Dynamic Analysis and Control of the MJR Robot

Palletizer

Fengyan Ning, Chao Yun, and Xinyi Chen

School of Mechanical Engineering and Automation, Beihang University,

100083, Beijing, P.R. China
[email protected]

Abstract. The MJR120 palletizer is researched and manufactured by the paper.

Using the Second Lagrangian kinetics, the author analyses the dynamical fea-
tures of the robot, derives the generated driving torque of the arms, and sets up
its inverse dynamical equations. To improve its performance of the system, a
new controller is designed with a self-adaptive fuzzy PID control algorithm,
which makes use of the merits both of the fuzzy control and the self-adaptive
control algorithm. The effectiveness of the proposed control method is verified
by simulation based on Matlab. The controller has better performance and ro-
bustness than the conventional PID controller in the Palletizer system.

Keywords: palletizer; dynamical features; self-adaptive fuzzy PID control.

1 Introduction
In the past few years, Palletizing technology, especially robotic Palletizing technol-
ogy, had been developed rapidly, the throughput of on-line palletizers had been in-
creasing and the flexibility, processing velocity, and payload of robotic palltizers had
been upgrading [1]. Among the palletizers, their links are either revolving with the R-
joint, or moving by a geodetic coordinate system; By heavy coupling, the kinematical
and dynamical analysis of the forth are very complex; and the latter overcomes cou-
pling, and is easy to control, but it covers much space and has very bad flexibility. In
the current situation, after a great deal of analysis, computation and optimized design,
the MJR120 palletizer has been researched and manufactured by the Robot Research
of Beihang University for the Wenzhou Hualian Package Ltd., to transfer and pallet-
ize materials on the packing production line. The robot’s maximum gripping load is
120kg, the maximum rotation velocity is 1400r/h, and the maximum rotation angle of
the waist is 3300. The palletizer has been manufactured and equipped well, and now is
being debugged and test run. The whole structure is show in the Fig. 1. The links of
the palletizer consist of a closed parallel four-bar linkage, couples two groups of little
paralle four-bar linkages. Each link may revolve by driving of the level and vertical
sliders to control the spread and pitch of the wrist in the level and vertical direction,
respectively. The mechanism can make the actuators of the two arms and their reduc-
ing mechanism be be installed into the waist to improve its dynamic features. By
chosing the adaptive parameters, its kinematics may be decoupling, thus it is easy to
control by kinematics, but with the partical closed links, some constraint forces act on

C. Xiong et al. (Eds.): ICIRA 2008, Part II, LNAI 5315, pp. 713–722, 2008.
© Springer-Verlag Berlin Heidelberg 2008
714 F. Ning, C. Yun, and X. Chen

Fig. 1. MJR120 palletizer

the corresponding the links so that the robot is coupling in dynamics, thus it is very
difficult to analysis their dynamics.
Using the Second Lagrangian, the paper completes its analysis and derivation of its
dynamical model. The system is very uncertain, time-varying and nonlinear, so it is
difficult for the conventional PID to adjust automatically its parameters. However, the
fuzzy PID controller can not only satisfy the needs but also improves the capacity of
eliminating system’s steady-state error. In this paper, the self-adaptive fuzzy PID
method is used in the pallizater. The simulation results verify the feasibility of the
scheme presented in the paper.

2 Dynamic Analysis of the MJR120 Robot Palletizer Model

2.1 Simplied Model of the Robot

The main structure of the palletizer is a closed paralle four-bar linkage, and couples
two groups of little paralle four-bar linkages. Four motors are used to drive the re-
volving and moving of the wrist, the level slider and the vertical slider, respectively.
For simplicity, we only consider the dynamics of its waist and paralle four-bar link-
ages, and neglect the others. Thus, we can obtain the robot’s simplied structure, and
set up its coordinate system, shown in the Fig. 2. Now, denote the length of the front
arm AE, connection-rod AB, rear arm CD and the little arm DP by l2, l3, l5 and l6,
respectively; denote the distance between their mass centers and their rotating axis by
l2c, l3c, l5c and l6c respectively; and the length of BC, the length between the mass cen-
ter of the connection-rod and the B point, the length between the mass center of the
rear arm and its C point, between the mass center of the little arm and the E point, and
its end P point denoted by l0, r3, r5, r6and rr6; denote the level distance between O1
point and the vertical slider; θ1, θ2 and θ3 are the revolving angle of the waist, front
arm and little arm respectively, as shown in the Fig. 2.
 Dynamic Analysis and Control of the MJR Robot Palletizer 715

Fig. 2. MJR robot’s simplied model

Here, we analyse and derive the dynamical equations of the robot by the Second
Lagrangian kinetics, which is the difference of the kinetic and potential energy of the
system [2-4].

