Carbon capture optimizer (physics-informed AI)
Connects to the capture unit’s sensors and keeps every hour of operation as a run. A physics-informed neural network learns the unit from those runs while obeying its physics (absorber mass transfer, temperature effects, the energy balance), so it stays trustworthy where data is thin. It recommends the solvent flow and reboiler temperature for the purest CO₂ at the best capture and energy, and people approve every change before the gateway applies it. A public page shares live performance with partners.
2 connected tools: Capture runs, Model runs.
One click opens a free 30-day workspace with this app and sample data. Change anything by asking.
Try it now
What it does
Capture runs
Every hour of operation of the capture unit: feed gas, solvent, temperatures and pressure in; capture efficiency, CO₂ purity and energy per tonne out. Filled from the unit’s sensors (or lab results), and the data the physics-informed model learns from. The model’s predictions sit beside the measurem
- Tracks
- Run, Date, Feed gas flow (L/min), Inlet CO₂ (%), Solvent flow (L/min), Absorber temperature (°C), Column pressure (bar), Reboiler temperature (°C), Capture efficiency (%), CO₂ purity (%), Energy per tonne (GJ/t), Model: purity (%), Model: efficiency (%), Operating mode, Source, Notes
- Flags automatically
- Below purity target when CO₂ purity (%) is under 98.5
- Low capture when Capture efficiency (%) is under 85
- High energy when Energy per tonne (GJ/t) is over 3.4
- Actions and automations
- Tell the process engineer: Tell the process team about this run.
- Automation: When capture falls below 85%, tell the process engineer
- Views and charts
- All runs, Measured vs model, Operating conditions, CO₂ purity, measured and predicted, Capture efficiency, measured and predicted, Average purity by operating mode
- Its agent
- Watches every run for the flags above, drafts the follow-up and waits for approval, at the autonomy level you choose.
Model runs
Each time the physics-informed neural network learns from the latest runs: how well it fits, how well it obeys the physics, and the setpoints it recommends with the purity, capture and energy it expects. Recommendations go to people for approval before the unit’s gateway applies them.
- Tracks
- Model run, Trained on, Runs learned from, Purity error (RMSE, %), Capture error (RMSE, %), Physics residual, Current purity (%), Expected purity (%), Current capture (%), Expected capture (%), Current energy (GJ/t), Expected energy (GJ/t), Confidence (%), Recommended setpoints, Status
- Flags automatically
- Purity gain available when Status is Proposed
- Low confidence when Confidence (%) is under 60
- Actions and automations
- Write the optimisation report: A Word report and PowerPoint on this model run: fit, physics, recommendation and expected gains.
- Views and charts
- All model runs, Purity: current vs expected
- Its agent
- Watches every model run for the flags above, drafts the follow-up and waits for approval, at the autonomy level you choose.
Related apps and templates
Frequently asked
Can we start from our own spreadsheets?
Yes. Import Excel or CSV and Neuclon maps the columns to the fields, or attach the sheet when you ask for the tool and it is designed around it.
Can we change it ourselves?
Yes. Ask in plain words, for example “add a downtime reason field” or “flag anything open more than 7 days”. Every version is kept.
How long does it take to start?
Minutes for the trial: it opens with sample data you can replace. A 30-day Sprint proves one improvement on your own data.
Start with this in your own workspace
Free for 30 days. It opens with sample data you can replace with yours, and every Business feature is included.