Rolling Bearing Fault Diagnosis Using Enhanced Convolutional Neural Network with Compressed Sensing

It is challenging to learn features from vibration signal with strong noises and achieve the bearing fault diagnosis. Compressed sensing (CS), as a sparse representative method, breaks though the limitations of Nyquist's sampling law, and it can reconstruct bearing signals effectively in the condition of disturbances. Therefore, this paper proposes a method of bearing fault diagnosis based on compressed sensing and convolutional neural network (CNN). The sparsity adaptive matching pursuit detection algorithm (SAMP) is used to solve the sparsity selection problem of compressed sensing. In the convolutional neural network, the dropout strategy is introduced to avoid overfitting. By comparing with several traditional models, the superiority of this method in bearing fault diagnosis is verified.