Machine learning-driven web-post buckling resistance prediction for high-strength steel beams with elliptically-based web openings

Abstract

The use of periodical elliptically-based web (EBW) openings in high strength steel (HSS) beams has been increasingly popular in recent years mainly because of the high strength-to-weight ratio and the reduction in the floor height as a result of allowing different utility services to pass through the web openings. However, these sections are susceptible to web-post buckling (WPB) failure mode and therefore it is imperative that an accurate design tool is made available for prediction of the web-post buckling capacity. Therefore, the present paper aims to implement the power of various machine learning (ML) methods for prediction of the WPB capacity in HSS beams with (EBW) openings and to assess the performance of existing analytical design model. For this purpose, a numerical model is developed and validated with the aim of conducting a total of 10,764 web-post finite element models, considering S460, S690 and S960 steel grades. This data is employed to train and validate different ML algorithms including Artificial Neural Networks (ANN), Support Vector Machine Regression (SVR) and Gene Expression Programming (GEP). Finally, the paper proposes new design models for WPB resistance prediction. The results are discussed in detail, and they are compared with the numerical models and the existing analytical design method. The proposed design models based on the machine learning predictions are shown to be powerful, reliable and efficient design tools for capacity predictions of the WPB resistance of HSS beams with periodical (EBW) openings.