Prediction of the deformation and local buckling behavior of structural systems using the deep neural network direct stiffness method (DNN-DSM)

This paper presents a novel approach to carrying out a beam-element analysis that accounts for the nonlinear deformation behavior of various RHS and SHS sections, ranging from mild to highstrength steel prefabricated by hot-rolling or cold forming. The deep neural network direct stiffness method (DNN-DSM), which makes use of deep neural networks (DNN), a subgroup of machine learning algorithms and more general artificial intelligence approaches, is used to predict the nonlinear stiffness matrix terms in a beam-element formulation for the implementation in the direct stiffness matrix (DSM). Those predictions are made by trained DNN models resulting from an extensive pool of geometrically material nonlinear simulations with additional imperfections (GMNIA) using shell based models. First implementations of this method are able to accurately predict the nonlinear load-displacement and moment-rotation behavior of various sections with great accuracy, combining the precision of shell analysis and the computational efficiency of beam element analysis. Previous published investigations already showed the feasibility and advantages of this method but were restricted to the small-scale prediction of the local buckling. This paper will go one step further and apply the DNN-DSM method to members, more specifically to columns and beams dominated ether by normal forces or bending, accompanied by comparisons with equivalent Abaqus models.

View Content