Probabilistic model updating of civil structures with a decentralized variational inference approach

Abstract

The quantification of uncertainties in civil structures poses a significant challenge due to the need to update a large number of unknown parameters, which can significantly increase the computational workload. A potential solution to this problem is the use of Approximate Bayesian Computation (ABC) approaches, which have gained attention in recent years. However, most existing ABC methods are limited in their ability to handle a large number of unknowns. To address this limitation, this study proposes a decentralized approach for uncertainty quantification in civil structures. In this approach, the structure is divided into several parts, and the uncertainties in each part are estimated independently. The proposed method utilizes the wavelet domain technique to eliminate the interfacial force of the target part and employs a recently developed variational inference approach to quantify the uncertainty in the target substructure. A three-span beam, a 2D truss, and an eight-story building were tested to evaluate the efficiency and feasibility of this approach. The results show that uncertainties in large structures can be estimated and that the proposed method is robust to high levels of measurement noise.