Design of neural-genetic based optimal fuzzy logic controller and application to a DC servo-system

Research output: ThesisDoctoral Thesis

Abstract

The design process of a fuzzy logic controller (FLC) usually can be divided into four sub stages. These are determination and tuning of input and output membership functions, determination and tuning operation of the input and output scaling factors, design operation of the rule base and choosing the implication, inference and defuzzification methods. Among them, much more attention should be paid for determination of the input and output scaling factors since, the output-scaling factor has the most influence on stability and oscillation tendency whereas the input scaling factors have the most influence on basic sensitivity of the controller with respect to the optimal choice of the operating areas of the input signals. However, relative importance of the input and output scaling-factors to the performance of a FLC system is yet to be fully established.

In this work a new and systematic method for the determination of the optimal values of input and output scaling factors is proposed. The method uses Genetic Algorithms (GA’s) and Artificial Neural Networks (ANN’s). The method uses the GA for searching the optimal values of scaling factors whereas the ANN is used for the computation of fitness function in each generation.

On the other hand a new self-tuning FLC architecture (NGTOSTFP-ID) is proposed. Here, the output-scaling factor is adjusted on-line by an independent fuzzy controller according to the trends of the controlled process. The rule base for tuning the output-scaling factor is defined on error (e) and change of error ( e) of the controlled variable. Also, the rule base of the main fuzzy controller is designed by a gradient decent technique without using any idea of a skilled operator. Furthermore, in order to eliminate the drawbacks of the FLC in transients, a feedforward integrator and a feedback derivative controller blocks are added to the controller architecture.

In order to demonstrate the validity of the proposed optimization method, the method is applied to conventional PI, conventional fuzzy PI, hybrid type fuzzy PID, self tuning type fuzzy PI and NGTOSTFP-ID controllers. Also the performances of these controllers are compared on a real-time servo system including a permanent magnet DC motor and load. All the controllers including neural network based system identifiers and genetic optimizers are coded under C++ and applied to the system via a Pentium PC. The performances of the controllers including the proposed self-tuning FLC are compared in terms of several performance measures such as peak overshoot, settling time, rise time, integral of absolute error (IAE) and integral of the time weighted absolute error (ITAE), in addition to the responses due to ramp type trajectory and finally, in each case, the proposed scheme tuned by introduced optimization method shows a remarkably improved performance over the other above-mentioned controllers.
Original languageEnglish
QualificationPh.D.
Awarding Institution
Supervisors/Advisors
  • Cansever, Galip, Supervisor
Publication statusPublished - 2002

Fingerprint

Servomechanisms
Fuzzy logic
Controllers
Tuning
Neural networks
Genetic algorithms
DC motors
Membership functions
Permanent magnets

Keywords

  • fuzzy logic control
  • genetic algorithms
  • artificial neural networks
  • scaling factor
  • self-tuning control

Cite this

@phdthesis{67643734d3d441748fd6cfbdb32539d2,
title = "Design of neural-genetic based optimal fuzzy logic controller and application to a DC servo-system",
abstract = "The design process of a fuzzy logic controller (FLC) usually can be divided into four sub stages. These are determination and tuning of input and output membership functions, determination and tuning operation of the input and output scaling factors, design operation of the rule base and choosing the implication, inference and defuzzification methods. Among them, much more attention should be paid for determination of the input and output scaling factors since, the output-scaling factor has the most influence on stability and oscillation tendency whereas the input scaling factors have the most influence on basic sensitivity of the controller with respect to the optimal choice of the operating areas of the input signals. However, relative importance of the input and output scaling-factors to the performance of a FLC system is yet to be fully established. In this work a new and systematic method for the determination of the optimal values of input and output scaling factors is proposed. The method uses Genetic Algorithms (GA’s) and Artificial Neural Networks (ANN’s). The method uses the GA for searching the optimal values of scaling factors whereas the ANN is used for the computation of fitness function in each generation. On the other hand a new self-tuning FLC architecture (NGTOSTFP-ID) is proposed. Here, the output-scaling factor is adjusted on-line by an independent fuzzy controller according to the trends of the controlled process. The rule base for tuning the output-scaling factor is defined on error (e) and change of error ( e) of the controlled variable. Also, the rule base of the main fuzzy controller is designed by a gradient decent technique without using any idea of a skilled operator. Furthermore, in order to eliminate the drawbacks of the FLC in transients, a feedforward integrator and a feedback derivative controller blocks are added to the controller architecture. In order to demonstrate the validity of the proposed optimization method, the method is applied to conventional PI, conventional fuzzy PI, hybrid type fuzzy PID, self tuning type fuzzy PI and NGTOSTFP-ID controllers. Also the performances of these controllers are compared on a real-time servo system including a permanent magnet DC motor and load. All the controllers including neural network based system identifiers and genetic optimizers are coded under C++ and applied to the system via a Pentium PC. The performances of the controllers including the proposed self-tuning FLC are compared in terms of several performance measures such as peak overshoot, settling time, rise time, integral of absolute error (IAE) and integral of the time weighted absolute error (ITAE), in addition to the responses due to ramp type trajectory and finally, in each case, the proposed scheme tuned by introduced optimization method shows a remarkably improved performance over the other above-mentioned controllers.",
keywords = "fuzzy logic control, genetic algorithms, artificial neural networks, scaling factor, self-tuning control",
author = "Kucukdemiral, {Ibrahim Beklan}",
note = "Date of award (from thesis module record): 24/5/2002, ET Thesis title not found in ethos. Validated as from a Turkish university and they don't appear to have a repository. ET 18/6/19",
year = "2002",
language = "English",

