Sparse Bayesian learning for joint load-parameter-response identification

doi:10.1016/j.ress.2025.111831

Sparse Bayesian learning for joint load-parameter-response identification

2025-10-28

0

PRE

AI

DOI：10.1016/j.ress.2025.111831

Original

Original request for help

Save

Abstract

En 中文

The joint identification of the structural loads, parameters, and responses from the sparse measurements is crucial but challenging for assessing the operational conditions of civil infrastructures. Existing joint identification methods typically require the prior information, such as the loading locations and the covariance matrices of the process and observation noise, which are unavailable in practice. To address this limitation, a sparse Bayesian learning technique based on the transmissibility matrix is proposed for the joint load-parameter-response identification. The spatial sparsity of damage indices (defined as the structural parameter change) and loads is adopted as priors in the Bayesian framework. The Expectation-Maximum (EM) algorithm is used to iteratively estimate the most probable values of the damage indices, unknown force time history, measurement noise, and other variance parameters. A modified Metropolis-Hastings (MH) algorithm is developed to jointly generate posterior samples of structural parameters and loads for the expectation calculation in the EM algorithm. Upon convergence, some hyperparameters approach infinity and the associated damage indices and forces become zero, resulting in the sparsity of the identified damage indices and loads. Consequently, the damage indices and input forces are located and quantified simultaneously, and the full-field structural responses are sequentially reconstructed with suppressed uncertainties. The proposed method is verified by a numerical simply-supported beam, a laboratory-tested three-story frame model, and a three-span continuous bridge model. Results demonstrate that the proposed sparse Bayesian technique can accurately identify the damage indices, loads, and responses simultaneously, even when the load position is unknown.

Sparse Bayesian learning for joint load-parameter-response identification

Abstract

Journal

Organization