Seismic structural reliability using different nonlinear dynamic response surface approximations

Performance-based design in earthquake engineering should be approached as an optimization process for optimum design parameters, in order to achieve satisfactory performance over the service life. Each of the specified performance criteria should be met with a prescribed minimum reliability and, given these constraints, a minimum or optimum total cost may be sought. The reliability estimations involve a nonlinear analysis for the dynamic responses of the structure, calculated via a step by step procedure over the complete earthquake record. This task could be computationally demanding, making unfeasible the direct implementation of a standard Montecarlo simulation. Dynamic responses represented by response surfaces make the simulation and the optimization process much more efficient. This paper presents a comparison between three methods for the implementation of response surfaces: a global approximation of a deterministic database, local interpolation of that database, or using artificial neural networks. The comparison uses, as an example, a 5-story reinforced concrete building. The results show good agreement between the methods and the paper discusses their corresponding advantages and limitations.