You are on page 1of 9

A method for studying multi-body dynamics and its applications to masonry

structures analysis
Giuseppe Gariup , Igino Pitacco & Pere Roca Fabregat
*

**

Departamento de Ingeniera de la Construccon, U.P.C., Barcelona


Dipartimento di Ingegneria Civile, Universit degli Studi di Udine, Udine, Italy

**

ABSTRACT: A multi-body approach to solid dynamics is presented. The numerical method implemented is a
discrete element method (d.e.m.) specifically intended for the analysis of masonry structures. Contacts
between different blocks are treated using interacting forces. Contact forces are modelled using penalty method for the normal component and an elasto-plastic behaviour for the shear. The model considers 3 different
types of damping. A first quantitative validation phase is presented together with possible future developments.
1INTRODUCTION
A brief review of the theoretical details and the numerical implementation of a discrete numerical
method for the analysis of blocky structures is
presented together with an initial phase of validation
of the model.
Discrete element methods were first introduced to
study stability of jointed rock (Cundall 1992) and
were later applied to static and dynamic analysis of
granular material and finally to seismic analysis of
masonry structures (Mamaghani 1999).
Masonry, due to the presence of mortar or dry joints
separating stone blocks or bricks, can be considered
a discontinuous material (Loureno 1996).
Each brick or stone block is modelled as a single
block with infinitesimal deformability. The finite deformation of the structure, typical of a seismic response, is therefore ascribable to the relative rototranslation of the blocks (Fig. 1).
Experimental campaigns on loading walls and macro-blocks structures (Oliveira 2003) and observation of seismic effects on real structures (Ordua
2003) confirm that collapse of masonry structures is
mainly due to relative displacements of bricks or
stones.
The numerical method should be used at first to perform dynamic 2D and 3D analyses of structures
formed by a limited number of macro-blocks, like
stone arches or loading walls, modelling each stone
or brick with a single block. Applicability of this

method to complex structures has been considered in


the last section of this paper.

Figure 1. Collapse due to roto-translation of rigid locks:(a)


geometrical model; (b) degrees of freedom of the system; (c)
first collapse mode (overall excessive rotation); (d) second collapse mode (coupled effect between excessive sliding and rotation).

The model is characterized by the following aspects:

- Formulation of the physical background for the development of the model (Simo 1998). Choice of finite strain theory and hyper-elastic behaviour of
block material.
- Geometrical and finite element definition of the
blocks (Belytschko 2000).
- Definition of the contact algorithm including the
geometrical definition of contact, contact updating
through the analysis, the description of the normal
response using a penalty method approach and the
description of the shear response considering sticking and sliding and plastic behaviour of the shear reaction (Wriggers 2002).
- Definition of damping, considering mass, stiffness
and contact damping (Zhong 1993).
- Numerical method used to solve the system of differential equations representing the f.e. problem.
Choice of an explicit integration algorithm (Bathe
1996).
The phase of validation is focussed on two-block dynamic analysis. The two blocks are superposed, the
lower one serves as ground for the motion of the
upper one and displacements are constrained on its
base. Different cases are modelled and analyzed with
the numerical code. The results are confronted with
analytical ones or with the expected behaviour. Each
case has been specifically chosen to test a particular
characteristic of the model:

- Damping tests: the upper block is inclined with respect to the vertical and touches the lower one with
an edge at the beginning of the analysis. Varying the
damping coefficients different rocking responses are
simulated. Gravity is the only force considered.
In the next sections the theoretical introduction and
the results of the numerical tests are presented.
2CONCEPTUAL MODEL
Discrete element method analyzes the interactions
between a finite number of distinct blocks. Each
block, in the implementation used in this work, is
elastically deformable, while non-linear deformation
is concentrated in the contacts.
All the equilibrium and constitutive equations have
been obtained under the hypothesis of finite deformations. The deformation of a single block has been
considered infinitesimal but the one of the whole
system, due to possible roto-translation of its parts,
can be large compared to the usual finite elements
problems.
Each block or body is composed by one or more 8node hexahedrons depending on the complexity of
the geometry. In the contact routine each body is
identified and interact with the others with its external surface.
The external surface of a block is given by the external faces of the finite elements forming it (Fig. 2).

