Novel metrics and methodologies for the verification and validation of prognostic algorithms

Prognosis is a fundamental enabling technique for condition-based maintenance (CBM) systems and prognostics and health management (PHM) systems and therefore, plays a critical role in the successful deployment of these systems. The purpose of prognosis is to predict the remaining useful life of a system/subsystem or a component when a fault is detected. Although different prognostic algorithms have been developed and tentatively applied to various mechanical and electrical systems in the past decade, the verification and validation (V&V) remains a challenging open problem. The difficulties lie in the facts that first, there is usually no statistically sufficient data to do V&V and second, there is no rigorous and general V&V framework available. In this paper, several new metrics and methodologies stemming from weather forecast verification, nonlinear exact filtering, nonlinear uncertainty propagation and Monte Carlo method are proposed to validate user defined particle-filtering based prognostic algorithm. The presented metrics and methodologies are generic and can be extended to the V&V of other prognostic algorithms on different platforms. The methodologies are demonstrated on the prognosis of a real world application.