Web crippling resistance of cold-formed steel built-up box sections through experimental testing, numerical simulation and deep learning

Abstract

This study proposes a deep-learning framework, namely a deep belief network (DBN), for investigating the web crippling resistance of cold-formed steel (CFS) built-up stiffened box sections. A total of 12 new web crippling tests are presented initially, which were carried out under interior-two-flange (ITF) and end-two-flange (ETF) loading conditions. In the experimental tests, two cross-section dimensions were considered, and the overall web depth varied from 290 mm to 550 mm. The yield strength of the specimens varied from 371.7 MPa to 643.6 MPa. The specimens were characterised by different screw spacing, varying between 50 mm and 300 mm. An elasto-plastic finite element (FE) model was developed and validated against the test results. Subsequently, a total of 1,356 data points for training the DBN were generated based on the validated FE model. When a comparison was made against the test results, the absolute percentage error of the DBN predictions was found to be around 10%. Using the DBN predictions which were based on the validated FE model, a comprehensive parametric study was undertaken to investigate the effects of section thickness, yield strength, bearing length and screw spacing on the web crippling resistance of CFS built-up stiffened box sections. New design equations for predicting the ITF and ETF web crippling resistances of CFS built-up stiffened box sections were then proposed, and a reliability analysis was undertaken to assess the feasibility of the proposed equations.