A machine learning-driven prediction of lower-bound buckling design load for cylindrical shells under localized axial compression

Abstract

Thin-walled cylindrical shells are extensively used across various fields because of their exceptional load-carrying efficiency. In practical applications, these structures are typically subjected to localized axial compression rather than the uniform axial compression considered in traditional research. A reliable and efficient buckling design method for cylindrical shells under such localized loads has not been developed to date. To address this challenge, a machine learning (ML) approach is proposed in this study for predicting the lower-bound buckling design load of cylindrical shells under localized axial compression. The artificial neural network (ANN) is selected as the ML model. Based on the modified energy barrier approach (MEBA), 500 samples are obtained by numerical simulations and their results are used to train the ANN model. The ANN model takes six geometric parameters, three material parameters, and one localized axial compression parameter as the inputs, while the lower-bound buckling load and the knockdown factor are the outputs. The feasibility and accuracy of the proposed ANN model are demonstrated by comparison with existing design codes and experimental results. The results suggest that this ML-based approach can fully exploit the load-carrying capacity of shells under localized axial compression, enabling more efficient and lightweight designs.