You are on page 1of 8

Available online at www.sciencedirect.

com

ScienceDirect
Procedia IUTAM 22 (2017) 176 – 183

IUTAM Symposium on Nonlinear and Delayed Dynamics of Mechatronic Systems

Sway reduction of a pendulum on a movable support using a delayed


proportional-derivative or derivative-acceleration feedback
Zaihua Wanga,∗, Qi Xub
a State Key Laboratory of Mechanics and Control of Mechanical Structures, and

College of Aerospace Engineering, Nanjing University of Aeronautics and Astronautics, 210016 Nanjing, China
b School of Mechanics and Engineering, Southwest Jiaotong University, 610031 Chengdu, China

Abstract
A pendulum on a movable support has been used widely in the modeling of many mechanical systems, say, as the simplest model
of gantry cranes. This paper presents a comparison study of the stability of the controlled pendulum with a delayed proportional-
derivative feedback controller or a delayed derivative-acceleration feedback controller for sway reduction. Numerical calculation
shows that the delay combinations and the sign combinations of the gains have important impact on the stability of the controlled
pendulum. Although acceleration feedback is used few in applications compared with the widely used proportional-derivative
feedback, the case when the delay in the acceleration feedback is double of the one in the derivative feedback can be particularly
useful in enlarging the stable region in the gain plane, or in enlarging the stable delay interval if the feedback gains are fixed.

©c 2017
2017TheTheAuthors.
Authors.Published
Published byby Elsevier
Elsevier B.V.
B.V. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of organizing committee of the IUTAM Symposium on Nonlinear and Delayed Dynamics of
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review
Mechatronicunder responsibility of organizing committee of the IUTAM Symposium on Nonlinear and Delayed Dynamics of
Systems.
Mechatronic Systems
Keywords: Sway reduction; proportional-derivative-acceleration feedback; time delay; delay combination.

1. Introduction

A pendulum on a movable support, a two-dimensional model shown in Fig. 1, serves as a mechanical model in
many applications. For example, it is the simplest model of gantry cranes that are used for moving heavy objects
within shipyards, ports, factories and so on. The gantry crane is composed of a hoisting cable, a hook, and a support
mechanism that is composed of a trolley moving over a girder. While the hoisting mechanism lifts and lowers and
deposits the payload from one point to the target point, the support mechanism moves the suspension point around the
crane work space. If the gantry moves fast, the payload may start to sway, and the crane operator may lose control of
the payload. It is highly demanded to make the cranes to transport the payload to the desired position accurately as
well as to suppress and eliminate the payload swing rapidly.
Lots of control methods are available for the control of gantry cranes, see for example, the review paper 1 . These
methods can be classified into two major categories: open-loop controls including input shaping and optimal control,

∗ Corresponding author.
E-mail address: zhwang@nuaa.edu.cn

2210-9838 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of organizing committee of the IUTAM Symposium on Nonlinear and Delayed Dynamics of Mechatronic Systems
doi:10.1016/j.piutam.2017.08.024
Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183 177

F

F

Fig. 1. A pendulum on a movable support, which can be served as the simplest model of a gantry crane.

