Performance Analysis of Soft Computing Technique Based Adaptation Mechanism for Model Reference Adaptive Control

Abstract

Adaptive Controller is a suitable choice for plants with variable parameters or with uncertain dynamics. Reference trajectory or set point values for the controller can be specified with different ways. It can be specified as a set of range of parameters, reference input trajectory or reference model output with boundary conditions and physical constraints. In this work, Model Reference Adaptive Control (MRAC) has been explored. The effectiveness of soft computing technique based adaptation mechanism for MRAC is analysed in the present work. MRAC using classical techniques like MIT rule and Lyapunov stability based method is applied in the earlier part of this work to study plant response and collect adaptation parameter data. Then soft computing techniques based adaptation mechanism are applied to control the same plant. Soft computing techniques like Neural Network (NN), Adaptive Neuro Fuzzy Inference System (ANFIS) and Fuzzy Subtractive Clustering (FSC) are applied. Results are presented, analyzed and compared for applied strategies. Idea is to take benefit of inherent learning and adaptation ability of various soft computing techniques in controller design. Adaptation mechanism is a very important part of MRAC. Observation of simulation results with different adaptation mechanism discussed in thesis reveals the strength of soft computing based MRAC compared to classical MRAC methods. The discussed strategy is more useful for a plant with uncertainty or variable parameter systems. Comparison of response with soft computing based MRAC and classical technique based MRAC shows that soft computing techniques like NN, ANFIS or FSC are effective in following model output and handling uncertainty. Statistical analysis of error data and control effort also supports derived results. With results it has been observed that to ANFIS based approach improves results than only NN based approach due to addition of fuzzy logic with NN. Even though NN based and ANFIS based approaches reduce error and reduce controller effort compared to classical technique, they have certain limitations like longer training time and requirement of huge amount of data for processing. FSC has been explored to deal with such issues and for result quality analysis. FSC based adaptation for MRAC is not addressed so far. FSC technique efficiently extracts and minimizes a set of rules for adaptation with a required minimum number of clusters. In present work, FSC based adaptation for MRAC controller has been tested on various systems. Results are validated on DC motor setup. It has come out as the best of all applied strategies with improved results. Overall stability of proposed system is proven for the FSC based technique.