A study on the control performance improvement of pneumatic artificial muscle manipulator using intelligent control

Improvement of the control performance of pneumatic artificial muscle manipulator with an intelligent phase plane switching control 5.1 Introduction and characteristics of MRB 52 5.2 Con

공학박사 학위논문

울산대학교 대학원 기계자동차 공학부

Tu Diep Cong Thanh

지능제어를 이용한 공압근육 매니퓰레이터의 제어성능향상을 위한 연구

A Study on the Control Performance Improvement

of Pneumatic Artificial Muscle Manipulator

Using Intelligent Control

지도교수 안경관

이 논문을공학박사학위 논문으로 제출함

2005 년 6 월

울산대학교 대학원 기계자동차 공학부

Tu Diep Cong Thanh

지능제어를 이용한 공압근육 매니퓰레이터의 제어성능향상을 위한 연구

A Study on the Control Performance Improvement

of Pneumatic Artificial Muscle Manipulator

Using Intelligent Control

Acknowledgments

The realization of this thesis would not have been possible without the help and unlimited support from professors, colleagues, friends, and my love-family to whom I

am most grateful Thanks for giving me the opportunity and the confidence

This is the best opportunity for me to express many thanks and my deepest gratitude to my advisor, Prof Ahn Kyoung Kwan, for all of his guidance, advice and support during the course of my research and thesis writing I am very thankful for the opportunities he has provided me, for his constant support and his many helpful ideas and suggestions My respect for him will always be in my heart

I am also honored to have Prof Byung Ryong Lee, Prof Jee Seong Jang, Prof Cheol Keun Ha and Prof Soon Yong Yang in my committee, whose inspiration, support and perseverance made this dissertation become possible I would like to thank them for their interest and encouragement throughout this research

There are not enough words for me to express my sincere gratitude towards all my Korean and Vietnamese friends Not many people have the luck to have their best friend as me Thanks for helping me through the difficult moments, for yours deep thinking and yours contributions to the realization of my thesis, and especially for the many animated discussions on the subject

A special thanks goes to my elder sister and her husband for taking care my parents during the time I studied abroad

I am extremely grateful to my parents for their constant support throughout the years The same feelings go towards Nhat Hong, the love of my life

June 2005

Tu Diep Cong Thanh

manipulator 11

2 Configuration, experiment setup and characteristic of pneumatic artificial

2.2 Configuration, experiment setup and characteristic of pneumatic

2.2.1 Configuration of 1 axis pneumatic artificial muscle

2.2.3 Characteristic of pneumatic artificial muscle manipulator 17

3 Improvement of the control performance of pneumatic artificial muscle

manipulator using an intelligent switching control method 20

3.2 Intelligent switching control algorithm 22

3.2.1 The overall control system 22

3.2.2 Recognition the external load condition

3.2.3 Proposition of the smooth switching algorithm 30

4 Nonlinear PID control to improve the control performance of pneumatic

artificial muscle manipulator using neural network 39

4.2 Nonlinear PID controller based on neural network 41

5 Improvement of the control performance of pneumatic artificial muscle

manipulator with an intelligent phase plane switching control

5.1 Introduction and characteristics of MRB 52

5.2 Conventional phase plane switching control algorithm 58

5.2.1 Positioning control system 58

5.2.2 Conventional phase plane switching control algorithm 59

5.3 Experimental results and conclusion of conventional phase plane

5.4 Intelligent phase plane switching control algorithm based on neural

network 65 5.5 Experimental results and conclusion of intelligent phase plane

switching control algorithm based on neural network 70

5.5.2 Conclusion 73

6 An intelligent control based on neural network and applying for force control

of pneumatic artificial muscle manipulator 80

6.2 Intelligent control based on neural network 82

6.2.3 An intelligent switching control algorithm 87 6.2.4 Proposition of the smooth switching algorithm 91

6.3.2 Conclusion 97 6.4 Force control of pneumatic artificial muscle manipulator 104

8.3 Conclusion 130

9 Improvement of the control performance of 2 axes pneumatic artificial muscle

manipulator using magneto-rheological brake 145

9.2 New concept of phase plane switching control algorithm 146

Reference 167

List of Figures

(a) Working of PAM

(b) PAM – FESTO Product

(c) The structure of PAM

Figure 2.2 A general configuration of 1 axis PAM manipulator 14 Figure 2.3 The schematic diagram of the PAM manipulator 15 Figure 2.4 Working principle of the PAM manipulator 16 Figure 2.5 A photograph of the experimental apparatus 16 Figure 2.6 Characteristics of the PAM manipulator 18

Figure 3.1 The schematic diagram of the PAM manipulator 21 Figure 3.2 A photograph of the experimental apparatus 22 Figure 3.3 Structure of the newly proposed control algorithm 24

Figure 3.6 Experimental results for learning data generation 28 (a) Control Input

