Intelligent Condition Monitoring for Machining Processes Under Cross-Domain Scenarios via Semi-Supervised Feature Separation

Tool wear state classification under varying working conditions remains challenging due to domain discrepancy and the scarcity of labeled target samples. To address this issue, this paper proposes a Semi-Supervised Domain-Adaptive Feature Separation framework (SS-DAFS) for intelligent condition monitoring in machining processes. Experiments are conducted on the PHM2010 milling dataset, where tool conditions are categorized into three health states, including initial wear, normal wear, and severe wear. The proposed framework separates shared domain-invariant features from private domain-specific features to enhance feature representation capability and reduce domain-specific interference. Meanwhile, Maximum Mean Discrepancy (MMD) is incorporated to further align feature distributions between source and target domains. Semi-supervised transfer tasks under one-shot and three-shot settings are constructed to evaluate the effectiveness of the proposed framework with extremely limited labeled target samples. Experimental results demonstrate that the proposed framework consistently outperforms several representative domain adaptation methods under both settings. The average accuracy of the proposed method under three-shot and one-shot achieves 95.68% and 94.34%, respectively. The results verify the effectiveness of the proposed framework for intelligent condition monitoring for machining processes under cross-domain machining scenarios.