Neural Prediction of Buckling Capacity of Stiffened Cylindrical Shells

Authors

1 Assistant Professor, Qazvin Branch, Islamic Azad University, Qazvin, Iran

2 Graduate student, Qazvin Branch, Islamic Azad University, Qazvin, Iran

Abstract

Estimation of the nonlinear buckling capacity of thin walled shells is one of the most important aspects of structural mechanics. In this study the axial buckling load of 132 stiffened shells were numerically calculated. The applicability of artificial neural networks (ANN) in predicting the buckling capacity of vertically stiffened shells was studied. To this end feed forward (FF) multi-layer perceptron (MLP) neural networks were used. The shell height (H), shell diameter (D), shell thickness (t), width (b), thickness of stiffeners (ts), and unstiffened length of the cylinder (l) were considered as the input vector; the relative buckling capacities of stiffened shells to unstiffened ones (Pcr/Pcr0) were considered as the output of the MLP networks. The back propagation algorithm was used to train the networks. Different structures of ANN were trained by 100 of the outputs of the numerical analysis and tested by 32 remains. The best ANN structure was selected based on the mean square error (MSE) and correlation coefficient (R2). The results of this study revealed that MLP neural networks can remarkably predict the axial capacity of stiffened shells. A parametric study on the buckling loads of the stiffened shells was performed by using the neural prediction and FEM.

Keywords