Reliability is an important performance for permanent magnet synchronous machine (PMSM) exploited to industrial applications and electric‐energy vehicles. Both air‐eccentricity and inter‐turn short circuit (ITSC) could induce the machine running in unbalanced status. Once not restored timely, the fault will lead to the whole drive system and its application damage. To enhance the reliability for PMSM, an effective fault diagnosis and classification method is necessary in advance. For this purpose, an improved exponential discriminant analysis (IEDA) algorithm is proposed and optimized. Further, the distortion characteristics of phase current signals with air‐eccentricity and ITSC are extracted and trained by the IEDA, which are applied for diagnosing whether the motor healthy or faulty and distinguishing fault types. Besides, the IEDA can not only identify fault type accurately but also estimate the fault degree precisely. Finally, compared to other common learning methods, several simulations and practical tests accompanying with diverse operating status are implemented, and their results can prove the proposed method has more accuracy and robustness than other methods. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.