Study On Fault Diagnosis & Testability In Analog Circuits Based On Time-Frequency-Domain Analysis And Neural Network

Over the last several decades, fault diagnosis theory and methods of analog circuits are always the challenging topics in international test research domain. The rapid progress in modern electronic and computer technology promotes the advent of system-on-chip and mixed-signal integrated circuits, which present higher and newer circuit test request; intensive study on fault diagnosis theories and methods is urgent task. With many years' development, the analog circuit tests acquire some research progress, but there appear many new theory research topics, such as the diagnosis of multiple faults and soft faults in analog circuit, the fault diagnosis in nonlinear analog circuit, which made conventional or traditional theories and methods of fault diagnosis invalid. As typical representation of computation intelligence technology, neural network provides a potent way for faults diagnosis in analog circuit, which takes circuit theoreticians' more attention. This dissertation based on neural network studies fault feature extraction, fault diagnosis methodology and testability problem in analog circuit (linear and nonlinear) profoundly. Its' theoretical basis is modern test technology, signal processing, system identification and testability analysis, etc. Author's main work concentrates on four aspects as follows:1. Feature extraction and fault diagnosis in analog circuits.â‘ In frequency domain, fault features are extracted from efficient points in frequency response curve directly. Simulation results validate the fault diagnosis effect of analog circuit based on neural network.â‘¡In time domain, dimensionality reduction in circuit characteristic variables is realized by principal component analysis, the diagnosis efficiency improved effectively.â‘¢In time-frequency domain, the noise immunization processing and fault feature extraction from circuit response are realized by wavelet transform, A fault diagnosis approach based on wavelet neural network for analog circuits is presented. â‘£The feature evaluation and extraction methods based on neural network are presented. The complex classification problem on faults pattern recognition in analog circuit is transferred to feature processing stage by feature extraction based on neural network effectively. Diagnostic illustrations validate this method.Fault feature extraction methods on analog circuit are researched from above four facts, which further optimize fault features, fault diagnosis efficiecy improve.2. Fault diagnosis in analog circuit based on binary tree with maximum fault information volume.The binary tree diagnosis method in analog circuit is presented. faults are located from fault feature sets with maximum fault information volume by information transmission characteristic of binary tree. In addition, the fuzzy graph is applied to fault diagnosis in analog circuit, the fault source is found by seeking for the strongest route with maximum fault information volume, fault diagnosis effect is well.3. Fault feature extraction and diagnosis method based on frequency domain kernel in nonlinear circuit.The frequency domain kernels are extracted from nonlinear circuit identification by Volterra series, they are preprocessed to construct fault features sample sets, which are input into neural network for fault diagnosis. The difficult identification problem in nonlinear circuit is resolved well, superior diagnosis efficiency acquired.4. Testability problem in analog circuit.To branching diagnostic method, the testable topological structure and testable topological condition of circuit are analyzed; the methods on testability analysis and testability design are presented. Research shown: The method has important practical significance in guiding to the testability design and fault diagnosis of analog circuit.The experimental results of theoretical studies support proposed methodologies and obtained conclusions in this dissertation.