A study of deep neural networks transfer learning for fault diagnosis applications

Abstract

Intelligent fault diagnosis utilizing deep learning algorithms has been widely investigated recently. Although previous results demonstrated excellent performance, features learned by Deep Convolutional Neural Networks (DCNN) are part of a large black box. Consequently, lack of understanding of underlying physical meanings embedded within the features can lead to poor performance when applied to different but related datasets i.e. transfer learning applications. This study will investigate the transfer learning performance of a Deep Convolutional Neural Network (DCNN) considering 4 different operating conditions. Utilizing the Lou & Loparo (2004) Case Western Reserve University (CWRU) bearing dataset, the DCNN will be trained to classify 12 classes. Each class represents a unique different fault scenario with varying severity i.e. inner race fault of 0.007”, 0.014” diameter. Initially, zero load data will be utilized for model training and the model will be tuned until testing accuracy above 99% is obtained. The model performance will be evaluated by feeding vibration data collected when the load is varied to 1, 2 and 3 HP. Initial results indicated that the classification accuracy will degrade substantially. To improve the network generalization capabilities, this paper proposes the addition of white Gaussian noise to the raw vibration data. Results indicate that a very high level of additive noise can improve the transfer learning accuracy. The discussion will then focus on the influence of changing loads on fault characteristics, network classification mechanism, and activation strength in addition to the visualization of convolution kernels in time and frequency domains.