2.2 Generalized Coordinate of the System

To simplied computation, we neglect the movment of the wrist, thus, the system be-
comes a three-dimensional model with three free-degrees. We can transform the posi-
tion coordination x1 and y4 of the two sliders into angle variety θ2 and θ3 in the XOY.
Therefore, taking the θ1, θ2 and θ3 as the generalized coordination of the system.

2.3 Dynamical Model of the Robot

Because of the parallel four-bar mechanism, great constrait forces exit among the
links. To simply, we first derive Lagrangian equations of the system.
Supposing that the vector radius of every link in the coordinates OXYZ is ri
(i=1,2,…7). Referring to their geometrical relation [5] , yields

( )
⎡ l0 c2 − l3c3 − l c1 ⎤
⎢ ⎥
( )
r1 = ⎢ l0 c2 − l3c3 − l s1 ⎥ (1)
⎢ 0 ⎥
⎣ ⎦

⎡(l0c2 − l3c3 + l2 cc2 − l )c1 ⎤

r2 = ⎢⎢(l0c2 − l3c3 + l2 cc2 − l )s1 ⎥⎥ (2)
⎢⎣ l 2 c s2 ⎥⎦
716 F. Ning, C. Yun, and X. Chen

(
⎡ l0c2 − r3c3 − l c1 ⎤ )
⎢ ⎥
(
r3 = ⎢ l0c2 − r3c3 − l s1 ⎥ ) (3)
⎢ l3c s3 ⎥⎦
⎣

⎡ − lc1 ⎤
⎢ ⎥
r4 = ⎢ − ls1 ⎥ (4)
⎢l s − l s ⎥
⎣3 3 0 2⎦

⎡(l5c c2 − l )c1 ⎤
r5 = ⎢⎢(l5c c2 − l )s1 ⎥⎥ (5)
⎢⎣l3 s3 + r5 s2 ⎥⎦

⎡(l5c2 − l6 c c3 − l )c1 ⎤
r6 = ⎢⎢(l5c2 − l6 c c3 − l )s1 ⎥⎥ (6)
⎢⎣ l2 s2 − r6 s3 ⎥⎦

⎡(l5c2 − l6c3 )c1 ⎤

⎢ ⎥
r7 = ⎢(l5c2 − l6c3 )s1 ⎥ (7)
⎢l s − r ' s ⎥
⎣2 2 6 2 ⎦

where s1=sinθ1, c1=cosθ1, s2=sinθ2, c2=cosθ2, s3=sinθ3, and c3=cosθ3, the following is
as so.
Differenttiating the expressions of (1) to (7), we can have the velocity expression
of every link vi (i=1,2,…7).
Now differentiating (1)~(7), the kinetic energy of the system can be given by
7 ⎛1 • 2⎞
T= ∑ ⎜ mv 2+1I θ ⎟
⎜2 i i 2 i i ⎟
i =1 ⎝
(8)
⎠
where I2, I3, I5 and I6 are the rotational inertias of the link 2, 3, 5 and 6, respectively;
mi (i=1,2,...,7) denote the mass of the level slider, front arm, connection-rod, vertical
slider, rear arm, little arm and wrist, respectively.
Turning now to the potential energy, we have that
7
U = ∑ mi gyci (9)
i =1

where, yci is the vertical coordiation of the link i.

