The Research About The Mechanisms And Applications Of The Fault Diagnosis Based On Soft Computing

The security and reliability of a system have been increasingly catching people's attention, and are becoming the important parameters in designing and appraising systems. The significance of fault diagnosis lies on avoiding accidents or reducing the damages of a disaster. Fault diagnosis has great economic and social value in industry production, biomedical area, aeronautics, spaceflight, biology medicine and national defense and so on. The intelligent fault diagnosis based on soft computing is an emerging theory and technology that combines the soft computation with reliability theory, information theory, cybernetics, artificial intelligence and system theory. It utilizes modern testing instruments and computers as its technology means to implement the system fault diagnosis. The fault diagnosis of a nonlinear system brings a great challenge to control engineers. Focused on the fault diagnosis of non-linear systems, research and exploration described in this dissertation were conducted on mechanism and application of fault diagnosis based on soft computing.The main contributions of this research work are concluded as follows:(1) An adaptive neuro-fuzzy inference systems (ANFIS) was applied to the modeling of a nonlinear system. Reseach was conducted to study the mechanism of a model-based fault diagnosis by combining an ANFIS's model with a LVQ network. The key of the diagnosis was to extract the characteristic parameters from ANFIS models as the input vector of LVQ network. The output parameters including the main fault information were considered as the major part of the input vector of a LVQ network. The fault diagnosis was realized by then weighting the output parameters and the parameters of the input membership functions and the value domain to compose the input vector of a LVQ network...(2) Considering the need of on-line fault diagnosis, an on-line intelligent fault diagnosis mechanism based on the ANFIS multi-models was proposed. The residual signals were generated by simultaneously on-line running multi-ANFIS models to represent different working situation or different faults. On-line fault diagnosis mechanism comprehensively took into account the dynamic characteristics, robustness and sensitivity and, utilized the accumulations of absolute values of the residuals to detect the occurrence of the faults. Specially, a unique Fault Index (FI) was also introduced to indicate the occurrence and the type of the faults.(3) A dyadic spline wavelet with a compact support and a generalized linear phase was selected to carry out a binary wavelet transformation of ECG. Then corresponding strategies were utilized to extract ST segments out of ECG signals(4) Focused on the complexity in learning a multi-input-multi-output (MIMO) ANFIS, in order to enhance the accuracy of the diagnosis system, research was conducted on the diagnosis mechanism that was composed of training system and diagnosis system The training system established n ANFIS sub-models corresponding to n kinds of faults respectively. The diagnosis system inputed the test data into the n ANFIS sub-models and calculated the residuals e_i =|1-(?)_i| between the output(?)_i (i=1,...,n ,i denotes the number of subsystem) from each sub-inference system and the standard output 1 to illustrate the occurrence of the kth kind of faults.(5) Rresearch on the mechanism of fault diagnosis based on fuzzy adaptive resonance mapping theory (ARTMAP) neural network was also conducted. The training and classification algorithms of the network were derived. That network was able to classify the unknown faults.(6) Focusing on the shortcomings of BP network, research on the fault diagnosis mechanism of BP neural network intelligence diagnosis method based on a genetic algorithm was conducted. And the diagnostic method and procedures about optimizing the weight values and threshold values of genetic algorithm of a three layer BP network were provided.(7) In order to apply a fuzzy neural network to the dynamic fault diagnosis, a Takagi-Sugeno recurrent fuzzy neural networks(TSRFNN) was constructed. The temporal information were embedded in the TSRFNN by adding feedback connections between the states layer and inputs layer of the fuzzy neural networks(FNN). The identification of the network was composed of two steps: structure identification and parameter identification. In the process of structure identification, a clustering method was proposed to determine the number of fuzzy rules and the initial fuzzy model from the given input-output data. In the process of parameter identification, the Dynamic Backpropagation (DBP)was used to determine the parameters of the TSRFNN. The network has been proven to satisfy a universal approximation property for any real continuous function defined on a closed and bounded set by theory reasoning and simulation examples.