Demagnetization fault diagnosis is a significant issue in magnetically suspended permanent magnet synchronous motors (PMSM). The occurrence of a permanent magnet (PM) demagnetization fault reduces the motor efficiency and even disables the whole power system. Unfortunately, it is challenging to extract the fault feature and evaluate the exact severity precisely due to the rich environmental noise and the coupling faults. This article proposes a novel strategy using the improved variational mode decomposition (VMD) and energy moment relative entropy (EMRE) to achieve the demagnetization fault diagnosis, facilitating the timely implementation of the corresponding control strategies. First, the harmonic component caused by the demagnetization fault is studied. An improved VMD algorithm is proposed to preprocess the displacement signals and decompose them into a series of mutually exclusive intrinsic mode functions (IMFs). Second, signal reconstruction is performed on the selected IMFs, and fault feature extraction is accomplished utilizing the Teager enhancement spectrum. Third, EMRE is proposed as the evaluation indicator to evaluate the exact demagnetization severity. Finally, the system experimental platform with different demagnetization degree prototypes is constructed for validation.