Control Loop Case History 107

The horrible flywheel effect

I have been working in a smelting plant where they have severe control problems. To make testing more difficult and onerous, nearly all their processes have excessively slow dynamics, with many of the loops being extremely slow temperature controls.

As in the case of many mining plants, it would appear that the controls and control systems were designed and implemented by people who do not really have a great deal of understanding of the practicalities of control. I believe that many plants are designed with only the processes in mind, and with little or no thought initially as to how to control them. Once all the designs are done, then the controls are put in; sometimes literally as an afterthought.

I know that some large mining groups have separate project groups who design new plants. It would also appear that often these groups may not have all that much knowledge of the actual practical implementation and operation of plant controls. Discussing this at a recent conference with a group of senior metallurgists and control engineers from various mining plants, I was told that they find many design errors and faults in new plants, which they make note of, and then feedback that information to the project design group. However when the next plant is built, they find exactly the same problems have been repeated. When they complain, they are always fobbed off with some remark like “it’s a question of economics”, or “in our experience it works fine.” All very frustrating for these good people.

When it comes to the controls and control strategies it is obvious that in many cases very little thought and knowledge has been employed. Often poor, or even unsuitable control equipment is used, probably due to ignorance, or based solely on cost rather than on the control requirements.

Controls in many mining plants also often seem to have been designed by metallurgists. Many metallurgists I have met who do not come from a chemical engineering background, usually have not undergone control courses, and have picked up their control knowledge in the field. Although many of these people no doubt are very competent and do a great job, I have on occasion come across some rather weird and wonderful control strategies that they have concocted. Some of these would have even made Heath Robinson quite envious. One fairly common example which I have come across several times in plants is finding two control valves in series in the same line in close proximity to each other, the one being for controlling pressure, and the other for controlling the flow. This is very interactive and doesn’t work at all well, usually resulting in on of the loops being left permanently in manual.

An example of a strange control strategy in this particular plant, was a control for a very important air temperature, which is critical for satisfactory operation of a drier further downstream. Instead of putting a direct feedback control onto one of the important control loops on this temperature, the designer had come up with some extremely complicated and highly involved control system which involved using a PI controller as a sort of ramp generator that continuously tried to get its output up to maximum, but actually attempted to achieve the control by writing a complicated formula from a calculation block that adjusted the high output limit of the controller up and down in an attempt to keep the critical temperature operating between two limits. Unfortunately when it reached the high limit another control came in which rapidly closed down the damper, and caused a huge bump in temperature. This is disastrous for the drier control as this air temperature needs to be kept as constant as possible. The probable reason for this design will be discussed a little later.

The interesting thing about it is that the people in the plant who have been using this system for years have got used to working with it, and didn’t want it changed, even with the bad control. There were remarks like: “We want you to make it work better, but don’t change the way it works”, and “Can’t you tune it better?”, or “Possibly the damper has problems”. People are very resistant to change, even if you can make it much better by implementing a much simpler and better control strategy.

Then there were also things in the plant, like using multiple controllers to control a single variable, where one would have done the job much better.

Another problem we unearthed was with a pressure control where three fans are used in parallel each with its own up-stream damper to suck gases out of a furnace. The idea was that depending on load, one, two, or three fans could run. The problem was that the dampers were relatively crude flaps sitting in a duct of about 1.5 meter diameter. When closed they did not shut off tightly, to say the least. This results in the fact that if a fan is off-line with its damper shut, then part of the gases being pushed through by the working fans are bypassed, and recirculated through the non-working fan instead of all going on up the stack. It was interesting to see that the fan that was off-line was actually being spun backwards. Thus a tremendous amount of energy and effort is being wasted. It was not surprising that it was extremely difficult trying to get the pressure up to setpoint.

Returning to the air temperature control; I was informed by a senior metallurgist that it exhibited “a severe flywheel effect”. He explained that if a change was made it took a long time to for it to take effect, and if the change was too large and eventually resulted in the process variable overshooting the setpoint, it would take a long time before corrective action would take effect. This is very much the sort of behaviour you see when changing speed of a flywheel with a large mass, which exhibits a high degree of inertia.

