Advanced parametric and machine learning techniques for reducing embodied carbon in concrete structures

To mitigate global climate impacts, effective strategies for reducing embodied carbon in building structures are imperative. This study presents a novel integration of parametric modeling with machine learning to predict and minimize embodied carbon in concrete structural frames. First, we develop a physics-based parametric model capable of estimating structural material quantities for reinforced concrete buildings based on design variables, such as grid size, occupancy type, and soil conditions. This model then facilitates the generation of numerous data instances through a Monte Carlo approach, enabling robust training of a neural network model.

The trained neural network offers high-accuracy estimates of embodied carbon, with most predictions falling within an error margin of less than 20%. Compared to 30 seconds per case with the physics-based model, this machine learning model can provide the results near-instantly, making it a valuable tool for rapid iterative design and decision-making processes. In addition, by analyzing a vast dataset generated from simulated structural configurations, our study identifies critical design features—such as reduced column spacing—that significantly lower the embodied carbon of concrete buildings. These insights inform sustainable building practices by providing actionable strategies for early-stage design optimizations.

This research focuses on zero-energy and zero-carbon building design. By leveraging digital tools, such as machine learning and parametric modeling, we demonstrate significant advancements in whole-building design optimization. Our findings not only enhance the precision of embodied carbon assessments in concrete structures but also contribute to global efforts to reduce the construction sector’s carbon footprint.