Real-time Fault Diagnosis of Rotating Machinery Based on 1-D Convolutional Neural Network

Abstract

Reliability is one of the most significant aspects in evaluating rail vehicle system and rotating machinery accounts for most of the failures, which means the real-time condition monitoring of the rotating components is essential. As the key component of condition monitoring, diagnosis method has the problems of manual-needed fault feature extraction, low ability of complex fault recognition and high time-consuming, which unable to meet the real-time diagnosis demand of rotating machinery. This paper proposed a novel 1-D convolutional neural network (CNN) diagnosis model using modified 1-D convolutional kernel and pooling based on the classical CNN LeNet-5. Different from the typical intelligent diagnosis models taking feature extraction and classification as two distinct blocks, the proposed model fuses the two phases into a single learning body: adaptive feature learning by alternating convolutional and pooling layers while classification by full connected layer. Bearing and gearbox health monitoring results show that the proposed model can achieve fast and accurately fault diagnosis, which can satisfy the real-time diagnosis requirement of rail vehicles.