MATRICS: Materials AI Training and Research Infrastructure for Comparative Standards
MATRICS (Materials AI Training and Research Infrastructure for Comparative Standards) is innovative in that it creates the first openly accessible, multimodal causal benchmark for advanced materials engineering, jointly developed by the NCC, RootCause and Authentise. It uniquely combines a large-scale, real-world R&D corpus (over 2,000 NCC documents spanning 14 years) with frontier AI causal discovery methods and direct industrial deployment pathways into manufacturing execution systems (MES).
Unlike existing benchmarks, which are either synthetic, text-only, or lack causal structure, this work integrates unstructured reports, structured experimental data and engineering knowledge into a single multimodal ontology. This enables evaluation of advanced tasks such as causal inference, counterfactual reasoning, experiment outcome prediction, and inverse design in real engineering contexts.
World-leading Team
RootCause will provide the underlying multimodal causal discovery platform, enabling the extraction of evidence-weighted causal graphs from heterogeneous engineering data. NCC will contribute validated industrial datasets and domain expertise, ensuring realism and relevance to Technology Readiness Level (TRL) 3--4 industrial R&D. Authentise will provide the deployment layer, embedding outputs directly into manufacturing workflows via its FLOW MES platform, enabling end-to-end validation from research data to shop-floor decision-making.
Innovations
A key innovation will be the introduction of a unified causal ontology for composites R&D that supports all benchmark tasks. This ontology enables recovery of process--structure--property--performance relationships and supports cross-domain, longitudinal and multi-modal reasoning across manufacturing processes and materials.
The project will also introduce several methodological advances. It includes a pre-registered evaluation protocol with multi-layer leakage controls (temporal, domain and parameter-space), ensuring reproducibility and scientific rigor. A fully open evaluation harness will allow third-party benchmarking with deterministic scoring, calibrated uncertainty measures, and public leaderboards---capabilities absent from current AI evaluation frameworks.
In addition, the benchmark dataset will be fully FAIR-compliant and openly released under CC-BY-4.0, addressing a major gap in composites R&D where no comparable multimodal annotated dataset currently exists. The inclusion of a CompoST-compatible ontology (CompoST is NCC's open data schema for advanced composite materials) and MES integration adapter further extends innovation beyond benchmarking into operational industrial deployment.
Impact
Overall, the project is novel in combining real-world industrial datasets, causal AI, open benchmarking infrastructure and direct manufacturing integration into a single end-to-end system - cutting R&D duplication costs by ~40% - enabling both scientific advancement and immediate industrial impact in advanced materials engineering. Commercially, for the SMEs, immidate post-project revenue growth of \>£2.2M, with the potential for rapid scaling (\>£10M), with \>25 jobs created.
👉 Find out what other projects we are working on here.
