is the initial value

ing and simulation phase of the development of such systems

of the step, and ℎ1 is the final value of the step.

is becoming demanding. Since the robot and robot-human

The arrows, that can be observed in Figure 6, are repre-

interactions increase in complexity, the simulation with a

senting the actuation and tension forces during the simula-

single simulation tool is not effective anymore. Therefore,

tion.

in this research work the combination of MSC Adams with

Due to cables nonlinear geometric properties and a

LifeMode plug-in was used as the most appropriate.

complex static and dynamic behaviour, the challenge was to

model the cable segments which play an important role in

a force transmission system of the AMP-Foot 2.0. The cable 3.2.1. Methods. The creation of human models begins by

transmission system was modelled using TKC toolbox feature generating a base human segment set, followed by joints,

and the tension forces inside of the system were obtained [22]. soft tissues, and contact elements between the model and the

Advances in Mechanical Engineering 5

(a)

(b) (c)

Figure 6: (a, b) Simulation of AMP-Foot 2.0 behaviour. (c) Cable properties GUI.

(a) (b)

environment. In order to study the influences of AMP-Foot lower body (two legs and pelvis) which is rigidly attached

2.0 device on normal overground walking the mechanical to the right extremity AMP-Foot 2.0 prosthesis. Once the

model of the human body was built using the Lifemodeler segments of the lower body are established, joints are created

plug-in in framework of MSC Adams environment. This between the segments. Along an inverse-dynamics simu-

combination of tools supports the analysis of the “human- lation, joints learn angulation patterns while the model is

robot system” effectiveness and mutual interaction. The being driven by the motion capture data. The nominal joints

considered model, see Figure 8, includes a model of human’s stiffness for both legs at the hip, knee, and the ankle is

6 Advances in Mechanical Engineering

20.0

10.0

Angle (deg)

0.0

−10.0

−20.0

−30.0

0.0 0.5 1.0 1.5 2.0

Time (s)

AMP-Foot 2.0 ankle angle during normal walking

red graph with AMP-Foot 2.0; blue line using the healthy foot.

Figure 8: Model of human lower limbs cosimulated with AMP-Foot

2.0.

assumed to be 1e5 Ns/mm and the damping to be 1e4 Ns/mm. implemented in MSC Adams.

The joint consists of a triaxis hinge and passive or active For simulation, driving of the human body is based

forces acting on each of the three degrees of freedom. They on captured motion data obtained through marker-based

are implemented as an assembly of two “virtual” bodies motion capture system. Motion capture (MOCAP) systems

of negligible mass and inertia and three revolute joints; track the trajectories of markers attached at various locations

see Figure 7. For every joint there are three axes: sagittal, on the body. The marker trajectories are then used to train

transverse, and frontal. The Lifemode software offers the the human model. During the training, the response of the

possibility to define every axis in a way suitable to certain body is recorded and later used for a forward dynamics

application. The axes of lower limb joints in present human simulation. Marker trajectory data is used to drive elements

model are defined as in the table in Figure 7. called “motion agents,” which are massless parts fixed to the

The settings specified for every joint axis in the model are body segments using spring elements.

interpreted as follows. The right foot part was removed and replaced by the

AMP-Foot 2.0 model and connected to the right lower leg

Driven. Kinematically driven using data from a driv- by a fixed joint; see Figure 8. Then, contact forces between

ing spline. prosthesis toe, heel parts, and the ground were defined as an

Passive. A torsional spring force with user-specified IMPACT function model. The IMPACT function represents

stiffness, damping, angular limits, and limit stiffness a simple model for contacts. It evaluates a force that turns

values. These joints are used in an inverse dynamics on when a distance falls below a nominal free length (when

analysis to record the joint angulations while the two parts collide). The force has two components: a spring

model is being manipulated with motion agents. or stiffness component and a damping or viscous compo-

nent. The stiffness component opposes the penetration. The

Hybrid III. The Hybrid III strength model is created

damping component of the force is a function of the speed of

for the individual joint axis with a user-specified scale

penetration and opposes the direction of relative motion.

value. The Hybrid III strength model is based on

physical measurements of an actual crash dummy.

The strength model consists of nonlinear stiffness, 3.2.2. Results. The obtained simulation data were filtered

damping, and frictional values and also includes joint by low pass filter and processed accordingly. In Figure 9

limit stop stiffness with hysteresis. the ankle angle data of the human body model with the

Servo. This selection creates a trained PD-servo type same physical properties as in real experiment are illustrated

controller on the joint axis. The joint is commanded (weight 75 kg, height 175 cm, male) during normal over-

to track an angular history spline with a user-specified ground walking. In the first simulation the human was

gain on the error between the actual angle and the walking without AMP-Foot 2.0 prosthesis, using a normal

commanded error. A user-specified derivative gain is human foot model (see Figure 9, blue dot-line).

specified to control the derivative of the error. Afterwards, the normal foot was removed and replaced by

the AMP-Foot 2.0 model and, therefore, we can follow from

It is important to notice that the mass (75 kg) of the whole the graph (see red line representation) the consequences for

human body is considered in simulation, although just the this change in gait kinematics. Please notice that in the first

lower part is represented. The human body model generated slot of time the AMP-Foot 2.0 prosthesis is passive; there is

by the LifeMod plug-in stores the joint motion trajectories in no force acting on the motor part and pulling the spring sets.

MSC Adams. These trajectories are used during simulation as The single actuation is coming from the human leg during

Advances in Mechanical Engineering 7

80000 25.0

70000 20.0

60000

Torque (N-mm)

15.0

50000 10.0

Angle (deg)

40000 5.0

30000 0

20000 −5.0

10000 −10.0

0.0

0.0 0.5 1.0 1.5 2.0 −15.0

Time (s) −20.0

−25.0

Torque right Hip 0 10 20 30 40 50 60 70 80 90 100

Torque right Knee Stride (%)

Torque ankle AMP Foot 2.0

Figure 12: Ankle angle comparison: red line: simulation of human

Figure 10: Torque values for right limb. model during wearing the AMP-Foot 2.0; blue line: simulation of

human model without AMP-Foot 2.0; green line: real experiment

data of subject during wearing the AMP-Foot 2.0 device.

60.0

50.0

40.0

Angle (deg)

20.0

by this way creating an additional torque motion.

10.0

0.0 4. Discussion

−10.0

0.0 0.5 1.0 1.5 2.0 It is crucial for design and control developing of assistive

Time (s) devices to have a model of the robotic system interacting

with the human before the actual physical system is ready

Figure 11: Knee angle: red line, with AMP-Foot 2.0 model; grey line, for the use. Sometimes, performing the real experiments is

with human foot model. too exhausting, expensive, or almost impossible and in this

case modeling and simulation approach is considered very

effective. In the framework of this paper the authors have

walking only. In the grey zone, see Figure 9, the AMP-Foot 2.0 researched in virtual environment the differences between

becomes actuated and, as a result, we can observe an increase human walking during wearing an assistive device and

of the ankle angle, due to the provided push off propulsion normal walking of the healthy person. In both simulation and

force of the prosthetic device. real-life testing the same human model was considered. This

In Figure 10 are illustrated the obtained torque values experiment is impossible to be repeated in real life. It was

of the right lower limb with the AMP-Foot 2.0 prosthesis expected to have some differences in the ankle angle data,

attached. but the main conclusion is that the human walking is not

Also, a slight difference was observed at the knee level affected in negative way while wearing the AMP-Foot 2.0. The

joint. In case the AMP-Foot 2.0 is attached to the human powered device is providing the human leg with additional

leg model, the authors noticed a small knee angle increase; propulsion force that helps locomotion.

see Figure 11. This occurrence can be explained by the lower If we will refer to real experiments, performed with

compliancy in the TO moment. Then the locking mechanism an amputee person with comparison to results obtained

opens, providing the additional propulsion force to the leg. from simulations with human model during wearing the

As was discussed before, the authors were interested to AMP-Foot 2.0, we can notice many common similarities in

make a comparison between data from simulations with the walking pattern. Even, if the gait kinematics of people is

data, based on real experiments; see Figure 12. individual, (in simulations a similar subject with the same

Conform Figure 12, the time-based data on level ground physical characteristics was considered) the obtained results

walking, one can observe the similarities between the angle remain valid and can be interpreted. If one will do the

ankle data of the human model wearing the AMP-Foot 2.0 comparison between the normal walking simulation and

device and the walking pattern of real amputee subject with both experiments which uses the AMP-Foot 2.0 prosthesis

attached prosthesis. It can be noticed for the human model (Figure 12), one will observe some difference in TO moment.

with the AMP-Foot 2.0 that there is an increase of the ankle In case of normal walking without the device the transaction

angle around HO moment. This fact can be the result of the of the foot from TO to a swing moment is more compliant. In

influence of contact forces between the heel and the ground case of simulations there can be other external factors that can

part. In other words, when the heel is touching the ground the slightly influence obtained data results, such as the simplified

8 Advances in Mechanical Engineering

contact definition (which is not so complex as in case of real [6] H. Bateni and S. J. Olney, “Kinematic and kinetic variations of

foot-ground contact). below-knee amputee gait,” Journal of Prosthetics and Orthotics,

vol. 14, no. 1, pp. 2–10, 2002.

[7] D. A. Winter, “Biomechanical motor patterns in normal walk-

5. Conclusions ing,” Journal of Motor Behavior, vol. 15, no. 4, pp. 302–330, 1983.

The main focus in the framework of the presented study [8] A. H. Hansen, D. S. Childress, S. C. Miff, S. A. Gard, and K.

P. Mesplay, “The human ankle during walking: implications for

was to inspect how the attached trans-tibial prosthesis will

design of biomimetic ankle prostheses,” Journal of Biomechan-

influence normal healthy human walking kinematics and ics, vol. 37, no. 10, pp. 1467–1474, 2004.

how close the AMP-Foot 2.0 working behaviour comes to

[9] S. K. Au, P. Dilworth, and H. Herr, “An ankle-foot emulation

a real human foot subsystem. The effectiveness of human- system for the study of human walking biomechanics,” in

robot interaction was analyzed during the tests with a real Proceedings of the IEEE International Conference on Robotics and

subject and in virtual environment. The authors admit that it Automation (ICRA ’06), pp. 2939–2945, Orlando, Fla, USA, May

is difficult to compare individuals having a different walking 2006.

pattern, even if for simulation a subject with the same physical [10] D. A. Winter and S. E. Sienko, “Biomechanics of below-knee

parameters as the one tested in real experiment was consid- amputee gait,” Journal of Biomechanics, vol. 21, no. 5, pp. 361–

ered. Moreover, the simulation comparison (subject walking 367, 1988.

with and without AMP-Foot 2.0) is almost impossible to [11] G. K. Klute, J. Czerniecki, and B. Hannaford, “Development

arrange in real life. This fact that the simulation results more of powered prosthetic lower limb,” in Proceedings of the 1st

interesting and significant. National Meeting, Veterans Affairs Rehabilitation R&D Service,

Thanks to the built human-prosthesis model further Washington, DC, USA, October 1998.

improvements on prototype design parameters can be [12] S. K. Au, Powered ankle-foot prosthesis for the improvement of

achieved and the prototype can be adapted to certain subject’s amputee walking economy [Ph.D. thesis], MIT Department of

individual properties. The result data demonstrate that AMP- Mechanical Engineering, Cambridge, Mass, USA, 2007.

Foot 2.0 prosthesis can successfully replace the human ankle- [13] J. K. Hitt, R. Bellman, M. Holgate, T. G. Sugar, and K.

foot subsystem, provided some individual adjustments are W. Hollander, “The SPARKy (spring ankle with regenerative

kinetics) project: design and analysis of a robotic transtibial

made.

prosthesis with regenerative kinetics,” in Proceedings of the 2007

ASME International Design Engineering Technical Conferences

Conflict of Interests and Computers and Information in Engineering Conference

(IDETC/CIE ’07), pp. 1587–1596, Las Vegas, Nev, USA, Septem-

The authors, Grosu Svetlana, Pierre Cherelle, Chris Verheul, ber 2007.

Bram Vanderborght, and Dirk Lefeber, declare that there is no [14] R. Versluys, G. Lenaerts, M. Van Damme et al., “Successful

conflict of interests regarding the publication of this paper. preliminary walking experiments on a transtibial amputee

fitted with a powered prosthesis,” Prosthetics and Orthotics

International, vol. 33, no. 4, pp. 368–377, 2009.

Acknowledgments [15] R. Versluys, P. Beyl, M. Van Damme, A. Desomer, R. Van Ham,

and D. Lefeber, “Prosthetic feet: state-of-the-art review and

This work has been partially supported by the European

the importance of mimicking human anklefoot biomechanics,”

Commission 7th Framework Program as part of the CORBYS Disability and Rehabilitation: Assistive Technology, vol. 4, no. 2,

(Cognitive Control Framework for Robotic Systems) project pp. 65–75, 2009.

under Grant Agreement no. FP7 ICT-270219. [16] P. Cherelle, A. Matthys, V. Grosu, B. Vanderborght, and D.

Lefeber, “The AMP-Foot 2.0: Mimicking intact ankle behavior

References with a powered transtibial prosthesis,” in Proceedings of the

IEEE International Conference on Biomedical Robotics and

[1] O. Khatib, K. Yokoi, O. Brock, K. Chang, and A. Casal, “Robots Biomechatronics, pp. 544–549, June 2012.

in human environments: basic autonomous capabilities,” Inter- [17] G. van Oort, R. Carloni, D. J. Borgerink, and S. Stramigioli,

national Journal of Robotics Research, vol. 18, no. 7, pp. 684–696, “An energy efficient knee locking mechanism for a dynamically

1999. walking robot,” in Proceedings of the IEEE International Confer-

[2] A. de Santis, B. Siciliano, A. de Luca, and A. Bicchi, “An atlas ence on Robotics and Automation, pp. 2003–2008, May 2011.

of physical human-robot interaction,” Mechanism and Machine [18] P. Cherelle, V. Grosu, A. Matthys, B. Vanderborght, and D.

Theory, vol. 43, no. 3, pp. 253–270, 2008. Lefeber, “Design and validation of the ankle mimicking pros-

[3] M.-J. Hsu, D. H. Nielsen, S.-J. Lin-Chan, and D. Shurr, “The thetic (AMP-) foot 2.0,” Transaction on Neural Systems and

effects of prosthetic foot design on physiologic measurements, Rehabilitation Engineering, 2013.

self-selected walking velocity, and physical activity in people [19] http://www.youtube.com/watch?v=tXLQq9a5kyI.

with transtibial amputation,” Archives of Physical Medicine and [20] J. G. de Jalón and E. Bayo, Kinematic and Dynamic Simulation

Rehabilitation, vol. 87, no. 1, pp. 123–129, 2006. of Multibody Systems: The Real-Time Challenge, Springer, New

[4] A. Esquenazi and R. DiGiacomo, “Rehabilitation after amputa- York, NY, USA, 1994.

tion,” Journal of the American Podiatric Medical Association, vol. [21] http://www.mscsoftware.com/.

91, no. 1, pp. 13–22, 2001. [22] S. Grosu, P. Cherelle, C. Verheul, B. Vanderborght, and D.

[5] S. Ron, Prosthetics and Orthotics: Lower Limb and Spinal, Lefeber, “AMP-Foot 2.0 prosthesis dynamic behavior, prelim-

Lippincott Williams & Wilkins, Baltimore, Md, USA, 2002. inary computational multibody dynamics simulation results,”

Advances in Mechanical Engineering 9

body Dynamics, Zagreb, Croatia, July 2013.

[23] AnyBody, http://www.anybodytech.com.

[24] OpenSim, https://simtk.org.

[25] SIMM, http://www.musculographics.com.

[26] LifeMod, http://www.lifemodeler.com.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2014, Article ID 957684, 8 pages

http://dx.doi.org/10.1155/2014/957684

Research Article

Controller Parameters Tuning Based on Transfer Matrix Method

for Multibody Systems

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Copyright © 2014 Hossam Hendy et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Transfer matrix method for multibody systems (MS-TMM) is a rife method to multi-rigid-flexible-body systems dynamics model

deduction due to that there are no needs to establish the global dynamics equations of the system. Its basic idea is transferring a

state vector between the body input(s) and output(s); this idea is close to the linear theories in control analysis and design. In this

paper, three controllers’ parameters tuning techniques for the proposed system model using MS-TMM are utilized; one technique is

applied to get the stability regions via the frequency response of MS-TMM derived model. Another technique considers a classical

PID controller design through the analysis of step input response of the system, and the last technique can be applied in both time

and frequency domains if the model has a known mathematical model. A car suspension system is considered to represent modeling

and tuning problems. In-depth study of MS-TMM with control techniques and defining the controllers’ parameters stability regions

provide an opportunity to formulate a relationship between MS-TMM and control design for novel control applications due to the

powerful strength of MS-TMM dealing with more complex problems of the controlled multibody systems.

1. Introduction based on PID control and [5] indicated that more than 97%

of regulatory controllers utilize the PID algorithm. But the

With the increments of complexity of multibody systems tuning of the controller gains is a problem because many

and the development of their design and control methods, industrial models suffer some burdens such as nonlinearities,

the need for more elegant formulations of the equations higher order, and time delay [6]. The basic idea of MS-TMM

of motion becomes an issue of paramount importance. is transferring the state vector between the body input(s)

Many methods and theories for developing the model of and output(s); this idea is close to linear theories in control

the multibody system dynamics and control are presented analysis and design. The control and feedback variables could

for such reasons. In transfer matrix method for multibody be considered in the transfer equation of MS-TMM. The

systems (MS-TMM) there are two cases to deal with control tuning of the controller parameters is a great issue to get a

element, one is to express the control force with state of stabilized system. However Ziegler-Nichols technique is one

system of previous time such as the delay control, and the of the simplest techniques of PID tuning; there are several

second is that the control force is relative to present state of disadvantages, such as many trials are needed to find the

system, such as real-time control systems. Rui et al. have trials stability limits and also not all tuning values work efficiently

to develop new controlled systems depending on deriving for all situations [7–10]. Shamsuzzoha and Skogestad (2010)

the dynamics equations using MS-TMM [1, 2]. Bestle et al. developed a new procedure for PI/PID tuning method in

reformulated the car suspension system using MS-TMM as closed-loop mode, only one closed-loop step test is required

similar as classical control theory’s structural diagram [3]. to deduce PI controller values. In this method also it is simple

Proportional-Integration-Differentiation (PID) control- to obtain the PID tuning parameters in one step [11, 12]. This

ler is prevalent in industry applications. Reference [4] indi- paper is organized as follows: the proposed system is modeled

cated that more than 90% of feedback control loops are by MS-TMM including the control law in Section 2. The

2 Advances in Mechanical Engineering

0 Disturbance X4,0

𝒁4,0

m1

𝑼g1

K2 fc

𝑼m1

C2

𝑼g

m3

x fc + 𝒁2,1 Output X1,0

r=0 + 𝑼m

C(s) 𝑬c

+ 𝒁1,0

K4 − X1,0

𝑬m

0

Figure 1: Dynamic model of controlled car. Figure 2: The structural diagram of the controlled system [3].

1 0

where U𝑗 = [ 𝑚𝑗 Ω2 1 ] and Ω is the external excitation

design of PI controller values stability region that achieves a

specified gain and phase margins, in addition to tuning the frequency.

controller parameters using different techniques, is presented For the linear spring 4, the transfer equation is

in Section 3. Simulation and results of car active suspension

Z4,𝑂 = U4 Z4,𝐼 (2)

system model with the designed controllers are provided in

Section 4. Conclusions are offered in Section 5.

with transfer matrix U4 = [ 10 −1/𝐾

1

4 ].

2. Problem Formulation be

A two-degree-of-freedom car dynamic model is established Z2,𝑂 = U2 Z2,𝐼 + E𝑐 𝐹𝑐 , (3)

to simulate the system as shown in Figure 1. It does not only

simplify the system design and analysis, but also represents

where U2 = [ 10 −(1/(𝐾21+𝑖𝐶2 Ω)) ], E𝑐 = [ 1/(𝐾2 +𝑖𝐶2 Ω) ], and 𝐹 is

𝑐

most of full car features [13]. In Figure 1, 𝑚1 is the sprung 0

the control force.

mass that represents the car body, 𝑚3 is the unsprung mass

Considering the negative PID feedback control and

which stands for the wheel, 𝐶2 is the damping coefficient,

regarding 𝑥1,0 as the measurement signal, the control force

and 𝐾2 and 𝐾4 are the springs stiffness. In this section,

takes the form

this model is deduced using MS-TMM. According to MS-

TMM, the system is portioned into many elements which are 𝑡 𝑑𝑥1,0

numbered as 1 to 4 from up to down, where body 1 and 3 𝑓𝑐 = − (𝑘𝑃 𝑥1,0 + 𝑘𝐼 ∫ 𝑥1,0 𝑑𝑡 + 𝑘𝐷 ). (4)

0 𝑑𝑡

are lumped masses, element 4 is a linear spring, while the

controlled element 2 includes spring and damper in parallel By transformation 𝑥 = 𝑋𝑒𝑖Ω𝑡 , 𝑓𝑐 = 𝐹𝑐 𝑒𝑖Ω𝑡 , it yields

connections. The boundaries at two ends are noted as 0. The

positive direction of 𝑥 axis is also shown in the figure. It 𝑘𝐼

should be mentioned that the controlled force 𝑓𝑐 serves as an 𝐹𝑐 = − (𝑘𝑃 + + 𝑖Ω𝑘𝐷) 𝑋1,0 =: −𝐶 (𝑖Ω) 𝑋1,0 . (5)

𝑖Ω

internal force of the system.

The transfer direction is stipulated as the positive direc- Meanwhile there should be

tion of 𝑥 axis. The state vector of an arbitrary connection

point of the system is defined as Z𝑖,𝑗 = [𝑋, 𝑄𝑥 ]𝑇𝑖,𝑗 , where 𝑖 and 𝑋1,0 = [1 0] [

𝑋

] =: E𝑚 Z1,0 . (6)

𝑗 stands for the sequence number of the body element and 𝑄𝑥 1,0

hinge element, respectively. Thereby, the transfer equation of

each element is discussed as follows: Regarding the transfer direction and (1)∼(3) and (5)∼(6), the

Considering the rise and fall of the road, the boundary structure diagram of the controlled system can be drawn in

end 𝑥4,0 is assumed to be in the form of 𝑥4,0 = 𝐴𝑒𝑖Ω𝑡 , where Figure 2.

𝐴 is the complex amplitude and Ω is the frequency of the From Figure 2, the overall transfer equation of the con-

movement. Thus, in a steady-state sense, the whole system trolled system can be written as

experiences the forced vibration in the sine form.

According to MS-TMM for the lumped mass 1 and 3, their Z1,0 = U1 Z2,1

transfer equations are = U1 (U2 U3 U4 Z4,0 + E𝑐 𝐹𝑐 ) (7)

Z𝑗,𝑂 = U𝑗 Z𝑗,𝐼 , (𝑗 = 1, 3) , (1) = U1 (U2 U3 U4 Z4,0 − E𝑐 𝐶 (𝑖Ω) E𝑚 Z1,0 ) .

Advances in Mechanical Engineering 3

Magnitude (dB)

0

Name Value Company −20

𝑚3 50 kg −40

𝑚1 250 kg −60

−80

𝐾2 16 kN/m −100

𝐾4 160 kN/m 0

𝐶2 −45

Phase (deg)

1500 N⋅s/m −90

−135

−180

which finally ends up with −225

−270

−1

Z4,0 = (U1 U2 U3 U4 ) [I2 + 𝐶 (𝑖Ω) U1 E𝑐 E𝑚 ] Z1,0 . (8) 100 101 102 103

Frequency (rad/s)

This can be rewritten considering the boundary condition of

Classical

the system Z1,0 = [ 𝑋0 ]1,0 , Z4,0 = [ 𝑄𝐴𝑥 ]4,0 as MS-TMM

[ ] = [ 11 12 ] [ ] . (9)

𝑄𝑥 4,0 𝑈21 𝑈22 0 1,0

Thus the frequency response function from the disturbance

to the output can be read as

3. Controller Parameters Tuning

𝑋1,0 1 This section illustrates different approaches for designing

= . (10) both PI and PID controllers and some methods for tuning

𝐴 𝑈11

their parameters. Considering the control scheme shown in

Setting 𝑘𝑃 = 𝑘𝐼 = 𝑘𝐷 = 0, one can acquire the formula for the Figure 2, at the beginning a PI controller is designed; from

uncontrolled system. (5) the designed PI controller will be at the form of

From the classical approaches for modeling the transfer

function of a passive suspension system using ordinary dif- 𝑘𝐼 𝑘𝑃 𝑠 + 𝑘𝐼

𝐶 (𝑠) = 𝑘𝑃 + = . (13)

ferential equations, the equations which describe the system 𝑠 𝑠

can be found in [14], and the transfer equation is From the previous literature reviews, instability is the disad-

𝐾4 𝐶2 𝐾 vantage of feedback control system. There is a risk that the

𝐺 (𝑠) = (𝑠 + 2 ) closed-loop system becomes unstable when using feedback.

𝑚3 𝑚1 𝐶2

Thus, analyzing the closed-loop stability of system is essential

𝐶2 𝐶2 3 requirement for feedback control system. The mean obstacle

× (𝑠4 + ( + )𝑠 is to indicate the controller parameters or gains stabilizing

𝑚3 𝑚1

region [16].

