Robust Adaptive Fault Diagnosis For T-S Fuzzy Systems

Due to the widely existence of complex nonlinearity in practical applications,considerable research effort has been devoted to the investigation of nonlinear systems.Takagi-Sugeno(T-S)fuzzy model provides a general framework to deal with nonlinear dynamics,then it has been used in many industrial systems.With the increasing complexity of practical systems,the demands of reliability and safety for these systems are more and more rigid.Then fault diagnosis technique has received extensive concern and widely applied in many fields.This thesis studies the problem of fault diagnosis for T-S fuzzy systems.The main contributions of the thesis are summarized as follows:First,this thesis is concerned with the fault estimation(FE)problem for a class of interconnected nonlinear systems described by T-S fuzzy models.Different from the existing FE approaches,where the reference inputs are always viewed as external disturbances,the considered reference signals are injected directly in the controller to excite the system states to ensure persistent excitation(PE)condition.Within this framework,a bank of adaptive observers are then designed to effectively estimate the actuator fault parameters even in the presence of nonlinear interconnections.Moreover,in the disturbance-free case,by employing graph theory,a global Lyapunov function is constructed such that the corresponding fault parameter estimate errors are proved to be asymptotically convergent provided that the interconnected strengthes are less than a given constant.Finally,two simulation examples are provided to demonstrate the validity of the presented FE scheme.This thesis also considers the finite frequency fault detection(FD)problem for a class of T-S fuzzy systems with partly immeasurable premise variables.First,the equivalence class in set theory is introduced to describe the T-S fuzzy systems,which can explicitly separate the measurable premise variables from the immeasurable ones,then the available information is fully used for the FD observer design to reduce the conservatism.In this case,however,the premise variables of the fuzzy models and the observer to be designed are unsynchronized which may lead to the existing FD methods based on parallel distribute compensation(PDC)strategy are infeasible.To solve this problem,a novel non-PDC unknown input observer combining H?/H_performance is designed to detect faults.Finally,a simulation example is given to verify the merits of the presented FD scheme.