Research On Fault Diagnosis Of Rotating Machinery In Nuclear-powered System Base On HMM-SVM Hybrid Model

Nuclear-powered system is a complicated system, which may have seriousconsequences when the fault occurs. Condition monitoring and fault diagnosis plays avital role in guaranteeing the safety and the reliability of nuclear-powered plant. Inthis thesis, the Hidden Markov Models (HMMs) and Support Vector Machine (SVM)which are introduced to condition monitoring and fault diagnosis for the keyequipments of nuclear power system. This study has enriched the ways and means ofdiagnosis, and has achieved good results in simulation experiments.In the thesis, the feature extraction techniques and specific feature selectionalgorithms for fault signals in amplitude domain, frequency domain and time domainwer studied; as well as the basic theory and algorithm of the Hidden Markov Models(HMMs) and Support Vector Machine (SVM); an improved algorithm of HMM calledContinuous Gaussian mixture Hidden Markov Model (CGHMM) was applied toimprove the hybrid model and a HMM-SVM hybrid model was established finally.The main contents of this thesis are as follows:1. This thesis briefly introduced the meaning of studying the faults diagnosistechnologies in nuclear-powered plant and summarized the developing and the currentsituations of it. A feasibility analysis of fault diagnosis system for rotation machineryon nuclear-powered plant based on HMM/SVM model was also carried out.2. Feature extraction method for the vibration signal of rotating machinery andequipment in nuclear power from amplitude domain and frequency domain to timedomain were studied, specially, on the method that based on wavelet analysis andwavelet packet energy moment.3. It summarizes the common faults and features of rotating machinery andequipment in nuclear power. After that, the feature selection problems and the specific algorithm were studied.4. The basic principles and algorithms of the Hidden Markov Models (HMMs)and Support Vector Machine (SVM) have been described and researched, and a newHMM/SVM hybrid model was proposed.5. It has designed the simulation program, and the fault diagnosis method basedon HMM/SVM hybrid model has been tested and verified.The study of this thesis, which is based on the"Research on ConditionMonitoring and Fault Diagnosis Technologies based on hidden markov model andsupport vector machine for Equipment of Nuclear Power System"(National HighTechnology Research and Development Program"863", No. 2008AA04Z407), is anuseful expansion and supplement of nuclear power equipment fault diagnosismethods.