𝐾2 𝐾2 𝐾4 2 𝐾4 𝐶2 𝐾 𝐾 −1 Decomposing (12) numerator and denominator to their

+( + + )𝑠 + 𝑠+ 4 2) . even and odd parts, the equation can be rewritten as

𝑚3 𝑚1 𝑚3 𝑚3 𝑚1 𝑚3 𝑚1

(11)

𝑁𝑒 (−Ω2 ) + 𝑗Ω𝑁𝑜 (−Ω2 )

The described system elements values are given in Table 1 𝐺 (𝑗Ω) = . (14)

𝐷𝑒 (−Ω2 ) + 𝑗Ω𝐷𝑜 (−Ω2 )

[15].

For the uncontrolled case, Bode plots for both classical From the control scheme diagram shown in Figure 2, and

method and MS-TMM results are shown in Figure 3, consid- considering the control law in (13), the closed-loop character-

ering the elements values given in Table 1 and substituting istic polynomial derived in (10) can be separated and solved

into (10) after eliminating the controller gains, the transfer to zero; then the control parameters can be written as [17]

equation of MS-TMM system as a function of the excitation

𝑋 (Ω) 𝑈 (Ω) − 𝑌 (Ω) 𝑅 (Ω)

frequency Ω is 𝑘𝑃 = , (15)

𝑄 (Ω) 𝑈 (Ω) − 𝑅 (Ω) 𝑆 (Ω)

𝐺 (𝑗Ω) = (𝐾4 𝐾2 + (𝑗Ω) 𝐶2 𝐾4 )

𝑌 (Ω) 𝑄 (Ω) − 𝑋 (Ω) 𝑆 (Ω)

𝑘𝐼 = , (16)

× (𝑚3 𝑚1 Ω4 − 𝑗Ω3 𝐶2 (𝑚3 + 𝑚1 ) 𝑄 (Ω) 𝑈 (Ω) − 𝑅 (Ω) 𝑆 (Ω)

(12)

where

− Ω2 (𝐾2 𝑚3 + 𝐾2 𝑚1 + 𝐾4 𝑚3 )

−1

𝑌 (Ω) = −Ω𝐷𝑒 (−Ω2 ) , 𝑄 (Ω) = −Ω2 𝑁0 (−Ω2 ) ,

+ (𝑗Ω) 𝐾4 𝐶2 + 𝐾4 𝐾2 ) .

This equation represents the open loop uncontrolled part of 𝑋 (Ω) = Ω2 𝐷0 (−Ω2 ) , 𝑆 (Ω) = Ω𝑁𝑒 (−Ω2 ) , (17)

the system enclosed in the solid blue line in Figure 2; results

from Figure 3 indicate that both techniques are identical. 𝑈 (Ω) = Ω𝑁𝑜 (−Ω2 ) , 𝑅 (Ω) = 𝑁𝑒 (−Ω2 ) .

4 Advances in Mechanical Engineering

140 30

Stability boundary for 25

120

M = 1 and 𝜑 = 0

20

100 Stability boundary for

15

M = 1 and 𝜑 = 0

KI

80

10

KI

Stability boundary for M = 1 and 𝜑 = 60

0

40

M = 1 and 𝜑 = 45 w = 31.5 KI = 0 w = 41.5

−5

20 −1 0 1 2 3 4 5

KP

w = 57

0

KI = 0 w = 41.5 Figure 5: Stability region for two cases.

−20

−1 0 1 2 3 4 5 6

KP 10

Magnitude (dB)

5

Figure 4: Stability region for frequency range Ω → [0, 56.7]. 0

−5

−10

−15

−20

−25

To determine the missing expressions in (15), (16), and (17), 0

the transfer function in (12) can be divided into odd and

Phase (deg)

−45

even portions for both numerator and denominator, and −90

−135

after the comparison with (14), the following variables can be −180

substituted as −225

10−1 100 101

Frequency (rad/s)

𝑁𝑒 (−Ω2 ) = 𝐾1 𝐾3 , 𝑁𝑜 (−Ω2 ) = 𝐾1 𝐶3 ,

PI controller

𝐷𝑒 (−Ω2 ) = 𝑚2 𝑚4 Ω4 Uncontrolled

(18) Figure 6: Bode plot for the controlled and uncontrolled system.

− (𝐾1 𝑚4 + 𝐾3 𝑚2 + 𝐾3 𝑚4 ) Ω2 + 𝐾1 𝐾3 ,

Substituting (17) and (18) into (15) and (16) and consider-

ing (19) yields to

For the controller design problems which based on gain and

phase margins design, it is inevitable to estimate the stabi- 𝑁𝐴 cos [𝜑] + 𝑁𝐵 sin [𝜑]

𝐾𝑃 = ,

lizing region of the controller gains. The stability boundary −𝑀 (𝑁𝑒2 + Ω2 𝑁𝑜2 )

locus in proportional and integral gains plane can be obtained (20)

from (15) and (16) after substituting the missing terms from Ω𝑁𝐵 cos [𝜑] − Ω𝑁𝐴 sin [𝜑]

𝐾𝐼 = ,

(17) and (18), then verifying the presented model using the −𝑀 (𝑁𝑒2 + Ω2 𝑁𝑜2 )

table of contents in Table 1.

The region of stability of the proportional gain 𝑘𝑃 and where

the integral gain 𝑘𝐼 that cause the closed-loop polynomial

to satisfy Hurwitz stability test is shown in Figure 4, where 𝑁𝐴 = (Ω2 𝑁𝑜 𝐷𝑜 + 𝑁𝑒 𝐷𝑒 ) , 𝑁𝐵 = Ω (𝑁𝑜 𝐷𝑒 − 𝑁𝑒 𝐷𝑜 ) .

line 𝑘𝐼 = 0 divides the controller gains plane into stable and (21)

unstable regions.

For designing of controller to meet a certain demanded Using these formulae makes it easy to satisfy certain con-

gain and phase margins which are important frequency ditions and find the stabilization region for the controller

domain requirements for user specifications, consider a parameters. Assuming two cases, one as the phase margin is

control function: greater than 45∘ and gain margin is greater than 1, another

as the phase margin is greater than 60∘ and gain margin is

greater than 1.

𝐺𝑐 (𝑗Ω) = 𝑀𝑒−𝑗𝜑 , (19) Figure 5 shows two regions of stabilities for two different

conditions. The blue circles locus shows the stability region

for a demanded conditions gain margin =1 and phase margin

where 𝑀 and 𝜑 are gain and phase margins, respectively. 45∘ , the maximum frequency for this region is 41.5 rad/sec,

Advances in Mechanical Engineering 5

90 1.6

80 1.4

70

Phase margin (deg)

1.2

60

1

Amplitude

50

40 0.8

30 0.6

20 0.4

10 0.2

0

0 0.5 1 1.5 2 2.5 3 0

0 0.5 1 1.5 2 2.5 3 3.5

Damping ratio

Time (s)

Figure 7: Phase margin versus damping ratio.

KP = 1; KI = 7.5

KP = 1; KI = 25

Passive susp.

the red solid locus shows another design conditions with

gain margin =1 and phase margin 60∘ , and the maximum Figure 8: Step response of the system.

frequency for this region is 31.5 rad/sec.

It is an applicable method to achieve user specified

gain and phase margins to define the values of controller Returning back to (10) and substituting 𝑗Ω = 𝑠, another

parameters with the ability to justify the maximum frequency technique is applied in this section; a developed methodology

without the need of sweeping over the parameters and also by Shamsuzzoha and Skogestad requires only one closed-loop

without using programming for solving inequality problems. step test to obtain PI controller setting. In this method it is

This method can be expanded to estimate the stabilizing also simple to obtain the PID tuning parameters [11, 12].

region of PID controller gains as shown in [17]. Considering an ideal PID controller in the form, as shown

From Figure 6, the plotted line in red solid color indicates in (5)

the output response of the uncontrolled system and also

shows the stability region of frequencies shown in red points 1 𝐾

𝑐 (𝑠) = 𝐾𝑐 (1 + + 𝜏𝐷𝑠) ⇐⇒ 𝑐 (𝑠) = (𝐾𝑃 + 𝐼 + 𝐾𝐷𝑠) .

in the range Ω ∈ [10.2, 15.6] rad/sec, while the blue dashed 𝜏𝐼 𝑠 𝑠

locus indicates the system response after applying the PI (24)

controller which is chosen from the stability region shown in

Figure 6 as 𝐾𝑃 = 1 and 𝐾𝐼 = 2.5; it is obvious that the stability 𝐾𝑐 is the controller gain and 𝜏𝐼 , 𝜏𝐷 are the controller integral

region of frequencies shown in blue squares is expanded in and derivative time, respectively.

the range Ω ∈ [0, 15.5] rad/sec. The defined parameters can be calculated using the

The relationship between the phase margin Φ𝑀, the following basic formulae:

overshoot (OS), and the damping ratio 𝜁 can be expressed as:

2𝜏 + 𝜃 𝜃

𝐾𝑐 = , 𝜏𝐼 = min {(𝑟 + ) , 8𝜃} ,

√−2𝜁2 + √1 + 4𝜁4 3𝑘𝜃 2

Φ𝑀 = 90∘ − tan−1 , (22) (25)

2𝜁 𝑟𝜃

𝜏𝐷 = ,

2𝜏 + 𝜃

where

where 𝑘 is the gain, 𝜏 is the time lag constant, and 𝜃 is

√1−𝜁2 )

OS% = 𝑒−(𝜁𝜋/ × 100%. (23) the time delay. The controller gain is only depending on

the overshoot, while the integral and derivative times are

The relation between Φ𝑀 and 𝜁 is plotted in Figure 7. function of the system peak time. Applying this approach, the

So it is noticeable to get different overshoot values due to tuning results simulation for the designed problem assuming

the changes of the phase margins with the controller design different values for 𝜏/𝜃 is indicated in Table 2.

parameters; Figure 9 shows the effect of changing these values An improved definition of (25) parameters were con-

as follow. cluded in Shamsuzzoha et al. [12], a closed-loop test with

In Figure 8 the red dash-dot line indicates the uncon- P-controller was applied to adjust overshoot value of 0.3021

trolled system step response; the overshoot is about 46%, and for P-only control then assigned the other parameters which

choosing a test point from Figure 3 with values of 𝐾𝑃 = 1 leaded to gain 𝐾𝑐 = 2.5 with 𝜏𝐼 = 0.2040 and peak time 0.153

and 𝐾𝐼 = 25, it is clear that it locates outside the regions for the controller gains.

of stabilities from the used technique, so it is clear from the In spite of Zeigler-Nichols methods’ drawbacks which

figure that the green solid line response overshoot value is are mentioned in previous sections, it is still one of the

about 55%, and unlike the line in blue dot color it is clear that easiest methods for determining the proportional gain 𝐾𝑃 ,

the overshoot value is improved to be 7% when the test point integral time 𝑇𝑖 , and derivative time 𝑇𝑑 based on the transient

is chosen as 𝐾𝑃 = 1 and 𝐾𝐼 = 7.5. response of a given system. Zeigler-Nichols suggested the

6 Advances in Mechanical Engineering

1.5

System: Z-N PID

Controller parameters 1.4 Time (s): 0.0789 System: passive susp.

𝜏/𝜃 Amplitude: 1.33 Time (s): 0.33

𝐾𝑃 𝐾𝐼 𝐾𝐷

Amplitude: 1.45

1.3

0.1 0.4000 0.6667 0.0333 1.2 System: PI

Amplitude

Time (s): 0.23

1.1 Amplitude: 1.08

0.5 0.6667 0.6667 0.1667 System: shass PID

Time (s): 0.158

1 Amplitude: 1.08

2.5 2.0000 0.6667 0.8333

0.9

10 7.0000 0.8750 3.3333

0.8

20 13.6667 1.7083 6.6667 0.7

50 33.6667 4.2083 16.6667 0.6

100 67.0000 8.3750 33.3333 0.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Time (s)

Table 3: Zeigler-Nichols tuning rules. Passive susp. Shass PID

Z-N PID PI

Type of controller 𝐾𝑃 𝑇𝑖 𝑇𝑑

P 0.5𝐾cr ∞ 0 Figure 9: Step response of the system.

PI 0.45𝐾cr 0.83 𝑃cr 0

PID 0.6𝐾cr 0.5 𝑃cr 0.125 𝑃cr Table 5: System response analysis parameters.

Marker

Table 4: Zeigler-Nichols tuning values. 𝑇𝑝 𝑀𝑝 O.S%

Passive

Type of controller 𝐾𝑃 𝑇𝑖 𝑇𝑑 0.336 sec 1.45 45%

suspension

P 2.55 ∞ 0 Z-N PID 0.0789 1.33 33% x x

0.158 1.08 8%

PID 3.06 0.036568 0.013852 PID

PI 0.23 1.08 8%

values of the controller parameters according to the shown response for a vehicle coming out of a pothole can be

formulae in Table 3. simulated as a step input with amplitude of 0.08 as [18]:

For the systems which have a mathematical model, root-

locus method can be used to find the critical gain 𝐾cr and the 0 ⇒ 𝑡 ≤ 0,

sustained oscillation frequency 𝜔cr ; then get 𝑃cr = 2𝜋/𝜔cr . Amp = { (26)

0.08 ⇒ 𝑡 ≥ 0.

Also the systems’ open loop system (passive suspension) and

closed-loop system (Active suspension) can be analyzed in The system output displacement is shown in Figure 10.

frequency domain to find the bandwidth, gain, and phase Figure 10 shows that the peak value of the passive suspen-

margins, and then the design parameters can be found easily. sion system in red solid line is nearly 0.116 m although the

Considering the previous controlled car problem, the input step amplitude is 0.08 m; that is, the overshoot is 45%,

results are found as follows: the critical gain is 5.1, and the and after using the estimated values from the described tech-

sustained oscillation frequency is 56.7 rad/sec. It is the same niques, the overshoot and response of the system improved

result shown in Figure 4, where the stability frequency for although the transient time is increased for some of them but

the case of assuming gain margin 1 and phase margin 0 is the displacement action is smoothed; Zeigler-Nichols results

Ω → [0, 56.7]. are plotted in dash-dot line, Shamsuzzoha results are plotted

From the previous values, the P, PI, and PID Zeigler- in dash lines, and PI controller is plotted in x marker line.

Nichols controller values can be tabulated in Table 4. Another simulation model for random roughness road is

Applying the controller parameters obtained from the applied where the road model is expressed as a differential

different tuning techniques, the system response is plotted in equation with the formula [19]:

Figure 9.

From Figure 9 and related to the different tuning meth- 𝑍̈

(𝑡) + 𝑎V𝑍 (𝑡) = 𝑊 (𝑡) , (27)

ods, the systems’ time of the first peak, the peak values, and ̈is the road random incentive and the constant “𝑎”

the overshoot values can be tabulated in Table 5. where 𝑍(𝑡)

depends on road grade; the values of “𝑎” are shown in Table 6.

𝑊(𝑡) is the white noise, with zero mean value and

4. Simulations and Results amplitude equals 1, and V is the vehicle speed.

Focusing for more details can be shown in Figure 12.

After defining the controller parameters, it is necessary to test The desired problem is validated using the results of

the system performance for some situations. The suspension road simulation of grade B and assuming the velocity is

Advances in Mechanical Engineering 7

0.12

0.1

0.1

0.05

Displacement (m)

0.08

Amplitude

0

0.06 −0.05

0.04 −0.1

−0.15

0.02

−0.2

0 0 5 10 15 20 25 30

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Time (s)

Time (s)

Passive susp. Shass PID

Passive susp. Shass PID PI

Z-N PID

Z-N PID PI

Figure 11: Output displacement of the system.

Figure 10: The output displacement from the system.

0.1

Road grade A B C D E

𝑎 0.132 0.1303 0.12 1.1007 0.09 Displacement (m) 0.05

0

50 km/hr; the results show the effectiveness of the used

controllers to reduce the displacement motion as shown in −0.05

Figure 11 and zoomed in Figure 12, also results illustrated

that different control techniques can be applied to MS-TMM, −0.1

results of the passive suspension response are plotted in blue 0 1 2 3 4 5 6 7 8 9 10

solid line, Zeigler-Nichols result is plotted in blue dot plot, Time (s)

Shamsuzzohas’ result is plotted in green dash-dot plot, and Passive susp. Shass PID

finally the PI controls’ result is plotted in red dash plot. Z-N PID PI

5. Conclusion and Future Work

As the increments of complexity of multibody systems and

the development of their design and control methods, the advantage of indicating inputs and outputs easily such as

need for more elegant formulations of the equations of power spectral density measurements for such kinds of active

motion becomes an issue of paramount importance. The vibration isolation techniques.

results showed identical values for MS-TMM results with the

ordinary classical methods. Thus, ongoing researches, MS-

TMM strategies, and its related topologies are highly recom- Conflict of Interests

mended for analyzing and solving the controlled systems due

to easy formulations, being systematic to apply, and simple The authors declare that there is no conflict of interests

program coding with less computational time. The results regarding the publication of the paper.

also showed that different controller design methods can be

impeded with MS-TMM which leads to more availability for Acknowledgments

other controllers applications; one depends on the frequency

response analysis, another considers the step input response The research was supported by the Research Fund for

of the system, and the last technique can be applied for the Doctoral Program of Higher Education of China

both time and frequency domains if the model has a known (20113219110025), the Natural Science Foundation of China

mathematical model. All controllers’ parameters values are Government (11102089), and the Program for New Century

applied to a model of controlled car system with brief analysis Excellent Talents in University (NCET-10-0075).

for the obtained results. The future work will concern with

deeply impeding different control techniques with MS-TMM

to develop new techniques for the more complex multibody

References

controlled systems considering nonstationary inputs by tak- [1] X.-T. Rui, B. He, B. Rong, G.-P. Wang, and Y.-Q. Lu, “Discrete

ing into consideration the system’s equations of motion with time transfer matrix method for dynamics of a multi-rigid-

studying practical systems in frequency domain due to its flexible-body system moving in plane,” Proceedings of the

8 Advances in Mechanical Engineering

2009.

[2] B. Rong, X. Rui, G. Wang, and F. Yang, “Discrete time transfer

matrix method for dynamics of multibody system with real-

time control,” Journal of Sound and Vibration, vol. 329, no. 6,

pp. 627–643, 2010.

[3] D. Bestle, “Application of the transfer matrix method to control

problems,” in Proceedings of the Thematic Conference, Multibody

Dynamics, Zagreb, Croatia, 2013.

[4] S. Yamato and I. Hashimoto, “Present status and future needs

the view from Japanese industry,” in Proceedings of the 4th

International Conference on Chemical Process Control, Y. Arkun

and W. H. Ray, Eds., Padre Island, Tex, USA, 1991.

[5] C. C. Yu, Auto Tuning of PID Controllers, Springer, 2nd edition,

2006.

[6] G. J. Suva, A. Datta, and S. P. Bhattacharyya, “New results on the

synthesis of PID controllers,” IEEE Transactions on Automatic

Control, vol. 47, no. 2, pp. 241–252, 2002.

[7] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic

controllers,” Journal of Dynamic Systems, Measurement, and

Control, vol. 115, no. 2, pp. 220–222, 1993.

[8] T. S. Schei, “A method for closed loop automatic tuning of PID

controllers,” Automatica, vol. 28, no. 3, pp. 587–591, 1992.

[9] B. D. Tyreus and W. L. Luyben, “Tuning PI controllers for

integrator/dead time processes,” Industrial and Engineering

Chemistry Research, vol. 31, no. 11, pp. 2625–2628, 1992.

[10] K. J. Astrom and T. Hagglund, PID Controllers: Theory, Design

and Tuning, Instrument Society of America, Research Triangle

Park, NC, USA, 2nd edition, 1995.

[11] M. Shamsuzzoha and S. Skogestad, “The setpoint overshoot

method: a simple and fast closed-loop approach for PID

tuning,” Journal of Process Control, vol. 20, no. 10, pp. 1220–1234,

2010.

[12] M. Shamsuzzoha, M. Lee, and H. Seki, “Closed-loop PI/PID

controller tuning for stable and unstable processes,” in Pro-

ceedings of the American Control Conference, pp. 2368–2373,

Montreal, Canada, 2012.

[13] A. J. Barr and J. L. Ray, “Control of an active suspension

using fuzzy logic,” in Proceedings of the 5th IEEE International

Conference on Fuzzy Systems, pp. 42–48, September 1996.

[14] M. Cai, H. Liu, P. Du, J. Guo, and H. Chang, “Suspension

dynamics simulation based on simulink,” in Proceedings of the

International Conference on Electric Information and Control

Engineering, pp. 2715–2718, April 2011.

[15] N. Changizi and M. Rouhani, “Comparing PID and fuzzy

logic control a quarter-car suspension system,” The Journal of

Mathematics and Computer Science, vol. 2, no. 3, pp. 559–564,

2011.

[16] K. J. Åström and T. Hägglund, “The future of PID control,”

Control Engineering Practice, vol. 9, no. 11, pp. 1163–1175, 2001.

[17] N. Tan, I. Kaya, C. Yeroglu, and D. P. Atherton, “Computation of

stabilizing PI and PID controllers using the stability boundary

locus,” Energy Conversion and Management, vol. 47, no. 18-19,

pp. 3045–3058, 2006.

[18] J. Y. Wong, Theory of Ground Vehicles, John Wiley & Sons, New

York, NY, USA, 1988.

[19] J. Zhihong, E. Zhang, L. Zhonghua, J. Jing, and Z. Fan,

“Simulation and experiment research on human riding comfort

for vehicle vibration in dynamic environment,” in Proceedings of

the 9th International Conference on Computer-Aided Industrial

Design and Conceptual Design: Multicultural Creation and

Design, pp. 175–181, November 2008.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2013, Article ID 725315, 6 pages

http://dx.doi.org/10.1155/2013/725315

Research Article

All Terrain Vehicle Flexible Multibody Dynamic

Simulation for Fatigue Prediction

Department of Vehicle Engineering, National Taipei University of Technology, No. 1, Section 3,

Zhong-Xiao E. Road, Taipei 10608, Taiwan

Copyright © 2013 J.-S. Chen and H.-Y. Hwang. This is an open access article distributed under the Creative Commons Attribution

License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly

cited.

This study presents the flexible-body dynamic analysis and simulated stress recovery of vehicle components to predict their lifetime

when maneuvering on an uneven random road. The subject of this study was an all-terrain vehicle (ATV). The body frame and

suspension system were modeled as flexible elements for multibody dynamic simulation. The simulations in this study revealed the

stress from the flexible elements and predicted the component fatigue life using the retrieved stresses. This approach considers the

interaction between dynamic forces and structure deformation and achieves more accurate structure stress prediction and fatigue

life prediction.

accurate loads and stresses of the vehicle components than

Fatigue is one of the major concerns in automotive engineer- the general rigid-body simulation can. The good correlation

ing. The structure and components of a vehicle are constantly between simulation and experimental results was presented

under cyclic loading, and especially on rough roads. This as well in their study. Some studies have focused on vehicle

could cause the fatigue failure of the structure or components.

components by modeling them as flexible elements in multi-

Ohchida [1] showed that 60% of machine equipment fail-

body dynamic simulation. Moon et al. [5] presented a taper

ures are caused by component fatigue. Therefore, a reliable

leaf spring with hysteretic characteristics. This study devel-

method for predicting the potential fatigue failures of vehicle

components is highly desirable. In the design cycle of auto- oped a flexible multi-body dynamic simulation to inter-

motive vehicles, the majority of automotive manufacturers face the leaf spring finite-element model and computation

use empirical methods or dynamic simulations to predict the model and correlated simulation and experimental results.

component and structure loadings [2]. The FEA model was Shabana and Sany [6] presented a rolling contact theory

designed to predict the stress of components and fatigue with multi-body dynamics to simulate the effects because

failure [3]. However, the FEA analysis process does not of the structural flexibility of the vehicle component and

consider the effects of the dynamic loads caused by structural track. Zhu et al. [7] predicted the fatigue life of a truck

deformations. The dynamic loading that occurs when the frame. They modeled the frame as a flexible element and the

vehicle is running can cause structural deformation, which in vehicle system as rigid multi-body dynamic model. Yang et al.

turn may affect load conditions and further change the stress [8] predicted the fatigue life of a wheel by simulating the EMU

of components. Thus, dynamic analysis with a rigid body wheel as a flexible body in multi-body simulations. Rathod

system is insufficient. The dynamic loading changes caused et al. [9] presented a multi-body railroad vehicle system that

by structural deformation should be considered in structure accounted for the dynamic coupling between a 3D wheel and

and component stress analysis. Cuadrado et al. [4] presented the rail structure flexibility. However, these studies did not

flexible-body dynamic simulations considering the effects include the deformation of the entire vehicle structure.

