Fault Detection And Diagnosis For Dynamic Systems

Aiming to solve some existing problems and follow trends in fault detection and diagnosis, combined with new results obtained in relative disciplines, some kinds of schemes for dynamic systems are proposed in this dissertation.1. A modified denoising method based on VC dimension and wavelet package is presented, improving the shortcomings of denoising methods based on empirical risk minimization and wavelets thresholds. Using FFT and wavelet package analysis theory, the abrupt fault signals detection for power systems on amplitude and frequency is also mentioned to advance the wavelet MRA signal detection approach.2. Two distinct BP training algorithms on multilayer perceptron type neural-network are developed to improve the trained network's quickness, robustness and generality. They are brought out from the combination of existing investigations-dynamic learning rate, momentum terms and quadratic function of weights, and amendment of desired error function respectively. The simulative results of rotating machinery fault pattern classification illustrated the effectiveness of the new algorithms.3. The fault detection of model-based nonlinear systems is studied by wavelet neural networks(WNN). First, a constructive learning algorithm is presented for WNN based on MRA. It can improve the calculative efficiency by getting the weights and hidden-layer nodes simultaneously. The WNN is used to construct the I/O nonlinear observer for a class of nonlinear system, and the observer tracks the real I/O signals to carry out real-time fault detection. Then, a kind of WNN based on single-scaling multidimensional wavelet frames and its matching pursuit algorithm is introduced. It is applied to approximate the nonlinear terms with Lipschitz property of nonlinear systems to establish the adaptive state observer. The robust fault detection is realized by the observer, demonstrating the predominant performance of the WNN. While the RBFNN is also utilized to predict faults.4. Fault diagnosis based on rough sets theory is presented. Here, to solve the difficulty of knowledge acquisition, three modified-RS-model -based methods are given for fault diagnostic knowledge extraction respectively. They also make up for the deficiency of classical RS theory. Then, a framework based on variable precision rough sets theory is produced for fault diagnostic expert systems.5. Observer-based fault detection is researched for time-delay systems. Some robust faultdetection schemes are developed, concerned the typical time-delay systems-uncertain systems with time-delay in state, multiple state time-delay systems with nonlinear disturbance, neutral time-delay systems and connecting large-scale time-delay systems. The stable conditions and design of gain matrices of the observers are given and proved by Lyapunov function, LMI and H control theory etc..Numerical simulations and experiments demonstrated the effectiveness and feasibility of all the proposed methods.