Aero-engine Inter-shaft Bearing Fault Diagnosis via Hybrid Kolmogorov-Arnold Classifier

Fault diagnosis for inter-shaft bearings in aero-engines using vibration signals is challenging due to harsh operating conditions, which make vibration signals highly complex and prone to misdiagnosis. Deep learning has emerged as a promising solution by integrating feature extractors and classifiers to learn meaningful representations from vibration data automatically. However, most existing methods focus on improving feature extractors while relying on multilayer perceptrons (MLPs) as classifiers. Despite their effectiveness, MLP classifiers suffer from high computational complexity, large parameter sizes, and limited interpretability. Inspired by the recent attention given to Kolmogorov-Arnold Networks (KAN), which serve as a powerful alternative to MLPs due to their superior interpretability and ability to approximate complex nonlinear functions, we pioneer the application of KAN in fault diagnosis and propose a novel Hybrid Kolmogorov-Arnold Classifier (HKAC) that replaces traditional MLP classifiers. Extensive experiments demonstrate that the proposed method achieves higher fault classification accuracy with only 35% of the parameters of conventional MLP classifiers. This study highlights the potential of combining deep learning with KAN to overcome the limitations of traditional methods, providing a robust and efficient classifier for aero-engine inter-shaft bearing fault diagnosis and other complex mechanical systems.