Predictive modelling of hybrid composite laminates buckling behaviour using Finite Element Analysis, refined Response Surface Methodology and Artificial Neural Network models with different data sizes

Accurate prediction of buckling loads in composite structures is essential, as their anisotropic and inhomogeneous properties complicate structural analysis. However, traditional predictive models often face limitations such as the inclusion of statistically insignificant polynomial terms in Response Surface Methodology (RSM) and poor learning performance in Artificial Neural Networks (ANN) due to unprocessed and limited data sizes. This study aimed to develop and evaluate predictive models for the buckling load of hybrid graphite/glass epoxy composite laminates using different data sizes. Two datasets were employed, comprising 27 runs generated through a Full Factorial Design (FFD) under the Design of Experiment (DOE) approach and 100 customised experimental runs. Two modelling approaches, RSM and ANN, were employed to predict the buckling load obtained from finite element analysis (FEA). The overall range of computed buckling loads was wide, spanning from 3.627 kN to 1730.8 kN, confirming the strong sensitivity of the structure to the design variables. The highest buckling loads occurred at [45, 1, 3 mm] (angle, volume fraction, thickness), and for hybrid laminates at [45, 0.5, 3 mm]. The RSM predictions produced ratios close to one when compared with FEA results, while the ANN models showed both underprediction and overprediction tendencies. The t-test results indicated no statistically significant difference between the 27 and 100 experimental runs, suggesting that model accuracy was influenced more by modelling approach and data treatment than dataset size. This study may contribute to enhancing knowledge of the buckling behaviour and failure of hybrid graphite/glass composite structures, which will help engineers design safer structures by reducing the risk of buckling.