Calculation parameter correction of steel truss nodal plate based on machine learning theory

Abstract

To improve the accuracy of deformation calculations in the finite element modeling of steel trusses, a method is proposed to modify the rigid arm by accounting for the stiffness of the nodal plate. A digital twin is established for each control parameter of the nodal plate, and the rigid arm lengths are adjusted to achieve displacement matching with a high-precision plate element model. A Bayesian decision tree and an artificial neural network (ANN) are then used to map the relationship between the twin parameters and the rigid arm length coefficients, respectively, to correct for stiffness. The reliability of these two machine learning methods was confirmed through mutual validation, with results showing that the ANN produced more accurate rigid arm length coefficients, while the Bayesian decision tree performed slightly less well. Both methods, however, successfully overcome the limitations of traditional beam elements, which do not account for nodal plate stiffness. Finally, large-scale bridge engineering examples using steel trusses demonstrate that correcting the rigid arm length significantly improves the calculation accuracy of commonly used beam elements. This approach also reduces the complexity of beam-plate coupling, making the modeling process more convenient, faster, and efficient.