06 / 06 · Case studyPhysics-informed AI
A digital twin that reads a farm’s climate with fewer sensors.
Sensors cost money. Blind spots cost crops.
An indoor farm needs to know the temperature and humidity around every plant. Covering every corner with sensors is expensive, and sensors fail.
Without them, a hot corner goes unseen until the crop shows it.
Simulate the air. Check it against a few sensors.
A digital twin is a live computer model of a real thing. This one models the air in a greenhouse: how heat and moisture move from lamps, vents and plants.
A few real sensors anchor the model. The model fills in every point between them. This is called virtual sensing. The model also obeys physics: energy that enters a room must go somewhere.
- 01A few sensors
- 02Physics model
- 03Neural network
- 04Full climate map
- 05Fan and vent control
Data plus the laws of physics
- Digital twin
- A live model that mirrors the real greenhouse and updates as new readings arrive.
- Virtual sensing
- The model estimates values at points where no sensor sits.
- Physics-informed learning
- The network is trained to respect heat and mass balances, not only to fit data.
- Heat and mass transfer
- The same equations I used to design chemical plants.
- Closed-loop control
- The climate map drives fans and vents before a hot spot forms.
The idea in numbers
- 1 map
- of temperature and humidity for the whole room, not a handful of points
- 2 laws
- the model must obey: conservation of energy and of mass
- > 0
- real sensors always stay. They keep the model honest
What comes next
Build the first version for one room of a vertical farm, and measure how many sensors it can replace.