Abstract: Due to the randomness and unpredictability of intermittent faults, traditional fault diagnosis methods are difficult to apply to the diagnosis of intermittent faults. In order to improve the reliability of the traction inverter system, this paper proposes an intelligent diagnosis method for the intermittent faults of the DC side voltage sensor. First, combining the nonlinear autoregressive dynamic network structure and the extreme learning machine to quickly mine the nonlinear mapping relationship between historical data and the stator currents of the asynchronous motor, the stator currents predictor of the traction motor is obtained. Then, a sliding time window is designed to construct the currents residuals, and the occurrence and disappearance time of intermittent faults are detected to obtain the evaluation index that characterizes the severity of intermittent faults. The proposed method is verified based on the rapid control prototype (RCP) experimental platform. The results show that the current predictor has great robustness to dynamic conditions such as sudden load changes and sudden speed changes. The proposed diagnosis method can complete the detection of the occurrence times and disappearance times of intermittent faults within 0.65 ms and 0.9 ms, respectively, and can accurately identify the early, middle, and late stages of intermittent faults of sensors, realizing the evaluation of the severity of intermittent faults.