An improved system identification method for hardfill dams considering the spatial variability of material parameters based on random field theory

Calibrating the finite element model plays an important role in the accurate safety evaluation of dams. Recent studies show that the material heterogeneity has important impact on the safety evaluation of hardfill dams. However, due to the complexity of material heterogeneity, only limited researches involve the system identification of dams considering the material heterogeneity. In this study, an improved system identification method considering the material parameters heterogeneity is proposed for hardfill dams based on the Hariri-Ardebili et al.'s method. The proposed improved method is composed by two main steps: firstly, identification of the correlation length based on the support vector machine (SVM) and then estimation of the spatial varied hardfill modulus based on the combination of the Karhunen-Loeve expansion (KLE) and surrogate-based optimization (SBO) algorithm. A series of numerical simulations are performed to test the proposed method. The simulation results indicate that the proposed method can improve both the accuracy and computational efficiency. In addition, the effects of the hardfill heterogeneity on the modal responses are also investigated in this study. It is observed that the variation of the natural frequencies tends to increase with the increase of correlation length.