A novel machine learning model to predict the moment capacity of cold-formed steel channel beams with edge-stiffened and un-stiffened web holes

A novel machine learning model, eXtreme Gradient Boosting (XGBoost), was used for the purpose of evaluating the moment capacity of cold-formed steel (CFS) channel beams with edge-stiffened web holes subject to bending. A total of 1620 data points were generated for training the XGBoost model, using an elasto-plastic finite element model which was validated against 12 sets of test data taken from the existing literature. The R2 score of XGBoost predictions for the moment capacity was around 99%. The performance of current design equations was evaluated through the comparison of their results against those obtained from the XGBoost model. The moment capacities obtained from the XGBoost testing dataset were also compared with that determined from the existing design equations for un-stiffened holes (USH) and edge-stiffened holes (ESH). The moment capacities determined from the current design equations for USH and ESH were found to be excessively conservative by 38.3%, and unconservative by 36.2% on average, respectively. Therefore, new design equations were proposed based on the results of parametric study using the XGBoost model. In the detailed parametric analysis, the effects of web depth, section thickness, and beam length on the moment capacity of channel beams (CFSCB) with ESH were considered. From the results of XGBoost outputs, the absolute percentage error of new design equations for that based on the strengths of unperforated CFSCB was 8.78%, and for that based on the strengths of CFSCB with USH, the absolute percentage error was 13.7%. Additionally, a reliability analysis was performed to evaluate the accuracy of the proposed equations that were used to predict the moment capacity of CFS channel beams with ESH subject to bending. The reliability indices of all the proposed equations were greater than 2.5 which can be reliable as per the guidelines of AISI.