(b) Angular Velocity

(c) Pressure 1

(d) Pressure 2

(e) Pressure Difference

Fig 3.8 Experiment results of the pneumatic artificial muscle manipulator without switching control in the case of three different external inertia load 33 Figure 3.9 Experimental results with no external inertia load (class 1) 34 Figure 3.10 Experimental results when external inertia load is 280 [kg.cm2] (class 2) 35Figure 3.11 Experimental results when external inertia load is 560 [kg.cm2] (class 3) 36

Figure 3.12 Experimental results when external inertia load is 420 [kg∙cm2] 37

Figure 3.13 Comparison of experimental results between with and without the

LVQNN 38

Figure 4.1 Structure of the nonlinear PID controller using neural network 42

Figure 4.4 Experimental results of conventional PID controller without external

Figure 4.5 Comparison between conventional PID controller and nonlinear PID

controller using neural network without external inertia load 49

Figure 4.6 Experimental results of nonlinear PID controller using neural network

Figure 4.7 Experimental results of conventional PID controller with and without

Figure 4.8 Comparison between conventional PID controller and nonlinear PID

controller using neural network with external inertia load 50

Figure 4.9 Experimental results of nonlinear PID controller using neural network

Figure 5.1 The schematic diagram of the PAM manipulator 55

Figure 5.2 A photograph of the experimental apparatus 56

Figure 5.3 Construction of magneto-rheological rotary brake 57

Figure 5.4 Characteristics of magneto-rheological brake 57

Figure 5.6 Concept of phase plane switching control 61

Figure 5.7 Comparison between PID controller 1 and PID controller 2 63

Figure 5.8 Experimental results of PID controller 2 with various loads 64

Figure 5.9 Experimental results of

proposed controller with various parameter of KED 64

Figure 5.10 Experimental results of proposed controller with various parameter of h 65

Fig 5.11 The structure of phase plane switching control using neural network 67 Figure 5.12 The block diagram of neural network 67

Figure 5.14 Comparison between conventional PID controller 2, conventional phase plane switching control and phase plane switching control using neural network

Figure 5.15 Comparison between the phase plane switching control using neural network and the newly proposed intelligent phase plane switching control algorithm

Figure 5.16 Experimental results of the newly proposed intelligent phase plane

Figure 5.17 Comparison between conventional PID controller 2, conventional phase plane switching control and phase plane switching control using neural network

Figure 5.18 Comparison between the phase plane switching control using neural network and the newly proposed intelligent phase plane switching control algorithm

Figure 5.19 Experimental results of the newly proposed intelligent phase plane

Figure 5.20 Comparison between conventional PID controller 2, conventional phase plane switching control and phase plane switching control using neural network

Figure 5.21 Comparison between the phase plane switching control using neural network and the newly proposed intelligent phase plane switching control algorithm

Figure 5.22 Experimental results of the newly proposed intelligent phase plane

Figure 6.1 Structure of the newly proposed intelligent control algorithm 83 Figure 6.2 Structure of neural network controller 84

Figure 6.4 Learning data for the LVQNN 89

Figure 6.5 Simulation results for learning data generation of control input 92

Figure 6.6 Simulation results for learning data generation of system response 93

Figure 6.8 Comparison of the simulation results with and without neural network

controller 98 Figure 6.9 Simulation results of system response with variation external

environments 98 Figure 6.10 Simulation results with respect to external environment 1 99

Figure 6.11 Simulation results with respect to external environment 2 100

Figure 6.12 Simulation results with respect to external environment 3 101

Figure 6.13 Simulation results with respect to

Figure 6.14 Comparison of the simulation results with and without proposed

intelligent controller with respect to environment 2 103

Figure 6.15 Comparison of the simulation results with and without proposed

intelligent controller with respect to environment 3 103

Figure 6.16 Comparison of the simulation results with and without proposed

intelligent controller with respect to environment between 2 and 3 104

Figure 6.17 Schematic diagram of the pneumatic artificial muscle manipulator 107

Figure 6.18 Working principle of the pneumatic artificial muscle manipulator 108

Figure 6.19 Photograph of the experimental apparatus 108

Figure 6.20 Structure of the proposed control algorithm 110

Figure 6.21 Experimental results of PID controller with respect to condition 1 and

Figure 6.22 Comparison of experimental results between conventional PID controller

Figure 6.23 Comparison of experimental results between conventional PID controller

Figure 6.24 Experimental results of proposed controller (condition 1) 114

Figure 6.25 Experimental results of proposed controller (condition 2) 115

Figure 6.26 Experimental results of PID controller with respect to condition 3 and

Figure 6.27 Comparison of experimental results between conventional PID controller

Figure 6.28 Comparison of experimental results between conventional PID

controller and proposed controller (condition 4) 117 Figure 6.29 Experimental results of proposed controller (condition 3) 118 Figure 6.30 Experimental results of proposed controller (condition 4) 119

