Practical ANN Model for Estimating Buckling Load Capacity of Corroded Web-Tapered Steel I-Section Columns

This study develops an artificial neural network (ANN) to estimate the critical buckling load (CBL) of corroded web-tapered steel I-section (WTSI) columns in pre-engineered steel buildings. A total of 387 datasets are employed to develop the ANN model. The datasets are generated from the proposed analytical model and Newton–Raphson method. The input parameters of the developed ANN model contain the cross-sectional dimensions of the steel column (i.e., the top and bottom flange width, top and bottom flange thickness, maximum section height, minimum section height, and web thickness), elastic modulus of material, and the column height. Meanwhile, the CBL is the output parameter of the ANN model. A predictive process for the CBL of the corroded WTSI columns has been proposed based on the ANN model and previous corrosion model. Results reveal that the ANN model showed an excellent performance in predicting the CBL of the corroded steel columns. The R 2 values of the training, testing, and validation data are 0.99975, 0.99916, and 0.99951, respectively. The root-mean-squared errors of the training, testing, and validation data are 96.705 kN , 103.402 kN , and 103.200 kN , respectively. Additionally, the a20-index is very close to 1.0. Moreover, a graphical user interface tool is constructed to facilitate the CBL calculation of the corroded WTSI columns.