Intelligent Fault Diagnosis Based On The Audio Signal Rolling Study

This article takes rolling bearing failure diagnosis with audio signals as object of research. In this paper, we analysis the domestic and foreign research of present situation in the foundation, proposes the feature parameter those are the DWT-LPCC and the DWT-MFCC based on the wavelet packet analysis and cepstrum analysis, and gives a information fusion failure diagnosis method which unifies based on the neural network and the D-S evidence theory, and carries on the simulation experiment of bearing failure diagnosis with audio signal.Firstly, we have analyzed the significance and present situation of the rolling bearing failure diagnosis research, reviewed the traditional bearing failure diagnosis method as well as its good and bad points, and explained that urgently we needs the new diagnosis method research and the application day by day in the complex mechanical system.Then, we uses the wavelet packet theoretical to analyze various kinds of audio signal of rolling bearing in different frequency band signal characteristic, uses the signal energy coefficient which the wavelet packet decomposes to improve the traditional MFCC and the traditional LPCC feature vector, and has proven new characteristic parameter that are the DWT-LPCC and the DWT-MFCC validity through the simulation experiment.Next, we obtain from the neuron structural model, detailed analysis neural network structure and function mapping, BP neural network characteristic. We have carried on the sub-neural network failure diagnosis simulation experiment using the MATLAB software. And we obtained from the D-S evidence theory's basic concept and the formation rule, to analyze the evidence theory to apply in the information fusion validity.Finally, this article designs a information fusion failure diagnosis method based on the neural network and the D-S evidence theory, based on the characteristic of the information of the diagnosis system, to divide two breakdown characteristic indication territory, designs two neural networks to carry on the preliminary diagnosis separately to the system, then diagnoses the result to transform carries on the policy-making stratification plane again as the basic probability evaluation the information fusion by the D-S evidence theory, so that the fusion result accuracy has achieved 95% and proven that this diagnosed method was feasibility and validity.