Machine learning models for the elastic-critical buckling moment of sinusoidal corrugated web beam

Abstract

The torsional stiffness of I-beams with sinusoidal corrugated web is higher than that of flat web beams and the accuracy of the available hand-calculation methods to determine the elastic critical lateral-torsional buckling moment depends on the geometrical parameters of the beam and the web corrugation. This study proposes different machine learning models to determine the elastic lateral-torsional buckling moments of corrugated web beams. Various machine-learning algorithms such as Decision Tree, Random Forests, Gradient Boosting, Support Vector Regression, Catboost, and Deep Neural Network were employed to develop and train for predicting the elastic-critical lateral-torsional buckling moments of I-beams with corrugated web. An extensive dataset with 2250 pieces was constructed using linear buckling analyses on full-shell finite element models to determine the elastic-critical buckling moment of simply supported beams with sinusoidal web corrugation. Based on the statistical parameters of the predicted and test data, the accuracy and safety assessment of the different machine learning models are examined. The accuracy of the available hand-calculation methods is also investigated. The results of the parametric study showed that the overall performance of the different machine learning models is promising, although, not all are directly suited for the described problem.