Machine Learning-Driven Closed-Loop Optimization for the Accelerated Design of Sustainable Cements Utilizing Algal Biomatter

The high embodied carbon of cement, along with the growing demand for construction materials, drives the need for more sustainable cementitious materials to mitigate climate change. In contrast to the solutions relying on inorganic supplementary cementitious materials (SCM) produced by carbon-intensive industries, utilizing carbon-negative biological materials is an emerging strategy to reduce CO₂ emissions ascribed to this class of material. Designing novel green cement, however, is challenging due to the complex, nonlinear hydration-strength relationships and the combinatorial nature of the large design space. Here, leveraging machine learning, we develop a closed-loop optimization strategy to accelerate the design of green cement with minimal CO₂ emissions while meeting a strength criterion. We introduce a novel approach that distinguishes itself from existing machine learning strategies, which rely on large pre-existing databases, by utilizing real-time collected data for optimization. As a demonstration, green cements incorporating carbon-negative and locally sourced macroalgae are tested in real-time, with varying macroalgae concentration, water-cement ratio, macroalgae particle sizes, and curing humidity, to predict strength evolution using an amortized Gaussian process (aGP) model while an early termination criterion is applied to accelerate the optimization process. Using only 28 days of experiment time and 40 tested formulations, this approach achieves both a 30.8 MPa strength requirement and a 93% achievable improvement in global warming potential (GWP), resulting in a 21% reduction in cement's GWP. We further validate the model-informed hydration kinetics via analyses in thermal gravimetric analysis (TGA), scanning electron microscopy (SEM), and X-ray diffraction (XRD). This result highlights the potential of our design framework for developing novel materials with machine learning while enabling scientific understanding.