Optimizing the influence of fiber orientation on the mechanical properties of ultrahigh‐performance concrete via data augmentation and machine learning

Abstract Ultrahigh‐performance concrete (UHPC) exhibits outstanding mechanical properties, with the morphology and orientation of steel fibers playing a decisive role in tensile strength and ductility. This study integrates experiments, finite‐element simulations, and machine learning analyses to investigate the influence of fiber orientation on UHPC performance and proposes an optimization strategy using data augmentation with a Wasserstein generative adversarial network with gradient penalty (WGAN‐GP). Experimental results show that short, smooth straight steel fibers oriented at 30°–45° effectively enhance tensile strength and ductility, while corrugated (wave‐shaped) fibers provide superior reinforcement at 45°–60°. Finite‐element simulations corroborate these trends, indicating an optimal fiber orientation range of 45°–60° for maximizing load‐bearing capacity. Using an Extreme Gradient Boosting model, WGAN‐GP augmentation enables highly accurate predictions, achieving a nearly perfect fit ( R 2 ≈0.997). Building on these findings, we propose an electromagnetic fiber‐orientation device to tailor fiber distribution, thereby improving both strength and toughness. Overall, the results provide theoretical support and practical guidance for UHPC mix design and manufacturing process optimization.