A Feature-Enhanced Multi-Kernel Adaptation Framework for Cross-Domain Fault Diagnosis of Ball Screw Feed Systems

Ball screw feed systems are critical components of CNC machine tools, and their health condition directly affects positioning accuracy and operational reliability. However, variations in rotational speed under changing operating conditions can cause substantial shifts in fault characteristics, limiting the generalization capability of conventional fault identification methods developed under a single operating condition. To address this issue, a feature-enhanced multi-kernel domain adaptation framework is proposed for cross-domain fault identification of ball screw feed systems. First, an attention-based feature enhancement module is developed to adaptively reweight vibration features, thereby emphasizing informative feature components and suppressing redundant and noise-related information. Subsequently, a residual feature extraction and fusion module is employed to learn hierarchical fault representations and improve feature discriminability. Finally, a multi-kernel maximum mean discrepancy (MK-MMD) strategy is integrated with adversarial domain adaptation to reduce feature distribution discrepancies between source and target domains under different rotational speeds and ball screw specimens. Local spalling fault experiments were conducted on the developed test bench, where vibration signals were collected using an accelerometer under various operating conditions and subsequently used to validate the proposed framework. Experimental results demonstrate that the proposed method achieves an average identification accuracy of 92.9% across the 12 investigated cross-domain tasks, compared with 69.7% for the optimized source-only CNN, with the performance gap becoming particularly pronounced under cross-specimen transfer conditions. The results indicate that the proposed framework improves the learning of fault-discriminative and domain-invariant representations, thereby improving the robustness of ball screw fault identification under variable operating conditions.