Application of deep learning method in web crippling strength prediction of cold-formed stainless steel channel sections under end-two-flange loading

This paper proposes a deep-learning framework, specifically, Deep Belief Network (DBN), for studying the web crippling performance of cold-formed stainless steel channel sections (lipped and unlipped as well fastened unfastened) with centered offset holes under end-two-flange loading condition. G430 ferritic, S32205 duplex 304 austenitic grades are considered. A total 17,281 data points training DBN generated from an elasto plastic finite element model, validated 69 experimental results reported in literature. When comparison was made against further 53 literature, predictions were found to be conservative by around 10%. compared Backpropagation Neural (a typical shallow artificial neural network) linear regression model based on PaddlePaddle, it that proposed outperformed these two methods, using same big this study. Using predictions, parametric study is then conducted investigate effect holes, which unified strength reduction factor equations proposed. Finally, reliability analysis conducted, shown can predict