closed-loop controls including linear controls, nonlinear controls, fuzzy controls and adaptive controls. Input-shaping
and optimal control are the most widely used control schemes, which are actually sensitive to variations in the crane
parameter values about the nominal values and changes in the initial conditions and external disturbances. Fuzzy
logic and adaptive control techniques are very hard to tune. Closed-loop controls may lead to the development of limit
cycles that result in large amplitude payload swing, after the stability loss of a steady state. It is important to keep the
closed-loop stable within a wide range of some system parameters.
In vibration control, accelerometer has been widely used to sense dynamical responses of mechanical and struc-
tural systems, due to its merits such as small volume, light weight, low cost and easy installation. Acceleration
feedback, which uses the acceleration signal directly, is usually time-delayed, because accelerometer signals are fre-
quently perturbed by high-frequency noise that should be filtered out. Actually, time delay is a commonly arising
issue in controller, filter, and actuator. In a series of works 2,3 , delayed feedback, which does not require detailed and
precise identification of the system to be controlled, was shown to be an effective means of eliminating cargo sway
in gantry cranes resulting from operator commands. In studying the problem of human postural balance 5 , the de-
layed proportional-derivative-acceleration (PDA, for short) feedback was superior to the corresponding proportional-
derivative (PD, for short) controller in the sense that the PDA controller could stabilize systems with approximately
40 percent larger feedback delays. In the study of a wheeled inverted pendulum like Segway robots, self-balancing
can be achieved simply with a single accelerometer as the sensor even in the presence of time delay, by introducing an
additional damper that did not have a physical connection with the wheels 6 . More recently, some special features of
delayed derivative-acceleration (DA, for short) controller in active vibration control have been observed 7 : the com-
bination of the two delays in acceleration feedback and velocity feedback has a substantial influence on the stable
region in the gain plane. With properly chosen parameter values, a delayed DA controller makes the closed-loop not
only decaying fast but also having a large stable region to avoid the occurrence of self-excited oscillation.
As an application of the observations given in reference7, this paper shows that delayed DA controller or delayed
PD controller is effective in reducing the sway of a pendulum on a movable support.

2. The Motion Equation of the Closed-Loop Under Delayed PDA Feedback

For the simplified pendulum model of the gantry crane in Fig. 1, assume that the cable is inextensible, the crane
rides on rails with viscous damping c1 , the payload rotates about the pivot P with viscous damping c2 in the vertical
plane, and the force applied to the motor is F(t). Let the coordinate origin be fixed at O(0, 0), x(t) be the displacement
of the crane at time t in the horizontal direction, θ(t) be the angle deviation of the payload. The coordinate (xp , yp ) of
the payload is xp = x + l sin θ, yp = −l cos θ, thus, the kinetic energy is
1 2 1 1 1  
T= m ẋ + M( ẋp2 + ẏ2p ) = m ẋ2 + M ẋ2 + l2 θ̇2 + 2l ẋθ̇ cos θ (1)
2 2 2 2
178 Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183

The Rayleigh dissipation function takes the form of Ψ = 12 c1 ẋ2 + 12 c2 θ̇2 . If the potential energy at the pivot point P
is set to zero, the potential energy is V = −Mgl cos θ, where g is the gravitational acceleration. Then the Lagrange
equations yield
ẍ g
(M + m) ẍ + Ml(θ̈ cos θ − θ̇2 sin θ) + c1 ẋ = F, θ̈ + c2 θ̇ + cos θ + sin θ = 0 (2)
l l
It follows that the equation governing θ is a nonlinear differential equation:
 M  g M c1 cos θ 1
1− cos2 θ θ̈ + sin θ + θ̇2 sin θ cos θ − ẋ + c2 θ̇ = − F cos θ (3)
M+m l M+m l(M + m) l(M + m)
The rotational damping c2 is usually very small, also c1 /l(M + m) can be regarded as a very small number because of
the heavy payload. For simplicity, we neglect the term with ẋ in Eq. (3) and generate the linearized equation round
the equilibrium θ = 0 as
m g 1
θ̈ + c2 θ̇ + θ = − F (4)
M+m l l(M + m)
The objective is to design a proper F in feedback form to make the payload move rapidly and smoothly, namely θ = 0
of (4) is asymptotically stable. If the control force F takes the form: F(t) = −K p θ(t − τ1 ) − Kd θ̇(t − τ2 ) − Ka θ̈(t − τ3 ),
where τ1 , τ2 , τ3 ≥ 0 are the delays, a delay differential equation (DDE, for short) is obtained:
m g 1
θ̈(t) + c2 θ̇(t) + θ(t) = (K p θ(t − τ1 ) + Kd θ̇(t − τ2 ) + Ka θ̈(t − τ3 )) (5)
M+m l l(M + m)
It is a retarded delay differential equation (RDDE, for short) if Ka = 0, and a neutral delay differential equation
(NDDE, for short) if Ka  0. In order that θ = 0 of the NDDE be asymptotically stable, it is necessary to have 7
|Ka | m
< ⇔ |Ka | < ml (6)
l(M + m) M + m

