Model updating of engineering structures based on modified subset simulation methods and measured data

Abstract Subset simulation (SuS) as a reliability method is used to solve optimization issues in areas of civil engineering for relatively simple engineering structures, and several advancements in the generation approaches of samples in the subset to alleviate the generation of repeat samples have been made. However, the application of the SuS method to real, complicated engineering structures is rare. In this paper, firstly, the feasibility of four typical SuS-based methods — namely, SuS-modified Metropolis-Hasting (SuS-MMH), SuS-MMH with delayed rejection (SuS-MMHDR), SuS-adaptive conditional sampling (SuS-ACS) and SuS-Hamiltonian Monte Carlo — was assessed based on the updating of a damaged simulated cantilever beam. The differences in the first 10 frequencies of the beam between the updated finite element model (FEM) and the real FEM were all reduced from above 5% to within 1%, with the methods of SuS-MMHDR and SuS-ACS showing better efficiency with fewer iteration steps. Then, the application of SuS-MMHDR and SuS-ACS to the model updating of a real steel corridor structure was analyzed. The results show that the differences in the first five vertical vibration frequencies are reduced from approximately 20% to within 10%, and the result of SuS-MMHDR was around 5%. Finally, the updated model under human-induced vibration for the steel corridor structure was verified. Based on the experiment, the difference in the largest acceleration between the updated model and real structure was controlled at around 10%, which shows relatively good agreement with each other. The above model updating applications show the feasibility of the SuS-based optimization method for real engineering optimization, and SuS-MMHDR shows the strongest performance.