This paper proposes an Open-Circuit-Fault (OCF) detection method for surface permanent magnet synchronous motor (SPMSM) drives based on a Long Short-Term Memory (LSTM) network. In the proposed method, the DC-link shunt current is used as the primary feature for fault detection. From this shunt current and switching signal, the three-phase currents are reconstructed. The difference between the reconstructed currents in the αβ-frame is further used as a secondary feature for faulty switch localization. The proposed method can detect OCF in MOSFETs within approximately one-third of the motor current period with accuracy exceeding 99%. Moreover, using only two features from a single current sensor measurement reduces algorithmic complexity. The effectiveness of the proposed method is validated on a 500W SPMSM drive system implemented in a real-time digital simulator.
