Академический Документы
Профессиональный Документы
Культура Документы
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/309410116
CITATIONS READS
0 47
5 authors, including:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Lale Canan Dlger on 30 October 2016.
AbstractThree case studies are performed in robotics and including the actuator dynamics also.
industrial servo press mechanism by using systems nonlinear
dynamics. Robotic systems and servo crank press are currently Three case studies are presented in this paper. These are all
available in Mechanical Engineering Departments laboratory represented nonlinear dynamics in robotics and manufacturing
(Gaziantep University/Turkey). Identification and simulation in a crank press example. These studies are all performed in
studies are carried out to get dynamic behavior of the systems of Departments System Dynamics and Control Laboratory, and
interest. Different motion trajectories are applied according to Mechatronics Laboratory at Gaziantep University/Turkey. A
the requirements for all. Soft computing techniques are studied Scara robot is studied PNP (Pick and Place) operation with
on these systems. Traditional PID is also applied. Control PSO-NN first. Hydraulic robot is performed with PSO-NN.
techniques are implemented as PSO-NN (Particle Swarm PSO has been used to train the NNs. A crank press is designed
Optimization-Neural Network), BP (Back Propagation), PSO- and driven by an AC servo motor. Some conclusions are
PID; PID and also CasFF (Cascade Feed-Forward) control in presented at the end. Also a servo press application is
laboratory testing in this study. All of the studies are given as presented by using cascade feed-forward (CasFF) control.
experimental base herein. This study is performed as a part of an industrial Project.
KeywordsScara robot; hydraulic manipulator; soft
computing; servo crank press; motion control implementation;
PSO-NN; BP; PSO-PID; CasFF
II. Control of Robotic Systems
and Servo Press
I. Introduction A literature study is dated for recent applications with
modeling and control background especially on nonlinear
An open kinematic chain of rigid links become a robot dynamics. The system of second order ordinary differential
manipulator. In robot or mechanism control, especially in high equations is derived. The dynamic behaviors of these systems
speeds, some problems can happen depending on the are explored by solving these equations in Matlab platform. A
complexity of the dynamics. There are also uncertainties; general procedure is given for robotic systems and servo press
parametric and dynamic. Knowledge based kinematics; inertia mechanism together with its solution procedure. Here the
parameters are the parametric uncertainties. However, link equations of motion for these systems are based on the
flexibility, actuator dynamics (AC, DC) are counted as principles of Lagrange [1-3].
dynamic uncertainties. A methodology is necessary to get the
systems nonlinear dynamical model. The reference trajectory M (q)q
C(q, q)q F (q) G(q) Q (1)
is given first. Control techniques are needed to track any
required trajectories where they can be described as a set of where q includes the coordinates; positions, velocities and
algebraic equations [1-2]. accelerations. Q is the input generalized forces or torques,
M(q) is the inertia matrix and C(q,dq/dt) is the
By considering Euler-Lagrangian method, the equations of Coriolis/centripetal forces and G(q) is the gravity force or
motion are studied with actuator dynamics. DC servo motors torque and F(dq/dt) is the friction vector.
and hydraulic actuators are used in this study. Mathematical
models are derived with assumptions taken during laboratory Soft computing approaches are also proposed for solving
studies. A manipulator dynamic model is represented by nonlinear and mathematically unmodeled systems. Soft
nonlinear coupled differential equations. The experimental computing (SC) approaches are preferable over conventional
setups and the simulation results with experimental ones are methods. In the field of robotics and manufacturing
described. Mathematic models are derived for all systems by automation where nonlinear system dynamics is existed,
particular benefit is obtained solving the problems.
