Buckling critical load prediction of pultruded fiber-reinforced polymer columns and feature analysis by machine learning

For slender FRP columns, predicting the global buckling critical loads is crucial in structural design. However, there is a lack of a consensus prediction method based on specialized domain knowledge. To address this issue, this study created a comprehensive database by collecting 365 experimental data related to global buckling of axially loaded pultruded FRP columns to predict buckling critical loads using such machine learning methods as extreme gradient boosting, artificial neural network, and support vector regression. The prediction accuracy and stability of the machine learning prediction methods were evaluated, and the interpretability of the features was analyzed in depth. The results show that the prediction accuracy of the traditional theoretical methods is low, while that of the machine learning methods is high. The contribution of geometric parameters to the buckling critical load is more than 80%. The contribution of material parameters to the buckling critical load is small, less than 20%. The cross-sectional moment of inertia has the most significant effect on the buckling critical load, while the shear modulus and compressive strength have a smaller effect.