Flexural behavior prediction of CFRP-strengthened circular CFST members via finite element method and artificial neural networks

Finite element (FE) models and artificial neural network (ANN) models are integrated to predict the flexural behavior of CFRP-strengthened circular concrete-filled steel tubular (CFRP-C-CFST) members. The FE models of flexural members are established based on the cohesive zone model (CZM) calibrated through CFRP-steel double lap shear (DLS) tests. These models successfully capture critical failure mechanisms, including CFRP debonding and fracture. The moment-deflection curves and longitudinal strain distributions of CFRP obtained from the FE models are compared with experimental results to validate their accuracy. Through systematic parametric analysis employing the validated FE model, 12 parameters governing member performance are quantitatively evaluated. These parameters encompass interfacial bonding properties, constituent material characteristics, and geometric configurations. Sensitivity analysis identifies nine dominant parameters, from which a comprehensive database of 1017 FE models is generated. Two multi-layered feed-forward back propagation network trained with the Bayesian Regularization (BR) algorithm are developed to predict the debonding strain and flexural capacity of CFRP-C-CFST flexural members, with corresponding closed-form analytical equations proposed. The prediction errors of the equations are within 13 % compared to experimental results. The MATLAB implementation code generated during this study are available in Mendeley Data at DOI: 10.17632/nr3ytkkp8k.1.