2 Advances in Mechanical Engineering

This study uses ADAMS to perform flexible-body where 𝑚𝑖 and 𝐼𝑖 represent the modal mass and modal inertia

dynamic simulation and models all structural components as of moving body, respectively.

flexible elements. The stress history of the structure was then The mass matrix of flexible body, 𝑀, can be expressed as

retrieved from the dynamic simulation for fatigue calcula-

tion. 𝑀𝑡𝑡 𝑀𝑡𝑟 𝑀𝑡𝑚

The remaining sections are as follows: the theoretical [ 𝑇 ]

[ 𝑀𝑡𝑟 𝑀𝑟𝑟 𝑀𝑟𝑚 ]

background of the flexible-body dynamics and fatigue anal- [

𝑀 (𝜉) = [ ], (6)

]

ysis, the simulation model, the simulation results, and the [ 𝑇

]

𝑀𝑡𝑚 𝑀𝑟𝑚 𝑀𝑚𝑚

conclusion. [ ]

where 𝑡, 𝑟, and 𝑚 represent the translational, rotational, and

2. Theory Background modal degrees of freedom, respectively.

2.1. Equations of Flexible-Body Dynamics. In equations of The stiffness matrix of the flexible body, 𝐾, is relatively

motion, the linear deformation of a structure can be repre- simple because there is no rigid body contribution:

sented by the combination of mode shapes and mode coordi-

𝐾𝑡𝑡 𝐾𝑡𝑟 𝐾𝑡𝑚

nates as follows: [ 𝑇 ] 0 0 0

[ ]

𝑀 𝐾 (𝜉) = [ 𝐾𝑡𝑟 𝐾𝑟𝑟 𝐾𝑟𝑚 ] = [0 0 0 ] . (7)

𝑢 = ∑𝜙𝑖 𝑞𝑖 , [ ]

[0 0 𝐾𝑚𝑚 ]

(1) 𝑇 𝑇

𝑖=1 𝐾𝑡𝑚 𝐾𝑟𝑚 𝐾𝑚𝑚

[ ]

where 𝜙 is the mode shape and 𝑞 is the mode coordinate.

The equations of motion with a Lagrange multiplier can

With the Cartesian coordinate (𝑥, 𝑦, 𝑧), Euler’s angle

be written as follows:

(𝜓, 𝜃, 𝜙), mode coordinate 𝑞, and the generalized coordinates

of the flexible element can be expressed as follows: 1 𝜕𝑀 ̇𝑇 ̇ 𝜕𝜓 𝑇

𝑀𝜉 ̈

+ 𝑀̇𝜉 ̇− [ 𝜉] 𝜉 + 𝐾𝜉 + 𝑓𝑔 + 𝐷𝜉 ̇+ [ ] 𝜆 = 𝑄.

𝑥} 2 𝜕𝜉 𝜕𝜉

{

{ }

{

{ 𝑦 }

} (8)

{

{ }

}

𝑥 { }

{𝑧}

{ } The term 𝐷 is the modal damping matrix, and 𝐾𝜉 and 𝐷𝜉,̇

𝜉 = (𝜓) = { 𝜓 } , (𝑖 = 1, 2, 3, . . . , 𝑀) . (2)

𝑞 {𝜃}

{ } respectively, represent the structural internal force caused

{

{ }

}

{𝜙}

{ } by the elastic deformation and velocity. The term 𝜆 is the

{ }

{ }

𝑞 Lagrange multiplier, 𝑓𝑔 is the structure weight, and 𝑄 is the

{ 𝑖} external force.

The position on the deformed body is written as The strain and stress can be retrieved using the mode

𝑟𝑖 = 𝑥 + 𝐴 (𝑆𝑖 + 𝜙𝑖 𝑞) , (3) coordinate [10]. This study calculates the stress in the struc-

ture using the modal stress recovery technique:

where 𝐴 is the transformation matrix between global coordi-

nates and the body local coordinates. The term 𝑆𝑖 represents {𝜀} = [𝐻] {𝑥} ,

the position in the body before deformation, and 𝜙𝑖 is the (9)

mode shape matrix. {𝜎} = [𝐸] {𝜀} .

From the position, the velocity of the deformed body is The term {𝜎} is the stress vector, {𝜀} is the strain vector,

𝑑𝑟𝑖 𝑑𝑥 𝑑𝐴 𝑑 (𝑠𝑖 + 𝜙𝑖 𝑞) and [𝐻] is the finite-element geometric deformation-to-

V𝑖 = = + (𝑠 + 𝜙𝑖 𝑞) + 𝐴 strain matrix. The term [𝐸] is the material property matrix.

𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑖 𝑑𝑡

= 𝑥̇− 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝜔 + 𝐴𝜙𝑖 𝑞 ̇ (4)

2.2. Fatigue Prediction. Metal material under cycling stress

can fail even when the stress is under the ultimate stress limit.

= 𝐹 − 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝐵 + 𝐴𝜙𝑖 𝜉.̇ The relationship between fatigue stress and life cycle is usually

The term 𝜔 is the angular velocity vector of the body represented by a stress-cycles curve (𝑆-𝑁 diagram). For steel,

coordinate, and 𝐹 is the transfer matrix between the time a cycle life greater than 106 is treated as a lasting life, and the

derivative of Euler’s angle and the angular velocity. stress corresponding to a 106 cycle is the endurance limit 𝑆𝑒

The kinematic energy of the body is [11].

The 𝑆-𝑁 diagram is plotted in log scale, and the formula

1 of 𝑆-𝑁 curve can be simplified as

𝑇= ∫ 𝜌V𝑇 V𝑑𝑉

2

log 𝑆𝑓 = log 𝑎 + 𝑏 log 𝑁 ⇒ 𝑆𝑓 = 𝑎𝑁𝑏 , (10)

1𝑁

≈ ∑ (𝑚𝑖 V𝑖𝑇 V𝑖 + 𝜔𝑖𝑇 𝐼𝑖 𝜔𝑖 ) (5)

2 𝑖=1 where 𝑏 is the slop of curve and 𝑁 is the life cycle.

During vehicle maneuvering, the structural stress is fluc-

1

= 𝜉𝑇̇𝑀 (𝜉) 𝜉,̇ tuating. The amplitude of stress variation varies with the driv-

2 ing and road conditions. The rain-flow counting technique

Advances in Mechanical Engineering 3

Revolute

joint

Bushing

Ball

joint

y x

Figure 1: Vehicle structure and suspension frames. z

can be applied to measure different stress variations during

driving. The Palmgrem-Miner cycle-ratio summation rule Table 1: Material property of vehicle structure.

can then be used to accumulate the damages of various loads

and cycles: Property Steel Steel

1015 1018

𝑖=𝑁

𝑛𝑖 Elastic modulus (MPa) 2.07𝐸05 2.07𝐸05

∑ = Damage, (11)

𝑁

𝑖=1 𝑓𝑖

Poisson ratio 0.29 0.29

Density (ton/mm3 ) 7.9𝐸 − 09 7.9𝐸 − 09

where 𝑛𝑖 is the number of cycles at the change of stress ampli- Ultimate strength (MPa) 414 455

tude (Δ𝜎𝑖 ) and 𝑁𝑓𝑡 is the number of cycles until failure at the

change of stress amplitude (Δ𝜎𝑖 ). The “Damage” parameter

usually ranges from 0.7 to 2.2. When the accumulated damage

[10], which were connected with a revolute joint. The upper

is larger than the “Damage” parameter, the component fails

torso, including the seat back, and the pelvis, including the

[12].

seat, were attached to the vehicle structure by springs and

dampers. Table 2 lists the spring stiffness and damping coef-

3. Vehicle Model ficient. The driveline was seated on the engine mounts, and

the engine mounts were modeled as rubber bushings. Table 2

The object of research in this study is a dune buggy. Figure 1 lists the parameters.

shows the geometric model based on the measurement on the

vehicle structure. For each component of vehicle structure,

MSC Nastran was used to perform modal analysis, and 3.2. Tire and Road Models. The simulations in this study

MSC Patran was used to process finite-element models and modeled tires as force elements, and these tire models

modal analysis data to create flexible elements for dynamic included the tire mass and inertia. The simulations in this

simulation [13]. These elements were imported to the vehicle study use the Fiala tire model, which was designed for vertical

model to assemble the whole vehicle. Then, we used ADAMS and longitudinal motion. Table 3 presents the tire parameters.

to perform the flexible-body dynamic simulations [14] and For simulation road condition, this study applies a ran-

determine the stress generated by random road maneuvers. dom 2D surface (Figure 3).

3.1. Components. The structure of vehicle model included 4. Simulation and Result

three primary parts: the body frame, the front suspension,

and the rear suspension (Figure 2). The powertrain and In the simulation, the vehicle maneuvered in straight runs

passenger weights were attached to the structure, and the total over a random uneven surface. The vehicle accelerated from

weight of vehicle was 423 kg. All the structure components standing still, and the data for analysis was collected when

were modeled as a flexible body. The tires were simulated as it reached a constant speed of 50 km/h. The duration of data

force elements in dynamic simulation [14]. Figure 2 shows collecting was 10 s. Several thicknesses of the structure frame

the connecting joint and forces applied between components tube were applied in the simulation: 1 mm, 1.5 mm, 2 mm, and

[14, 15], and Table 1 shows material properties of the frame 2.5 mm.

[16].

In addition to the weight of the vehicle structure and the 4.1. Dynamic Simulation. In flexible-body dynamic simula-

driveline, two dummies weighing 80 kg each were modeled tion, ADAMS can retrieve the structural stress which could be

and seated on the vehicle structure. The dummies were applied to failure or fatigue analysis. Figure 4 shows the high-

modeled as two parts (the upper torso; pelvis and legs) est stress locations during the dynamic simulation. Table 4

4 Advances in Mechanical Engineering

Upper torso (including seat back) 45 10 0.4

Hips and leg (including seat) 35 15 0.2

Powertrain 40 52 0.12

Tire model Fiala tire type Fiala tire type

Mass (kg) 10 11 Force

direction

Moment of inertia (kg-mm ) 2 Ixx, Iyy: 8𝐸5; Ixx, Iyy: 8𝐸5;

Izz: 16𝐸5 Izz: 16𝐸5

Tire radius (mm) 254 280

Tread width (mm) 200 200 Flexible

body

10.0 y

z x

V-axis length (mm)

5.0

Figure 5: Joint locations.

0.0

−5.0

simulation cannot determine the structural stress, the max-

−10.0 imum rear suspension joint forces were used to present the

0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 difference between these two types of simulation (Table 5).

H-axis length (m) Figure 5 shows the joint locations. The frame tube thickness

of the model was 2.0 mm in this simulation.

Figure 3: 2D random road profile.

In addition to the suspension and tire elements, the

flexible element in the vehicle structure also absorbed part of

582079 dynamic energy during the uneven road simulation. Thus, the

585076 joint force of the flexible-body model was smaller than that of

von Mises stress (MPa) the rigid-body model during the simulation.

592.32

533.09

470.86

586151

582887 4.2. Fatigue Analysis. The FEM model in this study adopts

414.62 several thicknesses of the frame tube. The flexible elements

355.39 were built based on these models, and flexible-body dynamic

296.16 simulations were run with different tube thicknesses. The

236.93 tube thicknesses were 1 mm, 1.5 mm, 2 mm, and 2.5 mm. The

177.7 maximum stress was retrieved and plotted in a stress-time

118.46 curve. Figure 6 shows the stress history of the structure with

59.23 a 2.5 mm tube thickness in 10 s simulation at a constant speed

5.48E − 014 of 50 km/h.

The rainflow counting method was applied to the stress

history curve, and the stress cycle number of different stress

amplitudes was used for fatigue analysis. Figure 7 shows the

Figure 4: Highest stress locations at random road simulation. stress range and cumulative cycle number of different frame

tube thicknesses.

Based on the stress cycle and material 𝑆-𝑁 diagram, the

structural life can be predicted using the Miner rule. Using

shows the maximum von Misses stress of these locations in a 10 s simulation cycle, Table 6 presents the structural life

uneven random road simulations. of different frame tube thicknesses. This duration time and

To compare the differences in dynamic load running on the maximum stress position listed in Table 4 show that

an uneven road surface, this study created both rigid-body the structure connecting points of two or more structure

and flexible-body models. Because the rigid-body dynamic members are the most vulnerable to fatigue damage.

Advances in Mechanical Engineering 5

Frame tube thickness (mm) Road surface Max. stress grid point Max. stress (MPa)

Flat 582887 215.36

1

Random 585076 498.30

Flat 582887 163.08

1.5

Random 586151 430.28

Flat 582887 137.87

2

Random 586151 361.84

Flat 582887 105.33

2.5

Random 585079 347.21

Frame tube thickness (mm) Right and left joints average force at rear Difference

suspension (N)

Flexible-body model 623.01 —

Rigid-body model 758.09 21.7%

Cycle number Less than 1 𝑁𝑓 = 218 𝑁𝑓 = 1265 𝑁𝑓 = 2780

250

Amplitude (MPa)

200

5. Conclusions

150

Conventional analysis methods model the vehicle body as a

100

rigid body in dynamic simulations. In this case, the dynamic

50 energy passing from the tire running over an uneven road can

only be absorbed and damped by tires, springs, and dampers

0 in suspension, and bushings. After this rigid body dynamic

0.0 2.0 4.0 6.0 8.0 10.0 simulation, the component load was applied to a finite-

Time (s) element model to determine the structural stress of the

Figure 6: Stress history of the max. stress grid point (585079).

component for further analysis. The approach used in this

paper is different from the conventional one. The dynamic

model in this study uses flexible bodies. The advantage of

using flexible-body dynamic simulation is that it simulates

450

structural stresses in dynamic analysis and can be retrieved

400

through the simulation. The dynamic response considers the

350 effect of structural deformation. This provides more realistic

Stress range (MPa)

250 structures. The structure can also absorb part of the energy.

200 Thus, the loads at joints were smaller compared with the loads

150 obtained in rigid-body simulation.

100 The stress received from flexible dynamic analysis can

50

be used in fatigue analysis. The structure connecting points

where two or more structure members were welded together

0

appeared to be the most vulnerable to fatigue damage. There-

1 10 100 1000

fore, these points should be reinforced during structure

Cumulative cycles

design and assembly to increase the expected lifetime.

2.5 mm (point 585079) Vehicle fatigue testing is costly and time consuming.

2 mm (point 586151)

1.5 mm (point 586151)

Flexible-body dynamic simulations can assist in the life pre-

diction of vehicle structures and components. The proposed

Figure 7: Cumulative cycle number of different frame tube thick- analysis tool can shorten the design cycle and can reduce

nesses. vehicle development cost and time.

6 Advances in Mechanical Engineering

tution of Mechanical Engineers K, vol. 223, no. 4, pp. 269–282,

𝐴: Transformation matrix between global 2010.

coordinates and the body local coordinates [10] T. M. Wasfy and A. K. Noor, “Computational strategies for flex-

𝐷: Modal damping matrix ible multibody systems,” Applied Mechanics Reviews, vol. 56, no.

𝐸: Material property matrix 6, pp. 553–613, 2003.

𝐹: Transfer matrix between the time [11] W. Yao, Structure Fatigue Analysis, Defence Inductry Publisher,

derivative of Euler’s angle and angular 2003.

velocity [12] J. E. Shigley and C. R. Mischke, Mechanical Engineering Design,

𝐻: Finite-element geometric McGraw Hill, 2010.

deformation-to-strain matrix [13] MSC Software, Nastran User’s Manual, MSC Software, Newport

𝐼𝑖 : Modal inertia Beach, Calif, USA, 2010.

𝐾: Stiffness matrix of the flexible body [14] MSC Software, MD Adams. ADAMS User’s Manual, MSC

𝑀: Mass matrix of flexible body Software, Newport Beach, Calif, USA, 2010.

𝑄: External force [15] M. Azadi, S. Behzadipour, and G. Faulkner, “Performance anal-

𝑆𝑖 : Position before deformation ysis of a semi-active mount made by a new variable stiffness

𝑓𝑔 : Structure weight spring,” Journal of Sound and Vibration, vol. 330, no. 12, pp.

𝑚𝑖 : Modal mass 2733–2746, 2011.

𝑞: Mode coordinate [16] R. C. Hibbeler, Mechanics of Materials, Prentice Hall, 2010.

𝜀: Strain vector

𝜆: Lagrange multiplier

𝜉: The generalized coordinate of flexible

elements

𝜎: Stress vector

𝜙: Mode shape

𝜙𝑖 : Mode shape matrix

𝜔: Angular velocity vector.

References

[1] H. Ohchida, “Analysis of service failures of Hitachi products

1970–1975,” Hitachi Company Report, 1979.

[2] S. Lu, Y. Li, and Y. Dong, “Dynamic modeling and virtual test

of a new all terrain off-road vehicle,” Applied Mechanics and

Materials, vol. 86, pp. 125–128, 2011.

[3] B. R. Miao, W. H. Zhang, L. M. Zhang, T. Zhu, and H. T. Yin,

“Research of modern railway vehicle structure fatigue design

method based on multibody dynamic simulation,” Advanced

Materials Research, vol. 118–120, pp. 327–331, 2010.

[4] J. Cuadrado, R. Gutiérrez, M. A. Naya, and M. González,

“Experimental validation of a flexible MBS dynamic formu-

lation through comparison between measured and calculated

stresses on a prototype car,” Multibody System Dynamics, vol. 11,

no. 2, pp. 147–166, 2004.

[5] I.-D. Moon, H.-S. Yoon, and C.-Y. Oh, “A flexible multi-body

dynamic model for analyzing the hysteretic characteristics and

the dynamic stress of a taper leaf spring,” Journal of Mechanical

Science and Technology, vol. 20, no. 10, pp. 1638–1645, 2006.

[6] A. A. Shabana and J. R. Sany, “A survey of rail vehicle track simu-

lations and flexible multibody dynamics,” Nonlinear Dynamics,

vol. 26, no. 2, pp. 179–210, 2001.

[7] S. Zhu, Z. Xiao, and X. Li, “Vehicle frame fatigue life prediction

based on finite element and multi-body dynamic,” Applied

Mechanics and Materials, vol. 141, no. 1, pp. 578–585, 2012.

[8] J. Yang, Q. Shi, G. Zhang, and J. Zhang, “The fatigue life simula-

tion of the wheel of CHR3 EMU in random loading,” Advanced

Materials Research, vol. 430–432, pp. 1424–1427, 2012.

[9] C. Rathod, R. Chamorro, J. L. Escalona, M. El-Sibaie, and A. A.

Shabana, “Validation of three-dimensional multi-body system

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2013, Article ID 153913, 12 pages

http://dx.doi.org/10.1155/2013/153913

Research Article

Projectile Impact Point Prediction Based on Self-Propelled

Artillery Dynamics and Doppler Radar Measurements

Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy

Institute of Launch Dynamics, Nanjing University of Science and Technology, 200 Xiao Ling Wei, Nanjing 210094, China

Copyright © 2013 Mostafa Khalil et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Any trajectory calculation method has three primary sources of errors, which are model error, parameter error, and initial state

error. In this paper, based on initial projectile flight trajectory data measured using Doppler radar system; a new iterative method

is developed to estimate the projectile attitude and the corresponding impact point to improve the second shot hit probability. In

order to estimate the projectile initial state, the launch dynamics model of practical 155 mm self-propelled artillery is defined,

and hence, the vibration characteristics of the self-propelled artillery is obtained using the transfer matrix method of linear

multibody system MSTMM. A discrete time transfer matrix DTTM-4DOF is developed using the modified point mass equations

of motion to compute the projectile trajectory and set a direct algebraic relation between any two successive radar data. During

iterations, adjustments to the repose angle are made until an agreement with acceptable tolerance occurs between the Doppler

radar measurements and the estimated values. Simulated Doppler radar measurements are generated using the nonlinear six-

degree-of-freedom trajectory model using the resulted initial disturbance. Results demonstrate that the data estimated using the

proposed algorithm agrees well with the simulated Doppler radar data obtained numerically using the nonlinear six-degree-of-

freedom model.

tion is needed to measure the in-flight projectile parameters.

There are different computation methods to estimate the Projectiles do not have enough space to install on-board

projectile ballistic trajectory and subsequent impact point. sensors to acquire these parameters. Further, considering the

These methods have three primary sources of errors, which severe environment faced by the projectile during flight such

are model error, parameter error, and initial state error. The as velocity and spinning rate, a low cost and accurate mea-

model error depends on how much the projectile equations surement method is required. The most promising method is

of motion are simplified. Increasing model complexity means using a Doppler radar system [4] due to its robustness, porta-

decreasing model error, but the parameter and initial state bility, accuracy, and providing nearly continuous velocity-

errors will increase [1]. In 1964, the projectile six-degree-of- time information on the projectile during the flight down

freedom (6-DOF) equations of motion in terms of direction range. A number of authors presented different algorithms

cosines [2] was developed by the ballistic research laboratory, based on Doppler radar measurements. In 1958, Shapiro [5]

which is the most accurate model that can simulate the rigid proposed three different estimation methods to compute the

body flight dynamics under the condition that all the aerody- six parameters that define the orbital parameters describing

namic forces and moments, and the initial state, are known the ballistic missile trajectory using single radar measure-

to a high degree of accuracy. Using this system of equa- ments which include range, range-rate, and the azimuth and

tions, a discrete time transfer matrix method DTTM-6DOF elevation angles. These methods are the maximum likelihood,

[3] has been developed to accurately compute the projectile the weighted least squares, and a deterministic method. In

ballistic trajectory and minimize the corresponding compu- 1985, a new iterative method had been developed [6] and

tational time. applied to a modified version of 6-DOF model to extract

2 Advances in Mechanical Engineering

actual thrust made during the flight of rocket-assisted projec- equation of self-propelled artillery was obtained to compute

tiles using Doppler radar measurements. In 2008, an iterative the corresponding vibration characteristics. In order to itera-

algorithm had been developed [7] using maximum likelihood tively compute the projectile attitude, a discrete time transfer

method to predict the drag characteristics of projectiles in matrix DTTM-4DOF is developed based on linearization

motion by processing data acquired by Doppler radar system, of the nonlinear modified point mass equations of motion

where the point mass model was used due to the projectile using second-order Taylor series [16]. The system state vector,

small angle of attack along the trajectory. which defines the projectile trajectory parameters during

The movements of artillery and projectile are very com- Doppler radar observations, is defined by 13 state variables,

plex in launch process because of the complex structure of which are the projectile position, velocity, spin rate, repose

self-propelled artillery and the severe environments, such angle, and wind velocity. During iterations, adjustments to

as, high temperature, high pressure, high speed, instanta- the repose angle are made until agreement with a certain tol-

neous state, multibody, and mutation, in launch process erance occurs between the Doppler radar velocity measure-

[8]. Therefore, gun launch dynamics has a primary concern ments and the value predicted using DTTM-4DOF. Finally,

in designing embedded navigation and guidance systems projectile impact point can be predicted precisely depending

in case of guided projectiles and evaluating the projectile on the length of radar data available, where the data estimated

initial disturbance in case of unguided projectiles. In 2000, at the end of radar observations is used as an initial state to

the launch dynamics of self-propelled artillery [9] based the 6-DOF trajectory model.

on transfer matrix method of multibody system MSTMM

was developed for studying the vibration characteristics and 2. Launch Dynamics Model for

dynamics of self-propelled artillery and the projectile’s initial Self-Propelled Artillery System

disturbance. In 2001, a finite element analysis [10] had been

done for the breech closure for the 155 mm Cannon M199, To accurately estimate the projectile initial state, launch pro-

which is normally mounted on the Towed Howitzer M198. cess has to be decided precisely by studying the artillery and

The analysis is for a 9 body problem with 13 contact surfaces projectile dynamics from the moment of firing to the state

and was solved for both static and dynamic load cases. In of muzzle point. Four interactions have to be considered, the

2004, a simplified model [11] had been done to predict joint interaction between projectile and artillery, the interaction

forces and accelerations along the length of a simplified between the artillery transverse and longitudinal motion, the

155 mm projectile using modal superposition. These pre- interaction between rigid and flexible body, and, finally, the

dicted loads used to locate all guidance equipment, sensors, interaction between space coordinate and time coordinate for

joints, and computers that must operate reliably after exiting describing motion parameters.

the gun. In 2005, a sophisticated 3D-finite-element model The launch dynamics model of self-propelled artillery is

[12] had been developed to investigate the survivability a multi-rigid-flexible system which has been studied in [8].

of embedded electrical systems by simulating the launch This system is composed of 51 elements (23 bodies and 28

dynamics of a surrogate Excalibur projectile. This study joints) as shown in Figure 1. It has 3 boundary ends including

determined free mounting locations trouble for sensitive ground, gun breech, and muzzle. The ground (0) is regarded

components and parametric investigation in identifying sen- as an infinity rigid body and the first input point, the gun

sitive factors affecting the muzzle-exit motion of projectile breech (44) is regarded as rigid body and the second input

substructures. In 2008, a new method [13] named as transfer point, and the muzzle (52) is the output point. The road

matrix method of linear multibody system MSTMM for lin- wheels (13–24) are regarded as lumped masses. The hull

ear hybrid multibody system dynamics was developed. This (37), revolving part (39), elevating part (41), and muzzle

method had been applied to a Multiple Rocket System as a brake (51) are regarded as rigid bodies; the barrel is divided

linear multi-rigid-flexible-body system. In 2011, the dynamics into 6 segments (45–50) and each is regarded as a beam

problem of a shipboard gun system had been solved [14] using with equal sectional area. The interactions (1–12) between

the discrete time transfer matrix method of multibody system ground and road wheel and the connections (25–36) between

MSDTTMM [15]. road wheel and the hull are, respectively, modeled with

In this paper, an iterative method is developed to deter- springs and dampers in 3 directions. Connections (38, 40,

mine the projectile kinematics including attitude and angular 42, and 43) are modeled with springs and rotary springs

motion using Doppler radar measurements. Doppler radar accompanying dampers to represent relative linear motion

measurements are only available during the first portion and relative angular motion in 3 directions at the same time.

of projectile trajectory including the range, the range-rate, All connections between barrel and gun breech, barrel and

and the azimuth and elevation angles (position and velocity muzzle brake, and each segment of barrel are regarded as

vectors), in order not to be recognized by the enemy coun- fixed. The masses of traversing mechanism, elevating mecha-

terattack systems. Simulated Doppler radar measurements nism, and equilibrator fall into elevating part, revolving parts,

are generated during flight time using 6-DOF trajectory and hull, respectively. According to the launch dynamics

model including the projectile initial disturbance problem. model of self-propelled artillery and the sequence number of

The projectile initial disturbance is computed by solving each element, there are 49 connection points which are 𝑃𝑖,𝑖−12 ,

the self-propelled artillery and projectile dynamics as a 𝑃𝑖,𝑖+12 (𝑖 = 13–24), 𝑃37,𝑖 (𝑖 = 25–36), 𝑃37,38 , 𝑃39,38 , 𝑃39,40 ,

multi-rigid-flexible system using Transfer Matrix Method of 𝑃41,40 , 𝑃41,42 , 𝑃41,43 , 𝑃44,45 , 𝑃45,46 , 𝑃46,47 , 𝑃47,48 , 𝑃48,49 , 𝑃49,50 ,

Multibody System MSTMM [8], where the overall transfer and 𝑃51,50 and 14 boundary points which are 𝑃0,𝑖 (𝑖 = 1–12),

Advances in Mechanical Engineering 3

Elevating part 41

Projectile

Breech 44 42 40 43 Beams 45–50 Muzzle brake 51

52

38

Revolving part 39

Hull 37

25 26 27 28 29 30

13 14 15 16 17 18

1 2 3 4 5 6 Road wheels 13–24

0

practical 155 mm self-propelled artillery shown in Figure 1 𝒙

was obtained using the transfer matrix method of linear V2

Vt 𝛼t

multibody system MSTMM [8]. p x2

During launch process, all forces and moments acting

between barrel and elevating part, elevating part and revolv-

X3

ing part, revolving part and the hull, the hull and road wheels, h2

and road wheels and the ground are regarded as internal V3

forces. The forces that acted on the barrel by projectile [8, 17] X1 h1 x3

BT

and the forces that acted on the barrel and muzzle brake V1 h3

x1

by propellant gas [8] are regarded as external forces of the O

artillery. The recoil resistance between the barrel and elevat-

ing part produced by recoil and counter-recoil mechanism is Figure 2: Coordinate system and directions of ballistic target.

decomposed into external force and internal force, and the

internal force is proportional to the relative displacement

𝑝𝛼 𝑝

between barrel and elevating part. where 𝐶𝐷, 𝐶𝐿𝛼 , 𝐶𝑁 , and 𝐶𝑙 are the aerodynamic drag

force, lift force, Magnus force, and spin damping moment

3. Projectile Flight Trajectory Model coefficients, respectively; V is the projectile velocity vector in

earth fixed coordinate; Vt is the projectile total velocity vector

Due to the complexity of projectile initial disturbance deter- with respect to air; V is the projectile total aerodynamic speed;

mination, a modified point mass 4-DOF model for an 𝜌 is the air density; 𝑚, 𝐼𝑥 , 𝐼𝑦 , 𝑑, and 𝑆ref are the projectile mass,

unguided rigid projectile was stated in [18, 19]. These equa- axial and transverse moments of inertia, reference diameter,

tions assume that the epicyclic pitching and yawing motion and reference area, respectively; and g and Λ are the earth’s

of the projectile are small everywhere along the trajectory; gravitational and Coriolis accelerations vectors.

therefore, the yaw and pitch moments considered in 6-DOF The projectile repose angle 𝛼𝑅 is defined by [19]

trajectory model [19] are neglected.

