Experimental Investigation of the Compressive Behavior of Hybrid Fiber-Reinforced Concrete and AI-Agent-Assisted Modeling

Abstract

Hybrid fibre reinforcement is increasingly used to improve the mechanical performance and failure behaviour of concrete by combining the complementary characteristics of different fibre types. However, the effects of different hybrid fibre combinations on compressive behaviour and mesoscale failure mechanisms, particularly when integrated with AI-assisted discrete element modelling, remain insufficiently understood. This study investigated the compressive behaviour and mesoscale failure mechanisms of concrete reinforced with steel, macro-synthetic, and polypropylene fibres, individually and in hybrid combinations, and evaluated an AI-agent-assisted PFC3D modelling workflow. Thirteen mixture groups comprising 39 cubic specimens were subjected to uniaxial compression testing. Steel-fibre concrete achieved a compressive strength of 39.7 MPa compared with 35.0 MPa for plain concrete, whereas macro-synthetic fibre concrete reached 35.7 MPa and polypropylene fibre concrete decreased to 25.0 MPa. Fibre incorporation generally changed failure from localised crack coalescence to more distributed cracking and improved post-peak specimen integrity. Three representative mixtures were modelled using PFC3D with an AI agent supporting task generation, script execution, feedback-based correction, and result recording. Simulated peak strengths of 34.2, 37.3, and 26.5 MPa for the N0, SM, and MP models, respectively, were within 5% of the corresponding experimental values. AI-agent-assisted modelling reduced the reported total modelling time from 142 to 34 min, corresponding to an efficiency improvement of 76.1%. The first-run task success rate was 61.1%, while all staged tasks were completed after correction. Repeated simulations produced low coefficients of variation for peak stress and crack number. The findings indicate that steel-macro-synthetic hybridisation provided comparatively favourable compressive performance, while the AI-assisted workflow supported efficient and reproducible discrete element modelling.