Geotechnical reliability-based design using generalized subset simulation with a design response vector

This study proposes an efficient full probabilistic design method using an advanced Monte Carlo simulation (MCS) algorithm, so-called generalized Subset Simulation (GSS), for which a concept of design response vector is proposed. The proposed design response vector contains responses of all possible designs and is used to, simultaneously, explore failure domains of different designs. With the design response vector, the proposed approach determines feasible design domains of different geotechnical limit states by a single GSS run. The GSS run can be terminated as the target reliability level (or other user-defined reliability levels) is reached to avoid completely exploring the whole design space. This provides significant computational savings, particularly as the design space is relatively large and the target reliability level is relatively high. The proposed approach is illustrated using a drilled shaft design example. Results show that it provides an efficient and robust vehicle to determine the feasible design domains for full probabilistic design compared with direct MCS and original Subset Simulation-based approaches. Moreover, based on failure samples generated by GSS, the design point is, approximately, determined and can be used to back-calculate case-dependent resistance (or deformation) factors, which links MCS-based full probabilistic design to semi-probabilistic design.

Automated summary

This study proposes an efficient full probabilistic design method using the advanced Monte Carlo simulation algorithm, known as generalized Subset Simulation (GSS), which introduces the concept of a design response vector to explore failure domains of different designs simultaneously. The approach determines feasible design domains for different geotechnical limit states by a single GSS run, providing significant computational savings and linking MCS-based full probabilistic design to semi-probabilistic design through back-calculating resistance factors based on failure samples.