Control Loop Case History 108

A previous article on tuning feedback controllers to decouple interactive processes was published in South African Instrumentation and Control in Loop Signature P2-5 in January 2006. (The series is also available on CD from the author for persons outside Southern Africa). To recap, in general feedback control loops do not work very well together in interactive situations.

Two interactive loops can be decoupled to a certain extent by tuning them in particular ways, but the best solution is to decouple them dynamically which involves using feedforward decouplers in each direction. The problem with tuning is that all one can do is to tune the one loop as fast as possible, and the other slowly. An excellent extreme example of this is often found in plants in cases where two loops are very interactive. The operators know exactly how to decouple them. They place the one loop in manual. A loop in manual is effectively an infinitely slow tune.

Another problem of decoupling loops by tuning is the fact that if you were to tune each of the loops with the other in manual, the dynamics of each loop alters as soon as the other loop is placed in automatic. This is because the coupling between the processes changes the dynamics of each loop when the controllers are in automatic. Therefore it is actually necessary to tune each loop with the other in automatic. This can create a further problem insomuch that if they are now both tuned properly to work with both in automatic, then if either loop is placed in manual, then the other loop that is still in automatic, could now react either very slowly, or very much faster.

A good example of successful tuning two interactive loops was found in a petro-chemical refinery where I recently performed some optimisation.

Fig. 1

As shown in Figure 1, the one process consists of an outlet gas pressure control on the top of a distillation column. The gas passes through the pressure control valve, and then goes through a scrubber, the outlet pressure of which is also controlled by another valve. Thus there are effectively two pressure control valves in series. This system is only really workable due to the fact that the scrubber has got quite a large capacity, which slows down pressure changes between the two valves.

With the loops being quite interactive, the original tuning of both controllers was really terrible. Controller PC1 was tuned very slowly but very cyclically, whilst Controller PC2 was tuned so slowly that it took almost hours to react to a change. Both controllers also had huge PV filters which as detailed in the Loop Signature first series, is highly undesirable.

_{Fig. 2}

Figure 2 shows an open loop test with both controllers in manual. A step change is made on the output of controller PIC1, whilst the output of controller PIC2 is kept steady. The test illustrated the following things:

Both loops reacted to the change, but Loop 1 is an integrating loop and Loop 2 is self regulating.

Further analysis of Loop 1 shows the process is a simple deadtime integrator with very small deadtime of about 10 seconds, and a process gain of 0.00057, which is equivalent to a retention time of almost 3 hours. (Purely out of interest this type of process was referred to in Loop Signature P2-19 as a "Slow Pure Integrator" which is very difficult to make unstable.)

Process 2 is extremely fast compared with Process 1 – see analysis below.

Fig. 3

Figure 3 shows another open loop test with both controllers in manual. A step change is made on the output of controller PIC2, whilst the output of controller PIC1 is kept steady. This test shows:

Both processes again react to the step change, but it takes a relatively long time for the pressure change to travel "backwards" through the scrubber and affect Process 1’s pressure.

The analysis of Process 2shows it is basically a double lag, self-regulating process with a very small deadtime, and the two lags have time constant of about 40 seconds each. A process with these relatively simple dynamics is not difficult to control fairly quickly. It is also a very much faster process than Process 1.

It was concluded from this, that if one was to tune Controller 2 as quickly as possible, and then put in a medium "slowish" tune on Controller 1, then interaction would be minimised.

To achieve this it was necessary to tune Controller 2 first using the step test shown in Figure 3. It was then placed in automatic, and a further step test (not shown) was then performed by stepping the output of Controller 1. Controller 1 was then tuned from this step. The response took the slightly altered dynamics of Process 1into account, which resulted as Controller 2 was in automatic. This enabled the tuning for Controller 1 to cope with these altered dynamics.

Fig. 4

The tunings worked wonderfully. Figure 4 shows a 16 hour performance trend with the plant operating under normal conditions. A trend like this where there are quite a few load disturbances occurring, is very useful to judge control performance. One can see the controllers’ outputs moving around to keep the PV’s on setpoint. In particular there was an absolutely huge load disturbance which can be seen near the end of the recording, where both outputs had to move dramatically up to catch the disturbances. It is wonderful to see how little variance occurred on the actual PV’s. Really great, and highly impressive control performance. (It should be noted that trends recorded when operating conditions are very steady, are not useful as a means of judging of control performance. It is only possible to judge the control in these sorts of tests, if load changes occur.)

It is also quite interesting to realise that a control system doesn’t ever eliminate variance. It just transfers it from the one side of the process to the other, so the PV is kept constant, and the variance now takes place on the PD (controller output, which is the input to the process.)

Here are the "before and after" controller tunings:

  One can see the vast differences between the old and the new. Also it's a complete mystery as to why they had put filters on the PV signals. There was absolutely no need for them. It is also quite interesting to see that they had used a relatively large derivative on the fast loop and not on the slow. Once again it is obvious that the people who had done the original tuning had not attended courses on practical control.

Index to articles

Michael Brown is a specialist in control loop optimisation, with many years of experience in process control instrumentation. His main activities are consulting, and teaching practical control loop analysis and optimisation. He gives training courses which can be held in clients' plants, where students can have the added benefit of practising on live loops. His work takes him to plants all over South Africa, and also to other countries. He can be contacted at:
Tel (011) 486-0567
Fax (011) 646-2385
E-Mail:  michael.brown@mweb.co.za