Hybrid modelling
A hybrid model combines physical knowledge from theory and simulation with machine learning. It computes in seconds and stays plausible where a pure data model would have to guess.
- Which inputs actually drive the result?
- Which process setting produces the required property?
- How many simulations does a reliable model need?
Why hybrid?
Simulations are accurate but expensive. Data models are fast but only as good as their training data. A hybrid model uses simulation to generate training data deliberately, and knowledge of the physics to choose inputs, value ranges and plausibility checks.
The result is a surrogate for components, processes and systems that you can use in parameter studies, in design, or even inside a controller.
Our approach in four steps
Define the problem: build the model, run FEM or CFD, assemble the data matrix.
Train: prepare training data, choose model type and configuration, train.
Refine: close data gaps, quantify influencing factors, raise accuracy where it matters.
Predict: compute results for new parameters, and inversely: which inputs produce the result you need?
Related projects
Does this match your question?
Send us a short description. We will tell you whether and how we can help.