Physics-Informed Speed-Integrated Hidden Markov Model for Bearing Fault Diagnosis under Variable Operating Conditions

Condition monitoring of rolling bearings is essential for ensuring the safe and reliable operation of rotating machinery and for reducing economic losses due to unexpected failures. Under variable-speed conditions, bearing vibration signals exhibit significant nonstationarity and time-varying behavior, posing dual challenges in uncertainty modeling and fault pattern recognition. To address these challenges, this study proposes a speed-integrated explicit duration hidden Markov model (SI-EDHMM), which embeds rotational speed information directly into the probabilistic structure of the model. By constructing a joint observation vector from vibration and speed signals, SI-EDHMM enables accurate characterization of fault features and effective state decoding in nonstationary environments. Fault detection is carried out via a likelihood ratio test. Experimental validation on the Ottawa variable-speed bearing dataset demonstrates that SI-EDHMM significantly outperforms the conventional EDHMM. The detection performance is assessed using the receiver operating characteristic curve and its associated area under the curve (AUC). SI-EDHMM achieves an AUC of 0.9783, representing a 7.12% improvement in detection accuracy over the traditional EDHMM.