Abstract: Permanent magnet synchronous motors (PMSMs) have emerged as the core drive components of electric vehicles owing to their merits of high energy efficiency and high power density. However, it is susceptible to permanent magnet demagnetization faults under complex operating conditions (e.g., flux-weakening control, collision and vibration), which seriously endangers driving safety and motor operational reliability. The signal injection method offers significant advantages in motor fault feature extraction due to its high characteristic signal distinguishability and robust anti-interference capability, while the random forest algorithm possesses strong multi-feature classification and generalization capabilities for accurate fault diagnosis. Based on the signal injection method, this paper collects motor characteristic signals under various demagnetization conditions and constructs an integrated demagnetization fault diagnosis system. The results demonstrate that the proposed hybrid method can precisely identify different demagnetization faults, providing a reliable technical solution for efficient PMSM demagnetization fault diagnosis in electric vehicles.