Michael Brown Control Engineering CC 

Practical Process Control Training & Loop Optimisation

 

CASE HISTORY 196

 

UNSTABLE CONDENSATE LEVEL CONTROL
 

The operators in a petrochemical refinery were having great trouble in trying to stabilise the condensate level in a vessel which was adversely affecting other loops downstream.  Several unsuccessful attempts had been made to retune the controller.


Level processes almost always fall into the integrating class of processes, which have been described in several previous articles, the most recent being in Case History 194 published a couple of months ago, so I will not repeat the basic description of the characteristics of such processes.  However it was mentioned that the PV of an integrating process like a level, will remain constant when the input and the output of the process are of equal value, and that the value of the PD (controller’s output) at that point is referred to as the “Balance Point”.  When one changes this value of the PD by a step, the PV starts moving in a constant ramp.


Now if one was to step the PD back to the previous Balance Point, the process should once again balance out and the PV should then stop ramping and remain constant at the value where it had reached.  If it did not remain constant and continued ramping, and most probably at a different ramp rate, then one or more of the following things would have occurred:

 

1. The valve had not moved back to the correct position due to positive or negative hysteresis.
2. There has been a load change in the process so the balance point had changed.
3. The process itself was non-linear, examples of which are  measuring levels over non-linear regions, for example in a horizontal cylindrical vessel like a boiler, or a gravity feed vessel (where there is no pump on a bottom outlet).  The balance point changes all the time as the level increases or decreases.


When we came to do the testing on the loop we did indeed find that the level was cycling slowly with the controller in automatic.  Now continuous cycling in integrating processes is generally due to one of the following causes:

 

a. Positive valve hysteresis plus the use of the I (integral) term in the controller. 
b. Negative valve hysteresis.
c. Bad tuning.

 

Now bad tuning of integrating loops is usually mainly due to incorrect settings of the I term, as most people do not have a “feeling” or understanding of how to properly tune integrating process controllers.  It is also generally very difficult to cause instability in the majority of integrating processes by using too high a P (proportional gain).  This particularly applies to levels which generally have quite large retention time. I have found that most people usually don’t use a high enough gain on integrating processes to get reasonable response in automatic.


This is very different than the tuning of self-regulating processes, particularly fast ones like flows, where too high a P gain can easily cause instability.   I have found that most practitioners have a better “feeling” when it comes to tuning self-regulating processes, and don’t really understand how to go about tuning integrating processes which require a completely different “mind set”.


If an integrating process is cycling due to one or both of the reasons listed in points a) and c) above, one can often easily stop integrating processes cycling merely by switching off the I term (and D term too if it has been used).  The process then usually stops cycling and finds its own balance point.  This may not leave the PV at set point, but this does not matter when one is running tests.  The equivalent of a P only controller on level control is the very ancient method of using a mechanical ball valve to keep the level constant.  These do not cycle.


Tests were performed on the condensate level.  The first was a closed loop “As Found” test which consisted of running the loop in automatic without changing any settings.  This test is shown in Figure 1.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 1

 


It was seen that the loop was cycling very slowly around set point.  The set point was then stepped down by 10%.  This appeared to have very little effect on the PV for quite a long time.  The level did then move down slightly but continued cycling.


Please note that the original tuning was P = 0.3, and I = 10 minutes/repeat.


An open loop test needed to be performed next.  However before such a test could be conducted, it was necessary to stop the cycling and find the balance point on the PD.  So the Integral term was switched off.  However the loop was so slow in responding it was decided to change the P gain to an arbitrary value of 2.0.  This worked well, and the process came down to balance quite quickly.  Once at balance the test was performed and is shown in Figure 2.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2


The value of the PD at the balance point was 15.13%.  Assuming the valve calibration was correct this is quite low down, which would mean that the valve was working at a very small opening. This would indicate that the valve was probably largely oversized.  (A well known general “rule of thumb” is that control valves under normal operating conditions should operate above 20% opening, as various problems can occur when a valve is trying to control close too close to seat.  One of these problems can be in machining the valve parts correctly so that the inherent valve characteristics are still followed at low openings.  Another problem is that the differential pressure across the valve could have increased and in some cases the actuator may not have sufficient force to overcome this pressure which could result in the valve cycling open and shut.)


Once the process was in balance the PD was stepped up by 5%.  The process responded with the PV dropping in a nice constant ramp.  The PD was then stepped back to the original Balance Point.  If all was working correctly, the PV should have levelled out.  However it still continued ramping down although at a much slower ramp rate.  As discussed above this could have been caused by either a load change or by positive valve hysteresis or by non-linearity in the process or in the valve characteristics.


The PD was then moved down in small steps of 1% to try and restore balance.  However after having closed it down by 3%, the PV started ramping up again, and this time it took 4 x 1% steps in the opposite direction to restore balance. The PD was then stepped down again and the PV started ramping up.  However after a while the ramp rate started dropping off slowly.  This was most probably due to a load disturbance occurring which would change the balance point.


Just out of interest this type of thing makes testing of integrating processes really challenging at times, as conditions in plants change frequently in integrating processes like levels.  For this reason it makes life so much simpler to do the testing if one can actually also record the flow though the valve.  This allows one to actually see what the valve is doing  I have always recommended that if possible a portable flow meter (like a clamp on ultrasonic flow meter) be used if there are no flow meters installed in line with the valve.  It can save a lot of time and also possibly result in plant savings by allowing integrating processes to be properly analysed.


In the present case the test allowed us to calculate reasonable tuning parameters on the two ramps that we could use after the valve had been stepped.  The new tuning we decided to use was P = 3.4, and I = 0.9 minutes/repeat, which is a huge difference from the original tuning of P = 0.3, and I = 10 minutes/repeat.


It is of interest to note that the open test also allowed us to determine the retention time of the level, and this was approximately 27 seconds, which means that the process was really quite fast for a level.  As a rough rule of thumb, level processes with retention times with 10 seconds or less are regarded as really fast and can be quite difficult to control particularly if the dead time is significant.  A “rule of thumb” is that the DT should be less than a tenth of the retention time on integrating processes.  The dead time in this case was about 2.3 seconds, so the controllability was just on the limit, and required very precise tuning.


The final closed loop test with the new tuning parameters is shown in Figure 3.  It can be seen that the control is now working pretty reasonably.  The response is nice and fast, and although a certain amount of cycling can still be expected because of the valve problem.  However the control could now react quickly enough to pretty much keep the PV at set point.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3


It is a good example how bad tuning as well as valve problems can result in cycling.