Control Loop Performance - what are users' experiences?

J

Thread Starter

Jim Ryan

I've read with interest all of the comments for PID Loops under 'Thermal Overload'. Most relate however to identifying the appropriate tuning constants for individual standalone controllers. Good questions and comments.

What if tuning is not the problem ? What if for example, a group of loops are cycling together, or two loops are interacting ? What if a control valve I fixed last week suddenly begins to stick ? What if my operators are messing around (frequent mode changes, tuning changes) with the controllers on nightshift. I've got all of the these problems and all of these conditions impact how my plant operates. How do I identify conditions like these in real time if I've got 2500+ control loops in the plant ?

I've read a couple papers on the subject and I'm wondering what other companies experiences have been with some of the commercial controller performance assessment packages that are now available on the market ? Which ones have you used ? Have you been able to measure an ROI ?

JR

P.S. I've posted a similar question under Process Control
 
R

Randy Miller

Jim,
I work for Honeywell in the Loop Scout team. I'd be willing to set up a call with one of our current customers if you want first hand answers to your questions.

email me at: [email protected]

With regards,
Randy
 
S

Sylvain Millette

Jim,
You bring a lot of great points and concerns. My company actually performs control loop performance evaluations. Our experience has been that only 1 loop in four actuaklly reduces process variability. The remainder either do nothing or actually increase. The reasons for this increase in process variabilirty in auto are are all listed in your questions, with a couple of other ones.

To answer your question, there is some software out there to perform some form of online monitoring, but they can only point you in the right direction. The typical variability index used is the Harris Index. However, as I mentioned, it only gives you a pointer. Still, the best way to pinpoint the actual problem is to directly measure the analog signal.

In short, Jim, online monitoring will lead you to the general problem area, but direct monitoring will isolate the problem.

Hope this helps.

Sylvain Millette, P. Eng.
[email protected]
(416) 452-9994
(705) 424-7261
 
B

Bruce Durdle

And a well-designed level loop will in fact make full use of the surge capacity available - so "no variabliity" in level usually means excessive variability in flow.

Flow variablility is usually made up of large amounts of noise which no amount of control tuning will remove.

Like all other problems in process control, you need to have a good understanding of the actual process, rather than set some arbitrary goals for "good performance" from an office.

Bruce
 
H

Heavner, Lou

Variability is inevitable and not all variability is equal. Property control of a sheet product like paper is more important than property
control of a blendstock like cat gas. Temperature control of steam on the inlet of a turbine is more important than on the inlet of a reboiler. Some variability is high frequency in nature and some is low frequency. Controllers can only attenuate variability up to a limiting frequency which is determined by the capability of the control valve, measurement instrument, and loop design. Process design (eg surge tanks) can be used to attenuate high frequency variability and take the variability that would be more damaging elsewhere. Variability can come from many sources and one is the deterioration over time of equipment and instrumentation. Many of the studies that document process variability and loop performance do indicate that a large percentage of loops actually amplify variability rather than attenuate it. Frequently the problem is that a loop is tuned when everything is installed and new. As control valves (and other equipment) wear, the original tuning no longer works. Occasionally, loop interaction will cause a loop to amplify variability. There are companies that perform loop audits and some of them will sell the tools and training necessary for plant personnel to perform their own loop audits. Intelligent instrumentation is making it easier to diagnose problems. It really doesn't
matter if you are comparing the increase in variability to manual control or automatic control at start-up. Many control loops do more harm than good, it can be a problem, and it is widespread. Bruce is correct, you need to define the objectives and carefully determine the problems hands-on.

Regards,

Lou Heavner
Applied Advanced Technologies Team
Emerson Process Management
Phone: (512) 834-7262
Fax: (512) 832-3199
e-Mail: [email protected]
 
A

Anthony Kerstens

Another source of variability are the plants full of button pushers and knob turners who insist on adjusting loop parameters without really having an understanding of the meaning of the adjustments.

Also, there are those who simply rely on an autotune function to come-up with parameters and then walk away without verifying the results.
For example, running an autotune on a heating device from cold to steady state and running an autotune on the same device from warm to steady state can provide drastically different loop parameters. Not to mention that the autotune only provides loop parameters that are close to optimal.

Anthony Kerstens P.Eng.
 