Moreover, let ωn = g(M + m)/(ml) be the natural frequency of the pendulum, and tˆ = ωn t, ϕ(tˆ) = θ(t), τ̂1 =
ωn τ1 , τ̂2 = ωn τ2 , τ̂3 = ωn τ3 , k p = K p /(mlω2n ), kd = Kd /(mlωn ), ka = Ka /(ml) and ξ = c2 (M + m)/(2mωn ). Since
, θ̈ = ω2n ddtˆϕ2 , the above DDE can be represented in the following dimensionless form
2
θ̇ = ωn dϕ
dtˆ

d2 ϕ(tˆ) dϕ(tˆ − τ̂2 ) d2 ϕ(tˆ − τ̂3 )


+ 2ξϕ(tˆ) + ϕ(tˆ) = k p ϕ(tˆ − τ̂1 ) + kd + ka (7)
dtˆ2 dtˆ dtˆ2
where |ka | < 1 is assumed true due to the condition (6). For simplicity, the hat in the above notations will be removed.

3. Determination of the Stable Regions in the Gain Planes

Under the condition (6), the equilibrium θ = 0 of the dimensionless DDE is asymptotically stable if and only if the
roots of the characteristic equation

p(λ) := λ2 + 2ξ λ + 1 − k p e−λτ1 − kd λe−λτ2 − ka λ2 e−λτ3 = 0 (8)

have negative real parts only. It is reasonable to assume that the equilibrium θ = 0 of the linearized closed-loop system
(5) is asymptotically stable when τ1 = τ2 = τ3 = 0. This is guaranteed if

k p < 1, kd < 2ξ, −1 < ka < 1 (9)

Because ξ is very small, kd ≤ 0 holds true below. In addition, the delays are assumed to be small.
For fixed delays, it is convenient to find the stable region in the (k p , kd )-plane or in the (ka , kd )-plane by using the
D-subdivision method 7 , on the basis of critical stability condition p(iω) = 0, which gives
ω2 cos(ωτ2 ) + 2ξω sin(ωτ2 ) − cos(ωτ2 ) ω2 sin(ωτ3 ) − 2ξω cos(ωτ3 ) − sin(ωτ3 )
ka = , kd = − (10)
ω2 (cos(ωτ2 ) cos(ωτ3 ) + sin(ωτ2 ) sin(ωτ3 )) ω(cos(ωτ2 ) cos(ωτ3 ) + sin(ωτ2 ) sin(ωτ3 ))
Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183 179

if k p = 0; or

ω2 cos(ωτ2 ) + 2ξω sin(ωτ2 ) − cos(ωτ2 ) ω2 sin(ωτ1 ) − 2ξω cos(ωτ1 ) − sin(ωτ1 )


kp = , kd = − (11)
cos(ωτ1 ) cos(ωτ2 ) + sin(ωτ1 ) sin(ωτ2 ) ω(cos(ωτ1 ) cos(ωτ2 ) + sin(ωτ1 ) sin(ωτ2 ))

if ka = 0. As ω increases from 0 to +∞, the curves defined by (10) and (11) divide the (ka , kd )-plane and the (k p , kd )-
plane into a number of sub-regions, respectively. In each sub-region, the closed-loop system has the same stability,
either asymptotically stable or unstable for all parameters taken in the same sub-region. Fig. 2 shows the case of the
curve (10) on the (ka , kd )-plane, where the sub-region containing (ka , kd ) = (0, 0) is labeled as stable. Further, it can

Fig. 2. The critical curves in (ka , kd ) plane determined by p(iω) = 0 for Eq. (7) with ξ = 0.05, τ2 = 0.1, τ3 = 2τ2 = 0.2 and k p = 0.

be show in Fig. 2 that the triangle-like sub-region containing (0, 0) is the only one that makes the closed-loop system
asymptotically stable. In fact, for the case of k p = 0, τ2 = τ and τ3 = 2τ, implicit differentiation gives
 −1
dλ 2ξ kd 2ka (λτ − 1)
= (2 + )eλτ + kd τ − + (12)
dkd λ λ eλτ

