In the long-term operation and service of machinery, new fault modes continuously emerge, placing higher demands on the continual learning and intelligent diagnostic capabilities of fault diagnosis models. Intelligent diagnosis driven by class-incremental learning offers a promising approach to ensuring safe operation throughout the equipment lifecycle. However, existing class-incremental learning methods fall short in addressing the challenge of efficient incremental transfer diagnosis under cross operating conditions. To overcome this limitation, this paper proposes a cross domain intelligent diagnostic method driven by class-incremental transfer diagnostic (CITD). Firstly, a novel knowledge distillation strategy is developed to mitigate catastrophic forgetting in incremental transfer diagnostic scenarios. Then, a generalization training and fast adaptation strategy is introduced to improve the generalization ability of the model for incremental transfer diagnosis. Experimental validation on a subway train transmission system dataset demonstrates that the proposed CITD method effectively adapts to cross-domain incremental transfer diagnosis tasks, delivering superior performance and outperforming several state-of-the-art class-incremental learning methods.

fast adaptation,class-incremental learning,knowledge distillation,transfer learning,lifelong diagnosis