Lou,
Good advice. Nothing will ever replace process understanding and knowledge, but I'm looking for something to point my guys in the right direction, so they are focused on priority problems. Your are right about more of my loops increaing than decreasing variability in my process. A number of my controllers remain in manual most of the time - not an ideal situation. To boot, my DMC+ controllers have been occasionaly switched off due to regulatory problems.

Variability is introduced in a variety of ways as you have suggested complicating matters - (valves, sensors, controller tuning, process upsets and distrubances, operator interactions with the controller, controller interactions) This is precisely the bigger problem I wish to address. I can fix the problem if I know what the problem is - isolating it and focusing my maintenance guys on the 'bad actors' is the issue. This is a complex problem magnified by the fact they have cut my maintenance resources and over 2500 loops in the plant.

The question remains, how do you know where to start when you have thousands of loops to manage ? Our guys do regular PM on many of the loops, and we've even commissioned consulting companies like yours to come in and audit and fix problems. The limitation with this approach is that control performance degrades and the problems continue to return.

I'm looking for end client users experiences using some of the existing technogies that are out there. I have read that there are a number of companies that have built in-house solutions - Eastman Chemical and Dupont come to mind. Do these tools work ? What has the cost/benefit been to the organization ?

good comments and thanks for responding,

JR
 
Jim:
Personally, I don't think there exists any software or method that can fix all kinds of control loop tuning problem. A very critical concept for loop tuning is its stability margin. We have to balance controllers' rubustness and its optimal performance, but they really need to be considered case by case, the opetrating point change? non-linearity? disturbance location? etc.

We have installed a real-time software, ProcessDoctor, by Matrikon Inc., I feel it is useful to get a whole picture of the controller performance. You can directly look at the Bode plot, analysis results in daily or weekly. It gives out a "relative" performance index to let you compare the current situation to the history. So you can easily to choice how fast the controllers.

I have used it for more than two years and like it very much.


Enbo Feng (Ph.D)
Senior Advanced Process Control Engineer
Celanese Canada Inc.
 
L

Leoncio Estevez-Reyes

Jim,

I have just seen this thread recently. In my opinion, there have been a lot of truisms expressed by several people but they have not specifically addressed your questions. I will try to do that briefly.

Yes, we have experience with commercial and in-house packages for loop performance assessment. In fact, our initial in-house package has helped refine some of the existing commercial packages. We have tested and used packages by several vendors and research organizations. We have settled on using a comprehensive Honeywell solution that was put together following our specs, and that includes LoopScout and AlarmScout. This solution is presently being deployed in the majority of our mills.

ROI improvement? Hard to measure a 100% accurately because accountants do not look at this level of details, but we have solid anecdotal evidence that says that we're impacting the bottom line. We have reduced maintenance expenses significantly in some mills by saving up to 75% in unnecessary control valve repairs, and PM's. We have focused the efforts (effectively reducing the time spent) in control maintenance during planned shut-downs, identified and prevented process incidents that could have caused down-time and would not have been identified without our assessment tools, and we have reduced occurrence of process alarms by more than 50%.

So far, the experience of the majority of users is "how was I able to live without this, for so long?"

Now, having said all those great things, let me close with a sobering reality. In my opinion, the secret of success have been 75% or more due to "elbow grease", discipline, encouraging and engaging the right people at the right level, dealing with the organizational inefficiencies, understanding the power and limitations of the technology (no analythical tool that I am aware of will be right a 100% of the time when evaluating control loops), and interjecting technology in a way that makes it become help and not hindrance. The new technologies have made possible to overcome barriers that would have been insurmountable some years ago, but putting the software and hardware in place will do you no good (none whatsoever) if you do not address all the factors I mentioned before.

After all, I believe that before you undertake any performance measurement program you have to know your process, and what the heck you're trying to accomplish with your controls. You have to understand variability and what you can do about it (and what you will not be able to do with controls, unless you change the process). You have to understand that wear and tear is a fact of life and must be dealt with before it makes your process go berserk, and you have to understand that if you do not measure performance, there is no way you will be able to know where you are at. If you don't internalize that, no amount of technology will ever help you.

Now I will step off my soap-box. Hope I helped. Regards,

Leoncio
 
S

sairam potaraju

Jim,
In my opinion understanding the source of variability (interactions, disturbances, etc.) and quantifying it is critical before starting on the tuning projects.

