Dynamic Bayesian Network for Operational Modal Identification

Abstract

Application of the dynamic Bayesian network (DBN) for modal identification under operational conditions is investigated. Under the assumption of Gaussian white noise, the stochastic state space model is converted to a DBN. Time-domain maximum likelihood estimation (MLE) is applied to learn the DBN. The expectationmaximization (EM) algorithm, initialized by parameters identified by the referencebased, data-driven stochastic subspace identification (SSI/DATA) method, is used to iteratively solve for the MLE. A numerical simulation example demonstrates the performance of the DBN-based identification method.