An Intelligent Onboard Condition Monitoring Method of Wheelset Bearings Under Time-Varying Speed

Generalized demodulation (GD) presents a novel approach for onboard condition monitoring of wheelset bearings under variable speed conditions. The identification of fault-related frequencies, however, still depends on expert experience and manual selection, which renders intelligent onboard condition monitoring impractical. In order to address this issue, an intelligent onboard condition monitoring method based on adaptive GD (AGD) and multidomain feature extraction (MDFE) techniques is developed. The proposed AGD method can map time-varying frequencies to optimal frequency positions, thereby mitigating spectral smear caused by speed fluctuations and providing a robust foundation for high-precision feature extraction. Subsequently, the useful fault information is extracted from the time domain, frequency domain, and demodulation domain by the proposed MDFE. Notably, the features from the demodulation domain are crucial for improving recognition accuracy and have not been reported in other studies. Ultimately, the health condition of the wheelset bearing is assessed using the support vector machines (SVMs). Experiments and field trials are conducted to validate the effectiveness of the proposed method. The results indicate that the features extracted by the MDFE technique not only successfully differentiate the health status of bearings but also accurately identify varying fault severities. Remarkably, the proposed method achieves a 100% accuracy rate in determining the absence of faults, while the accuracy for distinguishing between fault types exceeds 99%. Compared to other common features and traditional classification algorithms, the proposed method demonstrates superior fault classification capability and achieves the highest recognition accuracy.