Machine Learning Application of Structural Engineering Problems

Abstract

The cross-sectional properties of the basic structural system elements, such as columns and beams, are the basic structural design elements that need to be determined sensitively. For the optimum design of such structural system elements, it is necessary to minimize the displacement and volume by optimization. In this study, the design of a tubular column and I-section beam element has been optimized, and a section prediction model has been produced by the machine learning method, which has been successfully applied in the risk and damage detection of various engineering problems. For this purpose, optimum cross-section properties were determined for different load conditions with the Jaya Algorithm (JA), which is a metaheuristic algorithm. To minimize production errors arising from workmanship in the production of structural system elements, cross-section parameters are divided into classes covering certain dimensions. Different design combinations obtained by optimization were converted into a data set and training for machine learning was applied. With the trained data, a cross-section prediction model was produced that predicts the cross-sectional properties of column and beam samples on a class basis. When the results are examined, it is understood that the prediction models to be produced with the optimum design data are suitable for use in determining the cross-sectional properties of the structural system elements.