 Dynamic Analysis and Control of the MJR Robot Palletizer 717

The non-conservative forces of the robot are the joint troques τ1, τ2 and τ3, which
are the output torque of the motors of waist, level slider and vertical slider, respec-
tively [6] ：
τ = [τ 1 ,τ 2 ,τ 3 ]T (10)

The virtual workof the robot is

δQ = τ T δθ = τ1δθ1 + τ 2δx1 + τ 3δy4 (11)

Referring to the Fig. 2, we can derive the following geometry relation

⎧ x1 = l0 c2 − l3c3
⎨ (12)
⎩ y4 = l3 s3 − l0 s2
Differenttiating the above expression gives

⎧δx1 = −l0 s2δθ 2 + l3 s3δθ 3

⎨ (13)
⎩δy4 = l3c3δθ3 − l0c2δθ 2
Therefore, with the virtual work principle, we have

δQ = τ 1*δθ1 + τ 2*δθ 2 + τ 3*δθ3 (14)

Thus, the generalited force of the system is

τ * = Dτ (15)

After substituting Lagrange function and generalized force into the Lagrange, those
nonlinear dynamic equations of the robot can be written in matrix form as
•• • •
M (θ )θ + C (θ ,θ )θ = Dτ (16)

where the matrix M is a 3×3 positive definite, symmetry and time-varying generalized
quality matrix, M12= M13= M21=M31=0, M23=M32 , C12= C13=0 , G1=0
⎛ • •
⎞
C11 = 2m1 ( l0 c2 − l3 c3 − l ) ⎜ l3 s3 θ 3 − l0 s2 θ 2 ⎟
⎝ ⎠
+ 2m2 ( l0 c2 + l2 c c2 − l3 c3 − l ) ⎜ −l0 s2 θ 2 − l2 c s2 θ 2 + l3 s3 θ 3 ⎞⎟
⎛ • • •

⎝ ⎠
⎛ • •
⎞ ⎛ • •
⎞
+2m3 ( l0 c2 − r3c3 − l ) ⎜ −l0 s2 θ 2 + r3 s3 θ 3 ⎟ + 2m5 ⎜ l5c ls2 θ 2 − l52c s2 c2 θ 2 ⎟
⎝ ⎠ ⎝ ⎠
⎛ • •
⎞ ⎛ • •
⎞
+2m6 ( l5 c2 − l6 c c3 − l ) ⎜ −l5 s2 θ 2 + l6 c s3 θ 3 ⎟ + 2m7 ( l5 c2 − l6 c3 − l ) ⎜ −l5 s2 θ 2 + l6 s3 θ 3 ⎟
⎝ ⎠ ⎝ ⎠
•
⎛ • •
⎞ •
C21 = m1 ( l02 s2 − l0 ls2 − l0 l3 s2 c3 ) θ 1 + m2 ( l0 c2 + l2 c c2 − l3 c3 − l ) ⎜ −l0 s2 θ 2 − l2c s2 θ 2 ⎟ θ 1
⎝ ⎠
718 F. Ning, C. Yun, and X. Chen

• •
+ m3 ( l02 s2 c2 − l0 ls2 − l0 r3 s2 c3 ) θ 1 + m5 ( l52c c2 s2 − l5c ls2 ) θ 1
• • • •
+ m6 ( l0 l5 s2 + l5l6 c s2 c2 − l52 s2 c2 ) θ 2 θ 1 + m7 ( l0 l5 s2 + l5l6 s2 c2 − l52 s2 c2 ) θ 2 θ 1
•
C22 = s2 c2 θ 2 (m1l02 + m2 r22 − m2 l22c + m3l02 + m5l52c
− m5 r52 − m4 l02 + m6 l52 − m6 l22 + m7 l52 − m7 l22 )
•
C23 = s2 c3 θ3 ( − m1l0l3 − m2l3 r2 − m3l0 r3 − m6l5l6c − m7 l5l6 )
•
+ c2 s3 θ3 ( −m5l3 r5 + m4 l0 l3 + m6l2 r6 + m7 r1r6' )
• • •
C31 = m1 ( ll3 s3 + l32 s2 c3 − l0 l3 c2 s2 ) θ 1 + m6 ( ll6 c s3 − l62c c3 s3 + l5 l6 c c2 s3 ) θ 3 θ 1
• •
+ m7 ( ll6 s3 − l62 c3 s3 + l5 l6 c2 s3 ) θ 3 θ 1
• •
+ m2 ( l32 s3 c3 + ll3 s3 − l2 c l3 c2 s3 − l0 l3 c2 s3 ) θ 1 + m3 ( r32 s3 c3 + lr3 s3 − l0 r3 c2 s3 ) θ 1
•
C32 = c2 s3 θ 2 ( −m1l0 l3 − m2 l3 r2 − m3l0 r3 − m6 l5 l6 c − m7 l5 l6 )
•
+ s2 c3 θ 2 ( −m5 r5 l3 + m4 l3 l0 + m6 l62c + m7 r6'2 )
•
C33 = s3 c3 θ 3 ( m1l32 + m2 l32 + m3 r32 − m3l32c − m4 l02
− m5 l32 + m6 l62c − m6 r62 + m7 l62 − m7 r6'2 )
(
G2 = gc2 m2 l2 c − m4 l0 + m5 r' + m6 l2 + m7 l2
5
)
G3 = gc3 ( m3l3c + m4 l3 + m5l3 − m6 r6 − m7 r6' )
The other elements of the matrixes have been derived out, but they are ommitted
here because of their long expressions.

