Node-Incremental-Based Multisource Domain Adaptation for Fault Diagnosis of Rolling Bearings with Limited Data

Bearing fault diagnosis is essential for ensuring the safe and reliable operation of rotating machinery. However, accurate and timely fault identification with limited data remains a significant challenge. This study proposes a novel node-incremental-based multisource domain adaptation (NiMDA) approach for bearing fault diagnosis. The method employs a cloud model to adaptively extract fault-sensitive information while accounting for uncertainties across multiple wavelet packet decomposition levels. Subsequently, node incremental domain adaptation (NiDA) is used to construct a base classifier utilizing limited labeled data from both target and source domains. This approach reduces discrepancies between marginal and conditional distributions across different domain feature spaces during the node-increment process, resulting in a compact domain-adaptation structure. Robust diagnostic performance is achieved through parallel ensemble learning of NiDAs across multiple source domains. The experimental results demonstrate that NiMDA significantly outperforms state-of-the-art bearing fault diagnosis methods in few-shot scenarios, achieving improvements of 30.52%, 42.31%, 10.31%, 26.08%, 25.59%, and 7.98% over WDCNN, MCNN-LSTM, Bayesian-RF, DM-RVFLN, Five-shot, and ESCN, respectively, while maintaining satisfactory diagnostic speed.