Manufacturing and automation especially for hands and
Last part of this study is supported by Ministry of Science, Industry and
Technology for the financial support under SANTEZ project (Number:
01422.STZ.2012-I).
manipulators, mobile robots, multi agent robots are the Table I gives its mechanical and electrical data.A
potential applications seen [4]. In many recent studies soft mathematical model of SCARA robot is developed including
computing techniques are used for control of nonlinear servo actuator Dynamics. The equations of motion are derived
dynamical systems. Hybrid architectures are developed for by using Lagrangian mechanics which is an energy based
intelligent control systems. Soft computing techniques include approach. DC servo motors driving each robot joint is studied
fuzzy logic, swarm intelligence, neural networks and with a controller action. This robot is used for pick and place
evolutionary computation. They have been successfully operations. Scara robot is applied to point to point control.
applied to many robotic systems in practice. Satisfaction is
provided for controlling dynamic systems. Applications can
TABLE I. MECHANICAL AND ELECTRICAL DATA (SCARA)
be accounted as hybrid soft computing. Combinations can be
Fuzzy-NN, GA-NN, PSO-NN for example. These techniques Masses of links m1=1.90 kg, m2=0.93 kg
are quite successful in the field of control. These techniques Linkage inertias J1=0.0980 kg.m2 ,
have been used together in hybrid form to provide better J2=0.0115 kg.m2
solutions for the problems [5]. A study is seen in Fuzzy- Motor voltage constant Ke1= Ke2=0.047 V/rad/s
Wavelet Neural Networks (FWNNs) [6]. Motor torque constant Kt1= Kt2=0.047 Nm/A
Armature resistances Ra1= Ra2=3.5
III. Case Studies Armature inductances La1= La2= 1.3 mH
Gear box ratios N1=90 , N2=220
Three case studies are performed with available systems at
Laboratory. Scara robot is generally used for pick and place Gearbox inertias Jg1=0.0002 kg.m2,
Jg2=0.0005 kg.m2
operations in assembly lines. Electro hydraulic manipulator is
applied for heavier operations in industry. Both robots are
operated as 2 dof during experiments. A crank press is Robot-actuator-control system is studied with numerical
applied for sheet metal forming. simulation here. Experiments are carried on motion
trajectories as well. Evolutionary techniques are studied and
A. Case Study I: Scara Robot PSO-PID is designed for different motion characteristics [9].
In many SCARA (Selective Compliant Articulated Robot The number of particles in the swarm influences the
Arm) consists of 3 planar revolute joints as shoulder, elbow convergence of the algorithm. PSO algorithm [9,12] is given:
vik 1 wvi c1 r1 ( pbesti xik ) c2 r2 ( gbest xi )
k k
and wrist and a prismatic joint which operates in vertical plane (2)
for proper positioning of the work piece especially a good
choice for assembly tasks. Fig.1. shows this system. xik 1 xik vik 1 (3)
k k
Where; vi is the velocity of the particle, xi is the current
position of the particle (solution), w is the inertia weight
factor, pbest is the best solution value among the particle found
in a particular iteration, gbest is the global best solution
achieved ,r1 and r2 are the random numbers between [0, 1], c1
and c2 are the acceleration constants where; c1=c2=2. The
inertia weight factor w in the equation (1) can be calculated by
using the equation (3) where iter is the iteration number in a
particular instance and itermax is the maximum number of
iterations.
( wmax wmin )
w wmax .iter where 0.4 w 0.9 (4)
itermax
During the optimization, the number of particles in the
swarm, the objective function and the parameters c1 and c2 are
important. The following parameters in PSO are taken as
population size is 45, m=1000 iterations, c1=c2=2, wmax=0.9
and wmin=0.4 in this study. Performance of the controller can
be evaluated either in frequency domain or time domain. In
this study, performance of PID controller is analyzed in time
domain with the overshoot (Mp), the rise time (tr), the settling
time (ts) and the steady state error (ess). The performance
criterion W (K) is defined as
W ( K ) (1 e )(M P ess ) e (t s t r ) (5)
Fig 2 shows the results for scara robot with PSO-PID for
Fig.1. Scara robot; general view, top view and schematic representation [7,8] both motor axes.
Fig. 2. Simulation Results for Scara with PSO-PID
L1=0.0 mm L2= 752 mm This error is then accepted as the fitness function of a particle
that can be calculated by.