- Stability: the upper block is leaning on the lower


one at the beginning of the analysis. Gravity is the
only force considered.
- Impact: the upper block is placed at a vertical distance d from the lower one at the beginning of the
analysis. Gravity is the only force considered.
- Friction: the upper block leans on the lower one at
the beginning of the analysis. Gravity is applied together with a horizontal force. Friction is studied
varying the static friction coefficient of the contact.
- Updating of contact: the upper block leans in a instable position on the lower one whose upper face is
inclined. The evolution of motion is given by the
overturning and successive rolling of the upper
block. Gravity is the only force considered.

Figure 2. 2D representation of blocks: (a) single finite element;


(b) elements forming a block; (c) block composition; (d) block
boundary.

At the beginning of the analysis all the possible pairs


of contacting faces are determined and stored in an

array. The uncertainty is introduced by the roughness


of this first check: there can be surfaces included in a
contact pair which are not actually touching each
other but are simply close enough to trick the contact
criterion. During the calculation, at each time step,
each contact pair is considered separately and is
marked as active or not-active. A contact pair is considered active only if one surface is penetrating into
the other and some other criteria established to avoid
singular situations are satisfied. This procedure has
to be repeated during the calculation every n timesteps to follow the physical evolution of the system.
Each active contact pair represents a contact between
two blocks. The procedure to calculate the forces belonging to each contact consists in the following:
- The larger surface is identified as the target-surface
and the other as the hitting-surface.
- If the distance between the 4 nodes of the hitting-surface is higher than a tolerance, a regular grid
of virtual nodes is created on it.
- The position of each node, real or virtual, of the
hitting surface, relative to the target one, is calculated. If the node is not penetrating, the pair hitting-node/target-surface is considered non-active.
- If the node is penetrating the target surface, normal
and shear forces are calculated and added to the
forces acting on the two surfaces.

The normal force at contact is calculated using the


penalty method. In this approach a small penetration
is accepted during the impact of two blocks. After
the penetration, a reaction force, proportional to the
penetration and the penalty stiffness, orthogonal to
the target surface, starts to counteract the superposition of the two bodies. Choosing an adequate penalty
stiffness the penetration can be limited to an acceptable threshold.
The shear force has been modelled with elastoplastic
behaviour. Once a hitting point penetrates for the
first time in the target surface, the natural coordinates XH0 of the normal projection of the point onto
the surface are stored together with the point-surface
contact pair. At the successive time-steps has to be
checked if the point is still penetrating the surface. In
this case the new coordinates of the hitting-point XH1
have to be calculated. The shear force has been modelled with elasto-plastic behaviour. At first the distance d between XH0 and XH1 is calculated and multiplied by the shear stiffness KS. The value obtained is
confronted with the absolute value of the normal
penalty force Fn multiplied by the static friction coefficient ..
If dKS .Fn, the shear force is still in the elastic
range, its absolute value is given by dKS, the direction and verse is given by the vector connecting XH1
and XH0.
If dKS <.Fn, the shear force, calculated as an
elastic reaction to the displacement, exceeds the static friction force. In this case the direction and verse
of the shear force are the ones calculated in the elastic range while the absolute value is scaled to .Fn.
The contact forces, together with the external ones,
are used to calculate the displacements and the deformations of the blocks at the successive time-step.
Central difference method has been implemented to
solve the dynamic differential system given by
Fext Fc Fint
M u

Figure 3. 2D representation of non-active and active contact


pair: (a) lock and bounding box (it includes the block and permit a first rough contact test); (b) non-active contact pair (superposition of bounding-boxes, no block interpenetration); (c)
active contact pair (superposition of bounding-boxes and block
interpenetration).

