Current deep learning (DL) models applied to fault diagnosis of interrupters exhibit limited explainability. This reduces the trust of operation and maintenance personnel in the fault diagnosis model. In this study, an adaptive fault feature tracking method for vacuum interrupters based on ResNet-SHAP is proposed. The explanation technology is used to realize the adaptive detection of the fault feature area of interrupters. Firstly, the vibration signals of interrupters are quantitatively described by continuous wavelet transform (CWT), and the time-frequency scale feature set of vibration during the closing process is constructed. After comparing the matching effects of multiple DL models, it is shown that ResNet and SHAP have the highest matching degree. Experiments show that the fault feature distribution map generated by this method can explain the sensitivity of the diagnosis model to fault features. After the contact travel matching, the fault features can also be traced. Also, the fault feature tracking method can adaptively find hybrid fault features.