Introduction
Concrete is the most-used manufactured material on Earth, and one of the most carbon-intensive.
Cement production alone causes roughly 7-8% of global CO2 emissions, and the buildings-and-construction sector accounts for around 39% of energy-related CO2, with about 11% embodied in the materials and the act of construction itself. Making concrete far lower in carbon while making it last longer is one of the built environment's hardest sustainability problems, and it is exactly the kind of high-dimensional, data-rich, multi-objective challenge where modern AI can be decisive.
Today, sustainability, durability, and structural performance are designed in isolation: structural sizing follows prescriptive codes, durability is checked late, and Life Cycle Assessment (LCA) is bolted on after the design is fixed. This fragmentation hides the best low-carbon, high-circularity options, because the trade-offs are never explored together at the moment the decisions are actually made.
Based on the above background, the motivation for this study stems from several important challenges. First, while monitoring AI (micro) can eliminate information asymmetries in system, the high-emission firm is less incentivized to implement as benefit might be hurt. Second, forecasting AI (macro) can reduce uncertainties, but its effectiveness depends on model accuracy, which may incur a significant cost. Firms must balance this cost against the benefits of improved accuracy. Third, in competitive supply chains, rivals may adopt AI at different stages, creating strategic asymmetries that influence both market outcomes and environmental performance. These considerations highlight the need for a rigorous analytical study of AI adoption strategies in carbon markets.
Existing literature has studied carbon cap-and-trade in supply chains extensively (Kroes et al. 2012, Huang et al. 2021). There is also growing work on AI adoption in supply chains, primarily focusing on predictive analytics and operational efficiency (Mithas et al. 2022, Jazairy et al. 2025). However, the intersection of AI adoption and carbon market dynamics remains underexplored, particularly with regard to enhancing environmental sustainability in supply chains.
2. Challenges
This project asks whether a single AI system can reason across all three domains at once. It breaks down into three connected AI challenges.
Surrogate modelling: fast, accurate ML predictors of durability and service life for low-carbon concrete and structure, trained on experimental and mechanistic data, and quick enough to run inside an interactive optimisation loop.
Information extraction with LLMs: turning inconsistent, manufacturer-specific Environmental Product Declarations (EPDs), specifications, and papers into clean, standardised LCA data.
Multi-objective optimisation and explainable AI: an engine that balances conflicting goals (longest service life, lowest embodied carbon, structural adequacy), wrapped in an LLM-driven interface that lets a user explore the trade-off space and understand the reasoning behind each recommendation.
3. Core Research Questions
Can surrogate models predict properties of new concrete and structure accurately to drive design?
Can LLMs extract LCA data from unstructured, heterogeneous sources at the quality and consistency rigorous LCA demands?
How can a multi-objective engine reconcile genuinely conflicting goals, and present the trade-offs so that a practitioner trusts and acts on them?
4. Expected Outcomes
The outcome will be a validated, end-to-end toolset that demonstrates a practical route to decarbonising concrete construction without compromising structural integrity, and that supports the industry's shift from prescriptive codes towards performance-based, sustainability-driven standards. Crucially, it is designed to be used with industry, materials suppliers, contractors, and engineering consultancies, making it a strong fit for collaboration with non-academic partners.
This project is ideal for a student excited to apply AI to the physical world: someone who wants their machine learning and LLM work to translate into real, measurable carbon savings in the built environment. The student will graduate fluent across AI, sustainability assessment, and structural engineering, a rare and highly employable combination.
References:
Lehne, J., & Preston, F. (2018). Making concrete change. Innovation in low-carbon cement and concrete, 1-66.
IEA / GlobalABC, Global Status Report for Buildings and Construction
Li, F., Luo, D., & Niu, D. (2025). Data-intelligence driven methods for durability, damage diagnosis and performance prediction of concrete structures. Communications Engineering, 4(1), 100.
Kumar, A., Nazemi, F., Kodamana, H., Ramteke, M., & Bakshi, B. R. (2025). A Large Language Model-based Framework to Retrieve Life Cycle Inventory and Environmental Impact Data from Scientific Literature. Environmental Science & Technology, 59(42), 22533-22543.