TAKE the TOYS
FROM the BOYS
Since its inception in 1886, the aluminium production process consists of the electrolyzation of liquid salt containing dissolved alumina particles. As a standard bath, producers world-wide use a slightly modified cryolite melt (increased excess of aluminium fluoride) with 2% - 5% of alumina. This cryolite melt successfully dissolves alumina at temperatures below 1.000 °C. Unfortunately, under these conditions, it dissolves almost anything that makes it quite difficult to keep the melt in a container (carbon materials are quite stable but at temperatures around 1.000 °C they are buring easily when in contact with oxygen --> and the generation of oxygen or CO2 or CO is part of the process). One solution that has been used from the beginning is to create a natural crust of frozen cryolite along the inner perimeter of the reduction cell. Hence, one of the production requirements is a tight control of the superheat (typically btw. 5°C - 15°C) to keep this natural crust alive.
The chart shows a typical phase diagram of a real bath sample. The steepness of the curve at higher excess AlF3 makes the process more unstable and sensitive. However, this part of the curve also marks the process parameters where the process is most efficient.
With conventional methods (standard temperature measurements and chemical composition from bath samples) it is not possible to get tight control for successful modulation.
The answer to the challenge described above is an in-situ superheat measurement. This device measures the bath temperature and the liquidus temperature simultaneously with an accuracy of 1 ° C, which enables much more precise control of the process and enables process modulation.
The next diagram shows a control logic and the application of this new measuring method. The phase diagram is now converted into a concept with temperatures for bath and liquidus on the x and y axes. Based on the combination of these parameters, it is now possible to react precisely to the process parameters.
Real life scenarios have to deal with real life problems. A good measurement is the foundation of an excellent control logic. However, the measurement will be influenced by events that happened before the measurement in the individual pot, in the potline (typical potlines contain btw. 100 and 300 pots) or through external effects. To improve process reliability, it is necessary to transform the measured numbers with the known events by regression or AI methods. In this way, the process deviations can be significantly reduced and the process can be safely tuned and modulated.
The process methodology described is a prerequisite for the successful implementation of modulation and the "Virtual Battery Concept".