A metric for assessing and optimizing data-driven prognostic algorithms for predictive maintenance

Antonios Kamariotis, Konstantinos Tatsis, Eleni N. Chatzi, Kai Goebel, Daniel Straub

Abstract: Highlights•A novel decision-oriented metric for assessing data-driven prognostic algorithms.•The metric is defined in conjunction with a predictive maintenance (PdM) policy.•We investigate and improve certain PdM policies for two common decision settings.•The metric also serves for optimizing PdM policies and algorithms’ hyperparameters.•Numerical investigations are performed on the CMAPSS run-to-failure dataset.