Gear Fault Diagnosis Method Based On Hidden Semi-markov Model

Intelligent fault diagnosis technology is the development trend in the field of fault diagnosis. As one of the intelligent diagnosis technology, the Hidden Markov Model(HMM) has a strong pattern classification and modeling capabilities on time series of dynamic process. However, the model doesn’t consider the situations between the hidden states, which don’t agree with the actual conditions. Hidden Semi-Markov Model(HSMM), which is developed by introducing the state probability of dwell time, is closer to the actual conditions and has a good prospect of application in the field of fault diagnosis. Hidden Semi-Markov Model can be used for state recognition of the steady-state. Currently, the fault diagnosis of the machine equipment based on HMM has obtained good results, but there are few researches based on HSMM in the field of fault diagnosis. Thus in accordance with the existed achievements the further researches should be carried out, betting the Markov Model that closer to the actual conditions.Based on the research actuality above, the study is conducted arounding the Method on Fault Diagnosis Based on Hidden Semi-Markov Model and Application in Gear, which aims at narrowing down the selection of initial values of the observation probability matrix firstly, selecting the appropriate memory function to relax the Markov assumptions about the stability ineffectiveness and finding the method about multiple-faults and degradation degree simultaneous diagnosis. Therefore the main content of this dissertation can be expressed as below:1) Aiming at the selection about the observation probability matrix B, the calculating method for determining non-membership functions of Intuitionistic Fuzzy Sets(IFS) based on quartering is put forward, by which the target about the selection for B is obtained, laying a solid foundation for the further researches of HMM or HSMM.2) Aiming at the problem that the Markov stability ineffectiveness assumption is not consistent with the actual condition, the fault diagnosis model based on adaptive filtering and HMM is proposed, which considers the advantages of adaptive filtering in the field of processing the historic information fully. Subsequently, the gear experiment is taken as the example to prove the effectiveness and the feasibility of the model.3) Aiming at the irrationality that the state duration obeys the exponential distribution and the problem that HMM can not deal with the multiple-faults, based on Artificial Neural Network(ANN) and Hidden Semi-Markov Model(HSMM) a hierarchical diagnosis network is established with the respect to multiple-faults and the degradation degree simulation diagnosis, which divide a large fault diagnosis into several smaller patterns, so that the model not only can train the sub-network conveniently, but also can improve the classification performance of the whole network. Moreover, the model is capable of multiple-faults and their degradation degree simultaneous diagnosis. The experimental results show that the model can not only recognize multiple-faults diagnosis, but also achieve the degradation degree diagnosis for the corresponding faults, and the result is highly accurate.At the end of the dissertation, the summarization for the research work and the expectation about the future research direction are presented.