Diagnosing Hybrid Systems : a Bayesian Model Selection Approach

In this paper we examh the problem of monitoring and diagwsing noisy dmplex dynamical systems that m modeled as hybrid systems models of umtinuolrs behavior, interleaved by disnsts transitions. In particular, we examins cop experience abrupt, partial or full failme of component de vices. Building on our previous work in this area (MBCG99; matical formulation of the hybrid monitoring and diagnosis task maBayesian modeltracling andselection problem, and provision of a suitable tracking algmitbm. The wnlinear dynamics of many hybrid systems m t challenges to pmb abilistic tracking. Fur~ber, probabilistic tracking of a system els of the system cormponding to failure modes am numerous and g d y very unlikely. To focus tracking on these unlikely models and to rsduce the number of potential & elswkrconsideration, weexploitlogic-bawdtechniquesfbr qualitah model-baseddiagnosis to conjecture alimitedinitial set ofconsistent candidatemodels. In this paper wedisfxent classes of hybrid systems, focusing specifically on a metbod fbr traekiag multiple models of wnlioear bebavior sity propagdtion. To illustrate and motivate the epproach de and diagnosing NASA's Sprint AERCam, a small spherical robotic camera unit with 12 thrusters that enable both l k a r d r o t a t i ~ m o t i o n . tiauous systems with embedded supervisory COntrolleR that MBCGOO), our specific focus in this paper is on the mrrtha forthep~~ofdiagoosisirproblematicbeca~the& cuss alt€anative tracking techniques that are mlevant to difs i m W u s l y wing factomd sampling aad condtiioaal dmscribedinthis paper we examine t h e ~ ~ o f m o n i ~