Figure 7.1 A general configuration of 2 axis PAM manipulator 124 Figure 7.2 The schematic diagram of the PAM manipulator 125 Figure 7.3 Working principle of the PAM manipulator 126 Figure 7.4 A photograph of the experimental apparatus 127

Fig 8.1 Experimental results of conventional PID controller with various frequencies

Fig 8.2 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 1, frequency of reference f=0.05 Hz) 133 Fig 8.3 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 2, frequency of reference f=0.05 Hz) 133 Fig 8.4 Experimental results of nonlinear PID controller using neural network (Joint

Fig 8.5 Experimental results of nonlinear PID controller using neural network (Joint

Fig 8.6 Comparison between conventional PID controller and nonlinear PID

controller using neural network (frequency of reference f=0.05 Hz) 136 Fig 8.7 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 1, frequency of reference f=0.1 Hz) 137 Fig 8.8 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 2, frequency of reference f=0.1 Hz) 137

Fig 8.9 Experimental results of nonlinear PID controller using neural network (Joint

Fig 8.10 Experimental results of nonlinear PID controller using neural network

Fig 8.11 Comparison between conventional PID controller and nonlinear PID

controller using neural network (frequency of reference f=0.1 Hz) 140 Fig 8.12 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 1, frequency of reference f=0.2 Hz) 141 Fig 8.13 Comparison between conventional PID controller and nonlinear PID

controller using neural network (Joint 2, frequency of reference f=0.2 Hz) 141 Fig 8.14 Experimental results of nonlinear PID controller using neural network

Fig 8.15 Experimental results of nonlinear PID controller using neural network (Joint

Figure 9.3 Comparison between conventional PID controller and proposed controller with various frequencies of reference input (Joint 1) 151 Figure 9.4 Experimental results of the newly phase plane switching control algorithm

Figure 9.7 Experimental results of conventional PID controller with various

frequencies of reference input (Joint 2, load1) 154

Figure 9.8 Experimental results of conventional PID controller with various

frequencies of reference input (Joint 2, load 2) 155 Figure 9.9 Comparison between conventional PID controller and proposed controller with various frequencies of reference input (Joint 2, load 1) 156 Figure 9.10 Comparison between conventional PID controller and proposed

controller with various frequencies of reference input (Joint 2, load 2) 157 Figure 9.11 Experimental results of the newly phase plane switching control

Figure 9.11 Experimental results of the newly phase plane switching control

Figure 9.11 Experimental results of the newly phase plane switching control

Figure 9.11 Experimental results of the newly phase plane switching control

Figure 9.11 Experimental results of the newly phase plane switching control

Figure 9.11 Experimental results of the newly phase plane switching control

List of Tables

Table 3.1 Classification of the external inertia load 26

Table 3.3 Optimal parameters of the PID controller 31

Table 6.1 Optimal parameters of the PID controller 90 Table 6.2 Training success rate of the LVQNN (%) 94

A Study on the Control Performance Improvement of

Pneumatic Artificial Muscle Manipulator

Using Intelligent Control

Tu Diep Cong Thanh School of Mechanical and Automotive Engineering

Graduate School, University of Ulsan

Abstract The number of humans requiring rehabilitation due to fracture of a bone and joint disease caused by traffic accidents and cerebral apoplexy and for functional problems

in motion due to advanced age reaches several hundreds of millions on over the world Functional recovery therapy of rehabilitation is now normally carried out by medical therapists on a person-to-person basis, but automatic equipment has been put to practical use in physical therapy programs that repeat relatively simple operations Thus, the application of a robot to rehabilitation has become a matter of great concern For this function, seeking a novel actuator and applying it to human friendly therapy robot have become the challenging tasks as well as the modeling and control design for that appealing system

A study on the control performance improvement of pneumatic artificial muscle (PAM) manipulator in order realize the rehabilitation therapy robot in the near future is an urgent work that has led a mount of researchers attention on over the work in the recent years Dealing with that research, the intelligent controllers which are strongly encouraged in modern control are proposed and applied firstly to 1 axis PAM manipulator and updating in 2 axes PAM manipulator in the way of the step

by step in my study

Since, for the purpose of rehabilitation, it is impossible to distinguish between the robot working space and the human existing space, the robot must provide high levels of safety and flexibility for humans that are not common in general industrial

robots For this function, the robot must be constructional flexible However, this kind of robot has not been extensively studied due to the heavy and highly rigid because of its mechanical structure and motorization Therefore, the orientation of robotics toward applications needing greater proximity between the robot and the human operator has recently led researchers to develop novel actuator sharing some analogies with natural skeletal muscle A novel PAM actuator which has achieved increased popularity to provide these advantages such as high strength and high power to weight ratio, low cost, compactness, ease of maintenance, cleanliness, readily available and cheap power source, inherent safety and mobility assistance to humans performing tasks, has been regarded during the recent decades as an interesting alternative to hydraulic and electric actuators The PAM actuator is undoubtedly the most promising artificial muscle for the actuation of new types of human friendly therapy robots Thus, we use PAM actuators to operate the PAM manipulator drive for the functional recovery therapy However, the time variance, compliance, high hysteresis and nonlinearity, the air compressibility and the lack of damping ability of the PAM actuator bring the dynamic delay of the pressure response and cause the oscillatory motion Therefore, in the case of applying for the rehabilitation therapy robot, these following problems are urgent needed to concern:

