In Shape And Genetic Programming In Mechanical Fault Diagnosis

With the structural complication of machinery equipment, machinery equipment performs nonlinear behaviors, traditional fault diagnosis methods couldn't meet the requirement of engineering practice. It is an urgent and tough task to establish more intelligent and efficient state-monitoring and fault diagnosis methods. Fractal technology and genetic programming provide powerful tools to solve complicated nonlinear problems in machinery fault diagnosis, which are the focus of current research, also main content of this dissertation.Fractal dimension was studied firstly in the thesis. The comparison between correlation fractal dimension and variance fractal dimension revealed that the former was sensitive to the change of the signal in amplitude, while the later was sensitive to the variation in frequency. Following this discussion, a novel method for extracting the transient fractal feature of the signal was proposed in order to meet the further need of analyzing non-stationary signals. The effectiveness of the proposed method was verified by the application to diagnosing practices. Experimental results showed that the transient variation of the signal in both amplitude and frequency may be captured by the proposed technique. In comparison with the existing signal exact analyzing methods (e.g. exact wavelet analysis), the proposed method involved less computation and was therefore more feasible to be used in machinery fault diagnosing practice.Genetic Programming (abbreviated as GP) method was studied subsequently. Available machinery fault diagnostic methods showed unsatisfactory performances on both on-line and intelligent analysis because their operations involved intensive calculations and labors. Aiming at improving this situation, an intelligent fault diagnosis method based on Genetic Programming was proposed in this thesis. Attributed to the simple calculation of the mathematical model constructed, different kinds of machine faults may be diagnosed correctly and quickly. Moreover, human input was significantly reduced in the process of fault diagnosis. The effectiveness of the proposed strategy was validated by an illustrative example, in which three kinds of valve states inherent in a six-cylinder/four-stroke cycle diesel engine were identified. Different from existing GPs, the proposed method was more intelligent and efficient. Therefore, it was more feasible to be used in machinery fault diagnosing practice.