“A severe flywheel effect” is a fairly good way of describing processes with very slow dynamics. Essentially one tunes a controller to try and match the dynamics of a process as closely as possible. (My Part 2 Loop Signature series on control of more difficult processes, which were published in this magazine over the past couple of years, give detailed information on this subject for every type of dynamic that one might encounter in industrial process control.) The important thing here is that it means that processes with slow dynamics can only be controlled with a slow tuning, whereas fast processes can have fast control. This makes the control of slow processes more difficult from several aspects, including making the tuning more difficult because of the time it takes, and also from the importance of the need of the final control element (e.g. the valve) to do its job exactly as dictated by the controller. For example if you have a valve that has non-linear installed characteristics, or exhibits problems like hysteresis, then the controller has to do extra work to also correct for the valve problems, which may take a very long time, and result in rotten control.

In essence the purpose of the system on this smelter was to control the temperature of a flow of air which is being blown into the drier. The air is heated by passing it through a “hot gas generator” which utilises a fluidised bed. There are quite a few controls loops involved on this unit, but the two most important are:

a. Temperature of the fluidised bed of the generator, which is controlled by varying the amount of fuel being fed into the generator.

b. Outlet temperature of the hot air being heated as it passes though the generator. This is controlled by varying the flow of air coming into the generator through a damper on the input.

Figure 1 shows a sketch of the system, but with a more conventional control strategy than was actually employed in the plant.

Fig 1

As can be expected these two controls are highly interactive. If the output of either controller moved it had a severe effect on both of the PV’s (process variables).

Both of the temperature processes are very complex and interesting from the dynamic point of view. The bed temperature is an extremely slow integrating process with a large lag on it. Integrating processes are those that can only stay constant if one balances the input and output, which in this case is heat energy. If the input is larger than the output then the temperature will keep rising in a constant ramp, and will “run away”, and vice versa. (This is very similar to control of tank levels, except levels are much faster). Integrating processes are hard to control in manual because of this and require constant supervision. They really must operate in automatic. To illustrate the difficulty of manual control, we observed an operator trying for over 3 hours to stabilize this loop in manual without success. The slowness of this process coupled with the lag make it extremely difficult to control, and feedback control will perforce have to operate very slowly for stability. This is the loop that behaves with the “flywheel effect”.

The outlet temperature loop on the other hand is basically self-regulating (like a flow control) and is very fast relative to the bed temperature. A test was performed on it that proved it was extremely easy to control well using a conventional PI feedback controller. The existing control strategy on this process which was mentioned earlier using the “ramp generator” is very odd, and no one we spoke to could really explain properly how it works and why it was designed like that. The most probable reason that it was instituted is because there is a huge problem associated with this control as it interacts so badly with the bed temperature. If a normal control, operating with a medium fast response, was to be instituted on the outlet temperature, huge swings on the bed temperature result. There is a very tight high limit trip on the bed temperature, which makes the situation even more difficult. Therefore it becomes necessary that the outlet temperature control must operate extremely slowly. I believe the original strange control system was an attempt to slow the air temperature control down to prevent it causing bumps on the bed temperature with its “horrible flywheel effect”.

The recommendation was made to the plant that they should replace the existing outlet temperature control system with a conventional control loop as shown in Figure 1. This would, as mentioned above, need to be drastically detuned to minimise bumps on the bed temperature.

Another thing that would help a great deal would be to employ dynamic decoupling (feedforward in both directions) between the two loops. This has also been detailed in the Part 2 Loop Signature series.

These suggestions have been put forward to the plant people, but to date no decision on whether to try and implement the changes or not have been made. It would be very interesting if they agree to go ahead with the modifications, but as stated earlier they may prefer to live with the existing system which they are familiar with, even though it doesn’t work well. People get used to living with problems, and hate changes.

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