6 To utilize in therapy robot in the near future such as rehabilitation robot

for recovery function of limbs, which is the final goal of our research, it is necessary to realize the performance of transient response, even if the external load conditions change severely

6 To realize the precise position and the satisfactory control performance

without regard the various external load conditions as well as the effect of nonlinear factors contained in the PAM manipulator must be considered

6 To execute rehabilitation more efficiently, the robot needs to adjust

adaptively its own impedance parameters according to the physical condition of the patient

6 It is inevitable for a robot to have interactions with its environment In

particular, when the robot is to interact with humans, such as in medical or nursing care jobs, the key issue is safety to the humans It means that

such-robot should perform its task precisely as well as satisfy servo-type force control while maintaining stability and proper compliance, especially with greatly changing external environment

6 For practical physical therapy, the precise position control performance is

needed without regard the various frequencies of the repeat of PAM manipulator with multiple degrees of freedom (DOF)

As with the above problems with PAM manipulator and the requirements for safety to the human in case of the rehabilitation therapy robot, the intelligent controllers are proposed in this thesis as follow:

6 First, to overcome some limitations such as a deterioration of the

performance of transient response due to the changes in the external load conditions in the PAM manipulator, an intelligent switching algorithm of the control parameter using a learning vector quantization neural network (LVQNN) is newly proposed LVQNN is newly applied as a supervisor

of the traditional PID controller, which estimates the external inertia load and switches the gain of the PID controller It proves that the LVQNN

is an appropriate algorithm for the recognition of quickly-changing external load conditions and it has a little computation time as well as an easy application to the PAM manipulator

6 Second, the problems with the time variance, compliance, high hysteresis

and nonlinearity of pneumatic systems, which have made it difficult to realize precise position control with high speed, will be handed by using a nonlinear PID controller using neural network Superb mixture of conventional PID controller and the neural network, which has powerful capability of learning, adaptation and tackling nonlinearity, brings us a novel nonlinear PID controller using neural network This newly proposed controller is appropriate for a kind of plants with nonlinearity uncertainties and disturbances

6 Thirst, to execute rehabilitation more efficiently, the robot needs to adjust

adaptively its own impedance parameters according to the physical condition of the patient For this purpose, the mechanical impedance of the

human arm or limb under therapy is used as an objective evaluation index for recovery and an estimation method based on the adaptabilities of neural network is proposed In order to realize satisfactory control performance, a variable damper – Magneto-Rheological Brake (MRB), is equipped to the joint of the manipulator Superb mixture of conventional PID controller and a phase plane switching control using neural network brings us a novel controller, which proves that the stability of the manipulator could be improved greatly in a high gain control without regard to the change of external load condition and without decreasing the response speed and low stiffness of PAM manipulator

6 Fourth, the PAM manipulator has been expected to apply to contact tasks

because of the ability of collision force absorption, delicate force control and so on due to air compressibility In this study, a newly proposed intelligent controller which includes both neural network controller as compensator and an intelligent switching control algorithm based on learning vector quantization neural network (LVQNN) is used to control force of complex dynamics PAM manipulator without regard the changes

of external environments

6 Last, the same some control algorithms above such as nonlinear PID

control using neural network and phase plane switching control method using MRB are applied for position control of 2 axes PAM manipulator with respect to various frequencies of the repeat of the system

Research is, of course, never finished and it is the task of any researcher to indicate the focus of future attention In this dissertation, a study on the control performance improvement of PAM manipulator in order realize the rehabilitation therapy robot in the near future is considered The effectiveness of the proposed intelligent control algorithms are demonstrated through simulation and experimental results of fabrication of 1 axis and 2 axes PAM manipulator And the experiments show that PAM manipulator is well enough for constructing human friendly therapy robot with the proposed intelligent control algorithms

– 1 –

Part I Introduction

– 2 –

Chapter 1 Introduction

is electric system with very limited use of hydraulics or pneumatics But electrical systems suffer from relatively low power to weight ratio and especially in the case of human friendly robot, or human coexisting and collaborative systems such as a medical and welfare fields Therefore, sharing the robot working space with its environment is problematic

Table 1.1 Comparison of actuators

Pneumatics Cheap, quick response time,

simple control Position control difficult, fluid compressible, noisy Hydraulics High power/weight ratio, low

backlash, very strong, direct

drive possible Less reliable, expensive, servo control complex, noisy Electrics Accurate position and

velocity control, quiet, relative cheap Low power and torque/weight ratios, possible sparking

