A Framework for Evaluating PHM Models

Prognostic models for condition based maintenance are gaining favor with more available data. An evaluation construct presented here considers how well the model tracks degradation over time as well as how effectively it improves operations. Evaluation drives design, so signal conditioning and decision points can be optimized with the right cost function. The construct enables the design of a class of models which augment degradation physics with data-driven heuristics, supporting a more explainable recommendation. This framework presents a holistic way of estimating true degradation and evaluating model performance for operational value.