Rolling Element Bearing Fault Diagnosis Using Compressed Sensing and Convolutional Neural Network

Rolling bearing is one of the most commonly used components in rotating machinery. It's so easy to be damaged that it can cause mechanical fault. Therefore, it is of great significance for its condition monitoring and fault diagnosis. However, the traditional diagnosis methods still suffer from two problems, which are (1) the information density of the monitoring data is low because of huge monitoring data amount, and (2) the requirements of domain expertise and prior knowledge for sensitive feature extraction. Aiming at above problems, a new diagnosis method based on compressed sensing (CS) and convolution neural network (CNN) is proposed in this paper. The method consists of three key steps. First, the original monitoring signals are converted into compressed sensing domain for reducing data amount and improving its information density by using compressed sensing method. Second, the compressed signal is used as the input of the convolution neural network to extract sensitive features adaptively, and to realize the fault intelligence diagnosis. Finally, several groups of experiments are carried out to validate the feasibility of the proposed method in this paper, and the diagnostic accuracy achieves 93.7%, which is far higher than the traditional methods.