Hybrid AI-based prediction of mechanical properties of sustainable high-strength natural-fibre-reinforced concrete

In this work, plant fibres (hemp, sisal and ramie fibres) and human hair were incorporated into high-strength concrete (HSC) to assess their suitability as reinforcement materials. The mechanical properties of the high-strength natural-fibre-reinforced concretes (NFRCs) were experimentally evaluated and compared with those of conventional concrete. However, the inherent variability of natural fibres means it is difficult to predict complex mechanical properties, so advanced modelling techniques are required. To address this challenge, a hybrid artificial intelligence (AI) model is proposed to accurately predict mechanical properties and assess the suitability of different natural fibres in HSC applications. The hybrid model is based on a temporal convolutional network enhanced with an attention mechanism. The model achieved a high predictive performance with R² = 0.993 for compressive strength, R² = 0.993 for split tensile strength and R² = 0.9969 for flexural strength. The model exhibited low prediction errors for these strengths, with root mean square errors of 0.6897 MPa, 0.0834 MPa and 0.0778 MPa, respectively, indicating strong predictive capability. Compared with conventional methods, the proposed approach reduced the prediction error by approximately 20–30%. The combined experimental and data-driven approach provides a reliable framework for predicting the mechanical properties of high-strength NFRC.