(d
M0=79 kg M1= 49 kg M2= 50 kg
d desired ) 2
Fitness ( particle i ) ET
actual
J0=1.198 kg.m 2
J1=1.95 kg.m 2
J2= 6.8 kg.m 2 (7)
2
The mathematical model of manipulation system is
Controller presented for the manipulator configuration including actuator
type. The dynamic modeling and experimental implementation
are performed with the controller action. Lagrangian
X model mechanics is also applied and tested with applied controller.
Several trajectories are studied. System identification is
performed using PSO-NN for the robot. [10-13].
X reference
(2,L2 )
PSO NN STRUCTURE 1,2
Reference
1 BP
0,6
0
Fig. 4. Block diagram of the electro hydraulic robot
0 20 40 60 80 100 120
Number of inputs
Table II gives all mechanical parameters for robotic
system Fig.4 shows block diagram of the electro hydraulic
robot system. A circular trajectory is traced in the second test 1,2 Reference
by giving the coordinates of a circle to a hydraulic robot, RP BP
1
configuration is referred. For example, 240 data have been
Joint2 Response (m)
PSO-NN
recorded experimentally. Half of the data have been used for 0,8
the training. The rest of the data have been used in the test 0,6
session Test values for circular trajectory are given in Figs 5.a-
0,4
b-c.The system follows the reference signal with some error at
some points. A population of 45 particles is used for the PSO- 0,2
NN algorithm. Numbers of particles and hidden layers have
0
been tried on the system in different in various training. 0 20 40 60 80 100 120
Learning factors c1 and c2 are set to 2.0. Both the PSO-NN and Number of inputs
BP algorithm is trained for 5000 iterations. Learning rate and
momentum rate is chosen between 0 and 1. The initial weight
values are important on training results, if a priori knowledge
is available for weights. Initial weights of the both algorithms 1,4
are chosen between -1 and 1 randomly. Both algorithms are BP error rate
1,2
Total error rate
started with the same initial weight values. The performance 1 PSO-NN error rate
of the proposed algorithm is tried with many different initial 0,8
weights. For the solution of the problem a NN structure that
0,6
the input layer N has two nodes (N=2), hidden layer J with
0,4
twenty nodes (J=20) and output layer L with two nodes (L=2)
are used. The error E[n] is the sum of squares of the error 0,2
m 0
overall output units. ET E (n) is the total error, d actual is 0 1000 2000 3000 4000 5000 6000
n 1
Number of iteration
the actual output value taken from the system, and d desired is
the desired output value in each iteration.
Fig. 5. Joint response results for hydraulic robot
By looking at Fig. 5, PSO-NN algorithm is produced better
output values than BP-NN algorithm in this example. Similar
findings are taken dealing with different test signals Output
1(2) has got 96,407% correction rate in PSO-NN, but BP
gives 89,321%. Similarly, Output 2(L2) has given 95,574% for
PSO-NN, but 86, 4% for BP-NN. These errors can obviously
be reduced by using more training data, yielding increase in
computation in the network.
C. Case Study III: A Crank-Servo Press Mechanism
Metal forming applications are classified as bending,
stamping, deep drawing and shearing. Different motion
scenarios are necessary in metal forming sectors. Servo
presses offer motion flexibility and torque control. They
provide desired torques in desired velocities [14-19].
References
[1] J. Awrejcewicz, R. Szewczyk, M. Trojnacki, M. Kaliczyska,
Mechatronics: Ideas for Industrial Applications. Springer: Switzerland, 2015.
[2] J. Angeles, Fundamentals of Robotic Mechanical Systems: Theory,
Methods, and Algorithms, Springer-Verlag, 2003.
[3] A. Fratu and M. Fratu Equation of motion of the nonlinear robotic
systems with constraints, Bulletin of the Transilvania, University of Brasov,
Series I, Eng. Sci., vol 5(54), No.2, 2012, pp. 73-78.
[4] P. Melin and O. Castillo, Soft computing for Intelligent control of
Nonlinear Dynamical Systems, vol.2, No.1, 2004, pp. 45-78.
