A novel machine learning method to investigate the web crippling behaviour of perforated roll-formed aluminium alloy unlipped channels under interior-two flange loading

Abstract

This study presents a novel machine-learning model using the eXtreme Gradient Boosting (XGBoost) tool, for assessing the web crippling behaviour of perforated roll-formed aluminium alloy (RFA) unlipped channels (both fastened and unfastened) under interior-two-flange loading. A total of 1080 data points were generated for training the XGBoost model, utilizing an elastoplastic finite element (FE) model that was validated against 30 experimental results from the literature. A comparison against the numerical failure loads was conducted, and it was found that the prediction accuracy of XGBoost model was 94%. When compared with Ramdom Forest and Linear Regression methods, it was found that the proposed XGBoost model performed better than both the before mentioned methods. The web crippling strengths obtained from the XGBoost model, tests, and finite element analysis (FEA) were utilized to evaluate the performance of current design rules from the American Iron and Steel Institute (AISI), Australian/New Zealand Standards (AS/NZS 1664.1; AS/NZS 4600:2018) and Eurocode (CEN 2007). It is shown that the current design rules are not reliable to predict the web crippling strength of perforated RFA unlipped channels. As a consequence of the parametric analysis, new web crippling strength and web crippling strength reduction factor formulae for perforated RFA unlipped channels were proposed. A reliability analysis was then conducted, which confirmed that the proposed equations are capable of accurately predicting the ITF web crippling strengths of perforated RFA unlipped channels.