A novel two-stream convolution neural network with temporal information for bearing fault diagnosis

Abstract

Abstract Deep learning has been widely applied to intelligent fault diagnosis of rotating machines. In particular, convolutional neural network (CNN) improves the accuracy of diagnosis due to automatic feature extraction ability. However, CNN cannot make full use of the temporal information of the input signals. In order to overcome the above weaknesses, a new depth diagnositic framework called two stream convolution neural network with temporal information(TITSCNN) is proposed for bearing fault diagnosis. First, we obtain tempory signal through the difference of data points in the input signal. Next, tempory signal and original signal are used as input of two CNN branches respectively. Then, an attention mechanism is used to aggregate the features and assign weights to the features extrated by CNN adaptively. Finally, the learned high-level representations are fed into full connect layer to perform fault diagnosis. Experiments are conducted and the results showed that TITSCNN has higher accuracy compared to state-of-the-art methods. Further, experimental results demonstrated that TITSCNN has excellent noise immunity and transformation performance.