background image
Figure 15: Dissolved alumina concentration vs. the slope of the cell voltage

So clearly even if the 5 cycles are not identical, they almost perfectly match a
unique cubic correlation. For a more restricted range of alumina concentration variation
which corresponds to the typical range the In Situ controller would be able to operate the
cell and for a different value of the alumina dissolution constant, the linear correlation
presented in Figure 16 was obtained. This is the correlation that will be used to estimate
the alumina concentration once the slope of the noise free normalized cell voltage have
been estimated by the In Situ controller in Dyna/Marc test runs.

For example, the slope of 3.9 mV/min estimated after 10 minutes of no feed
observation by the quadratic fit of Figure 9 would correspond to an estimated alumina
concentration of:

-0.0279 * 3.9 + 2.3193 = 2.21 %

Using the correlation of Figure 15 would lead to a different estimate:

-750143 * 0.0039
3
+ 21065 * 0.0039
2
- 214.13 * 0.0039 + 2.914 = 2.35 %

This 0.14 % discrepancy between the real alumina concentration and the
estimated alumina concentration (assuming that Figure 16
correlation gives an exact
prediction) would not prevent the In Situ control logic to work quite well, it would
simply introduce a 0.14% offset on the targeted alumina concentration.