This paper addresses the issue of irreversible demagnetization in permanent magnet synchronous motors (PMSMs), which can occur in high temperature or heavy-duty working environment. Failure to diagnose demagnetization faults in time can lead to a continuous increase in stator current, raising copper losses and motor temperature and potentially causing more serious motor faults. We propose a diagnostic method for irreversible demagnetization in PMSMs, based on a combination of convolutional neural network (CNN) and long short-term memory neural network (LSTM). First, the change patterns of magnetic flux and stator current during demagnetization are analyzed, with the stator current selected as the signal input. Then, Wavelet threshold noise reduction is applied. Finally, fault diagnosis is conducted using the CNN-LSTM neural network. Simulation results proved that the method achieves high diagnostic accuracy for irreversible demagnetization fault diagnosis.