Machine Learning-Driven Optimization of Fiber-Reinforced Concrete Mix Design for Enhanced Flexural Performance in Precast Structural Beams, A systematic Review

https://doi.org/10.5281/zenodo.20279554

Abstract

Fiber-reinforced concrete (FRC) has emerged as a critical material for precast structural beam elements, yet its flexural performance is highly sensitive to mix design parameters, fiber characteristics, and curing conditions. This systematic literature review aims to synthesize current knowledge on machine learning-driven optimization of FRC mix designs specifically targeting enhanced flexural behavior in precast beams. We systematically collected and analyzed peer-reviewed studies published over the last two decades, focusing on the intersection of computational modeling, fiber reinforcement, and structural applications. Our methodology involved a structured search across major academic databases, followed by a rigorous screening process based on predefined inclusion criteria related to flexural testing, machine learning frameworks, and beam-scale validation. The review then categorizes the literature into eight thematic dimensions: predictive modeling of compressive and flexural strengths using neural networks and ensemble methods; the influence of fiber type, geometry, and hybridization on mechanical properties; sustainable formulations incorporating industrial by-products; performance under elevated temperatures; structural applications to beams, columns, and slabs; advanced hybrid algorithms and model interpretability; and emerging technologies such as 3D printing. Key findings indicate that artificial neural networks and random forest models consistently achieve high accuracy in predicting flexural strength, with fiber volume fraction and aspect ratio emerging as the most influential variables. Hybrid fibers, particularly combinations of steel and polypropylene, yield superior toughness and crack resistance. The conclusion highlights a growing trend toward integrating optimization algorithms—such as genetic algorithms and particle swarm optimization—with machine learning models to automate mix design while reducing trial batches. However, a notable gap exists in the transferability of these models from laboratory-scale specimens to full-scale precast beams, especially under dynamic or fire loading. This review provides a consolidated roadmap for researchers and engineers seeking to adopt data-driven methods for designing high-performance FRC in precast beam applications, thereby accelerating the transition toward more durable and resource-efficient infrastructure.