A Two-dimensional Convolutional Neural Network Optimization Method for Bearing Fault Diagnosis

Intelligent bearing fault diagnosis is a hot research field of mechanical big data condition monitoring. The traditional data-driven fault diagnosis method is extremely time-consuming and requires high expert experience for signal extraction-based feature extraction. In order to eliminate the pre-defined effects of parameters and improve the recognition rate while improving the feature extraction, this paper proposed a two-dimensional convolutional neural network optimization method for bearing fault diagnosis based on the research of one-dimensional convolution neural network fault diagnosis method. This method introduced a new data preprocessing method, which converted the original time domain signal data into a two-dimensional gray image to extract the transformed image features, eliminated the influence of manual features, and collects the experimental errors before verifying the classification. The data set adds noise reduction processing and optimizes the parameter adaptive learning rate for the convolutional neural network gradient descent algorithm. The simulation and experimental results show that the proposed two-dimensional optimized convolutional neural network fault diagnosis method is based on the signal-picture conversion format under 64×64. The AMSGrad algorithm can improve the accuracy of the fault prediction model to 98%, and the training speed is faster and higher. Classification accuracy and anti-noise performance can achieve a recognition accuracy of less than 5% loss in the actual speed range. © 2019 Chin. Soc. for Elec. Eng.