Exploration of Multi-Fidelity Co-Kriging and Real-Time Hybrid Simulation for Time-Dependent Seismic Performance Evaluation

Abstract

Real-time hybrid simulation (RTHS) divides the structure under investigation into experimental and numerical substructures thus facilitates large- or even full-scale experiments beyond laboratory constraints. Traditional analysis of time-varying uncertainties necessitates extensive RTHS experiments in laboratories when parts of structures are difficult for accurate modeling due to their complexity and/or nonlinearity. To address this challenge, this study explores Co-Kriging meta-modeling and adaptive sampling with RTHS aiming for efficient time-dependent performance evaluation with limited number of RTHS tests in laboratories. The expansion optimal linear estimation method is applied to discretize random processes for time-varying uncertainties. Key informative sample points are selected through an adaptive sampling technique, while the Co-Kriging meta-model combines low-fidelity numerical simulations with high-fidelity RTHS experiments to account for time-varying uncertainties in structural response prediction. A single-degree-of-freedom structure with time-varying stiffness degradation is used in this pilot study to experimentally evaluate the accuracy and efficiency of proposed multi-fidelity approach. It is demonstrated that the Co-Kriging meta-modeling provides a promising alternative for efficient time-dependent performance evaluation to account for time-varying uncertainties especially when parts of structure are difficult for numerical modeling. Compared with single fidelity based method such as Kriging, Co-Kriging provides faster convergence thus leading to less laboratory tests.