Reliability sensitivity analysis for RBSMC: A high-efficiency multiple response Gaussian process model

Abstract

In modern architectural design, reduced beam section moment connections (RBSMC) play a crucial role. However, quantifying the importance of random input variables and failure modes has not been performed to improve RBSMC reliability. Thus, we propose a sensitivity study that evaluates the effect of geometric dimensions on the reliability of the RBSMC. The analysis measures the significance of each random input variable on the reliability of the RBSMC. Then, a reliability predictor, which combines the multiple response Gaussian process (MRGP) with the Monte Carlo simulation (MCS), is introduced to efficiently produce surrogate models for failure surfaces. Based on the well-established surrogate models, the reliability is estimated without additional calling of the original limit state functions. The quadratic regression and analysis of variance are conducted to statistically evaluate the contribution of different variables to the reliability of the RBSMC. The proposal methodology accurately predicts reliability and identifies variables that significantly contribute to reliability. The results show that the proposed methodology is more efficient than existing methods at the early design stage. Furthermore, we applied this methodology to a radius cut RBSMC, which can minimize stress concentration and reduce the possibility of fracture to identify significant input variables for reliability. The computational results and discussions provide useful insights for the design of RBSMC.