The projectile modified point mass equations of motion 𝛼𝑅 = x − (cos 𝛼𝑡 ) i, (4)

with respect to earth fixed coordinate system (𝑋1 , 𝑋2 , 𝑋3 ) as

shown in Figure 2 are given by [19] where x is the projectile’s axis of symmetry unit vector and i

is the unit vector in the direction of Vt .

𝑇

And the total aerodynamic angle of attack is [3, 18, 19]

Ẋ= V = [𝑉1 𝑉2 𝑉3 ] , (1)

Vt ⋅ x

𝜌V𝑆ref 𝐶𝐷 𝜌𝑆 𝐶𝛼 𝛼𝑡 = cos−1 . (5)

V̇= − Vt + ref 𝐿 V2 𝛼𝑅 V

2𝑚 2𝑚

𝑝𝛼

(2) The simplified form of the repose angle is given by [19]

𝜌𝑆ref 𝑑𝐶𝑁

+ 𝑝 [Vt × 𝛼𝑅 ] + g + Λ,

2𝑚 2𝐼𝑥 𝑝

𝛼𝑅 = [V̇× Vt ] , (6)

2 𝜌𝑆ref 𝑑V4 𝐶𝑚

𝛼

𝜌𝑆ref 𝑑 V 𝑝

𝑝̇= 𝑝𝐶𝑙 , (3)

2𝐼𝑥 𝛼

where 𝐶𝑚 is the aerodynamic pitching moment coefficient.

4 Advances in Mechanical Engineering

For discrete systems, a good approximation was done for The equations of motion matrix A, summarizing (1)–(9)

the projectile repose angle [20] as multiplied by the time step Δ𝑡, are given by

2𝐼𝑥 𝑝

𝛼𝑅 = [g × Vt ] . (7) Δ𝑡2

𝜌𝑆ref 𝑑V4 𝐶𝑚

𝛼

[G3×3 I3 ⋅ Δ𝑡 O3×1 O3×3 O3×3 ]

[ 2 ]

The Coriolis acceleration due to earth’s rotation and gravita- [G ⋅ Δ𝑡 VV ⋅ Δ𝑡 V𝑝 ⋅ Δ𝑡 V𝛼 ⋅ Δ𝑡 VW ⋅ Δ𝑡]

̇ ̇ ̇ ̇

[ 3×3 ]

tional acceleration are given by [3, 19] A=[

[ O1×3 O1×3 𝑝𝑝̇ ⋅ Δ𝑡 O1×3 O1×3 ]

],

[ ]

0 − cos 𝜇 sin 𝜆 − sin 𝜇 𝑉1 [𝛼 G 𝛼V̇V̇V 𝛼V̇V̇𝑝 𝛼V̇ V̇𝛼 𝛼V̇ V̇W ]

[ V̇3×3 ]

Λ = 2Ω [cos 𝜇 sin 𝜆 0 cos 𝜇 cos 𝜆] [𝑉2 ] ,

ẆV ⋅ Δ𝑡 O3×3

[ O3×3 O3×3 O3×3 ]

[ sin 𝜇 − cos 𝜇 cos 𝜆 0 ] [𝑉3 ] (14)

𝑇

𝑋1 2𝑋2

g = −𝑔 ⋅ [ 1− 0] , where I3 is the 3 × 3 identity matrix.

𝑅𝑒 𝑅𝑒

All elements of matrix A in (14) are stated as follows:

(8)

earth’s angular velocity; 𝜇, 𝜆 are the corresponding latitude G3×3 = [ 0 2𝑔 − 𝑔𝑋2 0] , (15)

and longitude of the firing sight; and 𝑅𝑒 is the average radius [0 0 0]

of the earth (=6356766 m).

Due to spherical earth approximation, the instantaneous

projectile altitude 𝐻 is defined as [19] where 𝑔𝑋2 = 𝑔/𝑋2 , 𝑔 = 𝑔/𝑅𝑒 ,

𝑋12

𝐻 = 𝑋2 + . (9) V̇V = ΛV − 𝜌𝑎 1 𝐶𝐷 VI3 ,

⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

2 ⋅ 𝑅𝑒 (16)

V̇

W

Point Mass Equations of Motion where ΛV = 2 ⋅ Ω ⋅ [ cos 𝜇⋅sin 𝜆 0 cos 𝜇⋅cos 𝜆 ],

sin 𝜇 − cos 𝜇⋅cos 𝜆 0

To develop a discrete time transfer matrix DTTM-4DOF

using the modified point mass equations of motion men- 𝑝𝛼

V̇𝑝 = 𝜌𝑎2 𝐶𝑁 B1 ,

tioned above, a system state vector S is defined by 13 state

variables to define the projectile trajectory during flight time. V̇𝛼 = 𝜌𝑎1 𝐶𝐿𝛼 V2 I3 ,

These state variables are the projectile position, velocity, spin

rate, repose angle, and wind velocity, and are as follows: 𝑝 (17)

𝑝𝑝̇ = 𝜌𝑏1 V𝐶𝑙 ,

S13×1 −𝑏2 𝑝

𝛼V̇

= [Vt ]× ,

= [𝑋1 𝑋2 𝑋3 𝑉1 𝑉2 𝑉3 𝑝 𝛼1 𝛼2 𝛼3 𝑊1 𝑊2 𝑊3 ] .

𝑇 𝜌V4 𝐶𝑚𝛼

(10)

where 𝑎1 = (𝑆ref /2𝑚), 𝑎2 = ((𝑆ref 𝑑)/2𝑚), 𝑏1 = ((𝑆ref ⋅ 𝑑2 )/(2 ⋅

DTTM-4DOF is the relation between any two successive state 𝐼𝑥 )), 𝑏2 = (𝑑/𝑏1 ), and B1 =Vt × 𝛼𝑅 .

vectors, where

The change in wind velocity vector with the projectile

S𝑖+1 𝑖 position ẆV is given by

10×1 = DTTM-4DOF10×13 ⋅ S13×1 . (11)

The 2nd-order Taylor series is given by [16]

𝜕W/𝜕𝑡 𝜕W/𝜕𝑡 𝜕W

2 ẆV = = = . (18)

Δ𝑡 𝜕V 𝜕X/𝜕𝑡 𝜕X

S𝑖+1

10×1 = S𝑖10×1 + Ṡ

𝑖 ̈

𝑖

10×1 ⋅ Δ𝑡 + S10×1 ⋅ , (12)

2

where The Jacobian matrix J of the system, mentioned in (13b),

which represents discretization errors in matrix A, is given

Ṡ

10×1 ⋅ Δ𝑡 = I10×13 A13×13 ⋅ S13×1 , (13a) by

𝑖

̈ Δ𝑡2 𝜕Ṡ

𝑖

Δ𝑡2 𝜕Ṡ Δ𝑡2

𝑖

S10×1 ⋅ =[ ⋅ ] = [( ) ⋅ Ṡ

𝑖

13×1 ⋅ ] O3×3 I3 O3×1 O3×3 O3×3

2 𝜕𝑡 2 𝜕S 10×13 2 [O

[ j22 V̇𝑝 O3×3 j25 ]]

J = [ 3×3 ]. (19)

= [T10×10 J10×13 A13×13 S13×1 ] . [O1×3 j32 𝑝𝑝̇ O1×3 −j32 ]

(13b) [O3×3 j42 j43 O3×3 j45 ]

Advances in Mechanical Engineering 5

with respect to wind velocity and projectile velocity are given Determination Algorithm

by

The algorithm presented here is an iterative method to

V compute (1) the projectile muzzle velocity and line of sight

j25 = 𝜌𝑎1 𝐶𝐷VI3 + [𝜌𝑎1 ((𝐶𝐷 + 𝐶𝐷𝑀 𝑀) t by using the first projectile displacement observed by radar,

V

and (2) the projectile flight attitude and the corresponding

angular momentum by using the Doppler velocity observed

− (2𝐶𝐿𝛼 + 𝐶𝐿𝛼 𝑀 𝑀) 𝛼𝑅 ) by radar. At each time step, the difference between the data

𝑝𝛼

measured by radar and the estimated parameters by using

𝜌𝑎2 𝑝𝐶𝑁𝑀 𝑀 𝑝𝛼 DTTM-4DOF are linearly related as follows:

− B1 ] V𝑇t + 𝜌𝑎2 𝑝𝐶𝑁 [𝛼𝑅 ]× ,

V2

ΔX

(20)

ΔX m11 m12 m13 m14 [[ ΔV ]

].

[ ] =[ ]. (28)

where 𝐶𝑖 𝑀 = 𝜕𝐶𝑖 /𝜕𝑀 is the rate of change of any aerody- ΔV 𝑖+1 m21 m22 m23 m24 [ Δ𝑝 ]

namic coefficient with respect to Mach number [Δ𝛼𝑅 ]𝑖

j22 = ΛV − j25 . (21) At the beginning of flight time, the initial guess of the

projectile velocity V𝑜 is iteratively corrected by adding ΔV𝑜

Similarly, using (3), the rate of change of projectile spin rate using the first radar data available (𝑖 = 1) until the position

with respect to its velocity is given by difference ΔX1 between the estimated and observed value is

𝜌𝑏1 𝑝 𝑝 small enough (less than a given tolerance) as follows:

𝑝

j32 = (𝐶𝑙 + 𝐶𝑙 𝑀) Vt . (22)

V 𝑀

−1

ΔV𝑜 = [m12 ] ΔX(𝑖=1) . (29a)

The partial derivatives of the projectile yaw of repose in (7)

with respect to projectile velocity and spin rate are as follows: The projectile muzzle spin rate is corrected by adding Δ𝑝𝑜

𝛼 𝛼

using the velocity difference ΔV𝑜 as

𝑏2 𝑝 𝑏2 𝑝 (4𝐶𝑚 + 𝐶𝑚 𝑀)

j42 = 4 𝛼

[g]× + 2

𝑀

B2 V𝑇t ,

2𝜋 ⋅ V𝑜 − V𝑜 + ΔV𝑜

𝜌V 𝐶𝑚 𝜌V6 (𝐶𝛼 ) 𝑚

(23) Δ𝑝𝑜 = , (29b)

𝑛⋅𝑑

𝑏2

j43 = B,

𝛼 2 where 𝑛 is the gun twist rate.

𝜌V4 𝐶𝑚

At the beginning of each iteration cycle, the DTTM-

where B2 = g × Vt . 4DOF matrix is computed and hence the projectile states

The time matrix T mentioned in (13b) is given by S𝑖+1

10×1 in (27) to obtain the projectile velocity difference ΔV𝑖+1

which will be reduced until reaching a predefined tolerance

Δ𝑡

I . O3×3 O3×1 O3×3 by adding Δ𝛼𝑅𝑖 to the initial state as

[3 2 ]

[ Δ𝑡 ]

[O I . O O ] −1

T10×10 [

=[ 3×3 3

2 3×1 3×3 ]. (24) Δ𝛼𝑅𝑖 = [m24 ] ΔV𝑖+1 , (30a)

Δ𝑡 ]

[ ]

[ O1×3 O1×3 O1×3 ] 𝛼𝑅𝑖 = 𝛼𝑅𝑖 + Δ𝛼𝑅𝑖 . (30b)

2

[ O3×3 O3×3 O3×1 I3 ]

Then, the estimated parameters are used as initials for the

By Applying (14), (19), and (25) into (12), the overall discrete next time step.

time transfer matrix is given by As shown in Figure 3, during each time step, the projectile

attitude and the corresponding angular momentum can be

S𝑖+1 𝑖

10×1 = [M10×13 + [I10×13 + T10×10 ⋅ J10×13 ] ⋅ A13×13 ] ⋅ S13×1 , estimated using the following procedure.

(25)

I O7×6

(1) The projectile total angle of attack can be computed

where, M10×13 = [ O7×7

3×7 O3×6 ], by

m11 m12 m13 m14 m15

𝛼𝑡 = 𝛼𝑅 . (31)

[m21 m22 m23 m24 m25 ]

DTTM-4DOF = [

[m31

], (26)

m32 m33 m34 m35 ]

(2) The projectile’s axis of symmetry unit vector x is

[m41 m42 m43 m44 m45 ]

computed by [19]

S𝑖+1 𝑖

10×1 = [DTTM-4DOF] ⋅ S13×1 . (27)

x = 𝛼𝑅 + (cos 𝛼𝑡 ) i, (32)

All elements m𝑖𝑗 of the matrix DTTM-4DOF are listed in the

appendix. where i = Vt /|Vt |.

6 Advances in Mechanical Engineering

Radar data

Si[10×1] DTTM-4DOF [ X V p 𝛼R ]T Si+1 = DTTM ∗ Si −+

[Xi+1 Vi+1 ]T

Yes

No

[Xo Vo 𝜃0 po ]T Calculate ΔVo No

ΔX (i=1) ≤ tolerance

Yes j=0

eq. (29a)

i = 0, j = 0 No

Calculate Δ𝛼Ri No ΔVi ≤ tolerance

Radar data length < i

eq. (30a)

Yes

No

i>0 Calculate 𝛼t , x 𝛼R

i= i+1

eq. (31)–(32)

Yes

Calculate h Calculate 𝛼̇ ̇

R , i, ẋ

eq. (36) eq. (33)–(35) Projectile impact point

No

Yes

6DOF model Altitude = 0

at the end of radar data

(3) The rate of change of the projectile total velocity unit Table 1: Corresponding relation between the input point and the

vector is given by [19] sequence number.

i ̇= 𝛼𝑅 + [i × (g × i)] . (33) 1 Ground

2𝑚 V

2–13 𝑃13,1 , 𝑃14,2 , . . . , 𝑃24,12

(4) Due to the small variation in projectile repose angle 14 𝑃37,25

during flight [19], its rate of change is approximated 15 𝑃37,38

to be linear during each time step as 16 𝑃39,38

17 𝑃39,40

𝛼𝑅𝑖+1 − 𝛼𝑅𝑖

𝛼̇

𝑅 ≈ . (34) 18 𝑃41,40

Δ𝑡 19 𝑃41,42

20 𝑃44,0

(5) The rate of change of projectile’s axis of symmetry unit

vector ẋis computed by [19] 21 𝑃44,45

22–29 45 (beam 1)

ẋ= 𝛼̇ ̇

𝑅 + (cos 𝛼𝑡 ) i − (sin 𝛼𝑡 ) i. (35) 30 𝑃45,46

31–68 46 (beam 2)

(6) The projectile angular momentum divided by 𝐼𝑦 in 69 𝑃46,47

earth fixed coordinate is given by [19] 70–104 47 (beam 3)

105–126 48 (beam 4)

𝐼𝑥 ⋅ 𝑝 127–148 49 (beam 5)

h=( ) x + [x × x]̇. (36)

𝐼𝑦 149–229 50 (beam 6)

230 𝑃51,52

6. Simulated Trajectory Generation

In order to validate the proposed algorithm, a simulated 6.1. Natural Vibration Characteristics of the Self-Propelled

noiseless ballistic trajectory for 155 mm high explosive HE Artillery. To compute the vibration characteristics for

spin stabilized projectile is generated to simulate the Doppler 155 mm self-propelled artillery, the input points correspond-

radar velocity measurements data. The projectile mass, ing to sequence numbers of connection points and elements

length, center of mass measured from the nose, and the axial shown in Figure 1 are identified in Table 1.

and transverse mass moments of inertia are 46.5 kg, 0.902 m, All simulation and test results of natural frequencies for

0.593 m, 0.1585 kg/m2 , and 1.8816 kg/m2 , respectively. This the first sixteen ranks are listed in [8], where the simulation

trajectory is generated using the nonlinear 6-DOF trajectory results have a good agreement with that obtained by exper-

model [2, 19] including projectile initial disturbance problem iments. Figures 4 and 5 show part of the resulted system

with 52∘ firing elevating angle. mode shapes, where the 1st mode mainly represents the recoil

Advances in Mechanical Engineering 7

1 0.75

Mode shape

0.5 0.5

0 0.25

−0.5 0

0 40 80 120 160 200

Input point

−0.25

1st mode, X-direction

2nd mode, Z-direction −0.5

0 0.25 0.5 0.75 1

Figure 4: The 1st and 2nd mode shapes in 𝑋- and 𝑍-direction, xo /L b

respectively.

Y-direction

Z-direction

1

Figure 8: Projectile lateral displacement.

0.5

Y-direction

10

0

−0.5

5

−1

0 40 80 120 160 200

Input point

0

3rd mode 0 0.25 0.5 0.75 1

16th mode xo /L b

0.0015

Swing angle (rad)

8

0

xo (m)

4

−0.0015

0 0.25 0.5 0.75 1

0

0 0.25 0.5 0.75 1 xo /L b

Nondimensional launch time Plumb

Sidewise

Figure 6: Projectile longitudinal displacement.

Figure 9: Projectile angular displacement during launch.

revolving motions of turret, cradle, and barrel, the 3rd mode

Vx𝑜 (m/s)

and the 16th mode mainly represents the transverse motion

of barrel and complex motions of other elements. The low

0 rank mode shapes do not represent the transverse elastic

0 0.25 0.5 0.75 1

deformation of the barrel that is only represented in high rank

xo /L b

[8].

6000

6.2. Computation of Projectile Initial Disturbance. The

A x𝑜 (gee)

to gun tube coordinate system are shown in Figures 6–10.

0

The gun tube coordinate system 𝑂3 𝑥𝑂𝑦𝑂𝑧𝑂 is defined as

0 0.25 0.5 0.75 1 follows: the origin 𝑂3 is the intersection point between

xo /L b axis of gun tube and the vertical plane of projectile’s axis of

symmetry, the 𝑥𝑂-axis is along gun tube axis of symmetry

Figure 7: Projectile longitudinal velocity and acceleration. and points to muzzle, the 𝑦𝑂-axis (plumb direction) is vertical

8 Advances in Mechanical Engineering

1500 0.6

Roll rate (rad/s)

1000 0.4

𝛼R|1 (rad)

0.2

500

0

0

0 0.25 0.5 0.75 1 −0.2

xo /L b

−0.4

0 0.1 0.2 0.3 0.4 0.5

6 Time (s)

Swing rate (rad/s)

0.3

3

0.15

0

𝛼R|2 (rad)

0

−3

−0.15

−6

0 0.25 0.5 0.75 1

−0.3

xo /L b 0 0.1 0.2 0.3 0.4 0.5

Plumb Time (s)

Sidewise 0.4

Figure 10: Projectile angular velocity during launch. 𝛼R|3 (rad) 0.2

−0.2

438

−0.4

433 0 0.1 0.2 0.3 0.4 0.5

V1 (m/s)

Time (s)

428

Estimated

6DOF-R

423

6DOF-N

418 Figure 12: Projectile total angle of attack vector versus flight time.

0 0.1 0.2 0.3 0.4 0.5

Time (s)

552 Table 2: Simulation results of initial disturbance of projectile.

547 Initial disturbance Plumb Sidewise

V2 (m/s)

Swing angle/(10−3 rad) 2.11265 −0.06758

537

Swing angle velocity/(rad/s) −2.92724 −3.31593

532

0 0.1 0.2 0.3 0.4 0.5

Time (s)

to 𝑥𝑂-axis and points up, the direction of 𝑧𝑂-axis (swing

3

direction) is determined by the right-hand role. The projectile

2 longitudinal displacement, velocity, and acceleration with

respect to gun tube axis are shown in Figures 6 and 7, where

V3 (m/s)

1

𝐿 𝑏 is the total barrel length. Figure 8 shows the projectile

0 center of gravity plumb and sidewise displacement with

−1 respect to gun tube axis as function of nondimensionalized

−2

longitudinal displacement.

0 0.1 0.2 0.3 0.4 0.5 The projectile axes orientation with respect to line of sight

Time (s) LOS, which includes the projectile roll, plumb and sidewise

swing components, angular displacement, and velocity, are

Estimated

6DOF-R

illustrated in Figures 9 and 10, respectively. The correspond-

6DOF-N ing projectile angular displacement and velocity components

at muzzle are shown in Table 2, where the deflection angle

Figure 11: Projectile velocity vector versus flight time. is the angle between the projectile velocity vector and LOS

Advances in Mechanical Engineering 9

𝑉1 (m/s) 𝑉2 (m/s) 𝑉3 (m/s) 𝛼 (10−3 rad) 𝛽 (10−3 rad) 𝑝 (rad/s) 𝑞 (rad/s) 𝑟 (rad/s)

429.3847 551.9871 1.021021 1.40134 −1.52758 1279.72 −2.92724 −3.15983

1

30

0.8

15

0.6 0

x1

0.4 −15

−30

0.2 0 0.1 0.2 0.3 0.4 0.5

0 0.1 0.2 0.3 0.4 0.5

Time (s)

Time (s)

20

1

10

0.8

0

x2

0.6 −10

−20

0.4 0 0.1 0.2 0.3 0.4 0.5

0 0.1 0.2 0.3 0.4 0.5

Time (s)

Time (s)

30

0.4

𝛿𝛼R3 /𝛿t (rad/s)

15

0.2

0

0

x3

−15

−0.2

−30

−0.4 0 0.1 0.2 0.3 0.4 0.5

0 0.1 0.2 0.3 0.4 0.5

Time (s)

Time (s)

Estimated Estimated

6DOF-R 6DOF-R

6DOF-N 6DOF-N

Figure 13: The projectile’s axis of symmetry unit vector. Figure 14: Rate of change of the projectile AOA.

while projectile is just flying off from muzzle, and the swing projectile’s position and velocity vectors, using the initial

angle is the angle between projectile axes and LOS at the same flight conditions listed in Table 3. The Doppler radar output

moment. data frequency is assumed to be 100 Hz. The 6DOF-R data

The projectile initial flight conditions corresponding to is assumed to be the real projectile trajectory data. The

the self-propelled artillery launch dynamics, and hence, the corresponding impact point range, drift, maximum altitude,

projectile initial disturbance calculations proposed before are velocity, and flight time are 19689.2 m, 1027.85 m, 8180.9 m,

listed in Table 3, which include the projectile initial velocity 338.76 m/s, and 81.932 s, respectively.

