Abstract:Based on the gearbox experimental device of multiple defects, a new approach to fault diagnosis of gearbox was proposed in this paper. First, in the improved distance measure method, two restraint factors were presented to adjust the evaluation function by which the superior features can be selected from the original feature set. Second, based on the HMM (Hidden Markov Model) theory, the superior features were used to train the HMM until it was convergent, and the HMM under each gearbox state can be obtained. Finally, calculating the logarithmic likelihood probability of each HMM by the use of testing data, and the fault state can be identified based on the probability value. The testing results have shown that the combination of HMM and improved distance measure can effectively identify different gearbox failure patterns with high recognition accuracy, and can be applied to the real gearbox system for fault diagnosis.