Artificial neural networks prediction of inelastic displacement demands for structures built on soft soils

Abstract

Displacement is a better indicator of building damage than force; therefore, performance-based seismic design methodologies aim at controlling damage through limiting lateral inelastic displacement of structures. This study aims at predicting the inelastic displacement demand of structures through utilizing artificial neural networks and Bayesian regularization algorithm. A large set of inelastic displacement ratios was computed using nonlinear time history analyses of single-degree-of-freedom oscillators subjected to ground motions recorded on soft soils. Four types of hysteretic models were utilized to cover a wide range of structures, namely flag-shaped, elastic-perfectly plastic, small-Takeda, and large-Takeda hysteretic models. The inelastic displacement ratios were correlated with displacement ductility and structure period normalized to ground motion record predominant period. The study offered insight on inelastic displacement ratios of structures built on soft soils, the effect of structure period, ground motion predominant period, ductility level, and hysteretic model. The inelastic displacement ratios exceeded one for periods less than the record predominant period, while for periods more than 1.5 the predominant period, it was almost equal to one. Furthermore, in the vicinity of the ground motion record predominant period, inelastic displacement demands are smaller than elastic displacement demand. Dispersion of inelastic displacement ratio increases as the inelastic displacement demand increases, while larger dispersion was evident for structures with period ranges between 1.5 and 2.0 times the ground motion predominant period. Finally, the study provided the necessary parameters needed for artificial neural network models to regenerate inelastic displacement ratio models needed in designing structures built on soft soils.