(1)

where u is the displacement vector, its second


time derivative, M the mass matrix, Fext is the vector
of external force, Fc is the vector of contact forces,
Fint = Ku is the vector of internal forces and K is the
stiffness matrix.
The reasons leading to the choice of an explicit
method in the integration of Equation 1 are mainly
its robustness and its simplicity, suitable characteristics to deal with problems with many non-linearities.
In performing dynamic analysis with any code, it is
usually necessary to account for energy losses in the

physical system which are not considered in the numerical algorithm. In this formulation the
elastoplastic behaviour of contact shear is the only
possibility the system has to dissipate energy. Other
forms of damping have to be introduced in order to
get a better modelling of real structural dynamic behaviour.
The damped system of equations becomes

posing the blocks has been considered possessing the


mechanical properties of stone:

C u Fext Fc Fint
M u

The tangential or shear contact stiffness Ks has been


considered equal to the normal contact stiffness or
penalty parameter Kc, which has been chosen in order to limit the penetration to an acceptable
threshold:

(2)

where C is the damping matrix and u is the first derivative of the displacement vector.
Three types of damping have been introduced:
- Mass damping. The damping matrix is proportional
to the mass matrix:
C M

= 2600 kg/m3,
= 8.33 109 and
= 12.50 109.

Kc = 3.0 107 N/m3 and


Ks = 3.0 107 N/m3.

(3)
3.1 Stability

- Stiffness damping. The damping matrix is proportional to the matrix corresponding to finite element
stiffness:
C K

(4)

- Contact damping. The damping is proportional to a


contact stiffness matrix that takes into account a possible normal damping at contacts:
C Kc

Two cubic blocks A and B with 1.0 m side are superposed.


The lower one B has its base constrained. The gravitational force is the only force acting. Due to the deformability of the blocks this is not an equilibrate
configuration.
The geometrical configuration, forces and constraints and node numbering are summarize in Figure 4.

(5)

The values of constants and have to be determined to adjust the model to experimental observations.
The third form of damping seems the one with
stronger connection to the dissipative nature of the
impact. The first one has no physical meaning and it
is purely a numerical tool to dissipate energy. Stiffness damping is related to the dissipation due to
anelastic phenomenons in the deformation of the
blocks which are normally neglectable in masonry
analysis.
In the next section some examples of discrete element analysis are presented and validated.
3EXAMPLES
A first phase of testing and validation of the discrete
element numerical code has been carried out. Numerical simulations of different cases of two-block
interactions have been completed. The material com-

Figure 4. Stability: (a) geometrical model; (b) forces and constraints; (c) node numbering of block A.

A tends to press B which will react deforming and


lowering its upper face. The entity of this displacement is neglectable compared to the residual penetration imputable to the penalty approach. During the
impact A does not encounter a physical obstacle in
B, but experiences a force representing the contact.
This force tends to contrast the interpenetration,
when it reaches a certain value it has a repulsive effect and pushes A upward.
The equilibrium is obtained when the magnitude of
the penalty reaction d Ks equals the weight WA of A.
d is therefore the penetration that has to be tolerated
in the analysis and depends on the penalty stiffness
Kc.
If no damping is considered in the analysis, A will
oscillates indefinitely around the equilibrium position with more or less the same amplitude (Fig. 5).
The analysis was performed without damping to test
the numerical stability of the method.

of the penalty reaction would cause a visible change


in the motion of A.

Figure 6. (a) geometrical model; (b) forces and constraints; (c)


node numbering of block A.

Using the same geometrical configuration contact


and mass damping are included to see the effects of
the dissipation of energy.
A comparison between the undamped and a damped
solution is shown in Figure 7.
Figure 5. Stability: oscillation in the Z component of node 10
around the equilibrium position for an undamped model.