}

TY - THES

T1 - Design of neural-genetic based optimal fuzzy logic controller and application to a DC servo-system

AU - Kucukdemiral, Ibrahim Beklan

N1 - Date of award (from thesis module record): 24/5/2002, ET Thesis title not found in ethos. Validated as from a Turkish university and they don't appear to have a repository. ET 18/6/19

PY - 2002

Y1 - 2002

N2 - The design process of a fuzzy logic controller (FLC) usually can be divided into four sub stages. These are determination and tuning of input and output membership functions, determination and tuning operation of the input and output scaling factors, design operation of the rule base and choosing the implication, inference and defuzzification methods. Among them, much more attention should be paid for determination of the input and output scaling factors since, the output-scaling factor has the most influence on stability and oscillation tendency whereas the input scaling factors have the most influence on basic sensitivity of the controller with respect to the optimal choice of the operating areas of the input signals. However, relative importance of the input and output scaling-factors to the performance of a FLC system is yet to be fully established. In this work a new and systematic method for the determination of the optimal values of input and output scaling factors is proposed. The method uses Genetic Algorithms (GA’s) and Artificial Neural Networks (ANN’s). The method uses the GA for searching the optimal values of scaling factors whereas the ANN is used for the computation of fitness function in each generation. On the other hand a new self-tuning FLC architecture (NGTOSTFP-ID) is proposed. Here, the output-scaling factor is adjusted on-line by an independent fuzzy controller according to the trends of the controlled process. The rule base for tuning the output-scaling factor is defined on error (e) and change of error ( e) of the controlled variable. Also, the rule base of the main fuzzy controller is designed by a gradient decent technique without using any idea of a skilled operator. Furthermore, in order to eliminate the drawbacks of the FLC in transients, a feedforward integrator and a feedback derivative controller blocks are added to the controller architecture. In order to demonstrate the validity of the proposed optimization method, the method is applied to conventional PI, conventional fuzzy PI, hybrid type fuzzy PID, self tuning type fuzzy PI and NGTOSTFP-ID controllers. Also the performances of these controllers are compared on a real-time servo system including a permanent magnet DC motor and load. All the controllers including neural network based system identifiers and genetic optimizers are coded under C++ and applied to the system via a Pentium PC. The performances of the controllers including the proposed self-tuning FLC are compared in terms of several performance measures such as peak overshoot, settling time, rise time, integral of absolute error (IAE) and integral of the time weighted absolute error (ITAE), in addition to the responses due to ramp type trajectory and finally, in each case, the proposed scheme tuned by introduced optimization method shows a remarkably improved performance over the other above-mentioned controllers.

AB - The design process of a fuzzy logic controller (FLC) usually can be divided into four sub stages. These are determination and tuning of input and output membership functions, determination and tuning operation of the input and output scaling factors, design operation of the rule base and choosing the implication, inference and defuzzification methods. Among them, much more attention should be paid for determination of the input and output scaling factors since, the output-scaling factor has the most influence on stability and oscillation tendency whereas the input scaling factors have the most influence on basic sensitivity of the controller with respect to the optimal choice of the operating areas of the input signals. However, relative importance of the input and output scaling-factors to the performance of a FLC system is yet to be fully established. In this work a new and systematic method for the determination of the optimal values of input and output scaling factors is proposed. The method uses Genetic Algorithms (GA’s) and Artificial Neural Networks (ANN’s). The method uses the GA for searching the optimal values of scaling factors whereas the ANN is used for the computation of fitness function in each generation. On the other hand a new self-tuning FLC architecture (NGTOSTFP-ID) is proposed. Here, the output-scaling factor is adjusted on-line by an independent fuzzy controller according to the trends of the controlled process. The rule base for tuning the output-scaling factor is defined on error (e) and change of error ( e) of the controlled variable. Also, the rule base of the main fuzzy controller is designed by a gradient decent technique without using any idea of a skilled operator. Furthermore, in order to eliminate the drawbacks of the FLC in transients, a feedforward integrator and a feedback derivative controller blocks are added to the controller architecture. In order to demonstrate the validity of the proposed optimization method, the method is applied to conventional PI, conventional fuzzy PI, hybrid type fuzzy PID, self tuning type fuzzy PI and NGTOSTFP-ID controllers. Also the performances of these controllers are compared on a real-time servo system including a permanent magnet DC motor and load. All the controllers including neural network based system identifiers and genetic optimizers are coded under C++ and applied to the system via a Pentium PC. The performances of the controllers including the proposed self-tuning FLC are compared in terms of several performance measures such as peak overshoot, settling time, rise time, integral of absolute error (IAE) and integral of the time weighted absolute error (ITAE), in addition to the responses due to ramp type trajectory and finally, in each case, the proposed scheme tuned by introduced optimization method shows a remarkably improved performance over the other above-mentioned controllers.

KW - fuzzy logic control

KW - genetic algorithms

KW - artificial neural networks

KW - scaling factor

KW - self-tuning control

M3 - Doctoral Thesis

ER -