Physics-Informed Artificial Neural Network Approach for Axial Compression Buckling Analysis of Thin-Walled Cylinder

Abstract

Listen

Translate article

Unstable buckling leads to the failure of thin-walled cylinders under axial compression. Because of the manufacturing and loading imperfections, a significant variation was observed between experimental and theoretical buckling load. The current design criteria use a knockdown factor to estimate the buckling load of thin-walled structures. However, the design criteria employed a very conservative knockdown factor, because it is difficult to be accurately predicted. To improve the certainty of the knockdown factor, in this paper, a physics-informed artificial neural network (PANN) was employed to predict the thin-walled cylinder buckling load using experimental data collected by Seide. PANN has the potential to improve prediction and enforce some physical laws. A beam problem was presented to demonstrate the PANN capability. Then, the cylinder buckling analysis was carried out with PANN. An inequality constraint was applied to force the neural-network-predicted buckling load to be equal or smaller than the corresponding experimental data. The result shows that PANN can reach excellent prediction accuracy and is able to predict the buckling load equal to or smaller than the corresponding experimental buckling load. Additionally, both ANN and PANN are able to take the unique influential parameters as inputs and do not have presumed expression and thus have the potential to yield a less conservative prediction than the statistical equation.