Global flange buckling strength prediction of I-stiffened laminated composite FRP box-beams using ANN

Abstract

The utilisation of stiffened fibre-reinforced polymer (FRP) box-beams is highly advantageous for lightweight bridge structures; however, their design is computationally intensive due to anisotropy and buckling sensitivity. This study presents a computationally efficient artificial neural network (ANN) model for predicting the global flange buckling strength of stiffened laminated composite FRP box-beams subjected to lateral loading. A comprehensive database was generated using validated finite element analysis (FEA) considering fibre orientation, stiffener geometry, and orthotropic stiffness effects. Five governing parameters (αsf, CEIw, D1/D2, (EA)fs/(EA)fp, and βsf) derived from laminate theory and stiffened panel mechanics were used as ANN inputs. Among the evaluated models, the Bayesian regularised ANN achieved an average prediction error of 3.01% and a maximum error of 4.77%, while reducing computation time from approximately 300 s (FEA) to less than 0.01 s per prediction. The proposed model provides a reliable surrogate tool for design optimisation of composite box-beams.