Domain Adaptation via CORAL for Robust Motor Bearing Fault Diagnosis

Abstract

The accurate diagnosis of motor bearing faults is a critical task in industrial condition monitoring systems. However, the performance of diagnostic models often degrades significantly when there is a discrepancy between the distribution of training data (source domain) and real-world operational data (target domain). Domain adaptation techniques have emerged as promising solutions to address such distributional shifts. This paper presents a comprehensive approach to motor bearing fault diagnosis using Correlation Alignment (CORAL), a domain adaptation method that aligns the second-order statistics of source and target domain features. Our approach integrates CORAL with a deep learning backbone to extract domain-invariant features and enhance fault classification robustness. We conduct extensive experiments using the Case Western Reserve University (CWRU) bearing dataset, simulating domain shifts through varying load and motor speed conditions. The results demonstrate that CORAL significantly improves cross-domain diagnostic accuracy compared to models trained without adaptation, underscoring its effectiveness in practical scenarios where domain mismatch is prevalent.