Research On Fault Diagnosis Of Rolling Bearing Based On Texture Feature Extraction

In modern industry,it is great significance for the health status monitoring and fault diagnosis of rotating machinery.Rolling bearing is one of the most common and critical components in rotating machinery.It plays an essential role on rotating machinery.If it breaks down,it will have a great impact on rotating machinery and inevitably cause economic loss.Therefore,the fault diagnosis of rolling bearing is of great engineering significance and application value.The paper takes rolling bearing as the research object,combines with discrete wavelet transform,power spectrum subtraction,texture analysis method and convolutional neural network,and carries out research work in the following aspects in view of the practical problems of difficulty in extracting bearing state information existing in the fault diagnosis of rolling bearing,such as:(1)A noise reduction algorithm of rolling bearing vibration signal based on wavelet transform and power spectrum subtraction is proposed.The method first performs discrete wavelet transform on the collected vibration signals to obtain different sub-bands,and calculates the noise variance of the detail sub-band for estimation of the noise spectrum.Denoising of the signal is done by power spectrum subtraction and the estimated noise spectrum.The effectiveness of the proposed method is verified by simulated and real signals.(2)A hybrid binary model is proposed to extract signal texture features of rolling bearing fault,and combined with wavelet analysis to obtain signal texture features at different scales.On the basis of above,the broad learning system is used as a classifier to achieve the rapid diagnosis.Two rolling bearing datasets are used to verify the effectiveness of the proposed multi-scale mixed binary pattern as a texture feature extraction method.(3)A fault diagnosis method based on multi-scale texture statistical convolutional neural network was proposed by combining statistical analysis method with deep learning.The method first performs multi-scale analysis of the signal through wavelet transform and takes it as the input of CNN,then obtains the local texture of the signal through each convolution layer of CNN,and finally uses the proposed texture statistics module to perform texture statistics on the convolution map and use it as the final feature for classification.The rolling bearing fault diagnosis scheme under multiple working conditions are used to verify the effectiveness and superiority of the proposed multi-scale texture statistical convolutional neural network.