Thus, at the intersect point of the critical curve on the kd axis where ka = 0, kd < 0 and λ = iω hold, one finds
     
dλ dkd
sgn  = sgn  = sgn(kd ) < 0 (13)
dkd dλ

as done in reference 7. Hence, for fixed ka = 0, regarding the characteristic root as a function of kd , the characteristic
equation increases a pair of conjugate complex roots with negative real parts as kd passes through a critical value.
Thus, the connected triangle-like sub-region containing (ka , kd ) = (0, 0) is the only asymptotically stable one.
In the literature, the two delays are usually assumed the same. As found in reference7, the delay combination has an
important impact on the stability of the closed-loop system. Figure 3 shows that the delay combination τ2 = τ, τ3 = 2τ
enables the closed-loop system to have the largest stable region (the bold curve colored in black) in the (ka , kd )-plane,
which is larger than the one encircled by the bold curve colored in red for the delay combination τ2 = τ3 = τ and the
one encircled by the thin curve colored in blue for the delay combination τ2 = 2τ, τ3 = τ, where the condition (6) is
assumed true. A larger stable region means a better robustness of the controller. That is to say, when a delayed DA
control is used in real applications, the case when the acceleration feedback delay is double of the velocity feedback
delay is preferable from the viewpoint of stable region. Figure 3 shows also that the stable region in the (ka , kd )-plane
is sensitive to the change of the delays.
Similar results can be obtained for the delayed PD controller, the case of ka = 0. As shown in Fig. 4 for ξ = 0.05,
ka = 0, τ = 0.1 and τ = 0.2, Among the three delay combinations: τ1 = τ2 = τ (the critical stable curve is colored in
red); τ1 = τ, τ2 = 2τ (the critical stable curve is colored in black); τ1 = 2τ, τ2 = τ (the critical stable curve is colored
in blue), the one with τ1 = 2τ, τ2 = τ enables the closed-loop system subjected to (6) to have the largest stable region
in the (k p , kd )-plane. Again, the stable region in (k p , kd )-plane is sensitive to the change of the delays.
180 Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183

(a) (b)
Fig. 3. Comparison of the stable regions of the closed-loop systems (7) subjected to (6) and with ξ = 0.05 and k p = 0 for the three delay
combinations. (a) τ = 0.1; (b) τ = 0.2, where the dashed line is given by the critical condition kd = 2ξ.

(a) (b)
Fig. 4. Comparison of the stable regions of the closed-loop systems (7) subjected to (6) and with ξ = 0.05 and ka = 0 for the three delay
combinations. (a) τ = 0.1; (b) τ = 0.2, where the dashed line is given by the critical condition k p = 1.

For given (ka , kd ) and τ, the stability of the closed-loop system (7) can be characterized by the real part of the
rightmost characteristic root(s), namely by α = max{(λ) : p(λ) = 0}. The solution θ = 0 is asymptotically stable
if and only if α < 0. The smaller the number α is, the better the stability of x = 0 is. Table 1 presents a number of
calculation results of the rightmost characteristic roots. It shows that both the delayed DA controller and the delayed
PD controller can be preferable, depending on the sign combination of the gain values and the delay combination.

4. Calculation of the First Stable Delay Intervals

When the feedback gains are fixed, the stable delay intervals can be obtained accordingly. When k p = 0 and
τ2 = τ, τ3 = 2τ, for example, by separating the real and imaginary parts of p(iω)eiωτ = 0, one finds

−2ξkd ω2 −kd ω((ka − 1)ω2 + 1)


cos(ωτ) = , sin(ωτ) = (14)
(ka2 − 1)ω4 − 4ξ2 ω2 + 2ω2 − 1 (ka2 − 1)ω4 − 4ξ2 ω2 + 2ω2 − 1

Because sin2 (ωτ) + cos2 (ωτ) = 1, the critical frequency ω must be a root of F1 (ω) = 0 with