𝑉1 , 𝑉2 , 𝑉3 , initial roll 𝑝, pitch 𝑞 and yaw 𝑟 rates, initial angle Secondly, a nondisturbed trajectory simulation has been

of attack 𝛼, and side-slip angle 𝛽. performed as nominal trajectory 6DOF-N, where the initial

flight conditions used are 700 m/s, 52∘ , and 1279.7 rad/s for

7. Estimation Results muzzle velocity, elevating angle, and spin rate, respectively.

The corresponding impact point range, drift, maximum

A computer program has been developed by using the non- altitude, velocity, and flight time are 19729 m, 995 m, 8162 m,

linear 6-DOF dynamic equations stated in [19]. All simula- 338.66 m/s, and 81.829 s, respectively.

tions were computed with fixed time step size of 0.001 s and Finally, the iterative algorithm shown in Figure 3 has been

integrated using Runge-Kutta-Gill method. applied to estimate the projectile attitude, using the generated

Firstly, a simulated noiseless Doppler radar velocity Doppler radar data 6DOF-R, where the initial flight condi-

measurements data 6DOF-R are generated, including the tions are as the data used to simulate the projectile nominal

10 Advances in Mechanical Engineering

90

30

80

𝛿x1 /𝛿t (s−1 )

15

h1 (rad/s)

0 70

−15 60

−30

0 0.1 0.2 0.3 0.4 0.5 50

0 0.1 0.2 0.3 0.4 0.5

Time (s)

Time (s)

20

100

10

𝛿x2 /𝛿t (s−1 )

90

h2 (rad/s)

0

−10 80

−20

0 0.1 0.2 0.3 0.4 0.5 70

Time (s) 0 0.1 0.2 0.3 0.4 0.5

Time (s)

30

20

15

𝛿x2 /𝛿t (s−1 )

10

0

h3 (rad/s)

0

−15

−10

−30

0 0.1 0.2 0.3 0.4 0.5

−20

Time (s) 0 0.1 0.2 0.3 0.4 0.5

Estimated Time (s)

6DOF-R Estimated

6DOF-N 6DOF-R

6DOF-N

Figure 15: Rate of change of the projectile’s axis of symmetry unit

vector. Figure 16: Projectile angular momentum per unit 𝐼𝑦 .

trajectory 6DOF-N. Figures 11, 12, and 13 show that the be −0.0205% and −0.0448%, respectively, as the radar data

estimated projectile velocity, AOA, and attitude, respectively, availability is 15% of total flight time.

are matching very well with the simulated Doppler radar

trajectory 6DOF-R, although the nominal (non-disturbed) 8. Conclusion

trajectory results 6DOF-N are mismatching the 6DOF-R.

Figures 14 and 15 show the rate of change of the projectile The main objective of this paper is determining the effect

AOA and attitude, respectively, during flight time, where the of self-propelled artillery dynamics on the projectile impact

estimated values are matched very well with 6DOF-R except point prediction accuracy, and hence, an iterative projectile

at some peaks/dips data points due to the large difference attitude determination algorithm was developed to accurately

between the data frequency obtained from radar and the compute the projectile ballistic trajectory and the corre-

simulation speed of the real trajectory data (1 : 10). Therefore, sponding impact point parameters by using a Doppler radar

the final estimated value shown in Figure 16, which rep- data. The projectile initial disturbance was computed by

resents the projectile angular momentum per-unit 𝐼𝑦 has solving the interaction between the self-propelled artillery

mismatched with 6DOF-R data during these peaks/dips data and the projectile dynamics during launch process. A discrete

points. time transfer matrix DTTM-4DOF was developed by lin-

For nominal trajectory data 6DOF-N, the impact range 𝜀𝑅 earizing the projectile nonlinear modified point mass equa-

and drift 𝜀𝐷 errors relative to 6DOF-R are 0.202% and −3.2%, tions using second-order Taylor series. The main advantages

respectively. All impact point parameters inaccuracies for the of using DTTM-4DOF are (1) linearly relating any two

proposed algorithm as function of simulated Doppler radar successive radar data, and (2) increasing the computation

data availability are shown in Figure 17, where these errors, speed due to solving a set of algebraic equations instead of

𝜀𝑅 and 𝜀𝐷, can be reduced by approximately 8 and 7 times, using traditional numerical methods. An iterative projectile

respectively, as the radar data is available for the first 10% attitude determination algorithm was developed to accurately

of total flight time and approximately by 10 and 70 times to compute the projectile ballistic trajectory and predict the

Advances in Mechanical Engineering 11

0

0.05

−0.01

0

𝜀R (%)

𝜀R (%)

−0.02

−0.05

−0.03

−0.1 −0.04

0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100

Data availability (%) Data availability (%)

10 1.2

5 0.6

𝜀D (%)

𝜀D (%)

0 0

−5 −0.6

−10 −1.2

0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100

Data availability (%) Data availability (%)

Figure 17: Impact range and drift error versus Doppler Radar data availability.

2

impact point parameters by using a Doppler radar data. In Δ𝑡

[m13 m14 m15 ] = [V̇𝑝 V̇𝛼 V̇W ] ,

order to validate the proposed algorithm, a simulated noise- 2

less trajectory data was generated as Doppler radar velocity Δ𝑡

measurements data 6DOF-R for 155 mm HE spin stabilized m21 = (I3 + j22 ) GΔ𝑡,

2

projectile using the nonlinear six-degree-of-freedom trajec-

2

tory model including projectile initial disturbance problem. Δ𝑡

m22 = I3 + V̇V Δ𝑡 + (j22 V̇V + j25 ẆV ) ,

A nominal ballistic trajectory 6DOF-N was generated with 2

52∘ firing elevating angle and without including the effect of 2

Δ𝑡

launch dynamics to show the importance of the projectile ini- m23 = V̇𝑝 Δ𝑡 + [j22 V̇𝑝 + V̇𝑝 𝑝𝑝̇] ,

tial disturbance. Based on the results obtained, the proposed 2

algorithm estimated the projectile velocity and attitude very Δ𝑡2

well compared to the 6DOF-R trajectory data. The impact m24 = V̇𝛼 Δ𝑡 + j22 V̇𝛼 ,

2

range and drift inaccuracies were reduced from 0.202% and 2

Δ𝑡

−3.2% in case of nominal trajectory 6DOF-N to be −0.0205% m25 = V̇W Δ𝑡 + j22 V̇W ,

and −0.0448% by applying the proposed algorithm as the 2

radar data availability is 15% of total flight time. As the projec- Δ𝑡2

tile dynamic stability increased, the radar data availability can m31 = j32 G ,

2

be decreased. Finally, the proposed algorithm can estimate

2

the projectile impact point accurately without computing the Δ𝑡

m32 = j32 [V̇V − ẆV ] ,

projectile initial disturbance, where different projectiles have 2

different initial disturbances based on many factors, such

Δ𝑡2

as, gun barrel wear and erosion, projectile initial position m33 = 1 + 𝑝𝑝̇Δ𝑡 + [j32 V̇𝑝 + 𝑝𝑝̇

2

] ,

and attitude inside barrel, and the propellant charge amount, 2

composition, and homogeneity. 2

Δ𝑡

[m34 m35 ] = j32 [V̇𝛼 V̇W ] ,

2

Appendix

m41 = [𝛼V̇

+ j42 Δ𝑡] G,

Modified Point Mass Discrete Time

+ j42 Δ𝑡] V̇V + j45 ẆV Δ𝑡,

m42 = [𝛼V̇

Transfer Matrix Elements

All DTTM-4DOF elements mentioned in (26) are expressed + j42 Δ𝑡] V̇𝑝 + j43 𝑝𝑝̇Δ𝑡,

m43 = [𝛼V̇

as follows:

+ j42 Δ𝑡] V̇𝛼 ,

m44 = [𝛼V̇

m11 = I3 + GΔ𝑡2 ,

2 + j42 Δ𝑡] V̇W .

m45 = [𝛼V̇

Δ𝑡

m12 = I3 ⋅ Δ𝑡 + V̇V , (A.1)

2

12 Advances in Mechanical Engineering

numerical derivatives for initial value problems,” Journal of

The research was supported by the Research Fund for Computational Methods in Sciences and Engineering, vol. 4, no.

the Doctoral Program of Higher Education of China 1, pp. 105–114, 2004.

(20113219110025), the Natural Science Foundation of China [17] B. Rong, X. Rui, and L. Tao, “Discrete time transfer matrix

Government (11102089), and the Program for New Century method for launch dynamics modeling and cosimulation of

Excellent Talents in University (NCET-10-0075). self-propelled artillery system,” Journal of Applied Mechanics,

vol. 80, no. 1, 2013.

[18] R. F. Lieske and R. L. McCoy, “Equations of Motion for a

References Modified Point Mass Trajectory,” Report R1314 (AD, 485869),

[1] F. Fresconi, G. Cooper, and M. Costello, “Practical assessment of U.S. Army Ballistic Research Laboratory, Aberdeen Proving

real-time impact point estimators for smart weapons,” Journal of Ground, Md, USA, 1966.

Aerospace Engineering, vol. 24, no. 1, pp. 1–11, 2011. [19] R. L. McCoy, Modern Exterior Ballistics: the Launch and Flight

[2] R. F. Lieske and R. L. McCoy, “Equations of Motion of a Dynamics of Symmetric Projectiles, Schiffer Publishing, 1999.

Rigid Projectile,” Report 1244 (AD, 441598), U.S. Army Ballistic [20] J. W. Bradley, “An Alternative form of the Modified Point-Mass

Research Laboratory, Aberdeen Proving Ground, Md, USA, Equation of Motion,” Report BRL-MR-3875 (AD A229514),

1964. U.S. Army Ballistic Research Laboratory, Aberdeen Proving

[3] M. Khalil, X. Rui, and H. Hendy, “Discrete time transfer Ground, Md, USA, 1990.

matrix method for projectile trajectory prediction,” Journal of

Aerospace Engineering, 2013.

[4] J. D. Pinezich, J. Heller, and T. Lu, “Ballistic projectile tracking

using CW doppler radar,” IEEE Transactions on Aerospace and

Electronic Systems, vol. 46, no. 3, pp. 1302–1311, 2010.

[5] I. I. Shapiro, The Prediction of Ballistic Missile Trajectories From

Radar Observations, McGraw-Hill, New York, NY, USA, 1958.

[6] E. M. Friedman and M. J. Amoruso, “Determining Thrust or

Drag of a Rocket-Assisted Projectile From Doppler Trajec-

tory Data,” Tech. Rep. ARAED-TR-85008, Army Armament

Research Development and Engineering Center, New Jersey, NJ,

USA, 1985.

[7] G. G. Dutta, A. Singhal, A. Kushari, and A. K. Ghosh, “Estima-

tion of drag coefficient from radar-tracked flight data of a cargo

shell,” Defence Science Journal, vol. 58, no. 3, pp. 377–389, 2008.

[8] X. Rui, L. Yun, Y. Lu, B. He, and G. Wang, Transfer Matrix

Method of Multibody System and Its Application, Science Press,

Beijing, China, 2008.

[9] X. Rui, Y. Lu, W. Lu, J. Qian, and G. Wang, “A study on the launch

dynamics of self-propelled artillery,” Acta Armamentarii, vol. 21,

supplement 1, pp. 38–40, 2000.

[10] G. P. O’Hara, “Dynamic analysis of a 155 mm cannon breech,”

Shock and Vibration, vol. 8, no. 3-4, pp. 215–221, 2001.

[11] J. A. Cordes, D. Carlucci, and R. Jafar, “Dynamics of a simplified

155-MM projectile,” in Proceedings of the 21st International Sym-

posium on Ballistics, vol. 2, pp. 1164–1170, Adelaide, Australia,

April 2004.

[12] M. Chowdhury, A. M. R. Frydman, J. Cordes, L. Reinhardt,

and D. Carlucci, “3-D finite-element gun launch simulation

of a surrogate Excalibur 155 mm guided artillery projectile-

modeling capabilities and its implications,” in Proceedings of

the 22nd International Symposium on Ballistics, pp. 259–267,

Vancouver, Canada, November 2005.

[13] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for

linear multibody system,” Multibody System Dynamics, vol. 19,

no. 3, pp. 179–207, 2008.

[14] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer

matrix method for dynamics analysis of complex weapon

systems,” Science China Technological Sciences, vol. 54, no. 5, pp.

1061–1071, 2011.

[15] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer

matrix method for multibody system dynamics,” Multibody

System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2013, Article ID 854583, 9 pages

http://dx.doi.org/10.1155/2013/854583

Research Article

Study on the Dynamics of Laser Gyro Strapdown Inertial

Measurement Unit System Based on Transfer Matrix Method for

Multibody System

Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang, and Qinbo Zhou

Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Copyright © 2013 Gangli Chen et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The dynamic test precision of the strapdown inertial measurement unit (SIMU) is the basis of estimating accurate motion of various

vehicles such as warships, airplanes, spacecrafts, and missiles. So, it is paid great attention in the above fields to increase the dynamic

precision of SIMU by decreasing the vibration of the vehicles acting on the SIMU. In this paper, based on the transfer matrix

method for multibody system (MSTMM), the multibody system dynamics model of laser gyro strapdown inertial measurement

unit (LGSIMU) is developed; the overall transfer equation of the system is deduced automatically. The computational results

show that the frequency response function of the LGSIMU got by the proposed method and Newton-Euler method have good

agreements. Further, the vibration reduction performance and the attitude error responses under harmonic and random excitations

are analyzed. The proposed method provides a powerful technique for studying dynamics of LGSIMU because of using MSTMM

and its following features: without the global dynamics equations of the system, high programming, low order of system matrix,

and high computational speed.

and error compensation. The simplified model which has two

As there is no need to use external information for measure- degrees of freedom is often used for the dynamics analysis of

ments, LGSIMU is widely used in tactical missiles, chariots, LGSIMU [7, 8]. Due to the excessive simplification, the model

torpedoes, warships, airplanes, and so on [1–4]. Inertial can not accurately reflect the influence of some realistic

measurement unit (IMU) usually operates in bad dynamic factors on the system dynamics performance. To analyze the

environment such as vibration, shock, and overload. For dynamics performance of LGSIMU, attention is gradually

instance, broadband random excitations are produced by paid to more realistic dynamics model, which is established

missiles in powered phase and reentry phase flights as a result based on the theory of multibody system dynamics. Liu et

of its engine thrust, jet noise, and turbulent boundary layer al. [9] used the theory of multibody system dynamics to

pressure. LGSIMU is fixed on the vehicle and undergoes establish the dynamics model of LGSIMU and analyze its

the dynamic environments directly, which makes its opera- dynamic characteristics. The ordinary dynamics methods

tional environment very bad. Thus, its dynamic accuracy is are complicated due to the necessity to deduce the global

declined. dynamics equation and inconvenience for engineering appli-

In order to improve the dynamic accuracy, passive control cations. Rui et al. [10] put forward the transfer matrix method

method is usually adopted to restrain the dynamics response for multibody system (MSTMM) which does not need to

of LGSIMU system in the dynamic environment. The meth- establish the global dynamics equation and is highly stylized.

ods such as optimizing the bracket structure, matching the MSTMM has been widely used in the fields of multiple launch

stiffness of isolator, and improving the sustaining mode are rocket system, shipboard gun system, spacecraft, underwater

the main ways of passive control for the moment [5, 6]. towed system, and so on [11–15]. Yang et al. [16] used

2 Advances in Mechanical Engineering

analyzed the effect of shock on the output precision. Ma et

1 m

al. [17] used MSTMM to simulate the output power spectral

density (PSD) of LGSIMU undergoing random excitation.

Their works are both based on the body dynamics equations 2 k d

in MSTMM and modal superposition. w

of gyroscope centroid and the drift produced by structure

deformation greatly reduce the accuracy of strapdown IMU. Figure 1: One degree of freedom system used to introduce

However, due to a lack of the error analysis approach, ordi- MSTMM.

nary inertial instruments can hardly compensate the error

caused by dynamic environment. Savage [18] established the f1,0

technique for compensating the attitude, velocity, and posi-

tion errors under sinusoidal and random vibrations. Savage x1,2

also presented the simplified dynamics model of strapdown

f1,2

IMU system which had two degrees of freedom including

linear and angular motions. It provides an effective approach

Figure 2: Free body diagram of mass 1.

for error analysis and compensation, but the simplified model

cannot reflect the reality perfectly. So it has great engineering

significance to establish more realistic dynamics model for

error analysis of strapdown IMU. Applying Newton’s second law of motion to the lumped mass

In this paper, MSTMM is used to establish more realistic yields

dynamics model of LGSIMU system, elements transfer equa- ̈ = 𝑓1,2 − 𝑓1,0 .

𝑚𝑥1,2 (2)

tions, and transfer matrices. The overall transfer equation

and overall transfer matrix of the system are derived based For the system undergoing harmonic excitation 𝑤 =

on the MSTMM and its automatic deduction method [19]. 𝑊e , its steady state responses can be written as 𝑥1,0 =

iΩ𝑡

The method for computing frequency response functions

𝑋1,0 eiΩ𝑡 , 𝑥1,2 = 𝑋1,2 eiΩ𝑡 , 𝑓1,2 = 𝐹1,2 eiΩ𝑡 , 𝑓1,0 = 𝐹1,0 eiΩ𝑡

of LGSIMU system based on MSTMM is presented. The

[20]. Substituting these equations into (1) and (2), the transfer

responses of LGSIMU system undergoing various sinusoidal

equation of lumped mass 1 can be found as

and random excitations are obtained. Then the dynamics

performance is analyzed. Combining the formulae of errors Z1,0 = U1 Z1,2 , (3)

caused by vibration, the attitude errors of LGSIMU under

sinusoidal and random vibrations are discussed. The method- T

where Z = [𝑋 𝐹] is the state vector which consists of the

ology presented in this paper provides an effective technique complex amplitudes of displacement and internal force. The

for improving the dynamics performance of LGSIMU under transfer matrix is

dynamic environment.

1 0

U1 = [ ]. (4)

𝑚Ω2 1

2. Introduction to MSTMM: A Simple Example

For hinge 2, considering its constitutive relation, one can

In order to describe the proposed method presented in the obtain

following sections conveniently, the simple system shown in

Figure 1 is taken as an example. The system consists of lumped ̇ − 𝑥3,2

𝑓1,2 = 𝑓3,2 = −𝑘 (𝑥1,2 − 𝑥3,2 ) − 𝑑 (𝑥1,2 ̇ ). (5)

mass 1, spring-and-damper hinge 2, and base 3. The free

By substituting the steady state responses into (5), the

boundary is numbered as 0. The mass of the lumped mass

transfer equation of hinge 2 can be acquired as

is 𝑚, and the stiffness and viscous coefficients of hinge 2 are 𝑘

and 𝑑, respectively. The input and output of the system are the Z1,2 = U2 Z3,2 , (6)

displacements of base 3 and lamped mass 1, respectively. The

input of the system is assumed as harmonic excitation 𝑤 = T

where the state vector also has the form Z = [𝑋 𝐹] and the

𝑊eiΩ𝑡 . In the following of this section, the frequency response transfer matrix is

function of the system will be derived by using MSTMM.

1

Using the sign conventions in [10], the positive directions 1 −

of displacements and forces are shown in the free body U2 = [ 𝑘 + iΩ𝑑 ] . (7)

0 1

diagrams of Figures 2 and 3. For the lumped mass, the

displacements of output end and input end are equal, namely, Combining (3) and (6), the overall transfer equation of

the system can be obtained as

Advances in Mechanical Engineering 3

x1,2 3

6 5

f1,2

1 4 2

x3,2

7

10

f3,2 11

16

Figure 3: Free body diagram of spring-and-damper hinge 2.

15

8 12 y

9

where O

T 13 14 z x

Uall = [−I T] , Zall = [ZT1,0 ZT3,2 ] ,

Figure 4: Dynamics model of LGSIMU system.

1

1 − (9)

[ 𝑘 + iΩ𝑑 ]

T = U 1 U2 = [

[

].

2 𝑚Ω2 ] 1∼3, platform 7, and foundation 12 are considered as rigid

𝑚Ω 1 −

[ 𝑘 + iΩ𝑑 ] bodies, respectively; since the accelerometers are fixed on

the platform and their masses are relatively small, they

The boundary conditions of the system are are included in the platform; the connections 4∼6 between

T T platform and gyros are considered as spatial spring-and-

Z1,0 = [𝑋1,0 0] , Z3,2 = [𝑊 𝐹3,2 ] . (10) damper hinges; meanwhile, platform and foundation are

linked by isolators 8∼11, which are regarded as spring-and-

Substituting the boundary conditions into (8) yields damper hinges longitudinally vibrating in space. The motion

of the vehicle is the input of LGSIMU system. The isolators

Uall Zall = −B𝑊, (11) 13∼16 between foundation and platform are considered as

spring-and-damper hinges longitudinally vibrating in space.

where

The inertial coordinate system shown in Figure 4 is used

1 to describe the motion of elements. The state vectors of the

−1 −

[ 𝑘 + iΩ𝑑 ] input and output points of every element and the boundary

Uall = [ ],

T

Zall = [𝑋1,0 𝐹3,2 ] , [ 𝑚Ω2 ]

points take the form

0 1− (12)

[ 𝑘 + iΩ𝑑 ] T

Z = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ] . (15)

1

B=[ ].

𝑚Ω2 The state variables in the state vector are complex

amplitudes of displacements, angular displacements, internal

Solving (11), the unknown elements in the state vectors of forces, and internal torques when the system undergoes

boundary points can be obtained as harmonic excitation, respectively.

−1 𝑊 (𝑘 + iΩ𝑑) 1

Zall = −Uall B𝑊 = [ 2] . (13) 4. Topology Figure of the Dynamics

−𝑚Ω2 + 𝑘 + iΩ𝑑 −𝑚Ω

Model of LGSIMU System

As a result, the frequency response function of the system

is The dynamics model of LGSIMU system is constructed with

dynamics elements including bodies and hinges. In order

𝑋1,0 𝑘 + iΩ𝑑 to describe the transfer relationship among the state vectors

𝐻 (Ω) = = . (14)

𝑊 −𝑚Ω2 + 𝑘 + iΩ𝑑 of elements in the system, the topology figure of the model

will be used for deducing the overall transfer equation of

Equation (14) can easily be verified using classical fre- LGSIMU system. Isolators 8∼11 and 13∼16 can be treated

quency response function analysis (i.e., Newton’s second law as one equivalent hinge element, respectively, which will be

of motion and Fourier transform). shown in Section 5.3 ; thus the dynamics model of LGSIMU

system can be dealt with a tree system. Its topology figure can

3. Dynamics Model of LGSIMU System be got readily as illustrated in Figure 5.

Besides the sign conventions introduced in [10], the sign

According to the elements dynamics properties of LGSIMU conventions used in this paper are introduced as follows.

system, the dynamics model shown in Figure 4 is established

as a multibody system. The system can be divided into (1) A circle I denotes a body element and the number

elements which can be modeled as follows: three gyros inside it is the sequence number of the body element.

4 Advances in Mechanical Engineering

0 0 0 0 where Z𝑖,I and Z𝑖,O denote the state vectors of input and

output ends of element 𝑖, respectively. The transfer matrices

13 14 15 16

are

12 [ O

[ I O3×3 O3×3 ]

]

U𝑖 = [ 3×3 3

̃l ] ,

[𝑚Ω lCO −Ω (𝑚̃lIÕlIC + JI ) I3

2̃ 2

IO ] (17)

8 9 10 11 2

[ 𝑚Ω I3 −𝑚Ω2̃lIC O3×3 I3 ]

5 6 (𝑖 = 1, 2, 3, 12) ,

0 2 7 3 0

where C denotes the mass center and 𝑚 is the mass of rigid

4 body, JI is the inertia matrix relative to the input point I, ̃lIO

and ̃lIC are the cross product matrices from the input point I

Root

1 to the output point O and mass center C, respectively, and ̃lCO

0 is the cross product matrix from the mass center C to output

point O.

Figure 5: Topology figure of LGSIMU system.

Spring-and-Damper Hinges 4∼6. The transfer equations of

elements 4∼6 are

(2) An arrow → denotes a hinge element and the transfer

direction of state vectors, and the number beside it is Z𝑖,O = U𝑖 Z𝑖,I (𝑖 = 4, 5, 6) . (18)

the sequence number of the hinge element.

The transfer matrices are

I3 O3×3 O3×3 U14

(3) Body element 7 is dealt with three input ends and one [O3×3 I3 U23 O3×3 ]

output end and the other body elements are dealt with U𝑖 = [ ]

[O3×3 O3×3 I3 O3×3 ] (𝑖 = 4, 5, 6) , (19)

single input end and single output end.

[O3×3 O3×3 O3×3 I3 ]

(4) For a nonboundary end, the first and second subscript where

𝑖 and 𝑗 (𝑖, 𝑗 ≠0) in the state vector Z𝑖,𝑗 of the end 1

denote the sequence numbers of the adjacent body [− 𝑘 + iΩ𝑑 0 0 ]

element and hinge element, respectively. For a bound- [ 𝑥 𝑥 ]

[ 1 ]

ary end, the second subscript 𝑗 = 0 in the state vector U14 [

=[ 0 − 0 ],

𝑘 + iΩ𝑑 ]

Z𝑖,𝑗 , that is, the second subscript means boundary [ 𝑦 𝑦 ]

[ 1 ]

end, and the first subscript 𝑖 stands for the sequence 0 0 −

number of the element involved. [ 𝑘𝑧 + iΩ𝑑𝑧 ]

(20)

1

0 0

[ 𝑘𝑥 + iΩ𝑑𝑥 ]

(5) The boundary end corresponding to element 1 is [ ]

[ 1 ]

considered as the root, its state vector is denoted as U23 =[

[

0 + iΩ𝑑

0 ].

]

Z1,0 , other boundary ends are considered as the tips, [ 𝑘𝑦 𝑦 ]

and their state vectors are denoted as Z2,0 , Z3,0 , and [ 1 ]

0 0

Z13∼16,0 , respectively. The transfer directions are from [ 𝑘𝑧 + iΩ𝑑𝑧 ]

the tips to the root.

