Aiming at the problem of low recognition accuracy of inverter and current sensor faults in the vehicle three-phase permanent magnet synchronous motor drive system, a model integrating sparse auto encoder (SAE), convolutional neural net-work (CNN) and bidirectional long short memory network (BiLSTM) was proposed. Simulating and analyzing the open circuit and current sensor faults of three-phase two-level voltage source inverters; Using the three-phase stator current output by the motor as the object of fault feature extraction, the dataset is dimensionally reduced using SAE, and a new dataset is constructed using CutMix data augmentation method. Finally, the SAE-CNN-BiLSTM model is used to identify inverter open circuit and current sensor faults. The experimental results show that the SAE-CNN-BiLSTM model can effectively classify and identify faults, with an average accuracy of over 99 % for identifying inverter open circuit and current sensor faults, which is superior to the CNN-BiLSTM model.