THE OPERATING REGIME APPROACH FOR PRECISION HEALTH PROGNOSIS

摘要

Data-driven multiple-model prognosis is a superior approach for solving prognostic problems in complex non-linear systems, where analytic models that can cover all the operating conditions of the system are not available, or not easy to obtain. Traditionally, the operating regime (multiple model) approach is used for non-linear system modeling and control. Following the same principle, for control purposes, many multiple model fault detection and diagnosis approaches have been developed over the years. However, the applications of these methods are mostly limited to components that are under control, e.g. engines, pumps, etc. In this paper, the concept of a new data-driven multiple-model prognosis framework is presented, which attempts to emulate offline health inspection with controlled experiments through uncontrolled, online prognosis. Unlike the traditional approaches that pursue real-time performance for close-loop control, this new approach is aimed at precision health assessment, prognosis and diagnosis, especially for those components in a complex system that exhibit tractable degradation symptoms. In this paper, several key points about this new framework are discussed, including a definition of the operating regimes, regime partitioning strategies and algorithms, local model development and multiple-model information fusion.