𝑘𝑥 , 𝑘𝑦 , 𝑘𝑧 represent the stiffness coefficients of linear spring,

𝑘𝑥 , 𝑘𝑦 , 𝑘𝑧 denote the stiffness coefficients of rotary spring, 𝑑𝑥 ,

5. Transfer Equations and Transfer 𝑑𝑦 , 𝑑𝑧 stand for the damping coefficients of linear damper,

Matrices of LGSIMU Elements and 𝑑𝑥 , 𝑑𝑦 , 𝑑𝑧 denote the damping coefficients of rotary

damper, respectively.

5.1. Transfer Equations and Transfer Matrices of Gyros 1∼3 and

Foundation 12. The gyros 1∼3 and foundation 12 are rigid

bodies with single input end and single output end and their 5.3. Transfer Equations and Transfer Matrices of Isolators 8∼11

transfer equations are and 13∼16. As four isolators are sandwiched between two

rigid bodies, they can be treated as one new element, namely,

an equivalent spring-and-damper hinge with single input end

and single output end. The input and output points of the

Z𝑖,O = U𝑖 Z𝑖,I (𝑖 = 1, 2, 3, 12) , (16) new element can be selected at any positions on the two rigid

Advances in Mechanical Engineering 5

16 Foundation (𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 ) (𝑖 = 13 ∼ 16). The same goes for their output points

in 𝑂2 𝑥𝑦𝑧. The stiffness and damping coefficients of isolators

y 13∼16 are (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) and (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16),

x

O2 15 respectively. Simplify the system of forces acting on the input

z y

O1

ends of isolators 13∼16 into a system of forces only acting on

x the input point of the new element, and simplify the system

z

of forces acting on the output ends of isolators 13∼16 into a

13

Vehicle system of forces only acting on the output point of the new

14

element. Further, since the principle vectors and principle

Figure 6: Equivalent model of isolators. moments of the two new systems of forces should be equal,

the transfer equation of the equivalent spring-and-damper

hinge can be obtained as

bodies, respectively. Hereby, isolators 13∼16 are taken as an

example to deduce the transfer equation and transfer matrix Z13∼16,O = U13∼16 Z13∼16,I . (21)

of the new element. The input and output points of isolator

13 are selected as those of the new element, respectively. The transfer matrix is

Two coordinate systems 𝑂1 𝑥𝑦𝑧 and 𝑂2 𝑥𝑦𝑧 shown in Figure 6

are introduced for deduction, whose origins are at the −1

equilibrium positions of the input and output points of the U13∼16 = [ I6 (K13∼16 + iΩD13∼16 ) ] , (22)

new element. And their directions are the same with the O6×6 I6

coordinate system shown in Figure 4. The coordinates of

the input points of isolators 13∼16 in 𝑂1 𝑥𝑦𝑧 are noted as where Ω is the frequency of harmonic excitation,

K13∼16 =

16 16 16 16 16

[ 0 − ∑ 𝑘𝑦𝑖 𝑐𝑖 − ∑ 𝑘𝑧𝑖 𝑏𝑖 ∑ (𝑘𝑦𝑖 𝑐𝑖2 + 𝑘𝑧𝑖 𝑏𝑖2 ) − ∑ 𝑘𝑧𝑖 𝑎𝑖 𝑏𝑖 − ∑ 𝑘𝑦𝑖 𝑎𝑖 𝑐𝑖 ]

[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]

[ ]

[ 16 16 16 16 16 ]

[ 2 2 ]

[ ∑ 𝑘𝑥𝑖 𝑐𝑖 0 − ∑ 𝑘 𝑎 − ∑ 𝑘 𝑏 𝑎 ∑ (𝑘 𝑐 + 𝑘 𝑎 ) − ∑ 𝑘 𝑏 𝑐 ]

[ 𝑧𝑖 𝑖 𝑧𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑧𝑖 𝑖 𝑥𝑖 𝑖 𝑖 ]

[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]

[ 16 ]

[ 16 16 16 16 ]

[− ∑ 𝑘 𝑏 ∑ 𝑘 𝑎 0 − ∑ 𝑘 𝑐 𝑎 − ∑ 𝑘 𝑐 𝑏 ∑ (𝑘 𝑏 2

+ 𝑘 𝑎2 ]

) (23)

[ 𝑥𝑖 𝑖 𝑦𝑖 𝑖 𝑦𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑦𝑖 𝑖 ]

[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]

[ ].

[ 16 16 16 ]

[ ]

[ − ∑ 𝑘𝑥𝑖 0 0 0 − ∑ 𝑘𝑥𝑖 𝑐i ∑ 𝑘𝑥𝑖 𝑏i ]

[ 𝑖=13 ]

[ 𝑖=13 𝑖=13 ]

[ ]

[ 16 16 16 ]

[ 0 − ∑ 𝑘 0 ∑ 𝑘 𝑐 0 − ∑ 𝑘 𝑎 ]

[ 𝑦𝑖 𝑦𝑖 i 𝑦𝑖 i ]

[ 𝑖=13 𝑖=13 𝑖=13 ]

[ ]

[ 16 16 16 ]

[ ]

0 0 − ∑ 𝑘𝑧𝑖 − ∑ 𝑘𝑧𝑖 𝑏i ∑ 𝑘𝑧𝑖 𝑎i 0

[ 𝑖=13 𝑖=13 𝑖=13 ]

Replacing the stiffness coefficients (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) (𝑖 = 13 ∼ 16) platform and gyros 2∼3, respectively. The output point is

in K13∼16 by damping coefficients (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16), the connection point between the platform and gyro 1. The

one can immediately achieve D13∼16 . The transfer equation transfer equation of platform 7 is

and transfer matrix of the equivalent spring-and-damper

Z7,O = U7 Z7,I1 + U7,I2 Z7,I2 + U7,I3 Z7,I3 . (24)

hinge of elements 8∼11 have the same forms as (21) and (22).

The transfer matrices are

5.4. Transfer Equations and Transfer Matrices of Platform 7. I3 −̃lI1 O O3×3 O3×3

Platform 7 is considered as a rigid body with three input ends [ ]

[ O3×3 I3 O3×3 O3×3 ]

and single output end. The first input end is the connection U7 = [

[𝑚Ω2̃l 2 ̃ ̃ ̃l ]

],

point between platform 7 and isolator 8 and the second [ CO −Ω (𝑚lI1 O lI1 C + JI1 ) I3 I1 O ]

2 2̃

and third input points are the connection points between [ 𝑚Ω I3 −𝑚Ω lI1 C O3×3 I3 ]

6 Advances in Mechanical Engineering

O3×3 O3×3 O3×3 O3×3 theory of vibration [20], the steady state response can be

[O3×3 O3×3 O3×3 O3×3 ] expressed as 𝑦7,C = 𝑌7,C eiΩ𝑡 and the complex amplitude 𝑌7,C

U7,I𝑟 =[ ] (𝑟 = 2, 3) .

[O3×3 O3×3 I3 ̃lI𝑟 O ] is the frequency response of LGSIMU system.

[O3×3 O3×3 O3×3 I3 ] The boundary conditions of LGSIMU system are

(25) T

Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]1,0 ,

The geometrical equations of platform 7 are

T

H7 Z7,I1 = H7,I𝑟 Z7,I𝑟 , 𝑟 = 2, 3 (26) Z13∼16,0 = [0, 1, 0, 0, 0, 0, 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]13∼16,0 ,

T

where Z2,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]2,0 ,

I3 O3×3 O3×3 O3×3

H7 = [ ], T

Z3,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]3,0 .

O3×3 I3 O3×3 O3×3

(27) (32)

I ̃lI1 I𝑟 O3×3 O3×3

H7,I𝑟 =[ 3 ] (𝑟 = 2, 3) .

O3×3 I3 O3×3 O3×3 Substituting these boundary conditions into (28) yields

The notations in these matrices are similar to those in Uall Zall = −B, (33)

Section 5.1.

where Uall is a 24 × 24 square matrix composed of the 1st∼

6. Automatic Deduction of the Overall 6th, 19th∼30th, and 37th∼42nd columns of Uall , Zall is a

Transfer Equation of LGSIMU System column matrix consisting of the unknown elements in Zall ,

and B is the 14th column of Uall . For a damped system, Uall

According to the topology figure and the automatic deduc- is a nonsingular square matrix. Solving (33), the unknown

tion method of MSTMM, one can obtain the overall transfer elements in the state vectors of boundary points can be

equation of LGSIMU system: acquired as

Uall Zall = 0. (28) −1

Zall = −Uall B. (34)

The overall transfer matrix is

Using the transfer equations of elements, one can easily

−I12 T13∼16−1 T2−1 T3−1 obtain the frequency response function:

Uall = [O6×12 G13∼16−7 G2−7 O6×12 ] , (29)

[O6×12 G13∼16−7 O6×12 G3−7 ] 𝐻 (Ω) = 𝑌7,C = E7,I1 C U8∼11 U12 U13∼16 Z13∼16,0 , (35)

where where E7,I1 C = [0, 1, 0, −𝑐C , 0, 𝑎C , 0, 0, 0, 0, 0, 0], and (𝑎C , 𝑏C ,

T13∼16−1 = U1 U4 U7 U8∼11 U12 U13∼16 , 𝑐C ) denotes the coordinate of the platform’s mass center in the

body fixed coordinate system.

T2−1 = U1 U4 U7,5 U5 U2 , Similarly, the frequency response functions of other

points can be got. Using the frequency response function, one

T3−1 = U1 U4 U7,6 U6 U3 , can easily obtain the responses under sinusoidal and random

(30)

G13∼16−7 = −H7 U8∼11 U12 U13∼16 , excitations.

If the input is 𝑤 = 𝑊 sin(Ω𝑡+𝛼), the steady state response

G2−7 = H7,5 U5 U2 , is

G3−7 = H7,6 U6 U3 . 𝑦7,C = 𝑊 |𝐻 (Ω)| sin (Ω𝑡 + 𝛼 + 𝜑 (Ω)) , (36)

Zall is a column matrix consisting of the state vectors of system where |𝐻(Ω)| is the amplitude of 𝐻(Ω) and 𝜑(Ω) is the phase

boundary points, that is angle of 𝐻(Ω).

T If the input is a stationary random excitation 𝑤(𝑡) with

Zall = [ZT1,0 ZT13∼16,0 ZT2,0 ZT3,0 ] . (31) PSD 𝑆𝑤 (Ω), the response is also a stationary random process

and its PSD can be expressed as

7. Dynamic Responses of LGSIMU System

𝑆𝑦7,C (Ω) = |𝐻 (Ω)|2 𝑆𝑤 (Ω) . (37)

In order to introduce the method based on MSTMM to

calculate the frequency response, the input and output are For the case that the input is a stationary acceleration

assumed as the displacements of the vehicle and the mass random excitation 𝑤(𝑡)̈ with PSD 𝑆𝑤̈ (Ω) and the output is

center of platform along y-axis, respectively. Setting the input ̈ (𝑡) whose PSD

the acceleration of platform’s mass center 𝑦7,C

to be a harmonic excitation 𝑤 = eiΩ𝑡 and according to the is denoted as 𝑆𝑦7,C

̈ (Ω), regarding the relationship between

Advances in Mechanical Engineering 7

acceleration PSD and displacement PSD 𝑆𝑤̈ (Ω) = Ω4 𝑆𝑤 (Ω), If the response PSD of LGSIMU system undergoing

4

𝑆𝑦7,C

̈ (Ω) = Ω 𝑆𝑦7,C (Ω), and (37), one can obtain random excitation is 𝑆O (Ω), the mean value of the attitude

error caused by random vibration is

2

̈(Ω) = |𝐻 (Ω)| 𝑆𝑤̈

𝑆𝑦7,C (Ω) . (38)

𝐸 (𝛿Φ̇Algo )

According to the Wiener-Khintchine theorem, the auto- ∞ 1 sin ΩT𝑙

correlation function of the response can be obtained as = ∫ Ω [1 + (1 − cos ΩT𝑙 )] − 1 𝑆O (Ω) dΩ.

0 3 ΩT𝑙

(44)

1 ∞

𝑅𝑦7,C

̈(𝜏) = ∫ 𝑆 ̈(Ω) eiΩ𝜏 dΩ. (39)

2𝜋 −∞ 𝑦7,C

9. Numerical Example

Substituting (38) into (39) and setting 𝜏 = 0, the response

For the LGSIMU system shown in Figure 4, the mass and

mean square value can be written as

inertia matrix of foundation 12 are given as

1 ∞ 𝑚12 = 1.2 kg,

𝜙𝑦27,C

̈= 𝑅𝑦7,C

̈(0) = ∫ |𝐻 (Ω)|2 𝑆𝑤̈

(Ω) dΩ. (40)

2𝜋 −∞

0.0045 0 0 (45)

Similarly, the response PSDs and mean square values of J12,C = [ 0 0.0045 0 ] kg ⋅ m2 .

other points can be readily achieved. Then one can analyze [ 0 0 0.0045]

the dynamics performance and errors of LGSIMU system.

The mass and inertia matrix of platform 7 are

8. Error Analysis of LGSIMU 𝑚7 = 2 kg,

In order to improve the precision of LGSIMU in the dynamic 0.0075 0 0 (46)

environment, the errors must be compensated. According to J7,C = [ 0 0.0075 0 ] kg ⋅ m2 .

the error analysis method presented in [18], the attitude errors 0 0 0.0075]

are taken as examples to discuss the errors under sinusoidal [

and random vibrations in this section. The masses and inertia matrices of gyros 1∼3 are

Assuming that LGSIMU system is excited by a sinusoidal

excitation 𝑊 sin (Ω𝑡 + 𝛼), the angular motion response of the 𝑚𝑖 = 0.2 kg,

platform can be written as

0.0006 0 0

𝜃 (𝑡) = e𝑥 𝜃𝑥 (𝑡) + e𝑦 𝜃𝑦 (𝑡) + e𝑧 𝜃𝑧 (𝑡) , (41) J𝑖,C = [ 0 0.0006 0 ] kg ⋅ m2 (𝑖 = 1 ∼ 3) .

[ 0 0 0.0006]

where (47)

𝜃𝑥 (𝑡) = Θ𝑥 sin (Ω𝑡 + 𝜙𝑥 ) , The amplitude and phase frequency responses of the

LGSIMU system got by the proposed method and Newton-

𝜃𝑦 (𝑡) = Θ𝑦 sin (Ω𝑡 + 𝜙𝑦 ) , (42) Euler method are shown in Figures 7 and 8, respectively. It

can be seen that the results got by the two methods have good

𝜃𝑧 (𝑡) = Θ𝑧 sin (Ω𝑡 + 𝜙𝑧 ) . agreements, which verifies the effectiveness of the proposed

method. The attitude error rate under sinusoidal vibration

e𝑥 , e𝑦 , and e𝑧 are the unit vectors of the body-fixed coordinate is demonstrated in Figure 9, where the excitation frequency

frame of platform. Θ𝑥 , Θ𝑦 , Θ𝑧 and 𝜙𝑥 , 𝜙𝑦 , 𝜙𝑧 are the varies from 10 to 2000 Hz, the time interval is 20 ms, and the

amplitudes and phase angles of the angular motion steady amplitude of excitation is 0.2 mm. From this figure, it can be

state responses, respectively, which can be computed using concluded that when the excitation frequency is close to the

MSTMM as presented above. natural frequencies of the LGSIMU system (25 Hz, 46 Hz, and

The error caused by sinusoidal vibration is 85 Hz), attitude error rate reaches maximum values. So the

natural frequencies cannot be close to excitation frequency.

1 Currently, the vibration reduction system of LGSIMU is

𝛿Φ̇Algo = e𝑧 ΩΘ𝑥 Θ𝑦 sin (𝜙𝑥 − 𝜙𝑦 ) in the development phase. Typically, its vibration reduction

2

(43) index is the root mean square value of acceleration response

1 sin ΩT𝑙 should be less than 2 g, and the PSD curve comes down

× {[1 + (1 − cos ΩT𝑙 )] − 1} ,

3 ΩT𝑙 rapidly after 200 Hz [21]. In the following, the dynamics

response and the attitude error response of the LGSIMU

where 𝛿Φ̇Algo is the average attitude error rate, 𝑙 is high speed system under random excitation is computed and analyzed.

algorithm computation cycle index, and T𝑙 is time interval The PSD of the excitation is shown in Figure 10 and its

between 𝑙 cycles. root mean square value is 6.03 g. The PSD of the centroid

8 Advances in Mechanical Engineering

10 ×10−4

2

0

1

−10

|H(Ω)| (dB)

0

−20

−1

−30

−2

−40

−3

−50

101 102 103 104 −4

0 500 1000 1500 2000

f (Hz)

f (Hz)

MSTMM

Newton-Euler method Figure 9: Attitude error rate under sinusoidal vibration.

−50

PSD (g 2 /Hz)

0.02

𝜑(Ω) (∘ )

−100 −3 dB/dec

3 dB/dec

−150

20 80 1300 2000

−200 f (Hz)

101 102 103 104

f (Hz) Figure 10: Input PSD.

MSTMM

Newton-Euler method

Figure 8: Phase frequency response. In this paper, the method for calculating the frequency

response function of LGSIMU system is established by using

MSTMM. Based on this method, the responses of LGSIMU

acceleration of platform along y-axis is shown in Figure 11 system are obtained and the dynamics performance is dis-

and its root mean square value is 1.55 g. Since the root cussed. Furthermore, the attitude errors under sinusoidal

mean square value is less than 2 g and the PSD curve comes and random vibrations are calculated and the effect of

down rapidly after 200 Hz, the vibration reduction index damping coefficient on the errors is analyzed. The proposed

is achieved. The isolators 8∼11 are usually identical. The method can also be used to analyze the velocity and position

mean value of the attitude error rate versus the damping errors caused by vibration. The simulation results verify the

coefficient of isolators 8∼11 is shown in Figure 12. The mean effectiveness of the proposed method. Compared with the

value of the attitude error rate decreases with the increasing ordinary method, the proposed method does not need the

of the damping coefficient. Consequently, within the index global dynamics equation. It is also highly stylized, flexible

of vibration reduction, the damping coefficient of isolators for modeling, and easy to program and provides a powerful

should be as large as possible. technique for studying the dynamics of LGSIMU.

Improving the precision of LGSIMU in a dynamic environ- The authors declare that there is no conflict of interests

ment is an important problem over a long period of time. regarding the publication of this paper.

Advances in Mechanical Engineering 9

32–39, March 1992.

[5] J. I. Lahham, D. J. Wigent, and A. L. Coleman, “Tuned

10 −2 support structure for structure-borne noise reduction of inertial

navigator with dithered ring laser gyros (RLG),” in Proceedings

of the IEEE 2000 Position, Location and Navigation Symposium,

PSD (g 2 /Hz)

10−4 [6] J. F. Li, J. J. LIU, and X. L. Tang, “Aseismatic control of system

structure for laser strapdown IMU,” Journal of Chinese Inertial

Technology, vol. 14, no. 1, pp. 21–26, 2006.

10−6 [7] B. Wu, H. F. Wang, T. Qu et al., “Design of mounting plate

for a missile-bone laser gyro SINS,” Journal of Chinese Inertial

Technology, vol. 15, no. 2, pp. 132–135, 2007.

10−8 [8] Y. Xie, Y. J. Lei, J. P. Zhou et al., “Optimum parameter of INS

101 102 103 104 system absorber,” Structure and Environment Engineering, vol.

f (Hz) 32, no. 4, pp. 39–45, 2005.

[9] P. Liu, H. P. Liu, and C. Zhang, “Research on dynamical

Figure 11: Output PSD of platform along y-axis. characteristics of laser gyro strapdown IMU,” Optics and Opto-

×10−3 electronic Technology, vol. 9, no. 5, pp. 70–74, 2011.

3.5

[10] X. T. Rui, L. F. Yun, Y. Q. Lu et al., Transfer Matrix Method for

Multibody System and Its Applications, Science Press, Beijing,

China, 2008.

3 [11] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for

Attitude error rate (∘ /h)

no. 3, pp. 179–207, 2008.

2.5 [12] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer

matrix method for dynamics analysis of complex weapon

systems,” Science China Technological Sciences, vol. 54, no. 5, pp.

1061–1071, 2011.

2

[13] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer

matrix method for multibody system dynamics,” Multibody

System Dynamics, vol. 14, no. 4, pp. 317–344, 2005.

1.5 [14] B. Rong, X. Rui, G. Wang, and H. Yu, “Discrete time transfer

0 500 1000 1500 2000

matrix method for dynamic modeling of complex spacecraft

d (N·s/m)

with flexible appendages,” Journal of Computational and Non-

Figure 12: The mean values of attitude error rates. linear Dynamics, vol. 6, no. 1, Article ID 011013, 2011.

[15] G. P. Wang, B. Rong, L. Tao et al., “Riccati discrete time transfer

matrix method for dynamic modeling and simulation of an

underwater towed system,” Journal of Applied Mechanics, vol.

Acknowledgments 79, no. 1, Article ID 041014, 2012.

The research was supported by Research Fund for the [16] F. F. Yang, X. T. Rui, and L. Ma, “Dynamical model and

Doctoral Program of Higher Education of China (no. simulation of laser gyro strapdown inertial navigation system,”

20113219110025), Natural Science Foundation of China Gov- Journal of Chinese Inertial Technology, vol. 16, no. 3, pp. 301–305,

2008.

ernment (no. 11102089), and Program for New Century

Excellent Talents in University (NCET-10-0075). [17] L. Ma, X. Rui, F. Yang, and B. Rong, “Power spectrum analysis

for laser gyro damping system of strapdown inertial navigation,”

Journal of Vibration Engineering, vol. 22, no. 6, pp. 603–607,

References 2009.

[18] P. G. Savage, Strapdown Analytics, Strapdown Associates, Maple

[1] J. M. Oelschlaeger and L. O. Thielman, “GG1308 ring laser gyro Plain, Minn, USA, 2000.

inertial measurement systems: Honeywell’s low-cost solution

[19] X. T. Rui and J. S. Zhang, “Automatical transfer matrix method

for tactical applications,” in Proceedings of the Position Location

of multibody system,” in Proceedings of the 2nd Joint Inter-

and Navigation Symposium, pp. 528–536, March 1990.

national Conference on Multibody System Dynamics, Stuttgart,

[2] A. T. Daniel, G. John, and R. E. Mark, “Ring laser gyroscope Germany, 2012.

dither drive system and method,” United States Patent, No.

[20] W. T. Thomson and M. D. Dahleh, Theory of Vibration with

4981359, January 1, 1991.

Applications, Prentice Hall, Englewood Cliffs, NJ, USA, 5th

[3] R. D. Curby, G. H. McCammon, and S. Vallev, “Dither controller edition, 1993.

for ring laser angular rotation sensor,” United States Patent, No.

[21] H. F. Wang, B. Wu, M. Zhang et al., “Research on technique of

4740084, April 26, 1988.

aseismatic control for laser strapdown IMU,” Aerospace Control,

[4] J. I. Lahham and J. R. Brazell, “Acoustic noise reduction in the vol. 25, no. 6, pp. 81–85, 2007.

MK49 ship’s inertial navigation system (SINS),” in Proceedings

Hindawi Publishing Corporation

Advances in Mechanical Engineering

Volume 2013, Article ID 216014, 15 pages

http://dx.doi.org/10.1155/2013/216014

Research Article

Recursive Formulation for Dynamic Modeling and Simulation of

Multilink Spatial Flexible Robotic Manipulators

School of Sciences, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China

Copyright © 2013 Zhenjie Qian et al. This is an open access article distributed under the Creative Commons Attribution License,

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The dynamics for spatial manipulator arms consisting of n flexible links and n flexible joints is presented. All the transversal,

longitudinal, and torsional deformation of flexible links are considered. Within the total longitudinal deformation, the nonlinear

coupling term, also known as the longitudinal shortening caused by transversal deformation, also is considered here. Each flexible

joint is modeled as a linearly elastic torsional spring, and the mass of joint is considered. Lagrange’s equations are adopted to

derive the governing equations of motion of the system. The algorithmic procedure is based on recursive formulation using 4 × 4

homogenous transformation matrices where all the kinematical expressions as well as the final equations of motion are suited for

computation. A corresponding general-purpose C++ software package for dynamic simulation is developed. Several examples are

simulated to illustrate the performance of the algorithm.

