Gearbox Fault Diagnosis Under Variable Speed Condition Using Frequency Spectral Analysis with 1D Residual Neural Network

Abstract

Stringent classification of gearbox fault conditions is indispensable for industrial safety. Determining the best set of features by analyzing the statistical parameters of the signals is one of themost climacteric tasks in data-driven fault diagnosis. For variable speed conditions, variant fault types of gearboxes have dynamic characteristics and these statistical features fail to unveil them. To address this issue, deep learning algorithms are used to produce a better performance of the feature selection process. In this paper, a combination of frequency spectral analysis of the acoustic emission signals and a 1-dimensional residual neural network (1D-RNN) is proposed. Our proposed method through the processed 1D-RNN shows vigorous classification performance, resulting in up to 95.6% classification accuracy in all the considered scenarios.