Novel two-stage adaptive weight coefficients for Finite Element Model Updating of a cable-stayed bridge via sensitivity analysis and Kriging model

Abstract

The weight coefficients of the structural response residuals in an objective function are the key parameters affecting a Finite Element Model Updating (FEMU) results. Traditional equal weight methods cannot measure the importance and relative errors of different response residuals, resulting in low prediction accuracy of the updated Finite Element (FE) model. This study proposes a two-stage adaptive weight coefficient method according to the relative errors of the measured static displacements and modal frequencies. In the first stage, high weights are assigned to low-order frequencies and large displacements because they have higher accuracy and are less affected by environmental noises than high-order frequencies and small displacements. In the second stage, the weights for the responses with small relative errors are adaptively increased to reduce the prediction errors. This approach was comprehensively validated by numerical simulations of a damaged truss under various noise conditions and a full-scale cable-stayed bridge based on experimental measurements. Results indicate that the two-stage adaptive weight coefficients improve model updating accuracy and maintains robustness under various noise conditions compared to equal weight method. The updated FE model obtained from adaptive weight method for the cable-stayed bridge shows higher accuracy in predicting frequency and displacement than that obtained from equal weight method. After the model update, the maximum relative frequency error is reduced from −16.98% to −6.18%, and that of displacement is reduced from 17.85% to 5.87%. The two-stage adaptive weight coefficients improve the accuracy of FE model updating for bridge structures.