Statistical evaluation of the influence of manufacturing tolerances on the flexural cyclic behavior of I‐shaped hot rolled steel beams

This paper investigates the influence of manufacturing tolerances on the cyclic behavior of steel beams with European I‐shaped profiles. It also explores the applicability of machine learning methods in predicting this behavior. Previous research has focused on nominal sections, overlooking the effects of dimensional tolerances specified in the EN10034 standard. Through advanced finite element modeling, the study evaluates the influence of geometrical variability on the strength and deformation capacity of steel beams under cyclic flexural loading. A parametric study with Latin Hypercube sampling generated two thousand samples of various profiles and lengths, revealing that dimensional tolerances can lead to variation in overstrength ratio or reduced ductility. Finally, the paper assesses the accuracy of several machine learning models in predicting moment and rotation capacity, including nonlinear and linear regression analysis, neural network, decision tree, and random forest.