AI-Powered Optimization of Ultra-High-Performance Concrete (UHPC) Formulations

Use Case:

Leveraging AI and machine learning to assist in the design and selection of optimal UHPC concrete mixtures reducing costly experimentation.

Outcome:

Comprehensive feasibility validation of the AI-assisted design workflow; Development of an early prototype using Jupyter notebooks for ML model evaluation; Establishment of a roadmap for upscaling and extended testing with over 100 formulation trials planned.

Ecosystem Support:

Supported through the StairwAI program with tailored mentoring by AI and business experts, access to applied research coaching, and a structured feasibility and business planning framework.

AI Relevance:

This story showcases how AI can revolutionize traditional sectors by: I) enabling low-cost, high-impact experimentation in material science; II) supporting SME internal innovation cycles with adaptable ML models; III) encouraging broader uptake of AI in underexplored domains like construction materials.

Summary:

TESELA, a Spanish SME at the intersection of material science and heritage conservation, faced the challenge of optimizing the design of ultra-high-performance concrete (UHPC) mixtures—a process traditionally reliant on extensive laboratory trials and expert knowledge. Through the StairwAI support framework, the company launched av-AI-lable, an initiative aiming to embed AI into the early stages of UHPC mix design. The project focused on developing a decision-support tool that employs machine learning algorithms—particularly decision trees and random forests—to assist in selecting optimal mixtures from among a vast number of raw material combinations. A key component of the feasibility phase was the development of a functional prototype in Jupyter Notebook, created with the support of AI experts. This early-stage tool helped TESELA test assumptions, identify data bottlenecks, and outline a path toward operational deployment. The team also delivered a detailed business plan, supported by iterative mentoring and expert review. Although the AI model requires further validation through real-world testing—planned through over 100 experimental UHPC formulations—the project demonstrated strong feasibility and a clear pathway to market-readiness. The solution will be tested internally and eventually offered to laboratories, industry partners, and academic groups. av-AI-lable exemplifies the transformative potential of AI in traditional engineering disciplines, offering a low-barrier entry point for innovation while laying the groundwork for long-term sustainability and market integration.