3.2 Impact
Two cubic blocks A and B, with side equal to 1.0 m,
are considered. The blocks are aligned in the horizontal directions. The lower one B has its base constrained.
In the initial configuration the bottom of block A is
placed 10 cm above the upper part of B.
The gravitational force is the only force acting.
The geometrical configuration, forces and constraints and node numbering are summarize in Figure 6.
At first no damping is considered.
A should bounce on B going back periodically to the
initial configuration. At first no damping is considered. A should bounce on B going back periodically to the initial configuration.
This configuration should test the reaction of the
contact, as any error or deviation in the calculation

Figure 7. Z displacement of node 10: comparison between a


model with contact damping equal to 0.1 and an undamped
model.

3.3 Friction
A cubic block A with side equal to 0.5 m, is placed
upon the upper face of a parallelepiped B with base
equal to 0.7 m 4.0 m and height 1.0 m. Besides the

gravitational force, a horizontal force H is considered.


H magnitude is equal to 0.5 WA, where WA is the absolute value of the gravitational force acting in A,
and its direction is the same of the longer edge of B
base.
The geometrical configuration, forces and constraints and node numbering are summarize in Figure 8.
No damping is considered in this analysis.

1
a t2 ,
2

(6)

where a = 0.1g.
The analytical result can be compared with the numerical one (Fig. 9).
Using the same configuration the static frictional
coefficient is set equal to 0.6. The elastic limit of the
shear force is therefore exceeding the applied force
and no plastic sliding should occur. The elastic sliding is neglectable due to the high shear stiffness considered in the model.
The elastic response is clearly depicted in Figure 10.

Figure 10. X displacement of node 10: elastic shear response.


Figure 8. Friction: (a) geometrical model; (b) forces and constraints; (c) node numbering of block A.

This configuration is useful to check the behaviour


of the shear force at contact.
At first the static frictional coefficient is set equal to
0.4. The force applied horizontally is higher than the
elastic limit of the shear force and sliding is expected.

Figure 9. X displacement of node 10: comparison between the


analytical and the numerical solution.

The horizontal distance covered in the time-interval


(0, t) can be easily computed with

3.4 Contact updating


A cubic block A with side equal to 0.7 m lies onto a
45 inclined face (1.0 m 4.2 m) of a hexahedron B.
B has its base constrained. The gravitational force is
the only force acting.

Figure 12. Rolling: visual output of the numerical code at instant t = 0.

Figure 11. Rolling: (a) geometrical model; (b) forces and constraints; (c) node numbering of block A.

The geometrical configuration, forces and constraints and node numbering are summarize in Figure 11.
The static frictional coefficient is set equal to 1.1.
Contact and mass damping have been considered in
the analysis. The initial tilting couple, due to friction, equals the righting one, due to WA, leading to
an instable configuration. A small perturbation, like
the one induced by the deformability of blocks or by
the penalty approach, can start the motion.
A should roll over the inclined face of B. The face of
A impacting with B is changing at each overturn
(Fig. 12).
This configuration is therefore useful to check the
contact updating routine.

3.5 Rocking and damping


A slender block A with base equal to 0.2 m 0.7 m
and height equal to 1.0 m lies with one of the longer
edges of its base onto a parallelepiped B.
B has its base constrained. The shorter edge of A
base forms a 30 angle with the horizontal plane.
The gravitational force is the only force acting.
The geometrical configuration, forces and constraints and node numbering are summarize in Figure 13.
At first no damping is considered. This configuration
should lead to a rocking behaviour of A. The oscillations should not change amplitude during the analysis. Using the same geometrical configuration, contact damping is included. Depending on the magnitude of damping considered, the system should
evolve with different velocities towards the equilibrium.

- The frictional behaviour of the numerical model is


the one anticipated from the elasto-plastic shear-slip
relationship used in the implementation of the code.
- Contact updating is working properly, perceiving
changes in the contact configuration.
- Damping cause significance effects on the oscillation response of the block. Contact damping seems
to be the more suitable and manageable for structural
analysis, at least when the elastic deformations of the
blocks are neglectable.

