Adaptive Dual‐Attention Learning Frame for Bearing Fault Diagnosis

ABSTRACT Real‐time fault diagnosis of high‐speed train bearings is critical for operational safety but remains challenging due to the complex, non‐stationary nature of vibration signals. While Transformer‐based models excel at capturing long‐range dependencies, they often struggle to extract localized fault features and lack the adaptability required for continuous learning in dynamic industrial environments. To address these limitations, this paper proposes a novel Adaptive Learning Inferential Former (ALIF) framework. The core innovations of ALIF include: (1) a hybrid dual‐attention mechanism that simultaneously captures global trends and local details; (2) a dynamic routing strategy to adaptively switch focus based on input relevance; and (3) Design of an adaptive feedback loop that enables the model to perform context‐sensitive and real‐time predictions by continuously refining its internal representations based on incoming sensor signals. Experimental evaluations on the High‐Speed Train (HST) bearing dataset and the Case Western Reserve University (CWRU) dataset demonstrate the superior performance of the proposed method. ALIF achieved a classification accuracy of 99.79% on the HST dataset and 100% on the CWRU dataset, significantly outperforming state‐of‐the‐art deep learning models and standard Transformers.