AI Framework Delivers 96.5% Accurate Geopolymer Concrete Mix Designs

Context Portland cement accounts for a large share of global CO₂ emissions, prompting the industry to explore low‑carbon alternatives. Geopolymer concrete (GPC) replaces cement with industrial by‑products such as fly ash and ground‑granulated blast‑furnace slag, activated by alkaline solutions. Designing GPC mixes is complex; dozens of variables interact to determine strength, curing temperature, and durability. Traditional development relies on costly, time‑consuming trial‑and‑error.

Key Facts Researchers built a two‑stage AI system using 820 literature‑derived mix records. In the first stage, three predictive models estimated compressive strength from inputs like precursor ratios, alkaline molarity, and curing temperature. The genetic‑algorithm‑optimized XGBoost model achieved a root‑mean‑square error of 2.882 MPa and a mean absolute error of 1.905 MPa, explaining 96.48% of the variance (R² 0.9648). A TabTransformer model also performed well, reaching R² 0.9453, confirming the value of self‑attention for tabular material data.

In the second stage, a fine‑tuned generative large language model (based on Facebook’s OPT‑350M) produced human‑readable mix designs for a target strength. Text‑quality metrics showed a BERTScore of 0.9754 and a ROUGE‑L score of 0.8794, indicating strong alignment with reference recipes. Component‑level predictions for fly ash, slag, and coarse aggregates yielded R² values above 0.985, confirming chemical validity.

What It Means The framework shifts mix design from iterative experimentation to rapid, data‑driven synthesis. Engineers can input a desired compressive strength and receive a ready‑to‑test recipe, reducing material waste and laboratory cycles. By accelerating GPC adoption, the approach could lower construction‑sector emissions, especially where high‑temperature curing (≈65 °C) enables 70% strength gain within 24 hours.

Looking Ahead Future work will test the generated mixes in field trials and expand the model to incorporate durability and lifecycle‑assessment metrics, paving the way for broader low‑carbon concrete deployment.