A bearing fault diagnosis method using a dynamic model with deep domain adaptation

Abstract

Abstract The failure of the rolling bearing, one of the core components widely used in industry, will lead to the abnormal operation of the machine and even cause safety accidents. To avoid this situation, fault diagnosis become more and more important. The traditional fault diagnosis model uses the classical deep learning model as the backbone depends on a bulk of labeled fault data, which is difficult to acquire in real industrial scenarios. To address the above problems, a fault diagnosis method using a failure mechanism with deep domain adaptation is proposed. A failure mechanism-based bearing dynamic model is constructed to simulate different types of labeled fault data and deep domain adaptation is used to extract common fault features of simulation data and real data. The fault dataset including different types of failure modes running on bearing test rigs from Paderborn University is used to verify the performance of the proposed method. It follows that the introduction of deep domain adaption can improve the accuracy of diagnosis even though the labeled fault data is not enough.