Bayesian state estimation in partially-observed dynamic multidisciplinary systems

Multidisciplinary systems comprise several disciplines that are connected to each other with feedback coupled interactions. These coupled multidisciplinary systems are often observed through sensors providing noisy and partial measurements from these systems. A large number of disciplines and their complex interactions pose a huge uncertainty in the behavior of multidisciplinary systems. The reliable analysis and monitoring of these partially-observed multidisciplinary systems require an accurate estimation of their underlying states, in particular the coupling variables which characterize their stability. In this paper, we present a probabilistic state-space formulation of coupled multidisciplinary systems and develop a particle filtering framework for state estimation of these systems through noisy time-series measurements. The performance of the proposed framework is demonstrated through comprehensive numerical experiments using a coupled aerostructural system and a fire detection satellite. We empirically analyze the impact of monitoring a single discipline on state estimation of the entire coupled system.