I have two questions about loops control tuning problematic and would greatly appreciate the opinion of the experts on the subject.

The first question refers to the economic loss caused by improper tuning of PIDs. are there any statistics on that?

The second question refers to the skills that should develop Engineering students in order to be in a better position to face today's demands. Is it convenient for example to reinforce the fundamentals of the theory of control? Or is it wise to spend more time managing automated and intelligent software tools?

 

Just MHO.

People obsess a lot about things that do not matter much in the long run. Tuning loops so they are "perfect" is one of those things that is all but impossible to actually do, and maybe even counterproductive. Most times you are better off tuning them so they are a little under damped so they don't go crazy if some kind of unforeseen disturbance happens.

You can have a loop that is tuned so it works great under normal operating conditions but goes off the deep end if you get a valve that is sticky.

Yes, some times it is also important to have a loop tuned as tight as possible. But, if you do that, you need to account for the possibility of it oscillating due to some kind of process disturbance.

I am not sure that the so called "theory of control" is worth spending a whole lot of time on, given how rarely it is relevant in the real world. In a plant with hundreds or thousands of PID loops, no one can spend all that much time analyzing any one loop. You are almost forced to use some kind of software tools to deal with them just due to the sheer number of loops.

The other thing is that it is not unheard of for loops to interact with each other in unexpected ways. It is not always real obvious just what potential effects outside of the controlled loop itself there might be.

You also need to realize that in many cases PID loops have been misapplied, and no amount of tuning will solve that problem.

--
Bob
http://ilbob.blogspot.com/

 

a) Quite a few studies have been conducted over time on the performance of control loops. The results are always quite disappointing: On average about 30% of the loops are permanently on manual, only a fifth is actually delivering the requested performance, etc.

The associated loss is of course very much dependent on the service and the importance of the loop: A poorly performing temperature loop on a big distillation tower or a bad reactor temperature control of batch reactor can cost easily several ten thousands of Euro or Dollar per year - every year!

By the way: When assessing the performance we should not only look at the controlled variable (CV) behavior but also on the manipulated variable. The question is simply: How is the performance achieved? Just think about a little comparison: You drive your car and want to keep your speed as constant as possible. Two extreme approaches are: Either you try by making small, very precise actions on the gas pedal or you work like an on-off (“bang – bang“) controller – full down, no gas, full down. You would be surprised about the performance a real expert can achieve with a bang-bang controller – but at what price!

b) Most education is totally overloaded with theory, too much math, too little understanding of the real world and its problems and “imperfections”. Thus the best advice I can give from 40 years in this field is: Get on some really and fully practice oriented training course (I can’t resists to point at our offerings http://www.act-control.com/training.html ) and to get some really useful tuning tool such as our TOPAS (which, by the way, also allows you to calculate the incentives achieved from improves control performance).

Hans H. Eder
www.act-control.com

 

> The first question refers to the economic loss caused by improper tuning
> of PIDs. are there any statistics on that?

I am a motion control guy. Precision and speed is most important. One can't make parts that are out of specification and slow speed hurts production.

> The second question refers to the skills that should develop Engineering
> students in order to be in a better position to face today's demands. Is it
> convenient for example to reinforce the fundamentals of the theory of control?

It depends on which ones. I personally think that half the time is wasted teaching techniques that don't work in the real world. However, control theory does work. Simple pole placement works well and is easy to implement and keep the closed loop poles in a safe position.

> Or is it wise to spend more time managing automated and intelligent software tools?

This is a necessity. I have had to write auto tuning programs because I can't count on the customers to be able to tune a system on their own. The first question our tech support guys got before we added auto tuning was "where do I first set my gains too?".

In general, engineers forget what they have learned about control when the graduate and basically use trial and error to tune a system. Very few really understand what they are trying to tune. Z-N doesn't require knowledge of what is being tuned but Z-N only works for certain types of systems.

Control theory works but not as it is in the text books. One must model the system using differential equations. Laplace transforms assume the system is linear and we all know that plant gains may be non-linear but do you know that time constants may not be constant? PID are simple and not capable of controlling these non-linear systems without help. However one can use feed forwards and recalculate PID gains every scan to adjust for changing gains and time constants. I can do it but I didn't learn it in college. My control books didn't even come close to explaining how to use differential equations. Everything was explained in terms of Laplace and Z transforms.

Finally, students usually have access to programs like Matlab. Matlab is good for getting answers but Matlab has a lot of knowledge built into it so that students don't need to truly understand the required math.

 

Loop tuning is rarely an issue.

Either it is a simple loop, single input, single control variable, with no outside disturbances, and easily tuned. or it is a multi-variable system with wild disturbances, variable time delays, and process gains that are a function of the operating rate in the latter. loop tuning (either manual or self adjusting) is a secondary issue given the importance of your multi-variable control scheme and the form of the wild variables outside of the control scheme.

 

Thanks very much for your attention!

Rob

 

In the process industry (oil, gas, chemicals, food, pulp & paper, &#8230 loop tuning DOES matter. This has been confirmed by numerous studies over the last 40 years. This inherently means that it can become easily an issue even for relatively simple loops when top performance yet robustness is demanded.

Just to mention one mechanism: Better tuning allows moving variables closer to constraints, i.e. to better utilize the capabilities of the equipment which results in – depending on the situation – more product yield and / or quality, higher throughput, etc. Example: By improving a batch reactor temperature controller we could reduce the batch cycle time from 24 hours to 17 – quite an increase in plant capacity.

Furthermore, where a controller acts on an energy supplier (oil, gas, steam, … stream) tuning can clearly reduce the energy consumption. This is comparable to driving a car: We can try to keep a constant speed either by fine adjustments of the gas pedal (a “well-tuned” leg) or by bang-bang control – full throttle / none / full…In the latter case the fuel consumption can easily be twice as much as for the first case.

@ “d”: You just pointed at the two extreme cases - simple, easy loops and a MV situation - but the vast majority of the loops / control schemes lies in between and there are many loops that are not so easy to tune due to long deadtime, open loop overshoot, inverse response, etc. or simply due to the necessity of careful action on the manipulated variable in order not to upset the downstream equipment - on top of the demand for good performance of the controlled variable.

Hans H. Eder
www.act-control.com