Prediction of critical buckling load in pultruded composite columns using a new hybrid finite element–GMDH neural network model

This study presents a predictive model based on the finite element–Group Method of Data Handling (GMDH) neural network to estimate the critical buckling load of pultruded composite columns, explicitly accounting for the effects of fiber orientation. Composite column models with different cross-sections are developed using the Design of Experiments (DOE) technique to efficiently generate an optimized input dataset for neural network training. Laminate layups with varying fiber orientation angles are modeled using the finite element method, and their effects on the critical buckling load, layer damage, and column interaction curves, which define the load-carrying capacity of the columns, are investigated. The proposed model establishes a new predictive relationship between fiber orientation and the critical buckling load of composite columns by employing predictive algorithms based on a neural network. The novel aspect of the proposed model is that it enables the evaluation of different fiber orientations without the need for additional numerical simulations or costly experimental tests. The results indicate that closed-section composite columns can increase the critical buckling load by approximately 40%. Furthermore, validation against experimental data shows an average error of approximately 9.8% in the predicted critical buckling loads, demonstrating the accuracy of the proposed GMDH neural network model.