Pre-Processing-Free Gear Fault Diagnosis Using Small Datasets with Deep Convolutional Neural Network-Based Transfer Learning

摘要

Early fault diagnosis in complex mechanical systems such as gearbox hasalways been a great challenge, even with the recent development in deep neuralnetworks. The performance of a classic fault diagnosis system predominantlydepends on the features extracted and the classifier subsequently applied.Although a large number of attempts have been made regarding feature extractiontechniques, the methods require great human involvements are heavily depend ondomain expertise and may thus be non-representative and biased from applicationto application. On the other hand, while the deep neural networks basedapproaches feature adaptive feature extractions and inherent classifications,they usually require a substantial set of training data and thus hinder theirusage for engineering applications with limited training data such as gearboxfault diagnosis. This paper develops a deep convolutional neural network-basedtransfer learning approach that not only entertains pre-processing freeadaptive feature extractions, but also requires only a small set of trainingdata. The proposed approach performs gear fault diagnosis using pre-processingfree raw accelerometer data and experiments with various sizes of training datawere conducted. The superiority of the proposed approach is revealed bycomparing the performance with other methods such as locally trainedconvolution neural network and angle-frequency analysis based support vectormachine. The achieved accuracy indicates that the proposed approach is not onlyviable and robust, but also has the potential to be readily applicable to otherfault diagnosis practices.