Predicting Flexural Capacity of Ultrahigh-Performance Concrete Beams: Machine Learning–Based Approach

Abstract

Despite ongoing research efforts aimed at understanding the structural response of ultrahigh-performance concrete (UHPC) beams, there are very limited provisions for structural design of UHPC beams, specifically under flexure. To address the current limitations in design provisions for UHPC beams, which notably hinder the applicability of UHPC in concrete structures, it is essential to develop simplified design methodologies that can optimize the superior range of properties offered by this new class of material. In this study, a machine learning (ML)–based approach for predicting the flexural capacity of UHPC beams was developed. ML algorithms such as support vector machine regression (SVMR) and multigene genetic programming (MGGP) were applied to analyze data and observations collected from large set of tests on UHPC beams with different geometries, fiber properties, and material characteristics. SVMR was applied to predict the flexural capacity of UHPC beams, whereas MGGP was employed to derive a simplified expression, through data-driven analysis of actual test data, for predicting the flexural capacity of the beams. A parametric study was also performed to validate the proposed approach with varying input parameters. The results indicated that the proposed data-driven approach using ML was effective in predicting the flexural capacity of UHPC beams with different material and configuration characteristics, paving the way for the development of guidelines for the design of UHPC beams. The results of this comprehensive analysis showcase the merit of employing ML in structural engineering applications and thereby promoting the applicability of UHPC in concrete structures.