DEVELOPMENT OF FORMULAE FOR THE SECTION ROTATIONS DUE TO BENDING OF CURVED STEEL I-BEAMS THROUGH AI AND ML ALGORITHMS

Abstract

Currently, there is no standardized method nor a reliable and easy to implement analytical relationship for obtaining the section rotations of horizontally curved steel I-beams that could be specified in design guides and used by practicing engineers. Analysis of horizontally curved I-beams is especially difficult to perform using traditional methods because, even under the influence of gravitational loads, the beams experience a combination of primary flexure and non-uniform torsion that result in vertical deflections that are coupled with twist rotations. This research work proposes a systematic approach that was used to develop a simple analytical formula that could predict the sectional rotation of horizontally curved steel I-beams that were fully fixed at each end and subject to a vertical point load at their midspan. The development of such a formula had never been attempted before in international literature yet by combining data derived from 3D finite element analysis with a machine learning algorithm that makes use of higher-order nonlinear regression, a predictive formula was developed that could calculate the section's rotation of horizontally curved steel I-beams with a mean absolute error of just 1.44%.