Experimental Evaluation of Low Fidelity Models on Co-Kriging Metamodeling of Global Structural Response through Real-Time Hybrid Simulation

Real-time hybrid simulation (RTHS) provides a cyber-physical technique for large- or full-scale experiments in size limited laboratories when parts of the structure are difficult for accurate modeling. Traditional practice of RTHS assumes deterministic structural properties therefore could not account for uncertainties in global response prediction in an efficient and effective way. Previous studies have shown that metamodeling enables efficient uncertainty quantification through limited number of expensive physical experiments or computational simulation. More recent studies indicate that multifidelity Co-Kriging can achieve better accuracy with fewer experiments or less simulation. This study presents an experimental study of the influence of low-fidelity model accuracy on Co-Kriging metamodeling for uncertainty quantification. Laboratory RTHS through are considered as high-fidelity (HF) simulation and conducted in parallel with low-fidelity (LF) computational simulation of the same structure. The Co-Kriging metamodeling is then applied to integrate multifidelity simulation to render accurate response prediction over the entire sample space of uncertainty input variables. Different parameter values are used for same computational model to emulate different LF simulation for Co-Kriging metamodeling. RTHS tests are conducted for a single-degree-of-freedom (SDOF) structure with self-centering viscous damper (SC-VD). The Co-Kriging metamodels established from experimental results are then evaluated through validation tests and further compared with corresponding Kriging metamodels. A multifidelity Co-Kriging with LF model updating is further proposed to improve the convergence and accuracy in response estimation for uncertainty quantification.