– 3 –

Conversely, the human arm is not very accurate, but its lightness and joint flexibility due to the human musculature give it a natural capability for working in contact For the purpose of rehabilitation, it is impossible to distinguish between the robot working space and the human existing space, the robot must provide high levels of safety and flexibility for humans that are not common in general industrial robots For this function, the robot must be constructional flexible The orientation of robotics toward applications needing greater proximity between the robot and the human operator has recently led researchers to develop novel actuator sharing some analogies with natural skeletal muscle

A novel actuator, which has been regarded during the recent decades as an interesting alternative to hydraulic and electric actuators, is pneumatic artificial muscle (PAM) [3], which has achieved increased popularity to provide these advantages such as high strength and high power to weight ratio, low cost, compactness, ease of maintenance, cleanliness, readily available and cheap power source, inherent safety and mobility assistance to humans performing tasks It is superior to other actuators in the safety for human because of its flexibility and in the lightness, of which property is preferable in contacting tasks with human [4]

The PAM actuator is undoubtedly the most promising artificial muscle for the actuation of new types of human friendly therapy robots Thus, the PAM manipulator has been applied to construct a therapy robot in the case of having high level of safety for humans required For example, 2-dof robot for functional recovery therapy driven by pneumatic muscle [5-7], artificial muscle actuators for biorobotic systems [8-11], and a pneumatic muscle hand therapy device [12], in more recent year However, the time variance, compliance, high hysteresis and nonlinearity, the air compressibility and the lack of damping ability of the PAM manipulator bring the dynamic delay of the pressure response and cause the oscillatory motion Therefore,

it is not easy to realize a human-friendly therapy robot

Applying PAM actuator to rehabilitation therapy robot has become a challenging task and many control strategies have been proposed The limitations of the PAM manipulator have promoted research into a number of control strategies, such as a Kohonen-type neural network for the position control of robot end-effector within 1

– 4 –

cm after learning [13] Recently, the authors have developed a feed forward neural network controller, where joint angle and pressure of each chamber of pneumatic muscle are used as learning data and accurate trajectory following was obtained, with an error of 1o [14] After applying a feed forward + PID-type controller approach[15], the authors are turning to an adaptive controller [16-18], which is based on the on-line identification of a model with five parameters and three time delays Recently, the authors have announced that the position regulation of the joints of their arm prototype is better than ±0.5o [19] In addition, PID control [20], sliding mode control [21-23], fuzzy PD+I learning control [24], fuzzy + PID control [25], robust control[21, 23, 26], feedback linearization control [27], feed forward control + fuzzy logic [28], variable structure control algorithm [29], H∞control [30] and so on, have been applied to control the PAM manipulator Though these systems were successful in addressing smooth actuator motion in response to step inputs, the manipulator must be controlled slowly in order to get stable, accurate position control and the external load condition were also assumed to be constant or slowly varying In order to utilize PAM manipulator in therapy robot in the future such as rehabilitation robot for recovery function of limbs, it is necessary to realize of transient response, even if the external load conditions change severely But of transient response cannot be obtained in most pneumatic control system At the same time, the external load conditions cannot always be known exactly Therefore, it is necessary to propose a new intelligent control algorithm, which is applicable to a very compressible PAM system with various loads

Many intelligent control algorithms based on neural network have been proposed

An intelligent control using a neuro-fuzzy network was proposed by Iskarous and Kawamura [31] A hybrid network that combines fuzzy and neural network was used

to model and control complex dynamic systems, such as the PAM system An adaptive controller based on the neural network was applied to the artificial hand, which is composed of the PAM [32] Here, the neural network was used as a controller, which had the form of compensator or inverse of the model It was not easy to apply these control algorithms to the PAM manipulator with high hysteresis which is not modeled in the nonlinear inverse, to the quickly-changing external load

1.2 Motivation

A study on the control performance improvement of PAM manipulator in order realize the rehabilitation therapy robot in the near future is mentioned with the motivations and solutions as follow:

To utilize in therapy robot in the near future such as rehabilitation robot for recovery function of limbs, it is necessary to realize the performance of transient response, even if the external load conditions change severely

An intelligent switching algorithm of the control parameter using a learning vector quantization neural network (LVQNN) is newly proposed LVQNN is newly applied as a supervisor of the traditional PID controller, which estimates the external inertia load and switches the gain of the PID controller It was already proven by experiment on the position control of the pneumatic rod-cylinder that the LVQNN was an appropriate algorithm for the recognition of quickly changing external loads and it had little computation time as well as an easy application to the PAM system [33] The object of the study is to improvement of the control performance of PAM manipulator using an intelligent switching control method without regard for the great changes in external environments

To realize the precise position and the satisfactory control performance without regard the various external load conditions, the effect of nonlinear factors contained in the PAM manipulator will be handed by using a nonlinear PID controller using neural network Superb mixture of conventional PID controller and the neural network, which has powerful

