Discrete Hidden Markov Model And Its Gearbox Dynamic Fault Diagnosis

Gearboxes are common transmission systems in mechanical equipment.The abnormality of gearboxes usually lead to the equipment not being able to run normally,economic losses and safety accidents are also caused.It is significant to study the fault diagnosis technologies of gearboxes.Hidden Markov models have powerful timing modeling capabilities and can theoretically handle timing pattern recognition problems in any length.In this paper,the discrete Hidden Markov model is used for the fault diagnosis of the gearbox,and the research on the dimension reduction and model optimization is carried out.The main contents are as follows:(1)Aiming at the feature dimensionality reduction of nonlinear data,a method based on the cascade of single core kernel functions is proposed.The kernel parameters are optimized according to the ratio of in-class distance and between-class distance,and the local and the global information of the original feature set are extracted in sequence.Information,experiments show that this method takes into account the dimension reduction of features and the discriminability of various types of data.(2)Focus on the optimization of the initial observation matrix of discrete Hidden Markov Models,taking into account that the strategy of optimizing for a single sample or a single type of sample may cause the model to degenerate,select typical samples and samples which are easy to be misjudged to construct new features set.Use the particle swarm optimization algorithm to optimize the initial observation matrix.This method can obtain a better initial observation matrix.(3)For the problem of insufficient self-adaptation of discrete Hidden Markov Model,this paper uses discrete Hidden Markov Model to construct new features set containing time sequence information by information fusion,and uses BP neural network algorithm to identify gear working conditions.This method combines the strong timing modeling ability of hidden Markov model and the self-adaptation of BP neural network,accuracy of diagnosis is significantly improved.