Arpha: an Fdir Architecture for Autonomous Spacecrafts Based on Dynamic Probabilistic Graphical Models

This paper introduces a formal architecture for onboard diagnosis, prognosis and recovery called ARPHA. ARPHA is designed as part of the ESA/ESTEC study called VERIFIM (Verification of Failure Impact by Model checking). The goal is to allow the design of an innovative on-board FDIR (Fault Detection, Identification and Recovery) process for autonomous systems, able to deal with uncertain system/environment interactions, uncertain dynamic system evolution, partial observability and detection of recovery actions taking into account imminent failures. We propose to base the inference engine of ARPHA on Dynamic Decision Network (DDN), a class of Probabilistic Graphical Models suitable to reason about system evolution with control actions, over a finite time horizon. The DDN model needed by ARPHA is assumed to be derived from standard dependability modeling exploiting an extension of the Dynamic Fault Tree language, called EDFT. We finally discuss the software architecture of ARPHA, where on-board FDIR is implemented.