Small-sample artificial neural network based response surface method for reliability analysis of concrete bridges

In the paper, an artificial neural network based response surface method (ANN-RSM) in combination with a small-sample simulation technique is proposed. ANN as powerful parallel computational system is used for approximation of limit state function (LSF). Thanks to its ability to generalize it is efficient to fit LSF even with small number of simulations compared to polynomial RSM. Efficiency is emphasized by utilization of stratified simulation for selection of ANN training set elements. Proposed method is tested using simple limit state function taken from literature as well as employed for reliability and load-bearing capacity assessment of concrete bridge within the framework of fully probabilistic analysis. Results are compared with those obtained by other reliability methods.