Incipient Fault Identification of Bearings in Electric Drive System Under Varying Speeds Based on Adaptive Feature Mode Decomposition

• Published in: IEEE sensors journal Vol. 25; no. 9; pp. 15975 - 15995
• Main Authors: Wang, Chaoge, Tian, Xinyu, Zhou, Funa, Wang, Ran, Li, Hongkun
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Background noise; Bearings; Condition monitoring; Decomposition; Domains; Effectiveness; Electric drives; Electric vehicles; Fault detection; Fault diagnosis; Feature extraction; feature mode decomposition (FMD); Filter banks; Filtering algorithms; Finite impulse response filters; incipient fault identification; Motor stators; Noise; Parameters; Particle swarm optimization; Resampling; Roller bearings; rolling bearing; Rolling bearings; variable speed; Vibrations
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2025.3532489

Rolling bearings constitute a core component in the electric drive system of electric vehicles, and their health status is crucial for the safe operation of these vehicles. Therefore, effective condition monitoring and fault detection for bearings is of paramount importance. However, under variable operating conditions, the incipient weak fault signatures of bearings are prone to be masked by intense noise, significantly increasing the challenge of fault identification. To tackle this issue, an adaptive feature mode decomposition (AFMD) approach is introduced for diagnosing incipient weak fault in bearings under variable operating conditions. First, the instantaneous rotational frequency is extracted from the motor stator current signal. Subsequently, angular domain resampling is performed on the synchronously sampled vibration signal. Second, to overcome the lack of adaptability in determining the key input parameters of the FMD algorithm, which typically relies on repeated manual trials based on experience, this research proposes determining the number of decomposition modes through scale-space spectral segmentation. On this basis, the spectral Gini index (SGI) is adopted as the objective function, and the particle swarm optimization (PSO) is utilized to automatically ascertain the filter number and filter length. With the optimal decomposition parameter combination, the AFMD is employed to perform optimal mode decomposition on the obtained angular domain bearing signal, and the component exhibiting the highest SGI value is chosen as the sensitive mode. Finally, significant fault characteristic orders are extracted from the envelope order spectrum (EOS) of the sensitive component to accurately identify the fault type. The efficacy and superiority of the proposed methodology are confirmed through variable-speed simulated bearing signal, experimental data, and actual electric vehicle bearing diagnosis cases. The analysis demonstrates that the proposed approach can clearly and comprehensively capture weak fault information even under significant background noise interference, thereby enhancing the representation ability and diagnostic accuracy of early fault characteristics in bearings under variable-speed conditions.