– 6 –

capability of learning, adaptation and tackling nonlinearity, brings us a novel nonlinear PID controller using neural network This newly proposed controller is appropriate for a kind of plants with nonlinearity uncertainties and disturbances In addition, not only this newly proposed controller is very effective in the accurate position control of the PAM manipulator, but also it made the system more robust with respect to the change of plant parameter

To execute rehabilitation more efficiently, the robot needs to adjust adaptively its own impedance parameters according to the physical condition of the patient For this purpose, the mechanical impedance of the human arm or limb under therapy is used as an objective evaluation index for recovery and an estimation method based on the adaptabilities of neural network is proposed The compliance, damping torque and satisfactory control performance can be obtained by an intelligent phase plane switching control using magneto-rheological brake (MRB), which proved that the stability of the manipulator could be improved greatly in a high gain control without regard to the change of external environments in high gain control and without decreasing the response speed and low stiffness of manipulator

To satisfy servo-type force control, a newly proposed intelligent controller which includes both neural network controller as compensator of conventional PID controller and an intelligent switching control algorithm based on learning vector quantization neural network are used to control force of complex dynamics systems such as PAM manipulator without regard the changes of external environments A superb mixture of conventional PID controller and the neural network’s powerful capability

of learning, adaptive and tackle nonlinearity bring us a good-tracking force control for such a kind of plants which are high nonlinearity and friction In addition, with the greatly changing external force environments, a learning vector quantization neural network is applied as a supervisor of the

– 7 –

conventional PID controller, which estimate the external force environments and switch to the optimal gain of the PID controller

The precise position control performance of 2 axes PAM manipulator can

be obtain without regard the various frequencies of the repeat by applying both nonlinear PID control using neural network and a new concept phase plane switching control method The experiments show that PAM manipulator is well enough for constructing human friendly therapy robot with the proposed control algorithms

1.3 Outline of thesis

Research outlines for theses goals, which are mentioned above, are as follows: Improvement of the control performance of 1 axis PAM manipulator is given in part 2 with 5 chapters which are the configuration, an intelligent switching control method, nonlinear PID controller based on neural network, an intelligent phase plane switching control using magneto-rheological brake and an intelligent control based

on neural network as compensator and LVQNN as intelligent switching control method

A general configuration of 1 axis PAM manipulator is given in Chapter 2 In addition, totally experiment setup for developing these control algorithms above and the characteristic of PAM manipulator are also given in detail An introduction to the structure and operation of PAM is included in order to explain its extremely complexity, high nonlinearity and hysteresis

Chapter 3 provides an intelligent switching control method to overcome these problems such as a deterioration of the performance of transient response due to the great changes in the external environments in the PAM manipulator The object of this study is to implement proportional valves, rather than expensive servo valves, to develop a fast, accurate, inexpensive and intelligent PAM control system without regard for the great changes in external environments Experimental results was carried out in 1 axis PAM manipulator and from the experiments of the position control of an PAM manipulator, it was verified that the intelligent switching control

– 8 –

algorithm is very effective to overcome the deterioration of control performances of transient responses even if the external inertia load changed for 3,000[%] in conclusion

To handle the serious nonlinearity and high hysteresis, as well as the compliance, time variance, which have made PAM manipulator quite difficult to realize precise position control with high speed and high accuracy, is explained in Chapter 4 A novel nonlinear PID controller based on neural network is given in detail in this chapter One link of the PAM manipulator is fabricated and the effectiveness of the proposed control algorithm is demonstrated through the experiment The proposed nonlinear PID controller using neural network with compensation for nonlinear plants has shown that not only it was very effective in the accurate position control

of the PAM manipulator, but also it made the system more robust with respect to the change of plant parameter

An intelligent phase plane switching control using magneto-rheological brake, which harmonizes a phase plane switching control method, conventional PID controller and the adaptabilities of neural network, is newly proposed in Chapter 5

In order to reconcile both damping and response speed in high gain control, a magneto-rheological brake was applied as semi-active rotary damper to the PAM manipulator, where damping torque was tuned adaptively and optimally by the capacity of learning and adaptability of neural network with auto-tuning shape function From the experiments results of the position control of the PAM manipulator, it was verified that the newly proposed control algorithm was very effective in high gain control, good control performance, fast response and strong robust stability with respect to the 1,000[%] change of external inertia loads In particularly, it is anticipated that the newly proposed intelligent phase plane switching control using neural network with auto-tuning shape function is one of the most effective methods, which can be applied to a practically available human friendly therapy robot by using the PAM manipulator

Chapter 6 describes an intelligent controller, which includes both neural network controller as compensator and an intelligent switching control algorithm based on learning vector quantization neural network (LVQNN), to control force of complex

