A new ijPHM paper: Metrics for Offline Evaluation of Prognostic Performance

The International Journal of Prognostics and Health Management (ijPHM) recently published its first full paper entitled Metrics for Offline Evaluation of Prognostic Performance.

The new paper by Abhinav Saxena and colleagues at NASA Ames Research Center is an important milestone in the maturing field of PHM. Business models for PHM are still being debated and established, and the value of remaining useful life (RUL) predictions is still unclear. Lack of standards in the area of prognostic metrics makes it even harder to realize commercial (or safety) value out of RUL forecasts. Most PHM research articles define their own methodology for generating RUL estimates, along with uncertainty bounds for these estimates and timelines for estimation.

The ijPHM article surveys offline prognostic metrics used in the field that in general incorporate probabilistic uncertainty estimates from prognostic algorithms, and attempts to standardize the various metrics under a unified prognostic framework. Guidelines are provided to assist researchers in using the metrics. The paper suggests future work, most importantly an extension of these metrics to “online” RUL estimation for live predictions of component or system failure. One might consider online prognostic estimation to be the Holy Grail of PHM: if done successfully and reliably, online RUL estimation methods will unleash the full economic and safety potential of PHM research.

Overall, I find the Metrics article to be an indispensable reference for researchers in PHM. The PHM Society is proud to make the paper available to the global community as an open access article. I encourage the PHM community to read the article and engage in further debate on PHM metrics through this blog or the standards forum.


Serdar Uckun
President, PHM Society