Currently, we are testing matrikons processdoc software in our plant site, that seems to be pretty effective. The idea is to track the status of the control loops (not just the tuning) and link this to the process bottomline (like yield, energy, capacity etc.)

The challenge we see is in getting these "advanced" tools into the workflow so that the benefits are sustained and not one time gains.

Sai, Bayer Corp.
 
S

Sirish Shah, University of Alberta

There is a growing awareness in the process industry that it is impossible to monitor the performance of all loops, critical or otherwise, at all times. So what does one do? In this context, the question that Jim Ryan is asking is a very real and valid one.

The problem of controller performance monitoring pertains to that of detection of poorly performing loops and thereafter to diagnosis and 'fixing' the problem. The former is relatively simple on an off-line basis but requires great care when implemented in an on-line manner as you do not want false alarms. There is now software out there that can do this.

The problem of diagnosis is non-trivial. A controller may not be performing well because of a host of problems such as: 1) poor tuning; 2) sticky valves; 3) large disturbances; 4) interaction with other loops; 5) change of process operating conditions and process nonlinearities; 6)How to monitor multivariate performance; and so on and on. A lot of interesting work in academia has answers, some good and complete and others in-complete, to these questions. For example, how do disturbances propagate from one unit to other and upset the whole plant. Can the root cause of such disturbances be identified easily? Sometimes, yes by the use of spectral tools. There is a lot of information out there....It is impossible to answer or get to the bottom of all of this in this tiny text window.

Let the discussion and debate continue......

Sirish Shah
University of Alberta
 
S

Sirish Shah, University of Alberta

There is a growing awareness in the process industry that it is impossible to monitor the performance of all loops, critical or otherwise, at all times. So what does one do? In this context, the question that Jim Ryan is asking is a very real and valid one.

The problem of controller performance monitoring pertains to that of detection of poorly performing loops and thereafter to diagnosis and 'fixing' the problem. The former is relatively simple on an off-line basis but requires great care when implemented in an on-line manner as you do not want false alarms. There is now software out there that can do this.

The problem of diagnosis is non-trivial. A controller may not be performing well because of a host of problems such as: 1) poor tuning; 2) sticky valves; 3) large disturbances; 4) interaction with other loops; 5) change of process operating conditions and process nonlinearities; 6)How to monitor multivariate performance; and so on and on. A lot of interesting work in academia has answers, some good and complete and others in-complete, to these questions. For example, how do disturbances propagate from one unit to other and upset the whole plant. Can the root cause of such disturbances be identified easily? Sometimes, yes by the use of spectral tools. There is a lot of information out there....It is impossible to answer or get to the bottom of all of this in this tiny text window.

Let the discussion and debate continue......

Sirish Shah
University of Alberta
 
H

Heavner, Lou \[FRS/AUS\]

With all due respect, I would argue that it is becoming possible to monitor more loops and devices than ever. The solution is to put enough
intelligence in the device to allow it to monitor itself. It would broadcast an alarm when it detects a problem. Otherwise it continues to
operate normally. This is what really should be motivating the adoption of Fieldbus and device networking. It is desirable to enable intelligent
devices to communicate their alarms as they occur and then allow "maintenance" whoever that may be to dig into specific identified problems.
Intelligent diagnostics at the device level or at the other end of a Fieldbus can monitor for hard failures and deteriorating performance which
might be an indicator of impending hard failure or simply an indication of wear.

If you are looking for tools to deploy in a traditional system, you will have to accept some compromises, simply because the technologies of the 80's and 90's were not designed with this paradigm. You will have more success when you deploy a system that is designed with this concept in mind. That means replacing dumb single-value instruments and valves with "intelligent" ones with respect to self diagnoses and connecting them with a digital network capable of tapping that intelligence.

Regards,

Lou Heavner
Consultant
Advanced Applied Technologies Team
Emerson Process Management
Phone: (512) 834-7262
Fax: (512) 832-3199
e-Mail: [email protected]
 
Leoncio,

Thanks for your reply. I read your article in InTech a month ago or so with interest. Thanks for sharing some of numbers related to your reduction in maintenance costs.

