The aim of this paper is to predict web-post buckling shear strength of cellular beams made from normal strength steel using the Artificial Neural Networks (ANN). 304 developed finite-element numerical models were used to train, validate and test 16 different ANN models. To verify the accuracy of the ANN model, the ANN predictions were compared with experimental and analytical results. Results show that ANN models that used geometric parameters as an ANN input were able to predict web-post buckling strength to a higher level of accuracy in comparison to models using only geometric ratios as an ANN input. An ANN-based formula with 4 neurons was proposed in this study. In comparison to existing design guidance, it is shown that an ANN model trained with the Levenberg-Marquardt backpropagation algorithm is capable of predicting the web-post shear resistance to a higher level of accuracy. The formula developed can be easily implemented in Excel or in user graphical interface. It can be a potential tool for structural engineers who aim to accurately estimate the web-post buckling of cellular steel beams without the use of costly resources associated with FE analysis.

本文的目的是使用人工神经网络（ANN）来预测由普通强度钢制成的蜂窝状梁的网柱屈曲抗剪剪切强度。使用304个开发的有限元数值模型来训练，验证和测试16种不同的ANN模型。为了验证ANN模型的准确性，将ANN预测与实验和分析结果进行了比较。结果表明，与仅使用几何比率作为ANN输入的模型相比，使用几何参数作为ANN输入的ANN模型能够将网柱屈曲强度预测到更高的准确性。在这项研究中提出了一个基于神经网络的具有4个神经元的公式。与现有的设计指南相比，结果表明，采用Levenberg-Marquardt反向传播算法训练的ANN模型能够以较高的精度预测网柱抗剪强度。开发的公式可以在Excel或用户图形界面中轻松实现。对于结构工程师而言，它可能是一个潜在工具，旨在准确地估计蜂窝状钢梁的腹板后屈曲，而无需使用与有限元分析相关的昂贵资源。