MPflow: Multiphase CFD powered by Physics-Aware AI

Objective-driven engineering

Engineering has traditionally been process-driven, where simulations are executed through predefined steps before meaningful insights emerge. With Engineering AI, this paradigm shifts. Engineers specify performance goals, constraints, and trade-offs upfront, and the system dynamically orchestrates the simulation and validation workflow around those objectives.

Execution is guided by intent, allowing engineers to focus on results rather than process.

Simulation powered by AI

MPflow enables the deployment of machine learning models to replace the most computationally intensive parts of the CFD workflow. By bypassing traditional bottlenecks—such as meshing, solver setup, turbulence modelling, and boundary condition tuning—simulation time is dramatically reduced.

Each validation cycle becomes repeatable, auditable, and scalable across teams, enabling consistent and reliable engineering workflows.

Users can train ML models using their own existing data through a process we call data recycling. No AI expertise is required: MPflow provides an intuitive interface with predefined templates, rules, constraints, and best practices, allowing engineers to focus on results rather than implementation.

On-the-fly deployment of ML

MPflow integrates trained machine learning models directly within the CFD solver, enabling real-time predictions during simulation. These models dynamically replace computational bottlenecks—such as turbulence closure or phase interaction terms—without disrupting the workflow, ensuring seamless, stable, and accelerated simulations while preserving physical consistency.

Full ML deployment

MPflow enables full replacement of CFD simulations with trained neural networks that predict key flow quantities directly—eliminating meshing and solver execution. Using existing simulation and experimental data, users train models that deliver instant, high-fidelity predictions across operating conditions, dramatically accelerating design exploration and decision-making.

Don’t let engineering effort be constrained by CFD bottlenecks and missing numerical models. When analysis relies on meshing, hard-to-define parameters, and convergence challenges, valuable expertise is diverted from performance and design. Streamline the workflow and save time for true product innovation.

How AI in CFD works in practice

• Use your internal CFD data
  Leverage your existing simulation data with full privacy—your CFD datasets and numerical know-how remain entirely within your organisation.

• Recycle and structure the data
  MPflow enables efficient reuse of historical data, preparing it for training without additional simulation cost.

• Train your ML-CFD model
  We support you in training your own neural network, resulting in a fully proprietary ML-CFD model tailored to your applications and entirely owned by you.