Bayesian data driven model for uncertain modal properties identified from operational modal analysis

In structural health monitoring (SHM), data driven models' are often applied to investigate the relationship between the dynamic properties of a structure and environmental/operational conditions. Dynamic properties and environmental/operational conditions may not be directly measured but are rather inferred based on measured structural response data. Conventional data driven models assume training data as precise values without uncertainty, but this may not be justified when they are identified by operational modal analysis (OMA) where identification uncertainty can be significant. The associated confidence or precision may also vary depending on their identification uncertainties. This paper develops a Bayesian data driven model for modal properties identified from OMA. Identification uncertainty is incorporated fundamentally through the posterior distribution of modal properties of interest given the ambient vibration measurements. A Gaussian Process model is used for describing the potential unknown relationship between the modal properties and environmental/operational condition, which is subjected to OMA identification uncertainty. An efficient framework is developed to facilitate computation. The proposed method is validated by synthetic and laboratory data. Typhoon data from two tall buildings illustrates the field application of the proposed method. (C) 2019 The Authors. Published by Elsevier Ltd.