In most case, the incipient fault feature of interturn short circuit fault is difficult to extract, thus this paper provides a novel fault diagnosis method for the permanent magnet synchronous motor (PMSM) based on the local tangent space arrangement (LTSA). Firstly, the two-dimension instantaneous finite element model of PMSM is established in Ansoft simulation, and the performance indexes are obtained accordingly. Afterwards, the performance indexes are decomposed to high dimensional fault features through the wavelet packet, and through LTSA and other manifold learning methods, it is reduced to gain the mapping in low dimensional space, which can classify faults and normal state. Finally, the experimental results show that the manifold learning method can effectively extract the incipient fault feature of interturn short circuit fault, additionally, LTSA can be used to distinguish the number of short circuit turns, which provide a new idea for fault diagnosis and prediction of PMSM. This work is supported by National Basic Research Program of China (No. 2015CB755805).