An Adaptive, State-Based Framework for Fault Prediction in Rotating Equipment

Predictive modeling of industrial rotating equipment is difficult due to a number of implementation challenges. Existing approaches are not well-equipped to adapt to the range of degradation trends that industrial equipment experience and typically depend on large historical datasets for model training. In this paper, a state-based framework for modeling multi-stage degradation in rotating equipment is introduced. With this approach, machine experts can define multi-stage models of equipment health based on first principles or pre-existing knowledge in place of extensive historical data. An adaptive methodology is presented to estimate a machine's health stage history and predict faults based on real-time sensor measurements. A case study with rolling element bearings shows that the proposed approach makes fault predictions with similar accuracy rates as data-driven models trained on multiple historical faults while providing additional degradation detection capabilities.