The design of earthquake-resistant buildings involves economic and safety evaluations, which are strictly related to the probability of occurrence of exceptional events in the considered geographical area. When high-seismicity countries are involved, exploiting extreme resources of structural materials, as their ability to deform plastically, can be a valid compromise between economic saving and people safety. In a world that is increasingly interested in sustainable choices, an optimal design of structures is becoming a mandatory need. Artificial Intelligence may be of great help for this purpose. This paper presents a cost-effective procedure based on the inversion of a Multilayer Perceptron Artificial Neural Network to obtain optimal design parameters for earthquake-resistant buildings. The study refers to three-dimensional multi-story concentrically braced steel frames with active tension diagonal bracings, which are assumed to be the dissipative elements of the structure. The data relevant to a set of cases obtained by varying four design parameters made the input matrix. Three earthquakes consistent with the Chilean code spectrum were selected to carry out nonlinear dynamic analyses over all the cases, which led to obtain the output dataset collecting the capacity-design performances corresponding to the combinations of design parameters. The trained network is inverted through to solve the inverse problem and an optimization algorithm, based on minimizing the total cost, is finally adopted to obtain the optimal design parameters. The final numerical checks proven the effectiveness of the proposed optimization method.
Javier et al. (Mon,) studied this question.