3 Design of the Self-adaptive Fuzzy PID Control

The parameters of PID can be adjusted on-line, which can be changed through the
inquiry to the fuzzy control rules table saved aforehand in the computer [7-10].
The conventional PID controller can be described as ：
∫
de(t )
u (t ) = K P e(t ) + K I e(t )dt + K D (19)
dt
where, KP, KI and KD are the proportional, integral and derivative gains respectively,
u(t) is the output of controller; e(t) is the error signal. The control performance can
become better through adjusting the KP, KI and KD according to the changing control
parameters condition. The design algorithm in this paper is to adjust the KP, KI and KD
parameters on line trough fzzy inference based on the current e and de(t)/dt to make
the controlled object attain good performances. The block diagram of self-adaptive
fuzzy PID controller is shown in Fig. 3 [12-13].
 Dynamic Analysis and Control of the MJR Robot Palletizer 719

Fig. 3. Block diagram of self-adaptive fuzzy PID controller

The design process of controller is divided into four aspects:

The input variables and output variables. In the design, speed error e and error
change rate ec are used as fuzzy input and the proportional factor KI , the integral
factor KP, and the differential factor KD are used as fuzzy output.
Fuzzy language of input and output variables. The fuzzy variable of input variable
is e and ec, the fuzzy variable of output variable is KP, KI and KD. Firstly, it must be
confirmed the universe of discourse, quantum rank and quantum factor (Ke, Kec, KUp,
KUi, and KUd) of the input variables e, ec and output variable KP, KI and KD. Here,
quantum rank of e, ec is chosen as thirteen ranks {-6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5,
6,}. The fuzzy sets of e and ec are all defined as {NB, NS, ZO, PS, PB}. The quantum
rank of KP, KI and KD is chosen as ten ranks {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}. The fuzzy sets
of KP, KI, and KD are all defined as {Z, S, M, B}. The membership function of E,
EC, KP, KI, and KD is the triangle distrbution function. The suitable quantum factor
of each variable is adopted by considering the experiment and simulation results of
the Palletizer. The quantum factor of e is Ke=10; The quantum factor ec is Kec=100;
The quantum factor of KP, KI and KD is KUp=2.5; KUi=0.6; KUd=0.4 [11-12].
Fuzzy rules regulation. The principle of designed fuzzy rules is that the output of
controller can optimize the output response of the system. The fuzzy rules are gener-
alized as table I, table II and table III according to the expert experiment in the pal-
lizer and simulation analysis of the system [13-14].
PID parameters regulation. KP, KI, and KD can be regulated after completing the
fuzzy rule table of KP, KI, and KD.

Table 1. The control rulers of KP

ec e NB NS ZO PS PB
NB B B B B B
NS M B S S S
ZO M B Z S B
PS S S S S S
PB B B B M B

Table 2. The control rulers of KI

ec e NB NS ZO PS PB
NB Z M B M Z
NS S M B B S
ZO S B B M Z
PS S B M M S
PB Z S M S Z
720 F. Ning, C. Yun, and X. Chen