F1 (ω) = c0 ω8 + c1 ω6 + c2 ω4 + c3 ω2 + 1

where c0 = ka4 −2ka2 +1, c1 = 4ka2 +2ka kd2 +8ξ2 −ka2 kd2 −8ξ2 ka2 −kd2 −4, c2 = 16ξ4 +2kd2 +6−4ξ2 kd2 −2ka kd2 −2ka2 −16ξ2 , c3 =
8ξ2 − kd2 − 4. Once a positive root of F(ω) = 0 is in hand, the critical delay values can be found from (14). Ranking
all the critical delay values from small to large, the interval [0, +∞) is divided into infinite number of sub-intervals.
Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183 181

Table 1. The rightmost characteristic roots of system (7) with ξ = 0.05, τ = 0.1 for different feedback gains.
τ3
Gain values System (7) with τ2 = 2 =τ System (7) with τ2 = τ3 = τ
k p = 0, kd = −0.5, ka = +0.1 −0.3417 ± 1.039i −0.3473 ± 1.033i
k p = 0, kd = −0.5, ka = −0.1 −0.2916 ± 0.9270i −0.2881 ± 0.9304i
k p = 0, kd = −0.5, ka = +0.2 −0.3720 ± 1.116i −0.3874 ± 1.102i
k p = 0, kd = −0.5, ka = −0.2 −0.2713 ± 0.8846i −0.2656 ± 0.8902i
k p = 0, kd = −0.5, ka = +0.3 −0.4036 ± 1.219i −0.4380 ± 1.189i
k p = 0, kd = −0.5, ka = −0.3 −0.2537 ± 0.8482i −0.2464 ± 0.8551i
k p = 0, kd = −0.5, ka = +0.4 −0.4269 ± 1.361i −0.5031 ± 1.308i
k p = 0, kd = −0.5, ka = −0.4 −0.2382 ± 0.8164i −0.2299 ± 0.8241i
k p = +0.1, kd = −0.5, ka = 0 −0.3262 ± 0.9190i −0.3205 ± 0.9202i
k p = −0.1, kd = −0.5, ka = 0 −0.3036 ± 1.032i −0.3093 ± 1.031i
k p = +0.2, kd = −0.5, ka = 0 −0.3377 ± 0.8563i −0.3261 ± 0.8591i
k p = −0.2, kd = −0.5, ka = 0 −0.2924 ± 1.084i −0.3037 ± 1.082i
k p = +0.3, kd = −0.5, ka = 0 −0.3491 ± 0.7882i −0.3317 ± 0.7933i
k p = −0.3, kd = −0.5, ka = 0 −0.2812 ± 1.132i −0.2982 ± 1.131i
k p = +0.4, kd = −0.5, ka = 0 −0.3607 ± 0.7130i −0.3373 ± 0.7213i
k p = −0.4, kd = −0.5, ka = 0 −0.2702 ± 1.179i −0.2926 ± 1.178i

In each sub-interval, the stability of the closed-loop system is the same, either asymptotically stable or unstable for
all delay in this interval. Let τ denote the smallest one in all the minimal critical delay values, where each minimal
critical delay corresponds to a positive root of F1 (ω) = 0, hence if the closed-loop system is asymptotically stable
when τ = 0, it is asymptotically stable for all τ ∈ [0, τ), which is called the first stable delay interval. Similarly, when
k p = 0 and τ2 = τ3 = τ, by separating the real and imaginary parts of p(iω) = 0, one has
ka ω2 + 2ξkd − ka (2ξka − kd )ω2 + kd
cos(ωτ) = , sin(ωτ) = (15)
ka2 ω2 + kd2 ω(ka2 ω2 + kd2 )
It follows that ω must be a root of F2 (ω) = 0, where
F2 (ω) = (1 − ka2 )ω4 + (4ξ2 − kd2 − 2)ω2 + 1
Therefore, the first stable delay interval can be determined as done above.
Table 2 are some numerical examples of the first stable delay intervals for the two delay combinations. For the
delayed DA controller, taking ξ = 0.05, k p = 0, kd = −0.5, ka = 0.2 as a example for demonstration, F1 (ω) = 0 has
exactly two positive roots ω1 = 0.7308 and ω2 = 1.189. From Eq. (14) with ω = ω1 , the minimal critical delay value
is found to be 6.274, while solving Eq. (14) with ω = ω2 gives the minimal critical delay value 1.943. Because the
closed-loop is asymptotically stable when τ = 0, thus, as τ increases from zero, the stability is kept unchanged till
the minimal critical delay value 1.943. Therefore, the closed-loop under delayed DA controller with τ2 = τ3 /2 = τ is
asymptotically stable for all τ ∈ [0, 1.943) (the first stable delay interval). Similarly, when k p = 0, kd = −0.5, ka = 0.2
and τ2 = τ3 = τ, the polynomial F2 (ω) has two positive roots ω1 = 0.7755, ω2 = 1.316. The minimal critical delays
are found to be 5.431 and 0.9606 respectively. Thus, the closed-loop under delayed DA controller with τ2 = τ3 = τ
is asymptotically stable for all τ ∈ [0, 0.9606) (the first stable delay interval). Obviously, the stable delay interval
τ ∈ [0, 1.943) is much larger than τ ∈ [0, 0.9606). This means that when the feedback gains are fixed, the delayed
DA controller with τ2 = τ3 /2 = τ is much more robust against the change of time delay than that with τ2 = τ3 = τ.
182 Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183

