MDATS―A Multi-Domain Adaptive Classification Strategy for Mechanical Fault Detection

Mechanical fault detection and classification play a critical role in maintaining the reliability and safety of industrial systems. This paper presents a novel multi-domain adaptive classification strategy (MDATS) designed to enhance fault detection accuracy in varying operational conditions. Using time series data from the XJTU-SY bearing datasets, our approach leverages TS2Vec for time series encoding, coupled with CNN, TCN, and Transformer architectures to extract multi-scale features. To address domain adaptation challenges, the Transferable Domain Adaptation Network (TDAN) is employed, improving model generalization across multiple datasets. Experimental results demonstrate that the proposed MDATS framework significantly improves fault classification performance, particularly in target domains with differing characteristics. The framework is evaluated against baseline models, achieving superior results in both accuracy and domain adaptation.