CONTROL LOOP CASE HISTORY 169
TUNING A VERY DIFFICULT TEMPERATURE CONTROL LOOP
As I have mentioned before, Greg McMillan, one of the top control experts in the world has said that he finds temperature control loops generally the worst optimised processes as most people try tuning them without any scientific tuning tool, and the average human being has not got a long enough life span to tune them properly by “trial and error”. I recently came across a good example of this when optimising some controls in a chemical processing plant.
The loop in question is a temperature control loop with which the operators in the plant were having great difficulty. It was terribly slow and never seemed to be able to be controlled. It is a critical temperature, and the operators were very scared of the temperature moving more than one or two degrees Centigrade. (The transmitter range is 0 – 150°). A plant control engineer had spent many days trying to get a good step in manual to be able to determine the dynamics of the process. However it seemed almost impossible to get any meaningful response from the steps. The temperature didn’t seem to move properly or repeatably in spite of numerous small steps that were made.
The operators were finally persuaded to let us make much bigger steps on the controller’s output, and after quite long tests taking nearly two days, we were able to get a process response that was representative of the dynamics. We could then try tuning.
The dynamics of the process turned out to be extremely unusual. Basically it is a self-regulating process with a deadtime of about 14 minutes and time constant of 8 minutes. This makes it decidedly very slow, and hugely deadtime dominant, characteristics regarded as making a process difficult to control. Such processes need to be controlled with a slow control to avoid instability. Even more interesting is the fact that the proportional gain of the process is 0.04 which is unbelievably small for a self-regulating process. (Normally it is a “rule of thumb” that self-regulating processes should ideally have a process gain between 0.5 and 2.0).
The process gain of 0.04 means that the process can only be controlled over a range of 4% of the measuring span, for a full movement of the valve from zero to 100%. To express it in another way it means that in this case control is only possible over a range of 6°C, which is very small, and could result in no control if the process load changes over a bigger range.
In this case the plant personnel previously had little luck in trying to tune the loop by trial and error, and the existing tuning parameters as found in the controller were P = 0.1, I = 4.5 min/repeat, and D = 1.1 minutes. The closed loop response to a setpoint step change of 3°C is shown in Figure 1. (These tests were done on an accurately modelled simulation, as it would have been difficult and time wasting to do them on the plant). It showed the process took an absolutely amazing 75 hours to reach the new setpoint! It is also interesting to see how far the controller output had to move to make the small process change, because of the ridiculously small process gain.
Once we had managed to do the good open loop step test, and obtain a response that was truly representing the process dynamics, we could then use the Protuner to tune it. The new tuning is P = 6.0, I = 12 min/repeat, and no derivative. The final closed loop simulated test is shown in Figure 2 on the same time scale as used in the first figure so one can compare the difference in response. The process got to setpoint now in 3 hours which is 25 times faster than previously.
Just out of interest in actual fact the control worked extremely well and the response was very similar to the simulated one. The loop has been running in automatic and keeping the operators happy with it, ever since.
To finish off this article I have included a test performed on a flow loop which tended to easily cycle. Figure 3 is of a closed loop test performed on the loop. The test shows that the valve has a very interesting and unusual characteristic, which I have not come across previously. It sticks very badly, but only when opening. In the figure one can see on the first two steps how the valve sticks for approximately 8 seconds before it starts moving. However on the third step when the setpoint stepped down, the valve reacted immediately, and started closing downwards.
It is possible to get some sort of control still even with this bad valve behaviour, which can, and did cause cycling with normal tuning, one must tune the controller very slowly to compensate for the 8 second sticking period, which should be considered as deadtime in the loop. Obviously this situation should not be left like this, as the control is now terribly slow, and the valve should be fixed.
One other thing that can be seen in the test is that the valve/positioner combination has problems (possibly in a linkage) which causes the valve to jump around sometimes as annotated in the figure.