Variance‐Based Sensitivity of Seismic Damage of the Containment Building Using Efficient Bayesian Additive Regression Trees

ABSTRACT In nuclear power plants (NPPs), reactor containment building (RCB) is an important structure due to its scope of use. The RCBs of NPP in Korea have already been in operation for decades. Risk, reliability, and fragility assessments are required for safety inspection and maintenance during service life. All these factors are highly dependent on the damage states of the structure that vary due to the uncertainty of the input material parameters. The uncertainties are induced due to workmanship and quality control issues and can result in incorrect estimation of damage states. This motivates the importance of identifying how sensitive the input parameters are to the RCB damage states. Pushover analysis is used to estimate the damage states (DS1 as concrete cracking, DS2 as steel yielding, and DS3 as concrete crushing). An intermittent surrogate model (Bayesian additive regression trees [BART]) is constructed since sensitivity analysis requires many simulations and is computationally intensive for the RCB structure. The performance and stability of the BART model are investigated. Finally, Sobol's sensitivity indices are estimated employing the constructed BART model. Concrete's tensile strength is found to be dominant for DS1, while the yield strength of steel is found to be most sensitive for the case of DS2 and DS3.