Research And Applications Of Fault Diagnosis Based On Speech Signal

In recent years, the fault diagnosis technology based on speech signal has been used in many fields, such as the fault diagnosis of fan, engine, partial discharge and so on. The application of the fault diagnosis is various and what’s more, it also gradually becomes a research hotspot in the area of fault diagnosis and attracts the attention of the scholars. But now, there are still some problems in fault diagnosis, such as:the signal noise ratio (SNR) of the collected signal is low, the classifier cannot achieve class incremental learning, most of the application of fault diagnosis system is too limited, etc. To solve the above problems, this paper lays emphasis on the research of the filtering method of fault signal, feature extraction of the signal and the algorithm of incremental learning. The main work is as follows:Firstly, this paper introduces the filtering methods of fault diagnosis. And as the collected signal’s SNR is low, therefore, some filtering methods are introduced, such as:SVD filtering, wavelet filtering and EMD filtering. Particularly, to solve the problem in traditional EMD filtering method, the improved EMD filtering algorithm is proposed. And the improved algorithm has better filtering effect.Secondly, to extract the feature of the speech signal using wavelet packet decomposition energy, EMD energy and MFCC method. Then train the SVM classifier using the feature vector. The experiment results show that the diagnosis system has high diagnostic accuracy. In addition, the general diagnosis system doesn’t have class incremental learning function. In this paper, we do some research work in incremental learning.Finally, the fault diagnosis platform is developed based on C, Matlab, and MS SQL. By using C and Matlab mixed compile technology, the function of Matlab can be compiled in C environment. It can greatly save the time to develop the system. The data can be managed by MS SQL, and what’s more, the user can query, delete, and add the data easily.