Rotor inter-turn short circuit fault is hard to be diagnosed because mathematical morphology filter process stator current is not ideal. The method of PSO-adaptive mathematical morphology filtering is used to wipe out the noise interference. Firstly, the best weight coefficient of open-close and close-open operation is found by using PSO, and the best de-noise mathematical morphology filtering is built. Secondly, a fault diagnosis method of taking the wavelet packet entropy as a fault character is proposed combined with wavelet packet analysis and information entropy . Finally, the filtering effectiveness of the de-noise result of PSO-ADMMF and that of the conventional mathematical morphology are compared, and the wavelet entropy after de-noise is figured out. The result shows that the method is useful and effective.