background image
Figure 14. Proportional control, 1 measure per day, no delay
Case no 2: Proportional control, 1 measure per day, no delay
Using a 0.3 kg/hrșC proportional constant reduces slightly that
permanent error but increased significantly the oscillation in the
response as we can see in Figure 15.
Figure 15. Proportional control, 1 measure per day, no delay
Clearly, controlling the excess AlF3 concentration using only bath
temperature measurement even with one measurement per cell per
day and without any time delay is not as stable as using bath
sampling even with a much lower feedback control frequency and a
quite significant time lag.
Control logic based on both bath sampling and bath
temperature measurement
Many modern bath excess AlF3 concentration control strategies
[4,5] are using both the time lagged bath sample analysis and the
bath temperature in their control feedback loop. Is this a good idea?
Case no 1: PI mixed control, 1 sample per day, 1 day delay, 1
temperature measurement per day, no delay
Figure 16 presents one of the best possible feedback control
response that can be obtained using that mixed control strategy.
The bath temperature target is set to 975șC, which is compatible
with a 8.5% excess AlF3 concentration target for that demonstration
cell operating at that current superheat and average alumina and
CaF2 concentrations. All those factors are important as the
correlation presented in Figure 13 only holds in those conditions.
Furthermore, the bath sample analysis results are only lagging the
temperature measurement by 1 day reducing the risk of being out of
phase and hence inconsistent for the cell controller to act upon. The
proportional constants are set to 1 kg/hr% and 0.3 kg/hrșC
respectively.
Figure 16. PI mixed control, 1 sample per day, 1 day delay, 1
temperature measurement per day, no delay
Clearly when compared with the response of Figure 6, using the
bath temperature in addition to the bath sample analysis only adds
instability to the response.
Case no 2: PI mixed control, 1 sample per day, 1 day delay, 1
temperature measurement per day, no delay using inconsistent
targets
It is important to notice that this is one of the best possible
feedback responses possible. A more probable response would be
the feedback response resulting from the tug of war itself resulting
from setting inconsistent targets between the 2 parts of the
controller.
This is a more likely scenario because there are many factors
affecting the correlation between the excess AlF3 and the bath
temperature. As example, Figure 17 presents the modified bath
excess AlF3 vs. bath temperature correlation when the CaF2
concentration increases from 3% (in figure 13) to 8%.
As we can see, there is about a 5șC shift down in the obtained
correlation. Similar shift in the correlation will occur if the cell
superheat is increasing due to cell amperage increase per example.
In fact, the list of factors affecting that correlation is extremely
long: bath chemistry excluding excess AlF3, cell amperage, ACD,
anode size, anode cover thickness etc.
Any shift in the correlation will mean that if not readjusted, the 2
controller targets will become inconsistents. Case 2 is identical to
Case 1 in all aspects except that the bath temperature target has
been set to 970șC instead of 975șC. Figure 18 presents the resulting
tug of war feedback control response. Clearly this is the worst type
of control response possible. Unfortunately, this is still a quite
common one in the industry even nowadays.