At the ICSOBA2023 conference I had the opportunity to talk to many process engineers working at aluminium smelters around the world. They validated that the following digital use cases would improve their electrolysis operations:
- Predicting anode effects 15 minutes before they occur.
- Daily bath and liquidus temperature predictions replacing a part of manual measurements.
- Detecting anode spikes.
- Predicting holes in pot shells preventing an early end of life for the pot.
The process variables influencing these problems are known. The mechanisms are understood. Both statistical and physical models exist.
So if aluminium smelters are sitting on a wealth of data, why hasn’t it been done before?
Things are not as simple as they seem. Starting from the problem, e.g. detecting anode spikes, and working back to the data we see the problem: we don’t have the relevant data. The current through the entire pot is known, but not through an individual anode.
Lots of R&D efforts (individual anode current measurements, fiber optic cables measuring temperature, fluorine emission sensors, etc.) are therefore aimed at obtaining relevant data.