Mode set focused hybrid estimation

Abstract

Estimating the state of a hybrid system means accounting for the mode of operation or failure and the current state of the continuously valued entities concurrently. Existing hybrid estimation schemes try to overcome the problem of an exponentially growing number of possible mode-sequence/continuous-state combinations by merging hypotheses and/or deducing likelihood measures to identify tractable sets of the most likely hypotheses. However, they still suffer from unnecessarily high computational costs as the number of possible modes increases. Hybrid diagnosis schemes, on the other hand, estimate the current mode of operation/failure only, thus leaving the continuous evolution of the system implicit. This paper proposes a novel scheme that uses a combination of both the approaches in order to define posterior transition probabilities between the specified modes of the hybrid system, hence focusing better on relevant hypotheses. In order to demonstrate the effectiveness of the proposed method, the algorithm is applied to a satellite attitude control system and compared with existing hybrid estimation/diagnosis schemes, such as the Interacting Multiple Model (IMM) algorithm, a purely parity based method (HyDiag), and an existing hybrid Mode Estimation (hME) algorithm.

Bridging control and artificial intelligence theories for diagnosis: A survey

“…Lately, the mode estimation rME approach (Bayoudh et al, 2008c) was combined with the state estimation approach hME (Hofbaur and Williams, 2004), resulting in mutual enrichment (Rienmuller et al, 2009;Rienmüller et al, 2013). hME brings an estimate of the continuous state that is not provided by rME.…”

Iterative hybrid causal model based diagnosis: Application to automotive embedded functions

“…They provide a particularly relevant framework for diagnosis because discrete fault occurrence can be represented by switching dynamics. Many methods have hence addressed discrete faults by adding fault modes to the nominal system model for each discrete fault (Hofbaur and Williams, 2004;Wang et al, 2007;Bayoudh et al, 2008;Benazera and Travé-Massuyès, 2009;Rienmüller et al, 2013). Other methods have addressed parametric faults like (McIlraith et al, 2000b;Narasimhan and Biswas, 2007b).…”

An Integrated Framework for Diagnosis and Prognosis of Hybrid Systems

“…The links between diagnosis and prognosis in case of ambiguity will be further investigated. Methods that increase the amount of available information concerning an ambiguous diagnosis are obviously useful to focus on a reduce set of hypotheses [15]. Another idea is that diagnosis and prognosis processes feed each other: prognosis allows the diagnosis to be less ambiguous and thanks to the diagnosis, the prognosis process can refine its prediction.…”