Robust and Accurate Bearing Fault Diagnosis Using an Enhanced Multi-Kernel Learning Model

Abstract

Accurate fault diagnosis of motor bearings is crucial to maintaining the reliability and safety of industrial machinery. Traditional approaches to fault diagnosis often suffer from poor generalization, sensitivity to noise, and limited adaptability to complex fault patterns. In this study, we propose a robust and enhanced Multi-Kernel Learning (MKL) model that leverages the complementary strengths of multiple kernels to effectively capture non-linear fault characteristics in vibration signals. By integrating multiple kernel functions into a unified learning framework, our model enhances the discrimination power of the classifier and adapts efficiently to diverse operating conditions. The proposed method is evaluated on standard bearing fault datasets, including various levels of fault severity, speeds, and load conditions. Results demonstrate significant improvements in classification accuracy, robustness to noise, and generalization to unseen conditions, outperforming traditional single-kernel models and several state-of-the-art machine learning techniques.