[4] studied a manipulator consisting of one flexible link and

Flexible multibody dynamics has become a key methodology one flexible joint based on the assumption that the link is

for various engineering fields, such as robotic manipula- constrained to move only in a horizontal plane, whereas the

tor arms, large radar antennas, solar panels, transportation gravity and some coupling terms between the equations of

vehicles, and manufacturing equipment as well as flexible motions have been dropped. It has been demonstrated by

ligament in human musculoskeletal system [1], and so forth. Wallrapp and Schwertassek [5], that the so-called “geometric

In particular, the flexible robotic manipulators are typi- stiffening” problem can be solved by keeping second order

cally precise and complex operated at high speed. They are terms in the expression of the deformations of the material,

designed to be light with low inertia in order to achieve cost- and the amplitude of the flexible motions must remain small

reduction, energy-saving, and high-performance. Therefore, in those formalisms which are generally based on kinematic

the dynamic analysis of flexible manipulator arms is compli- restrictions as regards the flexibility.

cated due to the coupling between the large rigid body motion Low [6] presented vibration analysis of a rotating beam

and deformation. carrying a tip mass at its end by using Hamilton’s principle

A large number of the literature related to the so-called and the associated boundary conditions. Yoo et al. [7] used a

dynamic stiffening effects have been published, which is first non-Cartesian variable along with two Cartesian variables to

proposed by Kane et al. [2] due to link’s high-order coupling describe the elastic deformation and investigate the dynamic

flexibility. It was observed that an industrial link under high- stiffness effect. Ryu et al. [8] put forward a criterion on inclu-

speed rotational motion would exhibit instability problems sion of stiffening effects such that it clarifies the limit of the

when its angular velocity exceeds a certain limit. The incor- validity of the linear modeling method. El-Absy and Shabana

rect simulation solutions are attributed to neglecting the [9] introduced the effect of longitudinal deformation due to

high-order coupling deformation terms in the dynamic equa- bending and studied the influence of geometric stiffness on

tions. Wu and Haug [3] modeled a flexible multibody system instability problem of nonlinear elastic model. Al-Bedoor and

by means of substructure synthesis formulation to account Hamdan [10], based on the condition of inextensibility to

2 Advances in Mechanical Engineering

point, studied a rotating flexible arm deformation undergoing ith link

large planar motion. Hong et al. [11, 12] established higher- Y

ith joint

order rigid-flexible coupling dynamic model of rotating

flexible beam undergoing large overall motion. When the

deformation of flexible beam or the deformation rate is larger,

the first-order or zero-order approximation coupling model ith link deformation

will appear with divergent results, nevertheless the com-

plete high-order coupling model’s results are still converged.

Schwertassek et al. [13] presented the fundamental shape

functions choice in floating frame of reference formulation,

by separating the flexible body motion into a reference

motion and deformation. X

Book [14] used the 4 × 4 homogenous transformation O

matrices and assumed modes to describe the kinematics of Z

the rigid-joint and flexible-link robots, whereas the dynamic

Figure 1: Flexible robotic arms.

model cannot deal with the torsional deformation of links. A

recursive formulation for the spatial kinematic and dynamic

analysis of open chain mechanical systems containing inter- as well as the gravitational potential energy of system, and

connected deformable bodies is given by Changizi and then focuses on the derivation of the recursive rigid-flexible

Shabana [15], and Kim and Haug [16]. Jain and Rodriguez [17] coupling dynamic equations of the system. In the modeling,

developed new spatially recursive dynamics algorithms for the high-order coupling terms related to the non-linear

flexible multibody systems by using spatial operators, which geometry deformation are retained and the recursive strategy

is based on Newton-Euler factorization and innovations fac- for kinematics is adopted. In Section 4, several examples

torization of the system mass matrix. Hwang [18] developed a are simulated to illustrate the performance of the algorithm

recursive formulation for the flexible dynamic manufacturing proposed in the paper. Finally, Section 5 summarizes the

analysis of open-loop robotic systems with the generalized results and draws conclusions from them.

Newton-Euler equations. Zhang and Zhou [19, 20] did a

further work based on Book’s work in [14]. Both the bending 2. Kinematics of Flexible Robots

and torsional flexibility of links were taken into account.

Dynamic simulation of a spatial flexible manipulator arm was As a point of departure, the system considered here is an

given as an example to validate the algorithm. However, the assembly of n flexible links connected by n rotary joints, as

dynamic stiffening effect was not considered yet. shown in Figure 1.

When the system’s operating speed becomes high,

neglecting the flexibility of joints is quite devastating, which 2.1. Simplified Model of Flexible Joint. Figure 2 shows the

will usually give rise to the errors of position precision. A typi- flexible joint model. We can model the flexibility of joint 𝑖 as

cal dynamic model for taking into account the joint flexibility a linear torsional spring with stiffness 𝐾𝑡𝑖 ⋅ 𝐽𝑟𝑖 is the moment

was presented by Spong [21]. In his work, the joint flexibility of inertia of rotor 𝑖 about its spinning axis. And 𝜏𝑖 is the

is modeled as a torsional spring with more emphasis on torque exerted at joint 𝑖. For simplicity, we neglect friction

simplifying the equations of motion for control purposes. or damping in the flexible joint. Let 𝑞1𝑖 be the theoretical

Other dynamic analyses of robots with joint flexibilities can rotational angle of link 𝑖, 𝑞2𝑖 be the real rotational angle of

be seen in the work of Wasfy and Noor [22], Dwivedy and link 𝑖, 𝜀𝑖 be the torsional angle of joint 𝑖, 𝜑𝑖 be the angular

Eberhard [23], and Na and Kim [24]. displacement of rotor 𝑖, and 𝜙𝑖 be the gear ratio, respectively.

As mentioned above, there are a lot of research work The relationships among them are as follows:

in the dynamic modeling and simulation. But it is still very

difficult for us to deal with the dynamics of the complex 𝜀𝑖 = 𝑞2𝑖 − 𝑞1𝑖 , (1)

multibody systems, such as the spatial flexible-link and

flexible-joint robotic manipulators with consideration of the 𝜑𝑖 = 𝜙𝑖 𝑞1𝑖 . (2)

axial, bending, and torsional deformation for links, and

the flexibility and mass effects for joints, and the so-called 2.2. Simplified Model of Flexible Link. Assume that the links

“dynamic stiffening” effects. In this paper, we will present the are slender beams. Analysis here is based on the Euler-

dynamic modeling methodology to include all such terms. Bernoulli beam theory in the elastic small displacements

In the following section, the kinematics of the system are field.

presented, in which coordinate frames are established, and

4 × 4 homogeneous transformation matrices are used to 2.3. Coordinate Systems and Transformation Matrices. To

describe the kinematics of flexible links and flexible joints. express the transformation between different coordinate

The approach of assumed modes is employed to describe the systems clearly, we establish four coordinate systems for link

deformation of the flexible links. Section 3 firstly deals with 𝑖. Fix the coordinate system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 at the proximal end

the description of the kinetic and elastic potential energy of link 𝑖 (oriented so that the 𝑋𝑏 coincides with the neutral

Advances in Mechanical Engineering 3

Here, 𝑥𝑖𝑗 , 𝑦𝑖𝑗 , and 𝑧𝑖𝑗 are the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 components of

the elastic linear displacement mode 𝑗 of link 𝑖 at the origin of

the coordinate (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝜃𝑥𝑖𝑗 , 𝜃𝑦𝑖𝑗 , and 𝜃𝑧𝑖𝑗 are

ith link the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 rotation components of the elastic angular

Kti displacement mode 𝑗 of link 𝑖 at the origin of the coordinate

(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝛿𝑖𝑗 is the time-varying amplitude

q2i of mode 𝑗 of link 𝑖, and 𝑁𝑖 is the number of modes used to

Jri 𝜙i2 describe the deformation of link 𝑖.

Tran

smi

ssio q1i

Define 0 W𝑖 or W𝑖 to be the 4 × 4 homogeneous transfor-

n mation matrix from the base coordinate frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0

𝜏i Roto to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then, we have

r

(i − 0 ̂ 𝑖−1 A𝑖 ,

1)th

lin k

W𝑖 = W𝑖 = W𝑖−1 E𝑖−1 Ai = W (7)

where W

from the base coordinate frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0 to the distal

coordinate system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖−1 of link 𝑖 − 1.

Figure 2: Flexible joint model. 2.4. Velocity of a Point of Link 𝑖. Let 𝑖 h𝑖 (𝜂) be the homo-

geneous coordinates in the system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 of a point of

axis of link 𝑖 in undeformed shape). This will be referred the deformed link 𝑖 at position 𝜂 with the link under an

to as the base reference frame of link 𝑖. Fix the coordinate undeformed condition from the origin of (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then,

system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 at the distal end of link 𝑖. This is the 𝑖

h𝑖 (𝜂) can be approximated as

distal frame of link 𝑖. When link 𝑖 is in its undeformed

state, the distal frame can be located by a pure translation 𝑁𝑖

𝑖 T T

of the base reference frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 along the length 𝐿 𝑖 h𝑖 (𝜂) = [1 𝜂 0 0] + ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]

of link 𝑖. Let (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 and (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 be two Denavit- 𝑗=1

𝑖) and the distal end (at joint 𝑖 + 1) of link 𝑖, respectively. 1 𝑖 𝑖 T

− ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 [ 0 𝑥𝑖𝑗𝑘 (𝜂) 0 0 ] .

When joint 𝑖 is motionless, (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 is coincident with 2 𝑗=1 𝑘=1

(𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖−1 , and matrix HH𝑖−1 𝑖 , that is, the transformation

(8)

matrix between them, is the function of 𝑞2𝑖 . Matrix H𝑖 ,

Here, 𝑥𝑖𝑗𝑘 is the nonlinear strain coupling term, also known

the 4 × 4 homogeneous transformation matrix between

frames (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 and (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 , is a constant matrix. The as the axial shortening due to the bending deformations of

the link. When the flexible links are undergoing a high speed,

transformation matrix Hb𝑖 of (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 and (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 this term will bring the so-called dynamic stiffening effect

is also a constant matrix. Define the joint-transformation which will have a great influence on the dynamic behavior of

matrix A𝑖 of joint 𝑖 to be the transformation matrix from flexible arms. In the dynamic modeling presented here, the

(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖−1 to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then, high-order terms related to the non-linear coupling term are

A𝑖 = dH𝑖−1 HH𝑖−1 retained, which are ignored in the first-order and zero-order

𝑖 Hb𝑖 . (3)

approximation coupling modeling.

Obviously, A𝑖 is a function of 𝑞2𝑖 . Define E𝑖 to be the In terms of the fixed inertial coordinates of the base

link-transformation matrix of link 𝑖 from (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0 , the position 0 h𝑖 or h𝑖 of the point is given as

(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 . According to the assumption of small deforma-

tion of the links, the small deformable angles can be added h𝑖 = W 𝑖 𝑖 h𝑖 . (9)

vectorially. E𝑖 can be written as

𝑁𝑖 Taking the time derivative of the position h𝑖 , we have the

E𝑖 = H𝑖 + ∑𝛿𝑖𝑗 M𝑖𝑗 , (4) velocity of the point as

𝑗=1

𝑑h𝑖 𝑖

in which = ḣ ̇𝑖 ̇

𝑖 = W𝑖 h𝑖 +W𝑖 h𝑖 . (10)

1 0 0 0 𝑑𝑡

[𝐿 𝑖 1 0 0 ]

H𝑖 = [ ]

[ 0 0 1 0] , (5) To accelerate the computation of the matrices Ẇ𝑖 or Ẅ𝑖 , we

use the recursive kinematics method. By differentiating (7),

[ 0 0 0 1] one obtains

0 0 0 0

[𝑥𝑖𝑗 1 −𝜃𝑧𝑖𝑗 𝜃𝑦𝑖𝑗 ] ̂̇𝑖−1 A𝑖 + W

Ẇ𝑖 = W ̂ 𝑖−1 Ȧ𝑖 ,

M𝑖𝑗 = [

[𝑦𝑖𝑗 𝜃𝑧𝑖𝑗

]. (6)

0 −𝜃𝑥𝑖𝑗 ] (11)

[ 𝑧𝑖𝑗 −𝜃𝑦𝑖𝑗 𝜃𝑥𝑖𝑗 0 ] Ẅ ̂̈ ̂̇ ̇ ̂ ̈

𝑖 = W𝑖−1 A𝑖 + 2W𝑖−1 A𝑖 + W𝑖−1 A𝑖 ,

4 Advances in Mechanical Engineering

that the torsional angle of link 𝑖, 𝜃𝑥𝑖 , is expressed as

Ȧ𝑖 = U𝑖 𝑞2𝑖

̇,

𝑁𝑖

(12)

Ä 2

̇+ U𝑖 𝑞2𝑖̈ 𝜃𝑥𝑖 = ∑𝛿𝑖𝑗 𝜃𝑥𝑖𝑗 , (17)

𝑖 = U2𝑖 𝑞2𝑖 .

𝑗=1

2

, and 𝑞2𝑖 is the joint vari- where 𝛿𝑖𝑗 and 𝜃𝑥𝑖𝑗 are mentioned in (4) and (6), respectively.

̇ ̈

able of joint 𝑖. Thus, W𝑖 and W𝑖 can be computed recursively 𝛿𝑖𝑗 is the function of time, whereas 𝜃𝑥𝑖𝑗 is the function of

from Ŵ 𝑖−1 and its derivatives. Here, one additionally needs position 𝜂.

̂ Substituting (10) and (17) into (16), expanding it, and then

W𝑖−1 and its derivatives. These can be computed recursively

summing over all 𝑛 links, one finds the links’ kinetic energy

from W𝑖−1 and its derivatives as follows:

to be

𝑛 𝑛

̂ 𝑖 = W𝑖 E𝑖 ,

W 𝐾𝑏 = ∑𝐾𝑖𝑏 = ∑ Tr {Ẇ𝑖 B3𝑖 WT𝑖 + 2Ẇ𝑖 B2𝑖 WT𝑖 + W𝑖 B1𝑖 WT𝑖 }

𝑖=1 𝑖=1

̂̇𝑖 = Ẇ𝑖 E𝑖 + W𝑖 Ė

W 𝑖, (13)

𝑛 𝑁𝑖 𝑁𝑖

̂̈ ̈ ̇ ̇ ̈ + ∑ ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘

̇𝑇 ,

𝑖𝑗𝑘

W𝑖 = W𝑖 E𝑖 + 2W𝑖 E𝑖 + W𝑖 E𝑖 , 𝑖=1 𝑗=1 𝑘=1

(18)

where

where

𝑁𝑖

Ė ̇ 1 𝐿𝑖

𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 , 𝑇𝑖𝑗𝑘 = ∫ 𝐽 𝜃 𝜃 𝑑𝜂,

𝑗=1 2 0 𝑥𝑖 𝑥𝑖𝑗 𝑥𝑖𝑘

(14) 1 𝐿𝑖 𝑖 ̇𝑖 ̇

𝑁𝑖 B1𝑖 = ∫ 𝜇 h hT 𝑑𝜂,

Ë ̈

𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 .

2 0 𝑖 𝑖 𝑖

(19)

𝑗=1 1 𝐿𝑖 𝑖 𝑖 ̇

B2𝑖 = ∫ 𝜇 h hT 𝑑𝜂,

2 0 𝑖 𝑖 𝑖

3. Dynamics of Flexible Robots 1 𝐿𝑖 𝑖 𝑖 T

B3𝑖 = ∫ 𝜇 h h 𝑑𝜂.

To use Lagrange’s equations, we need the kinetic and potential 2 0 𝑖 𝑖 𝑖

energy of the system. With the consideration of 𝑖 h𝑖 expressed in (8) and its

derivative, consider

3.1. The System Kinetic Energy. The system kinetic energy 𝐾 𝑁𝑖

contains two parts: the links kinetic energy 𝐾𝑏 and the joints 𝑖

ḣ ̇

𝑖 (𝜂) = ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]

T

(20)

𝑁𝑖 𝑁𝑖

𝐾 = 𝐾𝑏 + 𝐾 𝑟 . (15) − ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 [0 𝑥𝑖𝑗𝑘 (𝜂) 0 0] ,

T

𝑗=1𝑘=1

Assume that the links are slender beams, so the rotary

inertia and shear effects can be neglected. Therefore, the the matrices B1𝑖 , B2𝑖 , and B3𝑖 can be written as

present analysis is based on the Euler-Bernoulli beam theory. 𝑁𝑖 𝑁𝑖

Also assume that the links can undergo a large overall rigid B1𝑖 = ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘

̇D

𝑖𝑗𝑘

motion, however the elastic displacements are small. The 𝑗=1 𝑘=1

kinetic energy of the 𝑖th link is 𝑁𝑖 𝑁𝑖 𝑁𝑖

+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘

̇𝛿 (−F − FT )

𝑖𝑙 𝑖𝑗𝑘𝑙

1 𝐿𝑖 𝑖𝑗𝑘𝑙

𝐾𝑖𝑏 = ∫ 𝜇 (𝜂) Tr {ḣ ̇

T

𝑖 h𝑖 } 𝑑𝜂

𝑗=1 𝑘=1 𝑙=1

2 0

(16) 𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖

1 𝐿𝑖 𝜕𝜃 (𝜂, 𝑡)

2

+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙̇𝛿𝑖𝑠 E𝑖𝑗𝑘𝑙𝑠 ,

+ ∫ 𝐽𝑥𝑖 (𝜂) ( 𝑥𝑖 ) 𝑑𝜂, 𝑗=1 𝑘=1 𝑙=1 𝑠=1

2 0 𝜕𝑡

(21)

where Tr{⋅} is the trace operator; 𝜇(𝜂) and 𝐽𝑥𝑖 are the mass per

𝑁𝑖 𝑁𝑖 𝑁𝑖

unit length and the polar moment of inertia per unit length

of the link about the neutral axis 𝑥, respectively. For slender B2𝑖 = ∑ 𝛿𝑖𝑗̇D𝑖𝑗 + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 DT𝑖𝑗𝑘

𝑗=1 𝑗=1 𝑘=1

beams with uniform cross section area along 𝑥 axis, 𝜇(𝜂) = 𝜇𝑖 .

The first term in (16) is the kinetic energy of link 𝑖 accounting 𝑁𝑖 𝑁𝑖

for the rigid-body motion and the lateral and longitudinal + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 (−E𝑖𝑗𝑘 )

deformation due to flexibility, whereas the second term is the 𝑗=1 𝑘=1

Advances in Mechanical Engineering 5

𝑁𝑖 𝑁𝑖 𝑁𝑖

1 Case 2. Link 𝑖 consists of a flexible beam with the con-

+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 (− FT𝑖𝑗𝑘𝑙 − F𝑖𝑙𝑗𝑘 ) centrated mass 𝑚𝑖 at its proximal end. To account for the

𝑗=1 𝑘=1 𝑙=1 2

contribution of the concentrated mass to the kinetic energy

𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖 of link 𝑖, the extra term D𝑖 should be added to the matrix D𝑖 ,

1

+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑙𝑘𝑠 ), where

𝑗=1 𝑘=1 𝑙=1 𝑠=1 2

1

D𝑖 = 𝑚𝑖 [1 0 0 0] [1 0 0 0] .

T

(26)

(22) 2

𝑁𝑖 𝑁𝑖 𝑁𝑖

B3𝑖 = D𝑖 + ∑𝛿𝑖𝑗 (D𝑖𝑗 + DT𝑖𝑗 ) + ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 D𝑖𝑗𝑘 Should the concentrated mass 𝑚𝑖 locate at the position

𝑗=1 𝑗=1 𝑘=1

(𝑥0 , 𝑦0 , 𝑧0 ) near the proximal end, the extra matrix D𝑖 is

modified to be

𝑁𝑖 𝑁𝑖

1 1

+ ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 (− E𝑖𝑗𝑘 − ET𝑖𝑗𝑘 ) 1

D𝑖 = 𝑚𝑖 [1 𝑥0 𝑦0 𝑧0 ] [1 𝑥0 𝑦0 𝑧0 ] .

T

2 2 (27)

𝑗=1 𝑘=1 2

𝑁𝑖 𝑁𝑖 𝑁𝑖

(23)

1 1 Case 3. Link 𝑖 consists of a flexible beam with the concen-

+ ∑ ∑ ∑𝛿𝑖𝑗 𝛿𝑖𝑘 𝛿𝑖𝑙 (− F𝑖𝑗𝑘𝑙 − FT𝑖𝑗𝑘𝑙 ) trated mass 𝑚𝑖 at its distal end. Considering the contribution

𝑗=1 𝑘=1 𝑙=1 2 2

of the concentrated mass to the kinetic energy of link 𝑖, the

𝑁 𝑁 𝑁 𝑁 extra terms D𝑖 , D𝑖𝑗 , D𝑖𝑗𝑘 , E𝑖𝑗𝑘 , F𝑖𝑗𝑘𝑙 , and E𝑖𝑗𝑘𝑙𝑠 should be added

𝑖 𝑖 𝑖 𝑖

1

+ ∑ ∑ ∑ ∑𝛿𝑖𝑗 𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( E𝑖𝑗𝑙𝑘𝑠 ), to the corresponding matrix D𝑖 , D𝑖𝑗 , D𝑖𝑗𝑘 , E𝑖𝑗𝑘 , F𝑖𝑗𝑘𝑙 , and E𝑖𝑗𝑘𝑙𝑠 ,

𝑗=1 𝑘=1 𝑙=1 𝑠=1 4 respectively. Here,

where 1

D𝑖 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝐿 𝑖 0 0] ,

T

1 𝐿𝑖 2

D𝑖 = ∫ 𝜇 L LT 𝑑𝜂, 1

2 0 𝑖 𝑖 𝑖 D𝑖𝑗 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )] ,

T

2

1 𝐿𝑖

D𝑖𝑗 = ∫ 𝜇 L UT 𝑑𝜂, 1

D𝑖𝑗𝑘 = 𝑚𝑖 [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )]

T

2 0 𝑖 𝑖 𝑖𝑗

2

1 𝐿𝑖

D𝑖𝑗𝑘 = ∫ 𝜇 U UT 𝑑𝜂, × [0 𝑥𝑖𝑘 (𝐿 𝑖 ) 𝑦𝑖𝑘 (𝐿 𝑖 ) 𝑧𝑖𝑘 (𝐿 𝑖 )] ,

2 0 𝑖 𝑖𝑗 𝑖𝑘

(24) 1

E𝑖𝑗𝑘 = 𝑚𝑖 [0 𝐿 𝑖 0 0] [0 𝑥𝑖𝑗𝑘 (𝐿 𝑖 ) 0 0] ,

T

1 𝐿𝑖

E𝑖𝑗𝑘 = ∫ 𝜇𝑖 L𝑖 ST𝑖𝑗𝑘 𝑑𝜂, 2

2 0 1

T

1 𝐿𝑖 F𝑖𝑗𝑘𝑙 = 𝑚𝑖 [0 𝑥𝑖𝑗 (𝐿 𝑖 ) 𝑦𝑖𝑗 (𝐿 𝑖 ) 𝑧𝑖𝑗 (𝐿 𝑖 )]

F𝑖𝑗𝑘𝑙 = ∫ 𝜇 U ST 𝑑𝜂, 2

2 0 𝑖 𝑖𝑗 𝑖𝑘𝑙 × [0 𝑥𝑖𝑘𝑙 (𝐿 𝑖 ) 0 0] ,

1 𝐿𝑖 1

E𝑖𝑗𝑘𝑙𝑠 = ∫ 𝜇 S ST 𝑑𝜂, T

E𝑖𝑗𝑘𝑙𝑠 = 𝑚𝑖 [0 𝑥𝑖𝑗𝑘 (𝐿 𝑖 ) 0 0] [0 𝑥𝑖𝑙𝑠 (𝐿 𝑖 ) 0 0] .

2 0 𝑖 𝑖𝑗𝑘 𝑖𝑙𝑠 2

in which (28)

L𝑖 = [1 𝜂 0 0] ,

T

For the calculation of the kinetic energy of the joint 𝑖,

T

we can lump its mass to link 𝑖 − 1 in accordance with the

U𝑖𝑗 = [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)] , (25) assumptions made in [21] for simplicity. Thus, the kinetic

T energy needs to be included only in the part accounting for

S𝑖𝑗𝑘 = [0 𝑥𝑖𝑗𝑘 (𝜂) 0 0] . spinning kinetic energy of the rotor of joint 𝑖 as follows:

In (21)–(23), the terms with underline remained in high- 1

order approximation coupling model. 𝐾𝑖𝑟 = 𝐽𝑟𝑖 𝜑𝑖̇

2

. (29)

2

It should be noted that the link shape mentioned above is

restricted to be the slender beam type. In fact, the link shape Considering the relationship in (2). We can obtain the

can further be extended to the other cases. kinetic energy of the joints as follows:

Case 1. Link 𝑖 is the rigid-body with irregular shape. In this 𝑛 𝑛

1

case, 𝑁𝑖 = 0, B1𝑖 = B2𝑖 = 0, and B3𝑖 = D𝑖 . The term D𝑖 is 𝐾𝑟 = ∑𝐾𝑖𝑟 = ∑ ( 𝐽𝑟𝑖 𝜙𝑖2 𝑞1𝑖

2

̇) . (30)

the equivalent of the inertia moment of the rigid-link. It is 𝑖=1 𝑖=1 2

actually the pseu-matrix of the inertia moment in [25] with

its more complex form compared with that of (23). Thus, for 3.2. The System Potential Energy. The potential energy of the

rigid-link, the link shape can be arbitrary, and one should flexible manipulators 𝑉 mainly includes the elastic potential

only input the corresponding matrix D𝑖 . energy of flexible joints 𝑉𝑟 , the elastic potential energy of the

6 Advances in Mechanical Engineering

flexible links 𝑉𝑏 , and their gravitational potential energy 𝑉𝑔 . Finally, by using the same procedure given by Book [14],

Therefore, the whole system potential energy is the gravity potential of the system is

𝑉 = 𝑉𝑟 + 𝑉𝑏 + 𝑉𝑔 . (31) 𝑛

𝑉𝑔 = −gT ∑W𝑖 r𝑖 , (37)

Firstly, the elastic potential energy of joint 𝑖 is 𝑖=1

𝑉𝑖𝑟 = 𝐾𝑡𝑖 (𝑞2𝑖 − 𝑞1𝑖 ) . (32) base, and it has the following form:

2

Secondly, the elastic potential of the flexible links con- gT = [0 𝑔𝑥 𝑔𝑦 𝑔𝑧 ] ;

sidered here have three parts: one by bending about the

transverse 𝑦𝑖 and 𝑧𝑖 axes, one by compressing about the lon- 𝑁𝑖 𝑁 𝑁 (38)

gitudinal 𝑥𝑖 axis, and one by twisting about the longitudinal 1 𝑖 𝑖

r𝑖 = 𝑀𝑖 r𝑟𝑖 + ∑ 𝛿𝑖𝑘 𝜀𝑖𝑘 − ∑ ∑𝛿𝑖𝑘 𝛿𝑖𝑙 𝜅𝑖𝑘𝑙 .

