Professional Documents
Culture Documents
When analysing a control loop, one of the of the process by recording the response of valve and the measurement. As a general rule,
important things that one must do is to the process to a step change on the input. on a flow loop the PG is considered reasonable
determine the dynamics of the process. Process One then graphically tries to determine the if it lies between 0,5 and 2. Problems can
dynamics can be loosely defined as various values of the two or three dynamics, and then result if it is outside that range as described
types of mathematical components that may plug the values into a formula to establish the below. The value of the PG when measuring it
be present in the response of a process to a tuning parameters. This is called ‘model based on flow loops in particular can give valuable
change in the input. Industrial type processes tuning’ and has not proved very successful, information as to whether the valve is
may include one or more of the following even though many other people have also oversized or if the transmitter span is
dynamics: process gain, dead time, first and come up with a variety of formulae. It works too wide.
also higher order lags, multiple lags, and occasionally, but not often, and seldom gives Oversized valves (PG>1) can cause two
positive and negative leads. really good tuning. However it can be adapted problems:
Apart from gaining valuable information to serve with a certain amount of trial and error 1. All valve problems are effectively multiplied
about the components in the control loop, it is testing if you really understand what you are by the PG value so that in closed loop
essential to identify and possibly quantify these doing. control the control variance will be made
dynamic components in order to be able to In this article we are going to discuss the worse by that value, and any cycling
tune the controller scientifically. In passing, it process gain of a self-regulating process. Figure will cause the PV’s amplitude to be
should be mentioned that the various dynamic 1 shows how the PG (process gain) is measured correspondingly that much bigger than the
components generally affect self-regulating for the response on a simple self-regulating PD’s amplitude. For example if you have
processes differently from the way they do to process like a flow process. It is the ratio of the a stick-slip cycle occurring with the valve
integrating processes. This can make tuning of change in the PV (process variable) that results cycling over a range of say 2%, and the valve
processes of complex dynamics very difficult, from a change in the PD (controller output). is four times oversized, then the amplitude
and it really becomes necessary for one to Note that it is best done by making a step of the cycle on the PV will be 8%.
employ a good tuning package for these. change on the PD with the controller in manual. 2. Many fast self-regulating processes like
Fortunately however, approximately 75 to 85% The ideal value of PG is unity, as you are fast flow loops are tuned with a very small
of all processes encountered in normal industrial then making use of the full ranges of both the proportional gain, and an extremely fast
process plants can be treated as having only two
or more of the first three dynamics mentioned
above, namely process gain, dead time, and
a first order lag. These processes are relatively
simple and in general if they can be identified
properly, they can be tuned without the need
for such a tuning package.
Ziegler and Nichols quickly realised in the
1930s that the ultimate method of tuning using
dynamic operational testing which oscillates
the process sinusoidally over a wide range
of frequencies was not practical or feasible
on industrial processes. They then pioneered
research to establish other alternative methods
for tuning. Although one of their methods was
to use a limited cycle on the loop, and which
works well on ideal loops, their other methods
all involved making a mathematical model Figure 1.
Figure 2.
integral. For some unknown reason, some To overcome this, one needs to increase the down on the measurement scale. Nearly
makes of controllers (even some very well integral value, usually quite substantially. all transmitters have a limited rangeability,
known ones) have a very limited low end This adversely affects pole cancellation with very few being able to give accurate
proportional gain setting. Typically, very tuning, and can drastically slow down the or even valid measurements near the
commonly, it is limited to 0,1 (or 1,000% tuning. bottom of their range. (Also be aware that
proportional band). I have even come across many types of transmitters have accuracies
one make of PLC where the lowest value you Transmitters with too wide a span (PG<1) specified in terms of ‘% of full scale’, instead
can insert is 0,25. In many cases on these also can cause two problems: of ‘% of reading. This means that these
fast loops, and if the valve is oversized, then 1. The accuracy and quality of the transmitters are less accurate as you move
these controllers will not allow you to insert measurement can be badly affected. This is down the scale.)
a small enough gain to prevent instability. because you will always be working lower Continued on page 18
Figure 3.
Continued from page 17 the controls of an offshore oil rig over the range over a test lasting some 22 minutes.
2. The PG is now small and must be less than internet. The testing of the controls on this During this period there seemed to be no
unity. As the control speed of response to particular rig was very difficult due to the noticeable movement of the PV. It appeared to
SP (setpoint) changes when in automatic safety requirements. Any proposed test had me that the valve was sticking. I wanted them
is largely dependent on the product of to be first submitted to a panel for evaluation to move the output further but the operators
PG (process gain) and Kp (controller gain) and approval with values given of size of would not allow this as they thought that the
i.e. PG x Kp = Response, it means that as steps etc. All the tests had to be performed valve might suddenly jump and cause a big
PG gets smaller you will have to put in a by a senior operator and could not be done surge in flow.
proportionally bigger Kp to get the desired by the optimising team. This of course makes Later that day the C&I staff on the rig
response. Now, although it is sometimes optimisation very slow and frustrating. emailed me the tests so I could examine them
necessary to use high Kp values to get However it is understandable as oil rigs in more detail. On expanding the tests, to my
the control response you need, what is are very hazardous plants, and safety is amazement I saw that the flow was reacting to
happening here is that you are putting in a paramount. our changes fairly well but in tiny movements.
higher Kp than would be needed if you had The process in the example was a simple The PG turned out to be an amazingly low
a properly spanned transmitter. Remember, flow loop. The operators had been complaining 0,09. This is far too low and indicates there
the higher the Kp, the more you are going that they had to run the loop in manual as it is something seriously wrong with the
to move the valve around, and this will just didn’t seem to work properly in automatic. measurement scaling. The reason the process
result in a shorter valve life. I have come We first tried performing a closed loop test didn’t respond on the closed loop test was
across quite a few cases where with very with the existing parameters where one makes that their tuning was completely wrong, with a
over-spanned transmitters, they have had a step change in SP. They would only allow proportional gain some 40 times too low and
to insert such a high Kp in the controller to a 9% change to be made. It was also initially an integral three times too slow.
get the desired speed of control response, found that they had the wrong controller setup It should also be pointed out that the
and this has caused the controller’s output on the controller, so that on SP changes there flow was being measured using a differential
to fluctuate widely, with unacceptable valve was only I (integral) action. They then changed pressure transmitter over an orifice plate. The
movements. This is particularly the case this so that both P and I action came in. The flow was running below 20% of measurement,
where there is fairly high process noise on result was that even after a few minutes it which is considered a region too low for this
the PV. To overcome this, C&I technicians or appeared as if the PV had not reacted. type of measurement. Generally the rule of
artisans often either drastically filter the PV, An open loop test was then performed, thumb is that the measurement should be over
or else equally drastically reduce the Kp to and this is shown in Figure 2. It should be 25%. However this did not really influence our
‘save the valve’. This means that one may not noted that when performing these tests on observations.
obtain the good control needed. our recording system we only see the variables This is a very interesting example of how
A brilliant example of this recently occurred trending on a 0-100% scale. The PD was moved measuring process gain can give you a big
when I was supervising loop bump tests on up and down in a number of steps over a 20% insight into loop problems.