Machinery Radial Rub Fault Detection via Shaft Relative Vibration Measurement Using Hidden Markov Model

This research is focused on establishing an effective defect diagnostic process for a journal-bearing system. The method was developed using vibration data from a journal-bearing rotor simulator under two different scenarios (a normal condition and a rubbing anomaly condition). After applying a resampling procedure to the raw vibration data, cycle-based time domain features were recovered to improve diagnostic performance. Hidden Markov model (HMM) was used to detect anomalies related to rotating machinery radial rub faults and compared to other available models in detecting anomalies. The Anomalize anomaly detection model was also run on top of HMM for testing and evaluation purposes. HMM produces very competitive prediction results as good as more complex models.