Table 2. The first stable delay intervals of system (7) with ξ = 0.05 for different feedback gains.
τ3
Gain values System (7) with τ2 = 2 =τ System (7) with τ2 = τ3 = τ
k p = 0, kd = −0.5, ka = +0.2 τ ∈ [0, 1.943) τ ∈ [0, 0.9606)
k p = 0, kd = −0.5, ka = −0.2 τ ∈ [0, 1.133) τ ∈ [0, 1.697)
k p = 0, kd = −0.4, ka = +0.3 τ ∈ [0, 2.230) τ ∈ [0, 0.7397)
k p = 0, kd = −0.4, ka = −0.3 τ ∈ [0, 1.092) τ ∈ [0, 1.924)
k p = 0, kd = −5, ka = +0.5 τ ∈ [0, 0.4647) τ ∈ [0, 0.1748)
k p = 0, kd = −5, ka = −0.5 τ ∈ [0, 0.1547) τ ∈ [0, 0.3549)
k p = 0, kd = −10, ka = +0.5 τ ∈ [0, 0.2367) τ ∈ [0, 0.0902)
k p = 0, kd = −10, ka = −0.5 τ ∈ [0, 0.0783) τ ∈ [0, 0.1807)
k p = +0.2, kd = −0.5, ka = 0 τ ∈ [0, 1.241) τ ∈ [0, 1.601)
k p = −0.2, kd = −0.5, ka = 0 τ ∈ [0, 1.747) τ ∈ [0, 1.134)
k p = +0.3, kd = −0.4, ka = 0 τ ∈ [0, 1.250) τ ∈ [0, 1.856)
k p = −0.3, kd = −0.4, ka = 0 τ ∈ [0, 2.183) τ ∈ [0, 0.9956)
k p = +0.5, kd = −5, ka = 0 τ ∈ [0, 0.3010) τ ∈ [0, 0.3101)
k p = −0.5 kd = −5, ka = 0 τ ∈ [0, 0.3123) τ ∈ [0, 0.3027)
k p = +0.5, kd = −10, ka = 0 τ ∈ [0, 0.1558) τ ∈ [0, 0.1570)
k p = −0.5, kd = −10, ka = 0 τ ∈ [0, 0.1573) τ ∈ [0, 0.1560)

(a) (b)
Fig. 5. Comparison of the time histories of the closed-loops (7) subjected to (6) and with ξ = 0.05, k p = 0, kd = −0.5 and ka = 0.2. (a) 0 < τ =
0.9 < min(1.9429, 0.9606); (b) 0.9606 < τ = 1.8 < 1.9429, where the initial conditions are ϕ(0) = 0.1, ϕ̇(0) = 0, for all t ∈ [−2τ, 0].