[5] A. Saxera, Abhirav Saxera, Review of soft computing techniques used in
robotics application, Int. J. of Information and Computation TechnologyI
vol. 3, No.3, 2013, pp.101-106,.
[6] M. Amina, Dynamic Non-linear System Modelling using Wavelet-Based
Soft Computing (SC) Techniques, University of Leading the Way
Westminster, September 2011.
[7] M.T. Das, Motion control of a SCARA robot with a PLC unit, Ms Thesis,
Gaziantep University, 2003.
[8] M.T. Das and L.C. Dulger, Mathematical modelling, simulation and
experimental verification of a scara robot, Simulation Modelling Practice and
Theory, vol. 13, 2005, pp. 257-271.
[9] M.T. Das and L.C. Dulger, Control of a scara robot: pso-pid approach,
control and intelligent systems, vol. 38, No.1, 2010, pp. 24-31.
[10] S. Kapucu, Adaptive Control of a Robot Manipulator by visual data,
Ph.D. Thesis, Gaziantep University, Mechanical Engineering Department,
1994.
[11] M. Topalbekirolu, L.C. Dulger and A. Kirecci, An electrohydraulic
spherical manipulator-dynamic modelling and simulation: A case study, Pr.
Fig. 8. Servo press control results a) motor, b) ram of Symposium of Mechanism and Machine Science, AzCIFToMM-zmir
Instutiteof Technology, October 2010, pp.162-169.
Fig. 8 presents the position results of the servo crank press. [12] M.T. Das, L.C. Dulger, and S. Kapucu, Identification and control of an
electro hydraulic robot: Particle Swarm optimization-Neural Network (PSO-
The motion profile couldnt generated by conventional crank NN) approach, ICINCO 2011- 8th Int. Conf. on Inf. in Control, Automation
press. CasFF servo application provides higher accuracy on and Robotics Proceedings, Holland-Noordwijkerhout, July 2011, pp. 49-55.
sheet metal forming manufacturing. Maximum manufacturing [13] M.T. Das, L.C. Dulger and G.S. Das, Robotic applications with Particle
error is recorded as 0.04 mm [19]. Swarm Optimization, 2013 Int. Conf. on Control, Decision and Information
Technologies-CoDIT2013 , Hammamet/TUNUSIA, May 2013.
[14] R. Halicioglu, L.C. Dulger and A.T. Bozdana, Modelling and simulation
Conclusions
based on matlab/simulink: A press mechanism, ICMSQUARE: Prague,
IV. 2013.
[15] A. Groseclose. Part: 1 New Application for Servo-Driven Presses.
Soft computing for controlling our dynamic systems of Stamping of automotive components. Stamping Journal FMA publication.
interest are presented here. Studies on the subject are still June 2009.
continuing. PSO-NN, BP-NN, PSO-PID and CasFF control [16] Z.L. Huo, X.S. Wang, Position control of servo press system based on
Fuzzy-PID. Forging & Stamping Technology, vol. 5, 026, 2011.
are applied for trajectory tracking and system control. Three [17] R. Halicioglu, L. C. Dulger, and A. T. Bozdana, Mechanisms,
case studies are presented as Scara Robot, Hydraulic classifications, and applications of servo presses: A review with
Manipulator and a Servo Crank Press for forming applications comparisons, Proceedings of the Institution of Mechanical Engineers, Part B:
in industry. An approach based on the training of NN method Journal of Engineering Manufacture, doi:0954405415600013, 2015.
[18] R. Halicioglu, L.C. Dulger and A. T. Bozdana, Servo pres tasarimi ve
with PSO is used for identification of a hydraulic servo robot dinamik modeli: uygulamali bir rnek, 17.UMTS, zmir, Haziran 2015.
Implementation results after system modelling, and control [19] R. Halicioglu, Design, synthesis and control of a mechanical servo press:
issues are discussed. A servo crank press is designed and An industrial application, PhD Thesis, Gaziantep University, 2015.
tested. This study is performed in an industrial project. Its
control issues are also included. Studies are performed in
Mechatronics Laboratory.