Regression models for ultimate hogging moment prediction of composite cellular beams

Abstract

ABSTRACT This work shows the accuracy and feasibility of polynomial regression models for the ultimate hogging moment prediction of composite cellular beams utilising different numbers of simulations. It discusses an attempt to reduce the number of samples in a finite element parametric study, which consists of using only the combination of extreme levels of each factor to develop the regression models. The parametric study has five parameters: cross-section dimensions, hogging moment distribution, unbraced length, opening diameter, and web-post width. The regression models are generated based on 32 (extreme values), 162 (extreme and intermediate values), and 360 numerical model results (all values). An ANOVA was performed on each regression model to assess the parameters’ order of relevance. According to ANOVA, the profile cross-section dimensions and hogging moment distribution were the most significant parameters on the beams’ ultimate moment for all regression models. As good precision was achieved, because of the insignificant differences within the regression models, it was observed that the combination of arbitrated extreme levels for the factors could significantly reduce the number of numerical simulations in a parametric study. Still, this approach could be used for any other type of structure to reduce the computational cost of future numerical investigations.