background image
The main innovation at the core of the new In Situ feed control algorithm is realization that it is
possible to indirectly measure the concentration of dissolved alumina in the bath during a no feed
track by numerically establishing the relationship that exists between the slope of the normalized
cell voltage and the alumina concentration. In fact, that correlation is implicitly used in all
continuous tracking algorithms that monitor the slope of the pseudo-resistance (or the slope of the
normalized cell voltage) to decide when it is time to shift from underfeeding to overfeeding.
Verifying that there is a unique correlation between the concentration of dissolved alumina in the
bath and the slope of the normalized cell voltage and numerically establishing that unique
correlation if it exits is something that can be quite easily done using a cell simulator. Figure 11
presents the results by running the In Situ feed control algorithm in Dyna/Marc for 24 hours. A no
feed-track is called every 3 hours in order to evaluate the dissolved alumina concentration. Figure
12 presents the correlation between the slope of the normalized cell voltage and the dissolved
alumina concentration. The black line is the fit of the average path during the tracking, all 8 tracks
are following the same trajectory. This is why the In Situ feed algorithm can use the shown equation
to establish the alumina concentration at the end of each track. So, there is a unique correlation,
because each track start from identical conditions, the conditions the In Situ feed algorithm is trying
to maintain.
The second innovation at the core of the In Situ feed control algorithm is the usage of the primary
calibration surface [3], at the end of each track, to establish the ACD once the dissolved alumina
concentration has been established. Then, based on an estimated evolution rate of the ACD, that
same primary calibration surface is used as well as an assumed ACD value to estimate every 5
minutes the dissolved alumina concentration from the cell normalized voltage. Finally, a simple PID
controller is used to maintain the estimated dissolved alumina concentration on its target value. In
the example shown in Figure 11, that target concentration was set to 2.25%.
Figure 13 presents the results of a second run, calling for a track every 12 hours only, this time with
the normal anode change event that was removed in the previous run in order to keep things more
simple. Figure 14 presents the corresponding 24 hours averaged specific power consumption and
current efficiency: 13.02 kWh/kg and 94.77 % respectively. Those results are quite similar to those
obtained using continuous tracking feed control algorithm with shorter cycles, but with far less risk
of having anode effects.
Conclusions
The author hopes that this demonstration study highlights the value of using a dynamic cell
simulator to optimized existing cell controller algorithms or to test new ones without putting real
cells at risk. Dyna/Marc cell simulator used in this study is available to the whole aluminium
industry through GeniSim Inc. Version 13 included the linear and quadratic RMS noise filtration
algorithms and the In Situ feed controller algorithm. Dyna/Marc cell simulator can also be used as a
cell design tool as demonstrated in [10].

References
[1] I. Tabsh and M. Dupuis, Process Simulation of Aluminum Reduction Cells, Light Metals, TMS,
(1996), 451-457.
[2] M. Dupuis, I. Eick and F. Waldman, Modelling Thermal Dynamic Response to a 3-Hour Total
Power Shutdown Event, 9
th
Australasian Aluminium Smelting Technology Conference and
Workshops, (2007).