Figure 13. Rocking: (a) geometrical model; (b) forces and constraints; (c) node numbering of block A.

A comparison between the analysis of an undamped,


a 0.2 and a 0.4 contact damping model is shown in
fig. 14.

Figure 14. Block A oscillation angle: comparison between undamped and damped solutions (contact damping equal to 0.2
and 0.4).

4ANALYSIS OF RESULTS
The initial phase of validation has shown a satisfactory behaviour of the model with regards to the following aspects:
- The method seems to be numerically stable.
- Modelling normal contact force with the penalty
method has given good qualitative results.

Despite the overall good behaviour of the model, a


problem has been encountered in the rocking motion
example. The block reaches the horizontal position
for the first time, for an instant ti lower than the one
obtained with the analytical solution and the difference decreases with the refinement of the discretization of the block.
The error introduced by the numerical method is significative when the parallelepiped is modelled using
only 1 hexahedron and becomes tolerable when the
mesh consisted in 4 identical hexahedrons.
This behaviour is due to the undervaluation of the
moment of inertia and a resulting overvaluation of
the rotation of a block when it has not been modelled
with an adequate number of finite elements. That depends on the fact that the algorithm has been developed considering the mass concentrated in the
barycentre of the element.
Obviously quadrupling the number of the elements
would critically increase the computational cost for
complex structures.
5CONCLUSIONS
In this paper a discrete element method with deformable blocks has been presented. The method is intended for performing seismic dynamic analysis on 3dimensional models of masonry structures.
A problem has been encountered in the last section
when a block is modelled with an insufficient number of hexahedral elements. It is due to the approximation introduced considering a non-consistent
stiffness matrix K.
Possible solutions are using a consistent stiffness
matrix in the calculation or using rigid blocks dynamics, and thus describing roto-translations exactly.
Calculating a consistent stiffness matrix would increase the computational cost of the method.
Rigid-blocks analysis seems to be a viable option
considering that the deformation in bricks and stones

is negligible compared to block displacement. Nevertheless considering bodies deformable could be


used to take into account the deformations of the
mortar joints or possible damages of a block.
Future developments should explore these possibilities, confronting results and choosing the most suitable one.
A validation phase with experimental results (Pea
2005) is also needed and will be taken into account.
REFERENCES
Bathe, K. (1996). Finite element procedures. Prentice Hall.
Belytschko, T. & Liu, W. . M. B. (2000). Nonlinear finite elements for continua and structures. John Wiley & Sons.
Cundall, P. (1992). Universal distinct element code, Users
manual. USA: ITASCA Consulting Group Inc.
Loureno, P. (1996). Computational strategies for masonry
structures. Ph. D. thesis, Delft University of Technology,
Delft, The Netherlands.
Mamaghani, I.H.P. & Aydan, O. . K. Y. (1999). Analysis of
masonry structures under static and dynamic loading by discrete finite element method. Structural Eng. Earthquake
Eng., JSCE 16(2), 7586.
Oliveira, D. (2003). Experimental and numerical analysis of
blocky masonry structures under cyclic loading. Ph. D.
thesis, Universidade do Minho, Guimares, Portugal.
Ordua, A. (2003). Seismic assessment of ancient masonry
structures by rigid blocks limit analysis. Ph. D. thesis,
Universidade do Minho, Guimares, Portugal.
Pea, F. (2005). Dinmica de bloques de piedra.
Experimentacin y simulacin numrica. Laboratorio
Nacional de Engenharia Civil, Proc. 0305/17/14483.
Simo, J.C. & Hughes, T. (1998). Computational inelasticity.
Springer.
Wriggers, P. (2002). Computational contact mechanics. John
Wiley & Sons.
Zhong, Z. (1993). Finite element procedures for contact-impact problems. Oxford University Press.

You might also like