Fault Detection And Fault-Tolerant Control Based On Online Fuzzy Identification

In this thesis, a novel intelligent fault detection and fault-tolerant control scheme is proposed bycombining data-driven and model-based methods, based on-line fuzzy clustering and closed-loopfuzzy identification of nonlinear uncertain systems. Takagi-Sugeno (T-S) fuzzy systems, which havebeen proven to be universal approximators, are employed to model nonlinear systems. The T-S fuzzymodel is identified online based on input-output data identification, whose structure (number of fuzzyrules) is evolving. According to change of the number of fuzzy rules, we can determine whether thereis abnormal or system failure occurs. A nonlinear feedback fault-tolerant controller is design based onthe on-line identified T-S fuzzy model to compensate the impact of failure on the system to ensuresystem stability while preserving the system’s key performance.We have completed the following work:1. The identification methods of T-S fuzzy models are studied, including the choice inputvariables, the determination of the membership functions, the generation mechanism fuzzy rules, theparameter learning algorithm. We have examined how to recognize the division of spaceautomatically through the online fuzzy clustering algorithm, and how to update existing or create anew fuzzy cluster center. Simulation examples have verified that the online clustering algorithm canautomatically generate fuzzy rules, which builds a foundation for the following fault detection andfault-tolerant control design.2. The sensitivity of online clustering algorithm to the system error/failure and its robustness tothe system parameter perturbations has been studied. When the nonlinear system works under normalconditions and there are only small parameter fluctuations, the proposed algorithm does not adjust thestructure of the off-line identified T-S fuzzy model since simply adjusting the fuzzy model consequentparameters can reduce the approximation error; when the system parameters or a greater degree ofstructural change (the system can be considered a failure occurs), the online fuzzy clusteringalgorithm can identify it and automatically generate a new fuzzy rule by clustering analysis of thesystem input and output data. By observing the change of number of rules, the system failure ormalfunction can be detected.3. The fault-tolerant control strategy for single-input-single-output nonlinear systems basedon-line fuzzy identification has been studied. A feedback reconfigurable controller is designed basedon the on-line identified T-S fuzzy model, by using its structure and parameters to achieve fault-tolerant control. The stability of the closed-loop systems is proven based on Lyapunov stabilitycriterion, the parameter adaptation law is derived, a compensation controller is designed and theeffects of modeling errors and disturbance on system stability and tracking performance are analyzed.4. Based on the research results for single-input-single-input systems, the proposed faultdetection and fault-tolerant control scheme is extended to multiple-input-multiple-output systems.The on-line fuzzy identification algorithm is modified to accommodate multivariable nonlinearsystems. A fault-tolerant controller is designed based on the multivariable T-S fuzzy model and itsparameter adaptive law is derived. Finally the closed-loop system stability is analyzed and simulationresults verify the effectiveness of the proposed fault detection and fault-tolerant control strategy.