We too have tried the Loop Scout, but our company has a strict policy with respect to releasing data outside the company.

thanks again,

JR


> Jim,
>
> I have just seen this thread recently. In my opinion, there have been a lot of truisms expressed by several people but they have not specifically addressed your questions. I will try to do that briefly.
>
> Yes, we have experience with commercial and in-house packages for loop performance assessment. In fact, our initial in-house package has helped refine some of the existing commercial packages. We have tested and used packages by several vendors and research organizations. We have settled on using a comprehensive Honeywell solution that was put together following our specs, and that includes LoopScout and AlarmScout. This solution is presently being deployed in the majority of our mills.
>
> ROI improvement? Hard to measure a 100% accurately because accountants do not look at this level of details, but we have solid anecdotal evidence that says that we're impacting the bottom line. We have reduced maintenance expenses significantly in some mills by saving up to 75% in unnecessary control valve repairs, and PM's. We have focused the efforts (effectively reducing the time spent) in control maintenance during planned shut-downs, identified and prevented process incidents that could have caused down-time and would not have been identified without our assessment tools, and we have reduced occurrence of process alarms by more than 50%.
>
> So far, the experience of the majority of users is "how was I able to live without this, for so long?"
>
> Now, having said all those great things, let me close with a sobering reality. In my opinion, the secret of success have been 75% or more due to "elbow grease", discipline, encouraging and engaging the right people at the right level, dealing with the organizational inefficiencies, understanding the power and limitations of the technology (no analythical tool that I am aware of will be right a 100% of the time when evaluating control loops), and interjecting technology in a way that makes it become help and not hindrance. The new technologies have made possible to overcome barriers that would have been insurmountable some years ago, but putting the software and hardware in place will do you no good (none whatsoever) if you do not address all the factors I mentioned before.
>
> After all, I believe that before you undertake any performance measurement program you have to know your process, and what the heck you're trying to accomplish with your controls. You have to understand variability and what you can do about it (and what you will not be able to do with controls, unless you change the process). You have to understand that wear and tear is a fact of life and must be dealt with before it makes your process go berserk, and you have to understand that if you do not measure performance, there is no way you will be able to know where you are at. If you don't internalize that, no amount of technology will ever help you.
>
> Now I will step off my soap-box. Hope I helped. Regards,
>
> Leoncio
 
Sai,

Thanks for your comments. I agree that changing the companies maintenance culture and work processes to include tools like this will be a challenge.

Good Luck with your program.

JR
 
C

Craig Rochester

I'm not a controls engineer(MechE), but have had several years experience in BioPharm processing. I've had extensive involvement in trouble-shooting problems.

Regarding people fooling with control parameters, they don't do it for fun. If it works properly they leave it alone(this does not apply to engineers). If you solve the process controls problem, this goes away.

I agree that there is no substitute for understanding the process. Both engineering principles and your plant's equipment specifics.

For real sticky problems, I hook up enough data loggers to continuously monitor the system. While that's running, I go around investigating everything in that system's environment. This includes people(2nd,3rd shift too), utilities, weather...anything that might effect the system.

People say: "We don't have time to do that!" My response is: "If you don't have time to do it right, where are you going to find the time to do it over"

Finally, I've had remarkable success using "feed-forward" control. If I know the process well enough, and monitor the relevant variables, I can develop an algorithm to calculate the CO. That'll get me most of the way there, then a closed feedback loop can do the rest.
 
S

Sairam potaraju

Sirish,
You address a very practical aspect of control monitoring that pertains to diagnosis and remedies. Our experience has been that the use process knowledge via steady state models, simple dynamic models and operation experience usually helps to find the root cause among the host of problems that you describe. Thereafter using the 80/20 rule improvement steps are planned out. Again, the control improvement efforts focus on improving the bottomline w.r.t to the operational goals.

Spectral analysis techniques alone do not necessarily provide a complete picture w.r.t cause and effect relationship. The results often need to be backed up by some process based explanation.

In short, combining process and control seems to be a critical step.

Sai, Bayer Corp
 
G
Jim,
I am new to this thread, but wondered what your outcome, if any, was from the thoughts presented to your questions?

-Greg
 
Greg,

It was good to hear from the user community. We selected Matrikon in the end as they were able to connect to the variety of hardware we have across the organisation.

They have the only product that currently handles both single loop as well as multivariable controllers.

JR

> Jim,
> I am new to this thread, but wondered what your outcome, if any, was from the thoughts presented to your questions?
>
> -Greg
 
Top