Online Real-Time Rotating Unbalance Forecast Incorporating Model-Based with Machine Learning Techniques

Abstract

Prognostics and Health Management (PHM) is a promising method of fault diagnosis for making maintenance decisions. For system fault development trends, different statistical or machine learning methods are being used. Unbalance is a fault that causes excessive vibrations in rotary systems, yet it cannot be totally eliminated. Thus, monitoring, and timely maintenance are needed, and this has been a research topic for years. This research forecasts rotating system unbalance faults using machine learning and system mathematical models. A machine-learning-based prognostic approach for unbalance faults in rotary systems is developed. Furthermore, operational datasets from a local petrochemical company on an overhung rotor system are utilized to validate the results. The proposed model is compared with other machine learning or statistical-based models for accuracy using the least root mean square error (RMSE) as the performance criterion. The proposed method has been proven feasible for industrial rotor unbalance prognostics.