Intelligent fault diagnosis of rolling bearing using one-dimensional multi-scale deep convolutional neural network based health state classification

Abstract

Fault diagnosis of rolling element bearings based on vibration signal is the most popular way to avoid underlying damage for any unexpected fault. In recent years, intelligent fault diagnosis algorithms using machine learning technique have achieved much success, and many deep learning techniques have also found their way into fault diagnosis of rotating machines. Considering that convolution is the most important method to analyze signals in digital signal processing, a novel deep convolutional neural networks is developed to operate directly on the raw vibration signal. The proposed MS-DCNN model could broaden and deepen the neural networks to learn better and more robust feature representations owing to multi-scale convolution layer, meanwhile, reduce the network parameters and the training time. Fault classification experiments of rolling element bearings have been undertaken to indicate the effectiveness of the MS-DCNN model. Compared with 1d-DCNN and 2d-DCNN, MS-DCNN can not only achieve higher accuracy rate in the testing set, but also run more smoothly in the training process.