– 9 –

dynamics systems such as PAM manipulator and without regard the great changes external environments The newly intelligent controller presented in this chapter can making online control with better dynamic property, strong robustness and suitable for the control of various kinds of complex dynamics systems A more essential factor is that the proposed controller is easy applied to both accurate position control and force-control of various plants, including linear and nonlinear process and without regard greatly changing external environments

Next, improvement of the control performance of 2 axes PAM manipulator is discussed in part 3 with 3 chapters, which are configuration and experiment setup of 2 axes PAM manipulator, nonlinear PID controller based on neural network and intelligent phase plane switching control algorithm based on neural network In order

to apply PAM manipulator for the purpose of rehabilitation therapy robot in near future, 2 axes PAM manipulator is considered as the developing of my research and

it concludes a work that started as an assignment to find a suitable flexible robot and utilized in therapy robot in the near future such as rehabilitation robot for recovery function of limbs

The detail design of the prototype 2 axes PAM manipulator is treated in Chapter

7 with configuration and experiment setup In this system, it includes all components for the purposed of research in next 2 chapters 8 and 9

In Chapter 8, nonlinear PID controller based on neural network is one more again applied for 2 axes PAM manipulator If severe nonlinearity is involved in the controlled process, a nonlinear control scheme will be more useful, particularly in case of high nonlinearity of the PAM manipulator It is anticipated that the combination will take the advantage of simplicity of PID control and the neural network’s powerful capability of learning, adaptability and tackling nonlinearity Experimental results proved that the controller designed by this method does not need any training procedure in advance, but it uses only the input and output of the plant for the adaptation of control parameter and can tune the parameters iteratively New concept phase plane switching control algorithm is mainly discussed again

in Chapter 9 Two links of the PAM manipulator is fabricated and the effectiveness

of the proposed control algorithm is demonstrated through the experiments From

– 10 –

experimental results, we can conclude that newly proposed phase plane switching control is one of the most effective methods to develop a practically available human friendly robot by using PAM manipulator with MRB, which made a challenging and appealing system for modeling and control design

The final part is 4 with the conclusions and discussion is given in chapter 10 Summarizing the outcomes of this study is also mainly described in this chapter

Part 2 Improvement of the Control Performance of

1 Axis Pneumatic Artificial

Muscle Manipulator

Chapter 2 Configuration, Experiment Setup and Characteristic of Pneumatic Artificial Muscle Manipulator

2.1 Introduction

The pneumatic artificial muscle was developed in artificial limb research in the 1950s and 1960s [34-35] They have recently been commercialized by the Bridgestone Rubber Company of Japan for robotic applications [36], and re-engineered by Prof Jack Winters for construction of biomechanically realistic skeletal models Pneumatic artificial muscles consist of an internal bladder surrounded by a braided mesh shell (with flexible yet non-extensible threads) that is attached at either end to fittings or to some tendon-like structure The structure of the pneumatic artificial muscle is shown in Fig 2.1 When the internal bladder is pressurized, the high pressure gas pushes against its inner surface and against the external shell, and tends to increase its volume Due to the non-extensibility (or very high longitudinal stiffness) of the threads in the braided mesh shell, the actuator shortens according to its volume increase and/or produces tension if it is coupled to a mechanical load This physical configuration causes pneumatic artificial muscles to have variable-stiffness spring-like characteristics, non-linear passive elasticity, physical flexibility, and very light weight compare to other kinds of artificial actuators [37]

Thus, the development of one component of a rehabilitation therapy robot, pneumatic artificial muscle manipulator, is reported, which its configuration is explained in next section

Figure 2.1 The structure of the PAM

2.2 Configuration, experiment setup and characteristic of pneumatic

artificial muscle manipulator

2.2.1 Configuration of 1 axis pneumatic artificial muscle manipulator

A general configuration of 1 axis PAM manipulator is shown in Fig 2.2 This configuration is totally built for the purpose of researches such as an intelligent switching control algorithm using LVQNN, nonlinear PID control using neural network, an intelligent phase plane switching control algorithm with magneto-rheological brake and force control of PAM manipulator Thus, with respect to the purpose of research in next 4 chapters, some components in the PAM manipulator

(a) Working of PAM

(b) PAM – FESTO Product

(c) The structure of PAM

maybe redundance, but they do not make effect to total system More detail of PAM manipulator will be come in next section

