background image
Since the estimated noise free normalized cell voltage after 10 minutes of no feed
observation was estimated to be 4.1199 V from the quadratic fit presented in Figure 9,
we can calculate by a simple interpolation method the estimated ACD:

4 + (4.1199 ­ 4.071) / (4.346 ­ 4.07) = 4.18 cm

Using the combined estimated cell voltage and slope of the cell voltage after 10
minutes of no feed observation, it is possible to estimate both the concentration of
dissolved alumina in the bath and the ACD. From that point on, until the next
observation, it is possible to estimate the evolution of the dissolved alumina
concentration by simply following the evolution of the estimated noise free normalized
cell voltage. In order to be able to do that, two things are needed:

1) An estimate of the evolution of the ACD
2) The primary calibration surface
By neglecting any changes of the metal level coming from changes of the ledge
thickness or the accumulation of sludge, the ACD evolution depends only on the linear
accumulation of the metal and the linear consumption of the anodes. For a prebaked
anode cell, the net result is a slow linear decrease of the ACD.
In the demo Dyna/Marc test case, that linear rate of change of the ACD was
estimated to be about -0.00033 cm/min, which means that 3 hours after the observation,
the ACD can be estimated to be:

4.18 ­ 0.00033 * 180 = 4.12 cm

Of course, this prediction cannot be considered sufficiently accurate over a very
long period of time of many hours. Yet, it only needs to be approximate until the next
observation when ore feed is once again restricted for an in situ alumina concentration
and ACD measurement.

Assuming that we can estimate the noise free normalized cell voltage every 5
minutes and provide an estimate of the ACD based on an assumed linear rate of change,
we need to replace the primary calibration curve by a primary calibration surface.
Contrary to the calibration curve that can only be used to estimate the concentration of
dissolved alumina in the bath from the cell voltage at a given ACD, the primary
calibration surface can be used to estimate the concentration of dissolved alumina in the
bath from any combination of cell voltage and ACD.
It is possible to fit the evolution of the CoeffA and CoeffB coefficients as the ACD
is changing from 4 to 5 cm as seen in Figures 18 and 19. By using those two extra
equations, it is possible to calculate first the value of CoeffA and CoeffB that
corresponds to the current estimate of the ACD and then calculate the estimate of the