A case study on the practicability of using linear analysis results in a Bayesian inference model to predict nonlinear responses in performance-based design methods

Compared to traditional methods based on mean response evaluation of seismic parameters with significant confidence margin, the growing use of the new generation of performance-based design methods, which are based on loss and financial assessment, necessitates an increase in accuracy and reliability in probabilistic evaluation of structural response for all values of seismic parameters. Even with the same limited number of common nonlinear analyses, utilizing the Bayesian approach, which allows the use of diverse and even inaccurate data to form beliefs, is a powerful method to predict and enhance seismic response results. In this paper, the practicability of using linear analysis data in a Bayesian inference model to predict nonlinear responses is evaluated. A 20-story reinforced concrete special moment resisting frame is being considered, and a Bayesian model for prediction of the maximum story drift and the peak floor acceleration has been investigated. The Bayesian model was developed on linear results and finally updated with a limited number of nonlinear results. The predictability power of predictors, Bayesian model comparison among different likelihood functions, and common diagnostics tools in numerical solution of the Bayesian model developed on linear results, have all been examined. The results demonstrate a significant improvement in the outcomes, while proving the practicability of developing a stable and reliable model based on linear analysis data.