A Markov Probability Model-Based Framework for Bearing Fault Detection Under Variable Speed Conditions

Bearing fault detection is important for reliable and safe operation of rotating machinery. The operating speed of the bearing is usually variable, which makes fault detection more challenging. Existing works have not explored a structural framework that links angle resampling with Markov models, leaving room for improvement in fault detection under variable speed conditions. This paper proposes a Markov probability model-based fault detection method for bearing operating under variable speed conditions. First, angle resampling is used to resample the data obtained under different rotational speeds to demodulate frequency modulation. Then, the resampled vibration signal is modeled using the explicit-duration hidden Markov model (EDHMM). Finally, the fault is detected through the likelihood ratio test. The fault detection results are visualized through the receiver operating characteristic (ROC) curve, its quantification metric area under ROC curve (AUC) value, and the corresponding sequence plot. The performance is compared to the angle resampling-free Markov model likelihood ratio detection method. The results indicate that the AUC of the proposed method is 0.9110, much higher than that of the angle resampling-free Markov model, 0.6439.