LAUSR

ABSTRACT Two different back-analysis frameworks based on multivariate machine learning models used to determine the material dynamic parameters of concrete gravity dams are proposed. For the framework I, the back-analysis is performed by solving an optimization problem and a multivariate machine learning model is trained to replace the FEM calculation during the optimization process. While the framework II uses a multivariate machine learning model directly and the material dynamic parameters are predicted using the machine learning mode. By using a numerical example and an experimental investigation, the robustness, accuracy, computation efficiency of these proposed back-analysis methods is verified.