Table 3. The control rulers of KD

ec e NB NS ZO PS PB
NB B B B B B
NS M B S S S
ZO M B Z S B
PS S S S S S
PB B B B M B

4 Simulation Results

To demonstrate the effectiveness of the designed controller, the MATLAB is used to

the whole system simulation. With the block diagram in the Fig. 3 and the fuzzy con-
trol rules in the table I, II and III, the simulation results are shown in the Fig. 4. Here,
we only show the controls of the waist, the others are ommited. From the figure, we
can see when its input angle is 1 radian, the ascending time is 50ms, the adjusting
time is 10ms, and the error of the steady state is 1.3% with the self-adaptive fuzzy
PID control, while the tranditional PID control has the defects of lower response rate,
longer delay and regulating time, so the forth has much obvious advantage than the
later in response for the nonlinear system, and have no overshoot which is easy to
occur in the fuzzy control. Therefore, the controller can satisfy the performance re-
quirement of the palletizer system.
With frequently changing of the load, the controller is required to posses good ro-
bustness. The Fig. 5 shows the system’s response when there is disturbance from the
external world. We can see that the self-adaptive fuzzy PID controller can suspress
the disturbance much quickly than the tranditional one, and let the robot have small
fluctation and its restoring time is relatively short, hence the controller have strong
robustness.

Fig. 4. Comparison of the step response results for traditional PID controller and self-adoptive
fuzzy PID controller
 Dynamic Analysis and Control of the MJR Robot Palletizer 721

Fig. 5. Comparison of the response curve of the waist under the disturbance of 1.4s

5 Conclusion
The MJR120 palletizer can satisfy the requirement of high-speed and heavy load for
material handling, whose mechanism can improve its dynamic features. By the
method of Lagrangian, this paper analyses its dynamical performance, derives the
generated driving torque of the arms, and sets up its inverse dynamic model. The
paper introduces the self-adaptive fuzzy PID control algorithm into the control system
of the palletizer. From the simulation results, we can see that the self-adaptive fuzzy
PID control has more obvious advantage than the traditional PID control, and the
designed controller can eliminate overshoot and posses strong robustness.

References
1. Hu, H.G., Gao, J.H., Yang, R.Q.: The development of palletizing technology. Modular
Machine Tool & Automatic Manufacturing Technique 6, 54–59 (2000)
2. Zhu, Z.X., Zhou, Q.H., Yin, J.S.: Theoretical Mechanics, pp. 315–325. Beijing University
Press (1997)
3. Fei, X.: Advanced Dynamics, pp. 27–42. Zhejiang University Press (1993)
4. Fei, R.Y., Zhang, H.H.: Mechabical design and analysis of robots, pp. 123–128. Beijing
Industry University Press (1998)
5. Tang, X.Y., Li, W.: Effect of parallel four-bar mechanism on spotwelding robot. Journal of
China Agricultural University 8(4), 36–40 (2003)
6. Ma, X.F.: Robot Mechanisms, pp. 233–239. Medical Industry Press (1991)
7. Spong, M.W., Vidyasagar, M.: Robot Dynamics and control, United States of America, pp.
234–245 (1989)
8. Terano, T., Asai, K., Sugeno, M.: Applied Fuzzy Systems. AP Professional 1994, p. 302
(1994)
9. Tanaka, K., Sugeno, M.: Stability Analysis and Design of Fuzzy Control Systems. Fuzzy
Sets and Systems 45(2), 135–156 (1992)
722 F. Ning, C. Yun, and X. Chen

10. Adachi, G., Horikawa, S., Furuhashi, T., Uchikawa, Y.: A New Linguistic Design Method
of Fuzzy Controller Using a Description of Dynamical Behavior of Fuzzy Control Sys-
tems. In: Proceedings of ACC 1995, pp. 2282–2286 (1995)
11. Tao, Y.H., Yin, Y.X., Ge, L.S.: The control and application of late-model PID, pp. 211–
219. China Machine Press, Beijing (1999)
12. Buja, G.S., Todesco, F.: Neural network implementation of fuzzy logic controller. IEEE
Trans. Ind. Rlectron. 41, 663–667 (1994)
13. Zhang, J.H.: Study and compare for PID control system and fuzzy self-adaptive PID con-
trol system. Journal of Wuhan University of Technology 27, 286–290 (2005)
14. Liu, J.K.: Advanced PID control and Matlab simulink, pp. 125–190. Publishing House of
Electronics Industry, Beijing (2004)