Intelligent Fault Diagnosis Method for Rolling Bearings Based on SSA-CNN-Transformer

Abstract

Deep learning has become a key solution in intelligent fault diagnosis, as its ability to learn features directly from raw data addresses the challenges of modeling complex signals in rolling bearings. Traditional Convolutional Neural Networks (CNNs) are constrained by fixed receptive fields and static kernels, which limit their adaptability to dynamic, multi-scale features in vibration signals. Moreover, existing models often lack an adaptive mechanism for evaluating feature importance, which reduces diagnostic robustness in non-stationary and variable operating conditions. This paper introduces the SSA-CNN-Transformer model, which integrates the Sparrow Search Algorithm (SSA) with a self-attention mechanism to address these challenges in intelligent bearing fault diagnosis. The SSA globally optimizes key hyperparameters, improving the efficiency and performance of the model architecture. The CNN module extracts local time-frequency features from vibration signals and performs multi-scale fusion, while the Transformer module captures long-range dependencies, leading to a more accurate and comprehensive representation of fault patterns for precise classification. Empirical evaluations on three publicly available datasets—CWRU, XJTU, and DIRG—demonstrate that the proposed model outperforms current state-of-the-art methods in multiple performance metrics, exhibiting superior diagnostic accuracy and generalization. This work offers valuable insights and a solid foundation for developing intelligent health monitoring systems for real-world industrial applications.