In view of the limitations of existing transfer learning methods in the application of circuit breaker fault diagnosis

the data collection in the target domain is restricted by the installation positions and sampling rates of external sensors

and the labeled data is scarce. These challenges lead to insufficient model adaptability and low diagnostic accuracy when extending from the source domain to the incomplete target domain. We propose a cross-domain fault diagnosis method of circuit breakers based on recurrence plot and distillation learning assisted improved ResNeSt18 transfer learning model. This method transforms one-dimensional vibration signals into two-dimensional recurrence plot through time-frequency decomposition and nonlinear dynamic analysis

and it enhances source domain feature representation through multi-channel fused recurrence plot. Under the distillation learning framework

we construct an improved ResNeSt18 deep learning model to achieve learning and knowledge transfer of domain-invariant features between the source domain and target domain. Introducing a joint distribution adaptation mechanism reduces the distribution discrepancy between domains

thereby further facilitating the diagnostic model to learn cross-domain invariant features. The experimental results demonstrate that the proposed method achieves an average diagnostic accuracy exceeding 91% in 12 different tasks with varying levels of migration difficulty

confirming its reliability and applicability in cross-domain fault diagnosis for circuit breakers.