530+
Person-months
Consortium effort across 20 work packages and 48 months.
Horizon Europe • GenAI4EU
About SimuLingua
SimuLingua delivers an open, multimodal Scientific Foundation Model (SciFM) for materials that unifies text, structures, images and simulation data in a closed design–simulate–validate loop. A physics-in-the-loop generative engine and AI-accelerated multi-scale simulation pipeline (DFT → phase-field → CFD/FEM) verify candidates; results flow back through a FAIR knowledge graph with ontologies, provenance and leakage-safe splits. Natural-language interfaces let domain experts query the KG, launch virtual experiments and steer inverse design — progressing from TRL 1 to TRL 4 in 48 months across six use cases.
9
Partners
Universities, SMEs, and one NGO across the value chain.
7
Countries
Norway, Sweden, UK, Netherlands, Ireland, Lithuania, Ukraine.
Horizon Europe
Project facts
Horizon Europe RIA under the GenAI4EU cluster — grant agreement 101295295.
Start
1 June 2026
End
May 2030
Duration
48 months · 530 person-months
Consortium
9 partners · 7 countries
Action type
HORIZON-RIA
Grant agreement
101295295
Topic
HORIZON-CL4-INDUSTRY-2025-01-DIGITAL-61
AI Foundation models in science (GenAI4EU) →Mission
Project summary
Materials discovery democratized — a synergistic fusion of human expertise, multi-scale simulation, and SciFM that understands every dialect of materials data, with physics in the loop, not just in the training data.
Unified multimodal SciFM
Graphormer, SciBERT, and numerical encoders fused via cross-modal attention — >10 simultaneous prediction tasks.
Physics-in-the-loop
Staged constraint curriculum from charge neutrality to CALPHAD-feasible synthesis windows.
FAIR knowledge graph
>10M entities, >15 sources, Data Gate audits, leakage-safe splits — DOME-ML compliant.
Generative inverse design
Conditional diffusion/VAE with synthesizability scoring and multi-objective NSGA-II optimisation.
Uncertainty & trust
Deep ensembles, conformal prediction (ECE ≤ 0.05), hallucination-risk scoring, human-in-the-loop NLP UI.
Amorphous + crystalline
Electroceramics, glasses, HEA alloys, recycled wire cladding — plus UC5+UC6 digital olfaction for carbon capture & geothermal.
SciFM
Four capabilities
What distinguishes Scientific Foundation Models from general-purpose AI.
Domain adaptation
Fine-tune on new material classes with limited data — six expert models, one foundation.
Domain generalization
Zero-shot predictions on unseen chemical spaces, from perovskites to high-entropy alloys.
Problem adaptation
Transfer across related tasks — property prediction spanning atomistic to continuum scales.
Problem generalization
Extend to novel physics regimes without full retraining, grounded by simulation backstops.
Platform
Generation & extraction of data
AI-accelerated multi-scale simulation feeding the multimodal SciFM and FAIR knowledge graph.
AI-accelerated multi-scale simulation
Atomistic
ångström – nanometre
DFT · CALPHAD
Electronic structure, energies & forces; phase equilibria & thermodynamics.
GNN surrogate
Mesoscale
micrometre
Multi-Phase Field
Microstructure & grain / phase evolution during synthesis and processing.
PINN surrogate
Continuum
millimetre – metre
CFD melt-pool · thermo-mechanical
Melt-pool thermo-fluids (Navier–Stokes); residual stress & distortion.
POD / POD + NN ROM surrogate
Closed loop
Physics-in-the-loop discovery engine
From generative proposals through fast simulation verification to lab validation and knowledge-graph feedback.
Natural-language interface — ask in plain language
1
Propose
The SciFM generates promising material & process candidates from any mix of inputs.
2
Generate
physics-in-loopEmbedded physics + a synthesizability score keep candidates valid.
3
Simulate 100–1000×
GNN / PINN / ROM surrogates of DFT → MPF → CFD/FEM verify physics fast.
4
Validate
Top candidates are synthesized & characterized across six industrial use cases.
Goals
Project objectives
What SimuLingua commits to deliver over the project lifetime.
- 1.Deliver an open multimodal SciFM unifying text, structures, images and simulation data.
- 2.Close the design–simulate–validate loop with physics-in-the-loop generative models and AI-accelerated multi-scale simulation.
