Benchmarking the vehicle integrated prognostic reasoner

Abstract

This paper outlines a benchmarking approach for evaluating the diagnostic and prognostic capabilities of the Vehicle Integrated Prognostic Reasoner (VIPR), a vehicle-level reasoner and an architecture which aims to detect, diagnose, and predict adverse events during the flight of an aircraft. A number of diagnostic and prognostic metrics exist, but these standards are defined for well-circumscribed algorithms that apply to small subsystems. For layered reasoners, such as VIPR, the overall performance cannot be evaluated by metrics solely directed toward timely detection and accuracy of estimation of the faults in individual components. Among other factors, the overall vehicle reasoner performance is governed by the effectiveness of the communication schemes between the different monitors and reasoners in the architecture, and the ability to propagate and fuse relevant information to make accurate, consistent, and timely predictions at different levels of the reasoner hierarchy. To address these issues, we outline an extended set of diagnostic and prognostics metrics that can be used to evaluate the performance of layered architecture, and we discuss a software architecture as well as an evaluation plan for benchmarking VIPR.