Explainable Fault Classification and Severity Diagnosis in Rotating Machinery Using Kolmogorov-Arnold Networks.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Rolling element bearings are vital for rotating machinery, but bearing faults cause significant industrial failures and downtime.
• Effective machinery monitoring is crucial for preventing failures and ensuring operational efficiency.

Purpose of the Study:

• To develop an automated and interpretable deep learning methodology for bearing fault diagnosis.
• To utilize Kolmogorov-Arnold Networks for feature selection and hyper-parameter optimization in a unified approach.

Main Methods:

• The study employed Kolmogorov-Arnold Networks, a novel deep learning technique, as an alternative to Multilayer Perceptrons.
• A unified framework was developed for automatic feature selection and hyper-parameter optimization.
• Shallow network architectures and reduced feature sets were prioritized for model efficiency and interpretability.

Main Results:

• The framework achieved perfect F1-Scores for bearing fault detection and high performance in fault and severity classification on benchmark datasets.
• Models demonstrated adaptability in identifying diverse fault types, including imbalance and misalignment, within the same dataset.
• Symbolic representations and feature attribution provided model interpretability and insights into optimal feature selection.

Conclusions:

• The proposed Kolmogorov-Arnold Network-based framework offers a practical and efficient solution for real-time machinery monitoring and bearing fault diagnosis.
• The methodology's interpretability and adaptability make it suitable for both practical applications and scientific research.
• This approach facilitates the development of lightweight, explainable AI models for industrial diagnostics.