A dynamic bayesian network approach for electro-optical system performance monitoring digital twin

This paper proposes a digital twin model based on dynamic Bayesian network (DBN) for the electro-optical system for its performance monitoring. In this model, a system-level peformance indicator from the perspective of the energy domain using Modulation Transfer Function (MTF) is first developed, which avoids tedious modeling of performance interactions between the multiple subsystems of the electro-optical system. Then, a DBN is constructed from the evolution of MTF to denote the dynamic performance degradation process and the propagation of epistemic uncertainty. In order to make the digital twin model capable of tracking and predicting the system performance states, an improved Gaussian particle filter with kernel smoothing (GPF-KS) is proposed as the inference algorithm for DRAT. A real dataset collected in the laboratory environment is used to validate the feasibility of the digital twin model and verify the effectiveness of the GPF-KS inference algorithm. The results show that our method is effective for joint estimation of states and parameters, leading to reliable estimation and prediction of the electro-optical system on-line health-status.