Figure 2.2 A general configuration of 1 axis PAM manipulator

2.2.2 Experiment setup

Totally the schematic diagram of the pneumatic artificial muscle manipulator is shown in Fig 2.3 The hardware includes an IBM compatible personal computer (Pentium 1 GHz), which calculates the control input, controls the proportional valve (FESTO, MPYE-5-1/8HF-710 B) and Magneto-Rheological Rotary Brake (LORD, MRB-2107-3 Rotary Brake), through D/A board (Advantech, PCI 1720), and two pneumatic artificial muscles (FESTO, MAS-10-N-220-AA-MCFK) The braking torque of magneto-rheological rotary brake is controlled by D/A broad through voltage to current converter, Wonder Box Device Controller Kit (LORD, RD-3002-03) And the lists of experimental hardware are tabulated in Table 2.1 The rotating torque is generated by the pressure difference between the antagonistic artificial muscles and the external load is rotated as a result (Fig 2.4) A joint angle, θ, is

p Pneumatic Artificial Muscle

detected by rotary encoder (METRONIX, H40-8-3600ZO) and the air pressure into each chamber is also measured by the pressure sensors (FESTO, SDE-10-10) and fed back to the computer through 24-bit digital counter board (Advantech, PCL 833) and A/D board (Advantech, PCI 1711), respectively The signal from force sensor (SETECH, YC33-50k) fed back to the computer through the inline amplifier (SENSOTEC, 953489) and A/D board (Advantech, PCI 1711) in series The external inertia load is connected to joint directly in series and could be changed variously The experiments are conducted under the pressure of 0.4 [MPa] and all control software is coded in C program language A photograph of the experimental apparatus is shown in Fig 2.5

Figure 2.3 The schematic diagram of the PAM manipulator

PC

A/D Converter (PCI 1711 )

Counter Board (PCL – 833)

Magneto-Rheological

Rotary Brake

V/I Converter Wonder Box Device Controller Kit

Inline Amplifier

Force sensor

Proportional valve

Pneumatic line Electrical line Muscles

Figure 2.4 Working principle of the PAM manipulator

Figure 2.5 A photograph of the experimental apparatus

Muscle 1 Muscle 2

θ

2.2.3 Characteristic of pneumatic artificial muscle manipulator

The PAM is a tube clothed with a sleeve made of twisted fiber-code, and fixed at both ends by fixtures The muscle is expanded to the radial direction and constricted

to the vertical direction by raising an inner pressure of the muscle through a conversion mechanism of the fiber-codes The PAM has a property like a spring, and can change its own compliance by the inner pressure A few slide parts and a little friction are favorable to a delicate power control But the PAM has characteristics of hysteresis, non-linearity and low damping Particularly, the system dynamics of the PAM changes drastically by the compressibility of air in cases of changing external loads

power-When using the PAM for the control of manipulator, it is necessary to understand its characteristics such as the high hysteresis, and serious nonlinearity and so on Therefore the following experiments are performed to investigate the characteristics of the PAM Figure 2.6 and 2.7 demonstrate the hysteresis characteristics for the joint This hysteresis can be shown by rotating a joint along a pressure trajectory from P1=Pmax, P2=0 to P1=0, P2=Pmax and back again by incrementing and decrementing the pressures by controlling the proportional valve The hysteresis of the PAM is shown in Fig 2.7 The width of the gap between the two curves depends on how fast the pressures are changed; the slower the change in the pressures, the narrower the gap The trajectory, control input to the proportional valve, velocity, and pressure of each chamber of the PAM are depicted in Fig 2.6 The velocity is numerically computed from the position Near the extreme values, the joint velocity decreases since the increase in exerted force for a constant change

in pressure is less

Figure 2.6 Characteristics of the PAM manipulator

-30-20-100

θ o []

0510

o /s]

0.00.2

Figure 2.7 Hysteresis of the PAM manipulator

Table 2.1 The lists of experimental hardware

-35 -30 -25 -20 -15 -10 -5 0 5

∆ P [MPa]

1 Proportional valve MPYE-5-1/8HF-710 B FESTO

6 Wonder Box Device

9 24-bit digital counter board PCL 833 ADVANTECH

11 Inline amplifier SENSOTEC, 953489 SENSOTEC

Chapter 3

Improvement of the Control Performance of Pneumatic Artificial Muscle Manipulator Using an Intelligent Switching Control Method

3.1 Introduction

As the PAM manipulator is one of the well-known systems for safety with humans, it is preferable in contacting tasks with humans and many control strategies have been proposed As a result, a considerable amount of research has been devoted

to the development of various position control systems for the PAM manipulator In particularly, for widespread use of these actuators in the field of manipulator, a high speed, precise control of the PAM manipulator is required Among previous control approaches, an adaptive controller [15-19], fuzzy logic controller [24-25, 28, 38-39], robust controller [21-23, 26, 40] and so on, have been applied to control the PAM manipulator Though these systems were successful in addressing smooth actuator motion in response to step inputs, many of these systems used expensive servo valves and the external inertia load were also assumed to be constant or slowly varying The external inertia load is not always exactly known and the contact force with humans

is different in each case when the manipulator will be used as a therapy robot in the near future Therefore, it is necessary to propose a new intelligent control algorithm, which is applicable to a very compressible pneumatic muscle system with various loads

Many intelligent control algorithms based on a neural network have been proposed [13, 31-32, 41] Here, the neural network was used as a controller, which had the form of compensator or inverse of the model and it was not easy to apply these control algorithms to the quickly-changing inertia load systems