current efficiency is maximized by operating very lean in alumina, so very close to the anode effect
conditions and taking advantage of the fact that during underfeeding, the slope of the cell pseudo-
resistance starts to rise significantly before the anode affect. Figure 3 presents the results by running
that feed control algorithm in Dyna/Marc. The top graphic is showing the 24 hours evolution of the
cell pseudo-resistance. Metal is tapped out at noon and anodes are changed at 18 hours. It can be
noticed that the cell is more noisy after the anode change. The middle graph is showing the noise-
free evolution of the slope of the cell pseudo-resistance in blue. It also presents the estimated slope
evolution that results from using linear RMS fitting with 60 datapoints, each datapoint being the
results of 5 seconds cell pseudo-resistance evolution averages. At that time scale, the 2.5 minutes
delay between the noise free pseudo-resistance evolution of the estimated pseudo-resistance
evolution is not noticeable but do affect the timing of the feeding regime shift. The third lower
graphic is showing the feeding rate evolution resulting from the algorithm decision. The
underfeeding rate is 70% of the nominal feeding rate while the overfeeding rate is 140% of the
nominal feeding rate. The overfeeding rate duration was set to 1 hour. As a result, the resulting
evolution of the dissolved alumina concentration in the bath in the same graph is varying from
around 2% to around 2.5%, 2% being the alumina concentration that would trigger an anode effect.
starting to increase when the feeding rate is changed from underfeeding to overfeeding. That
delayed response will trigger an anode effect if the shift of feeding regime is done too late; hence
the importance of eliminating as much as possible the delay in the pseudo-resistance slope
estimation. Figure 4 presents the resulting 24 hours averaged specific power consumption and
current efficiency: 12.96 kWh/kg and 94.71 % respectively.
led to a significant current efficiency increased over the usage of feed control algorithms that were
using nominal feeding rate most of the time. It is also well known that the shorter feeding cycle also
lead to current efficiency increase; this can be tested using the cell simulator. Figure 5 presents
results obtained using a shorter 40 minutes overfeeding rate duration. As a result, the dissolved
alumina concentration only varies from around 2% to around 2.3%. In Figure 6, this leads to a
predicted improvement of the current efficiency to 94.78% and a slight increase of the specific
power consumption to 13.01 kWh/kg if the ACD is kept constant.
[6] is also available in Dyna/Marc. The same reduction of the feeding cycle study presented above
can be repeated using this time the demand feed control algorithm. Figures 7 and 8 present the base
case results: 12.91 kWh/kg and 94.67 % current efficiency, while Figures 9 and 10 present results
for the case with shorter feed cycles: 13.09 kWh/kg and 94.65 % current efficiency. Despite a very
similar increase of the feed cycles and reduction of the range variation of the dissolved alumina
concentration, results on the global process efficiency predictions are different this time: the current
efficiency is not affected and the specific power consumption is increasing. The difference is
explained by the fact that this time, it was not possible to keep the same average ACD and operating
temperature, they both increased for the shorter cycles case.
useful to develop, without putting real cells at risk, a completely new feed control algorithm. One
such innovative new feed control algorithm that was recently tested using Dyna/Marc cell simulator,
it is the In Situ feed control algorithm [3, 7, 8, 9].