Scale-up prediction for process and materials engineers
It works in the lab.Will it hold at pilot scale?
CauseMat predicts what a material and a machine will do together, including combinations nobody has run and scales you have never operated. You see the predicted window, and the reasoning behind it.
See it run
7 runs in. The whole window mapped.
CauseMat turns the runs you already have into a map of the full process window. You see where it holds, where it is marginal, and where it fails.
- Predicted in spec
- Marginal
- Predicted out of spec
- You ran here
A map of predicted outcome over line speed and coating temperature. Seven runs sit in one corner of the window, and CauseMat predicts the rest of it, including where the process is marginal and where it fails.
What you get
0 runs at 50 L. A window for every material.
Each bar is the operating window CauseMat predicts at a scale you have not run. Where a run has since been done, it lands inside the window.
- Predicted window
- Run since confirmed
- NMC-8114.2 to 5.1NMC-811: predicted 4.2 to 5.1 metres per second. A later run at 4.8 landed inside the window.
- LFP-A3.6 to 4.4LFP-A: predicted 3.6 to 4.4 metres per second. No run at this scale yet.
- Graphite-C4.9 to 6.0Graphite-C: predicted 4.9 to 6.0 metres per second. A later run at 5.2 landed inside the window.
- Binder-X3.3 to 4.1Binder-X: predicted 3.3 to 4.1 metres per second. No run at this scale yet.
The knowledge grid
Watch what one experiment teaches every other cell.
A replay of a real two-machine study, one run at a time, through the same inference that makes our predictions. Column and row labels carry each material's and machine's own representation; cells show what transfers. Machine B's row stays empty until anchor runs calibrate the dial map — then every characterized material's cell fills at once.
The difference
Other tools answer inside your data. CauseMat answers outside it.
CauseMat models what a machine does and what a material needs as separate causes, then composes them. The causal reasoning is explicit, so you can read it, challenge it, and carry it to the next machine.
| Conventional ML | CauseMat | |
|---|---|---|
| A setting close to runs you already have | yes | yes |
| The outcome of a combination nobody has run | no | yes |
| A scale you have never operated | no | yes |
| A process with no history to train on | no | yes |
| The causal reasoning behind the answer, in full | no | yes |
| What it cannot determine, and the runs that would | no | yes |
| Carries to your next material and machine | no | yes |
| An exact value rather than a windowinside its own data. CauseMat answers in windows and margins. | yes | no |
Contact
Tell us your bottlenecks.We'll find the answers.
Scale-up that behaves differently at the pilot plant. Qualification that takes weeks every time. Yield drops nobody can explain.
Whichever one is costing you most, start there.
info@causemat.aiYour data is yours. So are the outcomes.
Models are built for your process and stay with it. Nothing you send trains a model for anyone else, and security is an essential priority, not an afterthought.