- 3.Build a FAIR knowledge graph with ontologies, provenance, leakage-safe splits and Data Gate audits.
- 4.Provide natural-language interfaces for querying the KG, virtual experiments and inverse design.
- 5.Validate generalisation across six diverse use cases from TRL 1 to TRL 4 in 48 months.
- 6.Release non-restricted models, datasets and code under open licences with model/data cards, DOIs and APIs.
Structure
Work packages
20 work packages across three reporting periods — grouped by technical pillar.
Lead: FLOWPHYS
Management & governance
Coordination (FLOWPHYS), ELSI framework and Ethics Advisory Board (VDU), risk management, and international cooperation.
WP1WP2WP3
- WP1 — Project Management & Governance · Period 1M1–18
- WP2 — Project Management & Governance · Period 2M19–36
- WP3 — Project Management & Governance · Period 3M37–48
Lead: TU/e
Data & knowledge graph
Materials ontology, ingestion from 15+ sources, multimodal KG, FAIR/DOME practices (VDU leads DMP), and Data Gate audits.
WP4WP5WP6
- WP4 — Data Orchestration & Multi-Modal Knowledge Graph · Period 1M1–18
- WP5 — Data Orchestration & Multi-Modal Knowledge Graph · Period 2M19–36
- WP6 — Data Orchestration & Multi-Modal Knowledge Graph · Period 3M37–48
Lead: TU/e
SciFM training & generative models
Multimodal architecture (Graphormer, SciBERT), V0→V1→V2 training, physics-informed inverse design and PEFT per use case.
WP7WP8WP9
- WP7 — SciFM Architecture & Initial Training · Period 1M3–18
- WP8 — SciFM Large-Scale Training & Generative Models · Period 2M19–36
- WP9 — SciFM Large-Scale Training & Generative Models · Period 3M37–42
Lead: FLOWPHYS
AI-accelerated simulation
GNN, PINN and ROM surrogates for DFT, phase-field, CFD/FEM — integrated pipeline with documented API.
WP10WP11WP12
- WP10 — AI-Accelerated Simulation: Surrogate Models · Period 1M3–18
- WP11 — AI-Accelerated Simulation: Pipeline Integration · Period 2M19–36
- WP12 — AI-Accelerated Simulation: Pipeline Integration · Period 3M37–42
Lead: UoS
Use case validation
KPI definition, in-silico screening of millions of candidates, experimental synthesis and closed-loop KG feedback.
WP13WP14WP15
- WP13 — Use Case Definition & Initial Validation · Period 1M1–18
- WP14 — Use Case In-Silico Discovery & Validation · Period 2M19–36
- WP15 — Use Case Experimental Validation & Assessment · Period 3M37–46
Lead: FLOWPHYS
Platform & NLP interface
Microservices integration, NLP UI, UQ/active learning, Apache 2.0 open-source release on GitHub and Hugging Face.
WP16WP17
- WP16 — Platform Integration & NLP Interface · Period 2M18–36
- WP17 — Platform Robustness & Final Release · Period 3M37–48
Lead: AEI
Dissemination, exploitation & policy
DEC strategy, website and outreach (AEI), scientific dissemination, exploitation/IPR; policy feedback and standardisation (VDU).
WP18WP19WP20
- WP18 — Impact Maximisation: DEC · Period 1M1–18
- WP19 — Impact Maximisation: DEC · Period 2M19–36
- WP20 — Impact Maximisation: DEC & Exploitation · Period 3M37–48
WP18–WP20
Dissemination & impact KPIs
Quantitative targets from the DEC plan — publications, outreach, open science and community building through simulingua.eu.
Dissemination
- Gold open access publications10+
- Conference presentations20+
- Citations (key pubs, 12 mo)>50
- Final public conference150+ stakeholders
Communication
- Social media posts150+
- Quarterly newsletters12
- Press releases4+
- Website articles & updates40+
- Infographics / videos4+
- Popular science pieces10+
Community & platform
- Training workshops4 · >200 participants
- GitHub stars / forks500+
- Active platform users500+
- Usability satisfaction>4.5/5
Open science
- FAIR compliance (public data)>85%
- Knowledge graph entities10–15M
- LicensingApache 2.0
Ethics & cooperation
- Gender balance (teams & workshops)40–60%
- International MoUsTPC · NASA (geospatial)