Application of Transfer Learning to Diagnose Ball Bearing Failure of Two Axial System Propulsion Systems with Different Operating Environments

초록

This study aims to propose a fault diagnosis method that utilizes a model learned from existing equipment to diagnose faults in equipment with different operating conditions. When operating conditions or environments change, a Domain Shif occurs. Therefore, the decision boundaries previously learned may not apply correctly. To address this, a fault diagnosis method using domain adaptation was employed. Additionally, frequency-based diagnostic techniques were utilized to minimize differences arising from different operating conditions or equipment specifications. The proposed method was applied to fault diagnosis of ball bearings in rotor systems. After extracting fault frequency characteristics from rotor ball bearings with fault signals, deep learning was applied to verify the fault diagnosis performance. The diagnostic algorithm used the Conv1D model as the basic structure. The CWRU (Case Western Reserve University) Bearing Dataset was used as the target for fault diagnosis. As a result, it was confirmed that the fault diagnosis of subjects with different environments is applicablethrough the domain adaptation process.