Multi-direction Reconstruction for Fault Diagnosis of Train Bearings

Abstract

Bearing condition is important for the operation safety of the trains. Traditional rule-based method can only detect the fault after the bearing is seriously damaged when the bearing temperature is far higher than the normal situation. In this paper, data driven bearing fault diagnosis of train is discussed. Taking the operation dynamics into account, a dynamic inner principal component analysis (DiPCA) based bearing fault monitoring method is proposed. After that, in order to locate the fault, a DiPCA based multi-directional reconstruction method is proposed to identify the possible faulty variables. Results from case studies using the data collected from a real train operation demonstrate the effectiveness of the proposed methods.