Development of the approximate analytical model for the stub-girder system using neural networks

Abstract

This paper presents the methodology to develop a neural-networks-based model for approximate structural analysis. The approach is verified by modeling a stub-girder system to predict its behavior. The development of approximate analytical model by using neural networks is studied in an effort to find a method to efficiently generate credible outputs that are consistent with those by Vierendeel truss girder and finite element models. The criteria and rules for determining the neural network architecture are described using the performance evaluation with eight architecture design aspects and two subject variables to be compared. The criteria of accuracy are mostly focused because one of the objectives of an approximate analysis is to provide results that are within an allowable error.