This paper proposed a data model-based diagnosis method for voltage sensor intermittent faults (IFs) of motor drive system. The nonlinear mapping relationship between historical data and induction motor stator currents is quickly mined by combining nonlinear autoregression with exogenous inputs (NARX) and random vector functional link (RVFL) to obtain a currents estimator. Subsequently, a sliding time window is designed to construct the currents residuals to detect the appearance and disappearance times of IFs.