𝑥𝑖 axis. Along an incremental length 𝑑𝜂, the elastic potential 𝑘=1

2 𝑘=1𝑙=1

energy is

Here, 𝑀𝑖 is the total mass of link 𝑖 and r𝑟𝑖 is the homogenous

2

1 𝜕𝜃 2 𝜕𝜃𝑦𝑖 𝜕𝑥 2 coordinates of the gravity center of link 𝑖 (undeformed) in the

𝑑𝑉𝑖𝑏 = {𝐸𝑖 [𝐼𝑧𝑖 ( 𝑧𝑖 ) + 𝐼𝑦𝑖 ( ) + 𝐴 𝑥𝑖 ( 𝑖 ) ] frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 ,

2 𝜕𝜂 𝜕𝜂 𝜕𝜂

T

𝜕𝜃 2 r𝑟𝑖 = [1 𝑟𝑥𝑖 0 0] ,

+ 𝐺𝑖 𝐼𝑥𝑖 ( 𝑥𝑖 ) } 𝑑𝜂. 𝐿𝑖

𝜕𝜂 T

𝜀𝑖𝑘 = ∫ 𝜇𝑖 [0 𝑥𝑖𝑘 𝑦𝑖𝑘 𝑧𝑖𝑘 ] 𝑑𝜂, (39)

(33) 0

𝐿𝑖

Here, 𝜃𝑥𝑖 , 𝜃𝑦𝑖 , and 𝜃𝑧𝑖 are the 𝑖th link’s rotations of the neutral 𝜅𝑖𝑘𝑙 = ∫ 𝜇𝑖 [0 𝑥𝑖𝑘𝑙 0 0] 𝑑𝜂.

T

𝐸𝑖 is Young’s modulus of material of link 𝑖; 𝐺𝑖 is the shear Note that 𝜀𝑖𝑘 and 𝜅𝑖𝑘𝑙 can be found in the top row of D𝑖𝑘 and

modulus of material of link 𝑖; 𝐴 𝑥𝑖 is the 𝑖th link’s area of cross E𝑖𝑗𝑘 , respectively.

section about the 𝑥𝑖 axis; 𝐼𝑦𝑖 and 𝐼𝑧𝑖 are the area moment

of inertia of 𝑖th link’s cross section about the 𝑦𝑖 and 𝑧𝑖 axes,

3.3. Dynamic Equations of the System. We use the Lagrange

respectively; and 𝐼𝑥𝑖 is the polar area moment of inertia of the

method to derive the dynamics of the system and accord with

𝑖th link’s cross section about the neutral axis. Similar to the

the method of [14]. Then, the form of Lagrange’s equations

torsional angle 𝜃𝑥𝑖 of (17), 𝜃𝑦𝑖 and 𝜃𝑧𝑖 can be expressed as

will be as follows.

𝑁𝑖

For the joint variable 𝑞1𝑗 , the following simply joint

𝜃𝑦𝑖 = ∑ 𝛿𝑖𝑘 𝜃𝑦𝑖𝑘 , variable equation:

𝑘=1

(34)

𝑁𝑖 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉

𝜃𝑧𝑖 = ∑ 𝛿𝑖𝑘 𝜃𝑧𝑖𝑘 , ( )− + = 𝜏𝑗 (𝑗 = 1, 2, . . . , 𝑛) . (40)

̇

𝑑𝑡 𝜕𝑞1𝑗 𝜕𝑞1𝑗 𝜕𝑞1𝑗

𝑘=1

where 𝜃𝑦𝑖𝑘 and 𝜃𝑧𝑖𝑘 are mentioned in (6). By integrating 𝑑𝑉𝑖𝑏 For the joint variable 𝑞2𝑗 , the following simply joint

of (33) over the link, and summing over all 𝑛 links, one can variable equation:

obtain 𝑉𝑏 as 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉

𝑛 𝑁𝑖 𝑁𝑖

( )− + =0 (𝑗 = 1, 2, . . . , 𝑛) . (41)

𝑏

𝑛 𝐿𝑖

1 ̇

𝑑𝑡 𝜕𝑞2𝑗 𝜕𝑞2𝑗 𝜕𝑞2𝑗

𝑉 = ∑ (∫ 𝑑𝑉𝑖𝑏 ) = ∑ ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 𝐾𝑖𝑗𝑘 , (35)

𝑖=1 0 2 𝑖=1 𝑗=1 𝑘=1

For the deformation variable 𝛿𝑗𝑓 , the following simply

where deformation variable equation:

𝐾𝑖𝑗𝑘 = 𝐾𝑥𝑖𝑗𝑘 + 𝐾𝑦𝑖𝑗𝑘 + 𝐾𝑧𝑖𝑗𝑘 + 𝐾𝑡𝑖𝑗𝑘 , 𝑑 𝜕𝐾 𝜕𝐾 𝜕𝑉

( )− + =0

𝐿𝑖 𝜕𝑥𝑖𝑗 𝜕𝑥𝑖𝑘 ̇

𝑑𝑡 𝜕𝛿𝑗𝑓 𝜕𝛿𝑗𝑓 𝜕𝛿𝑗𝑓 (42)

𝐾𝑥𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐴 𝑥𝑖 (𝜂) 𝑑𝜂,

0 𝜕𝜂 𝜕𝜂 (𝑗 = 1, 2, . . . , 𝑛; 𝑓 = 1, 2, . . . , 𝑁𝑗 ) .

𝐿𝑖 𝜕𝜃𝑦𝑖𝑗 𝜕𝜃𝑦𝑖𝑘

𝐾𝑦𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐼𝑦𝑖 (𝜂) 𝑑𝜂, Upon the substitution of the system kinetic energy of (15),

0 𝜕𝜂 𝜕𝜂 (36)

the system elastic potential energy of (31), and the gravity

𝐿𝑖 𝜕𝜃𝑧𝑖𝑗 𝜕𝜃𝑧𝑖𝑘 energy of (37) into (40), (41), and (42), thus becoming as

𝐾𝑧𝑖𝑗𝑘 = ∫ 𝐸𝑖 𝐼𝑧𝑖 (𝜂) 𝑑𝜂,

0 𝜕𝜂 𝜕𝜂 follows.

𝐿𝑖 𝜕𝜃𝑥𝑖𝑗 𝜕𝜃𝑥𝑖𝑘 For the joint variable 𝑞1𝑗 ,

𝐾𝑡𝑖𝑗𝑘 = ∫ 𝐺𝑖 𝐼𝑥𝑖 (𝜂) 𝑑𝜂.

0 𝜕𝜂 𝜕𝜂 𝐽𝑟𝑗 𝑛𝑖2 𝑞1𝑗

̈ = −𝐾𝑡𝑗 𝑞1𝑗 + 𝐾𝑡𝑗 𝑞2𝑗 + 𝜏𝑗 (𝑗 = 1, 2, . . . , 𝑛) . (43)

Advances in Mechanical Engineering 7

𝑛 𝑁 𝑁 𝑁

𝜕W𝑖 𝑖

T ̈

𝑖 𝑖

̇𝛿̇

[B3𝑖 Ẅ

T 𝑁𝑖 𝑁𝑖

2∑ Tr { 𝑖 + ∑ H W 𝛿

𝑖𝑘 𝑖 𝑖𝑘 + ∑ ∑ L𝑖𝑘𝑙 WT𝑖 𝛿𝑖𝑘 𝑖𝑙

𝑖=𝑗 𝜕𝑞2𝑗 𝑘=1 𝑘=1 𝑙=1

H𝑖𝑗 = D𝑖𝑗 + ∑ 𝛿𝑖𝑘 DT𝑖𝑗𝑘 + ∑ 𝛿𝑖𝑘 (−E𝑖𝑗𝑘 )

𝑘=1 𝑘=1

𝑁𝑖

̇]}

+ 2 ∑ H𝑖𝑘 ẆT𝑖 𝛿𝑖𝑘 𝑁 𝑁

𝑘=1

𝑖 𝑖

1

+ ∑ ∑𝛿𝑖𝑘 𝛿𝑖𝑙 (− FT𝑖𝑗𝑘𝑙 − F𝑖𝑙𝑗𝑘 )

𝑛 𝑘=1 𝑙=1

2

𝜕W𝑖

= gT ∑ r𝑖 + 𝐾𝑡𝑗 𝑞1𝑗 − 𝐾𝑡𝑗 𝑞2𝑗 .

𝑖=𝑗 𝜕𝑞2𝑗 𝑁 𝑁 𝑁

𝑖 𝑖 𝑖

1

(44) + ∑ ∑ ∑𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑙𝑘𝑠 ) ,

𝑘=1 𝑙=1 𝑠=1

2

For the deformation variables 𝛿𝑗𝑓 (𝑗 =

𝑁𝑖

1, 2, . . . , 𝑛; 𝑓 = 1, 2, . . . , 𝑁𝑗 ), (46)

L𝑖𝑗𝑘 = −E𝑖𝑗𝑘 + ∑𝛿𝑖𝑙 (−F𝑖𝑙𝑗𝑘 )

𝑁 𝑁 𝑁 𝑙=1

𝑛

𝜕W𝑖 𝑖

T ̈

i 𝑖

̇𝛿̇

2 ∑ Tr { [B3𝑖 Ẅ

T

𝑖 + ∑ H 𝑖𝑘 W 𝑖 𝛿𝑖𝑘 + ∑ ∑ L𝑖𝑘𝑙 WT𝑖 𝛿𝑖𝑘 𝑖𝑙

𝑖=𝑗+1 𝜕𝛿𝑗𝑓 𝑘=1 𝑘=1 𝑙=1

𝑁 𝑁

𝑖 𝑖

1

+ ∑ ∑𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑘𝑙𝑠 + ET𝑖𝑗𝑙𝑘𝑠 − E𝑖𝑗𝑙𝑘𝑠 ) ,

𝑙=1 𝑠=1

2

𝑁𝑖

̇]}

+ 2 ∑ H𝑖𝑘 ẆT𝑖 𝛿𝑖𝑘 𝑁𝑖 𝑁𝑖 𝑁𝑖

𝑘=1 N𝑖𝑗𝑘 = DT𝑖𝑗𝑘 + ∑𝛿𝑖𝑙 (−FT𝑖𝑗𝑘𝑙 − F𝑖𝑘𝑗𝑙 ) + ∑ ∑𝛿𝑖𝑙 𝛿𝑖𝑠 ET𝑖𝑗𝑙𝑘𝑠 ,

𝑙=1 𝑙=1 𝑠=1

𝑁 𝑁 𝑁

{ 𝑗 𝑗 𝑗

+ 2 Tr {[ẄH + ∑ W N 𝛿 ̈ + ∑ ∑ ̇ 𝛿̇

W𝑗 P𝑗𝑓𝑘𝑙 𝛿𝑗𝑘 𝑁𝑖

𝑗 𝑗𝑓 𝑗 𝑗𝑓𝑘 𝑗𝑘 𝑗𝑙

𝑘=1 𝑘=1 𝑙=1 P𝑖𝑗𝑘𝑙 = −FT𝑖𝑗𝑘𝑙 + ∑𝛿𝑖𝑠 ET𝑖𝑗𝑠𝑘𝑙 .

{[ 𝑠=1

𝑁𝑗 𝑁𝑗

}

+ 2 ∑ Ẇ𝑗 N𝑗𝑓𝑘 𝛿𝑗𝑘̇ ] WT + 2 ∑ 𝛿 ̈

𝑗} 𝑗𝑘 𝑇𝑗𝑓𝑘 Finally, by complicated derivation and assembling, and

𝑘=1 𝑘=1

] } using the recursive scheme to reduce the number of calcu-

𝑛

𝜕W𝑖 lation, the dynamic equations of the system are obtained in

= gT ∑ ( r𝑖 ) + gT W𝑗 𝜀𝑗𝑓 the following formulation:

𝑖=𝑗+1 𝜕𝛿 𝑗𝑓

𝑁𝑗 𝑁𝑗 = R𝐼 ,

Jz̈ (47)

T

− g W𝑗 ∑ 𝛿𝑗𝑘 𝜅𝑗𝑘𝑙 − ∑ 𝛿𝑗𝑘 𝐾𝑗𝑓𝑘 .

𝑘=1 𝑘=1

(45) where the generalized coordinates column z is defined as

T

z = [𝑞11 𝑞21 𝛿11 𝛿12 ⋅ ⋅ ⋅ 𝛿1𝑁1 𝑞12 𝑞22 𝛿21 ⋅ ⋅ ⋅ 𝛿2𝑁2 ⋅ ⋅ ⋅ 𝑞1𝑛 𝑞2𝑛 𝛿𝑛1 ⋅ ⋅ ⋅ 𝛿𝑛𝑁𝑛 ] , (48)

coefficients of the generalized acceleration z̈ in the system 0 𝐽𝑖𝑗 𝐽𝑖𝑗1 ⋅ ⋅ ⋅ 𝐽𝑖𝑗𝑁𝑗

dynamic equations. It is positive-definite and symmetric. The 𝑗

J𝑖 = (0 𝐼𝑖1𝑗 𝐼𝑖1𝑗1 ⋅ ⋅ ⋅ 𝐼𝑖1𝑗𝑁𝑗 ) . (50)

elements of J are arranged in the order of the generalized .. .. .. ..

coordinate column z, the J𝑖 are diagonal elements, the 𝑗 J𝑖 . . . d .

are off-diagonal elements, and these elements are given as 0 𝐼 𝐼

( 𝑖𝑁𝑖 𝑗 𝑖𝑁𝑖 𝑗1 ⋅ ⋅ ⋅ 𝐼𝑖𝑁𝑖 𝑗𝑁𝑗 )

follows:

𝐽𝑟𝑖 𝑛𝑖2 0 0 ⋅⋅⋅ 0 (I) In the joint variable equation 𝑗 (𝑗 = 1, 2, . . . , 𝑛), the

0 𝐽𝑖𝑖 𝐽𝑖𝑖1 ⋅ ⋅ ⋅ 𝐽𝑖𝑖𝑁𝑖 ̈

inertia coefficient 𝐽𝑗ℎ of the second derivative 𝑞2ℎ of the joint

variable 𝑞2ℎ is

J𝑖 = ( 0 𝐼𝑖1𝑖 𝐼𝑖1𝑖1 ⋅ ⋅ ⋅ 𝐼𝑖1𝑖𝑁𝑖 ) , (49)

.. .. .. ..

. . . d . ̂ 𝑗−1 U𝑗 ) 𝑗 ̃ ℎ UT W

̂T } ,

0 𝐼𝑖𝑁𝑖 𝑖 𝐼𝑖𝑁𝑖 𝑖1 ⋅ ⋅ ⋅ 𝐼𝑖𝑁𝑖 𝑖𝑁𝑖 𝐽𝑗ℎ = 2 Tr {(W F ℎ ℎ−1 (51)

8 Advances in Mechanical Engineering

where for 𝑗 = 1, . . . , 𝑛 − 1; ℎ = 1, . . . , 𝑗,

𝑛

𝑗̃

Fℎ = ∑ 𝑗̃ ̃ T,

W𝑖 B3𝑖 ℎ W 𝑖 (52) 𝐼𝑗𝑓ℎ = 2 Tr {W ̂ ℎ−1 Uℎ [ 𝑗 𝜙̃T MT + ℎ W ̃ 𝑗 H𝑗𝑓 ] WT } ; (59)

ℎ 𝑗𝑓 𝑗

𝑖=max(ℎ,𝑗)

for 𝑗 = 1, . . . , 𝑛 − 1; ℎ = 𝑗 + 1, . . . , 𝑛 − 1,

ℎ̃ 𝑗

W𝑖 = Eℎ Aℎ+1 ⋅ ⋅ ⋅ E𝑖−1 A𝑖 . (53) 𝐼𝑗𝑓ℎ = 2 Tr {(W𝑗 M𝑗𝑓 ) 𝜙̃ℎ UTℎ W

̂T } .

ℎ−1 (60)

Here, W ̃ 𝑖 is the 4 × 4 homogeneous transformation matrix

from the proximal reference frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑗 of link 𝑗 to the ̈

coefficients 𝐼𝑗𝑓ℎ𝑘 of the second derivative 𝛿ℎ𝑘 of the deforma-

proximal reference frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 of link 𝑖. From (52), we tion variable 𝛿ℎ𝑘 (1 ≤ 𝑓 ≤ 𝑁𝑗 , 1 ≤ 𝑘 ≤ 𝑁ℎ ) are expressed as

have 𝐽𝑗ℎ = 𝐽ℎ𝑗 (symmetric). follows:

(II) In the joint variable equation 𝑗 (𝑗 = 1, 2, . . . , 𝑛), the for 𝑗 = ℎ = 𝑛,

̈

inertia coefficient 𝐽𝑗ℎ𝑘 of the second derivative 𝛿ℎ𝑘 of the joint

variable 𝛿ℎ𝑘 (for 1 ≤ 𝑘 ≤ 𝑁𝑘 ) is expressed as follows: 𝐼𝑛𝑓𝑛𝑘 = 2 Tr {N𝑛𝑓𝑘 } + 2𝑇𝑛𝑓𝑘 ; (61)

for ℎ = 𝑛; 𝑗 = 1, . . . , 𝑛,

for 𝑗 = 𝑛; ℎ = 1, . . . , 𝑛 − 1,

𝐽𝑗ℎ𝑘 ̂ 𝑗−1 U𝑗 ) 𝑗 W

= 2 Tr {(W ̃ 𝑛 H𝑛𝑘 WT } ; (54)

𝑛

𝐼𝑛𝑓ℎ𝑘 = 2 Tr {(Wℎ Mℎ𝑘 ) ℎ W𝑛 H𝑛𝑓 WT𝑛 } ; (62)

for ℎ = 𝑗, . . . , 𝑛 − 1; 𝑗 = 1, . . . , 𝑛 − 1, for 𝑗 = 1, . . . , 𝑛 − 1; ℎ = 𝑛,

𝑗

̂ 𝑗−1 U𝑗 [ 𝑗 Fℎ MT + W

𝐽𝑗ℎ𝑘 = 2 Tr {W ̃ ℎ Hℎ𝑘 ] WT } ;

ℎ𝑘 ℎ (55) 𝐼𝑗𝑓𝑛𝑘 = 2 Tr {(W𝑗 M𝑗𝑓 ) 𝑗 W𝑛 H𝑛𝑘 WT𝑛 } ; (63)

for ℎ = 1, . . . , 𝑗 − 1; 𝑗 = 2, . . . , 𝑛, for 𝑗 = ℎ; ℎ = 1, . . . , 𝑛 − 1,

̂ 𝑗−1 U𝑗 ) 𝑗 Fℎ MT WT } .

𝐽𝑗ℎ𝑘 = 2 Tr {(W (56) 𝐼𝑗𝑓ℎ𝑘 = 2 Tr {M𝑗𝑓 𝑗 𝜙ℎ MT𝑗𝑘 + N𝑗𝑓𝑘 } + 2𝑇𝑗𝑓𝑘 ; (64)

ℎ𝑘 ℎ

Here, for 𝑗 = 2, . . . , 𝑛 − 1; ℎ = 1, . . . , 𝑗 − 1,

𝑛

𝑗 𝑗̃

Fℎ = ∑ W𝑖 B3𝑖 ℎ WT𝑖 , 𝐼𝑗𝑓ℎ𝑘 = 2 Tr {W𝑗 [M𝑗𝑓 𝑗 𝜙ℎ + HT𝑗𝑓 WT𝑗 ] MTℎ𝑘 WTℎ } ;

ℎ

(65)

𝑖=max(ℎ+1,𝑗) (57)

ℎ

W𝑖 = Aℎ+1 Eℎ+1 ⋅ ⋅ ⋅ E𝑖−1 A𝑖 , for 𝑗 = 1, . . . , 𝑛 − 2; ℎ = 𝑗 + 1, . . . , 𝑛 − 1,

where 𝑗 W𝑖 is the 4 × 4 homogeneous transformation matrix

from the distal coordinate system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑗 of link 𝑗 to the Here,

proximal coordinate system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 of link 𝑖. Note that

the inertia coefficient for the deformation variable 𝛿ℎ𝑘 in the 𝑛

𝑗̃ 𝑗 ̃ T,

joint equation 𝑗 is the same as that for the joint variable 𝑞2ℎ 𝜙ℎ = ∑ W𝑖 B3𝑖 ℎ W 𝑖

𝑖=max(𝑗+1,ℎ)

in the deformation equation ℎ, 𝑘. This further extends the

symmetry of the inertia matrix and reduces the computation (67)

𝑛

necessary. 𝑗 𝑗

𝜙ℎ = ∑ W𝑖 B3𝑖 ℎ WT𝑖 .

(III) In the joint variable equation 𝑗 (𝑗 = 1, 2, . . . , 𝑛), the 𝑖=max(𝑗+1,ℎ+1)

̈

inertia coefficients 𝐼𝑗𝑓ℎ of the second derivative 𝑞2ℎ of the joint

variable 𝑞2ℎ are expressed as follows: For the symmetry of the coefficients, we have 𝐼𝑗𝑓ℎ𝑘 =

for 𝑗 = 𝑛; ℎ = 1, . . . , 𝑛, 𝐼ℎ𝑘𝑗𝑓 .

̂ ℎ−1 Uℎ ) ℎ W

̃ 𝑛 H𝑛𝑓 WT } ; (V) R𝐼 is generalized force, containing the remaining

𝐼𝑗𝑓𝑛 = 2 Tr {(W 𝑛 (58) dynamics and external forcing terms except impact force as

follows:

T

R𝐼 = [𝑅𝑟1

𝐼 𝐼

𝑅𝐽1 𝐼

𝑅11 𝐼

⋅ ⋅ ⋅ 𝑅1𝑁1

𝐼

𝑅𝑟2 𝐼

𝑅𝐽2 𝐼

𝑅21 𝐼

⋅ ⋅ ⋅ 𝑅2𝑁2

𝐼

⋅ ⋅ ⋅ 𝑅𝑟𝑛 𝐼

𝑅𝐽𝑛 𝐼

𝑅𝑛1 𝐼

⋅ ⋅ ⋅ 𝑅𝑛𝑁𝑛

] . (68)

𝐼 𝐼

The 𝑅𝑟𝑗 is the remains of (43) with the second derivatives The 𝑅𝐽𝑗 is the remains of (44) with the second derivatives

removed as follows: removed as follows: for 𝑗 = 1,

𝐼

𝐼

𝑅𝑟𝑗 = −𝐾𝑡𝑗 𝑞1𝑗 + 𝐾𝑡𝑗 𝑞2𝑗 + 𝜏𝑗 (𝑗 = 1, . . . , 𝑛) . (69) 𝑅𝐽1 = −2 Tr {U1 Q1 } + gT U1 P1 + 𝐾𝑡1 𝑞11 − 𝐾𝑡1 𝑞21 ; (70)

Advances in Mechanical Engineering 9

for 𝑗 = 2, . . . , 𝑛, ̂̈

The value of W𝑖 can be obtained similarly as follows:

𝐼

𝑅𝐽𝑗 ̂ 𝑗−1 U𝑗 Q𝑗 } + gT W

= −2 Tr {W ̂ 𝑗−1 U𝑗 P𝑗 + 𝐾𝑡𝑗 𝑞1𝑗 − 𝐾𝑡𝑗 𝑞2𝑗 . 𝑖 𝑁ℎ

(71) ̂̈

W ̂ ℎ

̈

𝑖 = ∑ (Wℎ−1 Uℎ W𝑖 𝑞2ℎ + ∑ Wℎ Mℎ𝑘 ℎ W ̈

̂ 𝑖 𝛿ℎ𝑘 ̂̈

)+W k𝑖 ,

ℎ=1 𝑘=1

𝐼 (76)

The 𝑅𝑗𝑓 is the remains of (45) with the second derivatives

removed as follows.

For 𝑗 = 1, . . . , 𝑛 − 1, where

ℎ

𝐼

𝑅𝐽𝑗 = −2 Tr {W𝑗 M𝑗𝑓 A𝑗+1 Q𝑗+1 W𝑖 = Eℎ Aℎ+1 ⋅ ⋅ ⋅ A𝑖 E𝑖 ,

ℎ̂

(77)

𝑁𝑗 W𝑖 = Aℎ+1 Eℎ+1 ⋅ ⋅ ⋅ A𝑖 E𝑖 .

+ [Ẅ ̇ ̇

k𝑗 H𝑗𝑓 + 2W𝑗 ( ∑ N𝑗𝑓𝑘 𝛿𝑗𝑘 )

𝑘=1

For accelerating the computation of the generalized

[ ̂̈ 𝑗

𝑁𝑗 𝑁𝑗

inertia matrix of the system, the terms Ẅ ̃

k𝑖 , Wk𝑖 , Q𝑗 , P𝑗 , Fℎ ,

𝑗̃ 𝑗

̇ 𝛿̇)]

+W𝑗 ( ∑ ∑P𝑗𝑓𝑘𝑙 𝛿𝑗𝑘 𝜙ℎ , ℎ F𝑗 , and 𝜙ℎ defined in (52)–(76), respectively, can be

𝑗𝑙

𝑘=1 𝑙=1 calculated recursively as follows:

] (72)

T

×W𝑗 } Ẅ 2