What happens if there is a change, which could be either a
setpoint change, or a load disturbance? In most cases in real
life, the controls are set up so slowly that the controller
cannot deal with the change in automatic, so in most plants the
operators immediately switch back to manual, and adjust the
controller’s output to get the process back to setpoint. They
then switch back to auto.
To illustrate the case in point at the plant
where I was working, all flow control loops had an integral
value set in them of 60 seconds. As integral value should be set
for self-regulating processes equal to the time constant of the
process. An integral value of 60 seconds for the average flow
loop immediately informs one that it is about 30 to 60 times to
slow, and there is no way that the control will be able to deal
effectively with changes.
Fig. 1
The comparison in tuning is dramatically
illustrated in Figures 1 and 2. In the first figure one can see
the response to a 10% setpoint change on one of the flow loops
with the existing “as found” tuning. The process has not
even reached setpoint a half an hour later. With the correct
integral in the tuning, and a gain that allows effectively
critical damping – which is considered a fairly slow and
extremely robust tune, the process gets to setpoint in about 30
seconds, after the same setpoint change. This is shown in Figure
2.
Fig. 2
The control people in the plant had always
considered that their flow loops were working extremely well!
However the fact is that the tuning was useless, and as they
only looked at trends when the flow process is stable they saw
everything steady on setpoint, and no cycling, so they were
happy. In fact one could say that they may just as well have
left all their flows on manual. What is the point of using
expensive control equipment that isn’t actually working
properly?
Their main concern was in fact the control on
their floatation bank levels, which were cyclic in the extreme.
A previous Case History No. 105, dealt with
some aspects of floatation bank flow controls. However the
control of the level in each cell is also critical. Figure 3
shows a typical bank with only 4 cells, but most banks in
platinum concentrators contain many more cells. Furthermore the
output from certain banks is also fed into other banks
downstream. The level in each cell is monitored by a level
transmitter, (usually ultrasonic), and this PV signal is sent to
a level feedback controller which modulates one (or sometimes 2
parallel) valves on the output of the tank. The output flow from
the cell then gravitates into the next tank, and so on down the
line.
Fig. 3
A future Case History will be dedicated only
to discussing the problems associated with the level control of
floatation banks. However for the moment let it suffice to say
that there are in fact numerous problems which make this
particular type of process extremely difficult to control, and
it is extremely hard to achieve accurate control and keep levels
in cells constant. One of the main problems which I wish to
discuss here is that the levels in the cells throughout the
banks tend to be very cyclic.
Levels are integrating type processes which
for all the reasons detailed in my Loop Signature Series are
inherently unstable and tend by nature to be very cyclic. Now it
is obvious that if the feed into the first cell in a bank is not
constant, then the feedback controller will react and will
adjust the valve on the cell’s output, which in turn will
cause the feed into the next cell to also vary. This will repeat
all the way down the bank.
A problem that will be explained in the
future article is that an unstable or non-constant feed into the
bank will almost certainly result in cycling, which will
probably occur in all the cells in the bank. Unfortunately,
again for reasons that cannot be dealt with now, the feedback
level control of floatation bank cells generally has to be
pretty slow, and it is almost impossible to keep the levels
constant if the flow through the bank is continually changing.
What generally results is almost an “amplifier” effect with
a smallish cycle in the first cell getting larger and larger as
you go down the bank.
Now the view of the metallurgists in the
particular plant I was working in, and it is also a view I have
found in many other plants as well, is that there are level
controllers in each cell and it should surely be a simple matter
to tune them properly so as to eliminate cycling completely.
Now whilst it is completely true that the
main reason for applying controllers to continuous process
plants is to minimize control variance on each process,
unfortunately these people fail to comprehend the fact that
control cannot completely make variance vanish completely. All
the controller does is in fact is to transfer the variance from
the measurement (output) side of the process to the other
(input) side of the process. This is nicely illustrated in
Figures 4 and 5. So although one may be able to try and keep the
level fairly constant in the first cell of the bank by using the
best and fastest possible tuning (which unfortunately as stated
above, is not all that fast), the variance is immediately
transferred into the second tank, and because of the dynamics of
the processes, and the tuning, it is usually actually increases
the variance in the next cell. Hence this “amplifier”
effect.
Fig. 4
Fig. 5
Feedback control cannot deal very effectively
with this problem. Therefore the most essential thing in this
type of process is to ensure that the feed into a floatation
bank is kept as steady and constant as possible. To achieve
this, it is necessary to absorb fluctuations in feeds occurring
upstream in the process by installing a surge tank situated
immediately before the bank. This tank must be sized large
enough to allow a constant feed to be made into the bank under
all load conditions. The surge tank has to have a special level
control applied which keeps the tank from overflowing or running
empty but at the same time tries to keep the output flow as
constant as possible; and if and when it does change, then the
change must be made to be very slow. The best way of achieving
this is to use an “error cube” technique which has been
described fully in Loop Signature P1-29 (available on CD for
persons outside South Africa). (The next Case History due for
publication in 2 months, will also be discussing some more
aspects of this.)
Once the flow into the first cell is nice and
steady, then the feedback control can deal happily with the
level, and does not have to keep on making variations which
would affect the next cell, etc. Unfortunately this plant had
been designed and built without any surge tanks, so most of the
banks in the plant were nearly always cycling badly. To compound
the problem, the outlet of some banks is fed into the input of
other banks, and the outlet of certain banks is also
recirculated into the same bank. These feeds which of course are
now very cyclic, compound the problem dramatically, and there is
very little chance that any simple fix can be made, such as
tuning. Of course the general impression of personnel in the
plants is that all that needs to be done is to tune the
controllers properly.
It should be mentioned that various more
advanced techniques such as feedforward control systems can be
used to try and minimise the problem, but these generally are
not all that successful, as dynamics and conditions change with
time and loads, and the dynamic models used then no longer work
well. One very “clued-up” plant metallurgist I worked with
some time ago, said that her advanced control floatation bank
level control system worked well provided it was completely
retuned at least once every 6 weeks. This would not be so
critical if the basic things like surge tanks to ensure constant
feeds were incorporated, which could then allow the base layer
controls to work properly.
If the plant designers and process experts
only had better understanding of the principles of practical
process control, they would be able to design and get their
plants to operate much better with vastly increased recovery
rates, and far less problems. Their perceptions of the operation
and capabilities of controls are generally completely wrong.
