Explainable fault and severity classification for rolling element bearings using Kolmogorov-Arnold networks

TL;DR

The paper tackles the challenge of reliable bearing fault diagnosis with explainable, edge-friendly models. It introduces Kolmogorov-Arnold Networks (KANs) that enable automatic feature selection via feature-attribution scores and replaceable symbolic activations, forming a unified pipeline for fault detection, fault classification, and severity estimation. A three-stage workflow—feature-library construction with Pareto-front feature selection, hyperparameter tuning of the KAN grid and adaptivity, and symbolic regression to produce interpretable models—yields lightweight models whose decisions can be expressed analytically. Evaluations on the CWRU and MaFaulDa datasets show near-perfect fault-detection performance and strong fault/severity classification results, with symbolic variants offering valuable interpretability and acceptable performance trade-offs. The approach demonstrates practical potential for real-time condition monitoring and provides a generalizable framework for explainable fault diagnosis beyond bearing faults.

Abstract

Rolling element bearings are critical components of rotating machinery, with their performance directly influencing the efficiency and reliability of industrial systems. At the same time, bearing faults are a leading cause of machinery failures, often resulting in costly downtime, reduced productivity, and, in extreme cases, catastrophic damage. This study presents a methodology that utilizes Kolmogorov-Arnold Networks to address these challenges through automatic feature selection, hyperparameter tuning and interpretable fault analysis within a unified framework. By training shallow network architectures and minimizing the number of selected features, the framework produces lightweight models that deliver explainable results through feature attribution and symbolic representations of their activation functions. Validated on two widely recognized datasets for bearing fault diagnosis, the framework achieved perfect F1-Scores for fault detection and high performance in fault and severity classification tasks, including 100% F1-Scores in most cases. Notably, it demonstrated adaptability by handling diverse fault types, such as imbalance and misalignment, within the same dataset. The symbolic representations enhanced model interpretability, while feature attribution offered insights into the optimal feature types or signals for each studied task. These results highlight the framework's potential for practical applications, such as real-time machinery monitoring, and for scientific research requiring efficient and explainable models.

Figures (12)

• Figure 1: Example of an iteration of node score attribution in a layer of 3 input and 2 output nodes.
• Figure 2: Schematic representation of the proposed framework, encompassing the feature selection, model selection and model evaluation processes.
• Figure 3: Feature attribution scores for all features within the feature library of CWRU. For the chosen threshold, $\tau = 7.16\cdot 10^{-2}$, only a single feature is selected.
• Figure 4: Confusion matrices for fault detection in the CWRU dataset by the regular (left) and symbolic (right) version of the trained KAN model.
• Figure 5: The symbolic KAN's output edges uniquely define class-exclusive regions separated by a decision boundary. The CWRU dataset's points are correctly classified in their respective categories.