Design of sustainable mortar incorporating construction and demolition waste through adaptive experiments accelerated by machine learning

Abstract

Traditional approaches to designing sustainable construction materials are slow, resource-intensive, and heavily reliant on trial and error. In this study, we present an integrated framework that combines experiments, machine learning (ML), and Bayesian optimization (BO) to drastically accelerate the materials design. A key component of the framework is an ML model that predicts 28-day compressive strength from mixture composition and early-age mechanical properties, which allows nearly 10x acceleration of classical BO-driven design cycles. We demonstrate the potential of this new framework on the designing sustainable mortars containing construction and demolition waste with minimal ordinary Portland cement (OPC) contents and late-age compressive strength satisfying standard requirements. Following the model guidance, we predict, fabricate, and test new mixtures with a Image 1 lower OPC content and compressive strength above Image 2. Experimental fabrication and testing of the optimized mixtures confirmed the accuracy of the ML predictions while correlations between mixture components and compressive strength agree with published literature.

Keywords

• Accelerated Bayesian optimization
• Machine learning
• Construction and demolition waste
• Sustainability