Yung-Yaw Chen

Design of self-learning fuzzy sliding mode controllers based on genetic algorithms

Abstract In this paper, genetic algorithms were applied to search a sub-optimal fuzzy rule-base for a fuzzy sliding mode controller. Two types of fuzzy sliding mode controllers based on genetic algorithms were proposed. The fitness functions were defined so that the controllers which can drive and keep the state on the user-defined sliding surface would be assigned a higher fitness value. The sliding surface plays a very important role in the design of a fuzzy sliding mode controller. It can dominate the dynamic behaviors of the control system as well as reduce the size of the fuzzy rule-base. In conventional fuzzy logic control, an increase in either input variables or the associated linguistic labels would lead to the exponential growth of the number of rules. The number of parameters or the equivalent length of strings used in the computations of genetic algorithms for a fuzzy logic controller are usually quite extensive. As a result, the considerable computation load prevents the use of genetic operations in the tuning of membership functions in a fuzzy rule-base. This paper shows that the number of rules in a fuzzy sliding mode controller is a linear function of the number of input variables. The computation load of the inference engine in a fuzzy sliding mode controller is thus smaller than that in a fuzzy logic controller. Moreover, the string length of parameters is shorter in a fuzzy sliding mode controller than in a fuzzy logic controller when the parameters are searched by genetic algorithms. The simulation results showed the efficiency of the proposed approach and demonstrated the applicability of the genetic algorithm in the fuzzy sliding mode controller design.

A GA-based fuzzy controller with sliding mode

In this study, the genetic algorithms are applied to find out a nearly optimal fuzzy rule-base for fuzzy sliding mode controller in the sense of fitness. In conventional fuzzy logic controllers (FLC), linearly increasing in either input variables or input linguistic labels would lead the number of rules grow up exponentially. Since the larger size of rule base would cause the longer string length and higher computing load, it becomes one of the difficulties of realizing genetic algorithms to search the suitable rules or membership functions for fuzzy logic controllers. This paper will show that the number of rules in fuzzy sliding mode controller (FSMC) is a linear function of input variables, such that the inferring load of the inference engine in FSMC is more light than that of FLC, and the string length of unknown parameters in FSMC is shorter than that in FLC. Therefore, using genetic algorithms to search fuzzy rules or membership functions for FSMC becomes more economical and applicable. The simulation results verify the efficiency of proposed approach.<<ETX>>

A self-learning fuzzy controller

A learning scheme for the fuzzy control systems is proposed. The scheme can find proper control actions in the rule-base of a fuzzy controller for a number of different dynamic systems, including the cart-pole system. The learning scheme, which involves the ideas of temporal difference and fuzzy control, can construct a fuzzy controller, which can serve the control purpose on its own after the learning process. A number of different plant dynamics have also been tested, and the results suggest that the method is universal enough to control processes other than the commonly seen inverted pendulum.<<ETX>>

RBF-network-based sliding mode control

A sliding mode controller (SMC) design method based on radial basis function network (RBFN) is proposed in this paper. Similar to the multilayer perceptron, the RBFN also known to be a good universal approximator. In this work, the weights of the RBFN are changed according to some adaptive algorithms for the purpose of controlling the system state to hit a user-defined sliding surface and then slide along it. The initial weights of the RBFN can be set to small random numbers, and then online tuned automatically, no supervised learning procedures are needed. By applying the RBFN-based sliding mode controller to control a nonlinear unstable inverted pendulum system, the simulation results show the expected approximation sliding property was occurred, and the dynamic behavior of the control system can be determined by the sliding surface.<<ETX>>

Rules extraction for fuzzy control systems

A method of extracting the control rules for a fuzzy controller is proposed. The method successfully constructs the rule base for a fuzzy controller, which is usually provided by human experts. By using the concepts of cell state space and performance evaluation, an optimization algorithm is able to achieve the proper control of a process. The control rules can then be retrieved from the output of this algorithm. This method extracts qualitative control rules from the quantitative processes.<<ETX>>

Real-coded genetic algorithm based fuzzy sliding-mode control design for precision positioning

A fuzzy sliding-mode controller was optimized through real-coded genetic algorithms and successfully implemented on an industrial XY table. The fuzzy sliding-mode controller is special type of fuzzy controller. By using the sliding surface, the fuzzy rule is simpler and the entire rule base is more compact. Thus, more easy to apply self-learning schemes. The real-coded genetic algorithm uses the internal floating-point representation of the computer system. With this advantage, the finite resolution problem of traditional genetic algorithm has been solved. The experimental results show the success of this approach.

Human Posture Recognition by Simple Rules

Recognition of human posture for home care system is studied in this paper. Home care system is to determine whether man is in danger or not by image of video and the emergent signal will be sent to hospital and family if man is in danger. Computer vision is a popular tool to solve this kind of problem. The human silhouette is an useful information for the purpose of recognition of human posture and it can be separated from video by imaging process. There is an assumption that the human silhouette has been obtained in this research. In this paper, human posture will be classified as stand, sit, kneel, and stoop by some parameters and simple rules. It is a real time system if using this method and the correct rate exceeds in 90%.

Vision based fruit sorting system using measures of fuzziness and degree of matching

Fuzzy approaches were used to determine optimal thresholding values of fruits images, and fuzzy degree of matching was applied to classify the color and size of fruit. Results showed that fuzzy method was superior to the traditional statistical methods, and a accuracy of 93.3% for combined sorting was reported. The errors due to miscategorization could thus be reduced if the fuzzy methods were used. The developed fuzzy algorithms were integrated with the machine vision guided robotic sorting system for fruits.<<ETX>>

A fuzzy scheduling controller for the computer disk file track-following servo

A fuzzy tuning algorithm is developed for the computer disk drive track following servo system. A Zentek 3100 disk drive was modified, and a controller scheduling capability was added to the servo loop to compensate for the plant variations as the actuator was locked on to different tracks. The mathematical models for the actuator on a number of tracks chosen were experimentally identified. The H/sub infinity / design technique is then employed to obtain a robust optimal controller for each operating point. A combined controller is then calculated using a fuzzy algorithm. The fuzzy algorithm is used to represent the complex relationship between the track number and the corresponding controller. It is shown that with the controller scheduling action, the closed-loop performance is improved for the actuator at every track position.<<ETX>>

A variable-based genetic algorithm

Genetic algorithms are very powerful search algorithms based on the mechanics of natural selection and natural genetics. As well known, one of differences from many other conventional search algorithms is that genetic algorithms require the natural parameter set of the optimization problem to be coded as a finite-length string. However, the encoding and decoding processes waste many computation time and lose the accuracy of the parameters. In this paper, a novel variable-based genetic algorithm is proposed. The algorithm processes the parameters themselves without coding. It can save the coding processing time and get more accurate values of the parameters. Finally, the system identification problem has been used to demonstrate the power of the algorithm.<<ETX>>