From research to simulations you can run.

Flowphys develops original multi-physics solvers, scientific AI and optimisation methods for ambitious European R&I projects — and turns approved research results into accessible, runnable demonstrators on SimuPort.

Explore SimuPort Partner with Flowphys

Flowphys research.

Coordinator

HORIZON-CL4-INDUSTRY-2025-01-DIGITAL-61

SimuLingua: Simulation-Powered Computational Linguistics for Materials Foundation Models

↳ Horizon Europe SimuLingua

WP leader of EU-funded R&I projects

↳ Explore more

Granted patents

15+

Simulators live on SimuPort

Structural Dynamics
Topology Optimization
Metal Forming
Photonics
Melt Pools
Jet Flows
Pipe & Channel flows
Thermodynamics
Chemistry
Computational Welding Mechanics

↳ Read more SimuPort

SimuPort

Research results you can run in a browser

SimuPort is our online simulation platform, developed and operated by Flowphys on its own infrastructure. Public simulators open in a browser with no installation and no licence request. Each one comes from a research project and carries its model description, assumptions, limitations and validation record, so it can be read as a scientific artefact rather than a demonstration.

Behind the simulators is an agentic framework that turns a described problem into a running model — building geometry, meshing, coupling the physics, solving and verifying. The agents write Flux, our simulation language, so every model they produce can be opened, inspected and re-run.

ACCESS NOTE Not every result can be made public. Where a consortium’s agreements allow it, we publish an open simulator. Where they do not, the same model can run as a controlled-access demonstrator for partners, evaluators and named stakeholder groups, or as a private deployment on isolated infrastructure.

From research method to runnable result

Research results usually stop being usable the day the funding stops. These four steps are how ours keep working. This route takes from a new numerical method to a simulator that stays open long after the project closes.

Method
A numerical method is developed in-house rather than assembled from external libraries.
Project
It is applied to a real problem inside a funded research consortium.
Validation
Results are compared against experiment or a published benchmark.
Runnable result
Where the consortium approves publication, the working method becomes a simulator on SimuPort that remains available after the grant closes.

Research frontiers

Advancing the field, not listing the catalogue

Why
A general language model has read about physics; it has not run any. Simulation generates unlimited correctly-labelled data on demand, and every answer can be checked against a solver — so the reward signal is verifiable rather than approximated.
Approach
FPS-SciFM, with variants on Granite 4.1-8B-instruct and other architectures, trained by reinforcement learning from verifiable rewards (RLVR). FLUX is tokenizer-optimised and our tools curate training files automatically, so the corpus grows from our own simulation output.
State
In continuing development. FPS-SciFM v1.0 model launched on SimuPort.

Why
The agents write FLUX, the same language our solvers execute. Every generated model is a complete, readable definition — open it, change one term, re-run it, and anyone gets the same result. Reproducible, not one-off.
What it does
Builds geometry and mesh, applies boundary conditions, runs and monitors the solve, batches runs, trains surrogates and optimises — iterating until the design meets your specification.
In parallel
Separate agents mesh, check feasibility before and after each run, and couple solvers. Geometry comes from conversation, CAD, or an image our recognition turns into a 3D model or pipe network.
State
Deployed on SimuPort today.

Why
In ALABAMA, RESTORE and WeldGalaxy, thousands of melt-pool and welding simulations trained the surrogates that optimisation then ran against. That workflow does not exist for a team whose single run takes a day.
Where it comes from
In-house sparse-matrix preconditioned conjugate-gradient solvers, adaptive multi-resolution topology optimisation, optimised large eddy simulation, and a cluster we own and tune ourselves.
Measured
published benchmark

Also advancing

Coupled multi-physics and multi-scale methods
Simulation chains spanning density functional theory, thermodynamics, microstructure, continuum flow and mechanics, with surrogate stages replacing full physics where accuracy allows.
Uncertainty-aware optimisation
Optimisation that accounts for variability in material properties and process parameters, so an optimised design still performs when recycled feedstock and real manufacturing tolerances are involved.
Open and interactive publication of computational research
Publishing results as runnable, versioned, citable artefacts rather than static figures.

What we develop

Every algorithm in the Flowphys suite is written in-house, without external numerical libraries. That gives us visibility into and control over every layer of the stack — solvers, coupling schemes, optimisation and surrogate training — and makes possible physics couplings and automation that are difficult to achieve when integrating third-party components. In a research project it means new physics can be added where a work package needs it, rather than approximated within the limits of an existing tool.

01
Solvers and numerical methods
The foundations, all developed and maintained by us.
- Fractional-step finite-element algorithms for incompressible and weakly compressible flow
- Optimised large eddy simulation for turbulent flow
- Arbitrary Lagrangian-Eulerian methods with novel mesh-rezoning algorithms
- Geometrically exact beam and shell elements formulated on Lie groups
- Nédélec edge-element electromagnetics
- Multi-phase-field solvers for microstructure evolution
02
Coupled multi-physics
Physics solved together rather than passed between tools.
- Fluid-thermal, thermo-mechanical and fluid-thermal-mechanical interaction
- Fluid-acoustic interaction with variable-density waves in viscous turbulent flow
- Both staggered and monolithic thermo-mechanical formulations
- 1D network models coupled directly to 3D CFD
03
Optimisation and uncertainty quantification
A five-stage chain, run iteratively until the optimised designs stop changing.
- Latin hypercube design of experiments
- Hybrid Kriging, gradient-boosting and physics-informed surrogates
- Genetic-algorithm optimisation returning Pareto fronts
- Topology optimisation for structures, fluids and electromagnetics
- Fatigue constraints and techno-economic models inside the loop
04
Scientific AI
Models that learn from simulation and feed back into it.
- Physics-informed neural networks as fast, physics-respecting surrogates
- Operator learning — DeepONet and Fourier neural operators
- Proper-orthogonal-decomposition reduced-order models
- A family of scientific foundation models

Within projects we build to requirement: solvers for physics with no existing implementation, integration of models contributed by other partners, and interfaces into a consortium’s existing tool chain. Most of the capability above was first developed inside a funded project.

Next step

Partner with Flowphys — Our contribution to a consortium

We write our own solvers. When a work package needs physics that no available code implements, we can implement it rather than design around the gap. Most of what is now in the Flowphys suite was first built inside a funded project for exactly that reason.

In practice we contribute in four ways.

We write proposals.
Not only a partner description. We lead and co-write proposals — the technical work packages, the methodology, the simulation and impact narrative — and we deliver to deadline. Of the proposals we have led or co-written, Flowphys has very high funding rate.
We build the method.
New solvers, new couplings, new optimisation formulations, developed to what the project actually requires rather than to what an existing tool allows.
We run the campaigns.
Thousands of simulations to train surrogates and drive genetic-algorithm optimisation — the workflow behind our contributions to ALABAMA, RESTORE and WeldGalaxy. This is only possible because our solvers are fast enough that one run is cheap.
We make results usable.
Where a consortium approves publication, approved results become runnable demonstrators on SimuPort, hosted on our own infrastructure and available after the grant closes. Where publication is not appropriate, the same model can run as controlled-access or on isolated infrastructure.

We can take the coordinator role — we currently coordinate the Horizon Europe project SimuLingua — or lead a work package, lead a task, or join as a technology provider.

Partner with Flowphys Explore SimuPort

Partners