Search

With the development of artificial intelligence algorithm, BP neural network algorithm is widely used in many fields, such as fault diagnosis, intelligent control and dynamic signal processing, because it has many advantages for example self-learning, self-organization and nonlinear mapping. Compared with BP neural network, the hidden Markov model is suitable for dynamic time series modeling and has strong temporal classification ability. However, the hidden Markov model has problems of initial model optimization and algorithm underflow when applied to pattern classification. In this paper, the discrete hidden Markov model (DHMM) and BP neural network algorithm are combined to apply to the fault diagnosis of gearbox. Firstly, the probabilities of failures were obtained by preprocessing of the fault samples. Then the probabilities are added to the time-frequency characteristics as new features. The BP neural network algorithm were used to classify the samples whose features had been extended. The experimental results showed that the proposed method was more conducive to fault diagnosis of gearbox.