Abstract: Insulator faults in transmission lines are the primary cause of power grid outages. Existing fault identification methods suffer from poor generalization with small samples and insufficient robustness in feature extraction. To achieve efficient and accurate detection of insulator defects in complex scenarios, this paper proposes a visual large model fault classification adaptation method driven by stochastic configuration networks. It integrates DINOv2 visual large model transfer feature extraction with stochastic configuration network classifiers, establishing an end-to-end insulator defect recognition process. The study demonstrates that this method effectively enhances the accuracy and generalization capability of power equipment fault classification, meeting the engineering application requirements of intelligent power grid inspections.