Similarly, when ξ = 0.05, k p = 0, kd = −0.5, ka = −0.2, it is found that when the feedback gains are fixed, the
delayed DA controller with τ2 = τ3 = τ is much robust against the change of time delay than that with τ2 = τ3 /2 = τ,
the two stable delay intervals are [0, 1.697) and [0, 1.133) respectively. Among the four cases, it is the case with
k p = 0, kd = −0.5, ka = 0.2 > 0 and τ2 = τ3 = τ that has the largest stable delay interval, which means that the
corresponding controller is the most robust against the change of delay values. Here one has kd < 0 and ka > 0.
Figures 4-5 show the time-histories of the sway angle θ for different delay combinations, which confirm that the stable
delay interval can be greatly enlarged by choosing proper sign of feedback gains.
Zaihua Wang and Qi Xu / Procedia IUTAM 22 (2017) 176 – 183 183

(a) (b)
Fig. 6. Comparison of the time histories of the closed-loops (7) subjected to (6) and with ξ = 0.05, k p = 0, kd = −0.5 and ka = −0.2. (a) 0 < τ =
0.9 < min(1.697, 1.133); (b) 1.133 < τ = 1.5 < 1.697, where the initial conditions are ϕ(0) = 0.1, ϕ̇(0) = 0, for all t ∈ [−2τ, 0].

For the delayed PD controller, as shown in Table 1, among four cases for k p = ±0.2, it is the case with k p = −0.2 <
0, kd = −0.5, ka = 0 and τ2 = τ3 /2 = τ that has the largest stable delay interval, which means that the corresponding
controller is the most robust against the change of delay values. For this case, one has k p < 0 and kd < 0.

5. Conclusions

“PID controllers are the most important class of controllers since 98% of the controllers in use in applications
worldwide are of this category.” 8 Actually, PD controller and PI controller are also major controls in sway reduction
of a pendulum on a movable support 1 . With the delays in feedback loops taken into account, this paper shows that the
delayed PD controller or delayed DA controller works effectively in reducing the sway of a pendulum on a movable
support. Numerical examples show that the control effect of delayed PD controller or delayed DA controller depends
not only on the delay combination but also on the combination of the sign of feedback gains. With a properly chosen
delay combination and sign combination, the closed-loop has not only a large stable region in the gain plane but also
a large stable delay interval. Experimental validation of this observation is left for future consideration.

Acknowledgements

The authors thank the financial support of National Natural Science Foundation of China under Grant 11372354,
and thank Prof. Gabor Stepan at Department of Applied Mechanics, Budapest University of Technology and Econ-
omy, Hungary, for his helpful discussion.

References

1. Abdel-Rahman EM, Nayfeh AH, Masoud ZN. Dynamics and control of cranes: A review. Journal of Vibration and Control 2003; 12: 863-909.
2. Masoud ZN, Nayfeh AH, AI-Mousa A. Delayed position-feedback controller for the reduction of payload pendulations of rotary cranes. Journal
of Vibration and Control 2003; 9: 257-277.
3. Masoud ZN, Nayfeh AH. Sway reduction on container cranes using delayed feedback controller. Nonlinear Dynamics 2003; 34: 347-358.
4. Nayfeh AH, Masoud ZN, Nayfeh NA. A smart sway controller for cranes from theory to laboratory to industry. In: Segla S, Tuma J, Petrikova
I, et. al. eds. Vibration Problems ICoVP 2011 Supplementary – The 10th International Conference on Vibration Problems. Technical University
Liberec 2011; 14-29.
5. Insperger T, Milton J, Stepan G. Acceleration feedback improves balancing against relex delay. Journal of the Royal Society Interface 2013; 10:
20120763.
6. Xu Q, Stepan G, Wang ZH. Balancing a wheeled inverted pendulum with a single accelerometer in the presence of time delay. Journal of
Vibration and Control 2015, doi: 10.1177/1077546315583400;
7. Wang ZH, Hu HY, Xu Q, Stepan G. Effect of delay combinations on stability and Hopf bifurcation of a SDOF oscillator with delayed
acceleration-derivative feedback. International Journal of Nonlinear Mechanics 2016, doi: 10.1016/j.ijnonlinmec.2016.10.008;
8. Datta A, Ho M-T, Bhattacharyya SP. Structure and Synthesis of PID Controllers. London: Springer-Verlag; 2000.

You might also like