How to be a World Traveler in Process Control

There are many different worlds of process control. Here you can learn to understand and appreciate the differences enabling you to extend your expertise in process control. You can benefit from the experiences in other worlds, expanding your capability and marketability helping to make the profession stronger by building on the talents and knowledge of exceptional experts, such as James Beall, Mark Coughran, Lou Heavner, Sigifredo Nino, Michel Ruel, Greg Shinskey, Jacques Smuts and Terry Tolliver.

The different worlds are the result of the extreme diversity of products in the process industry with different dynamics and objectives.  I came out of the world of continuous chemical processes, the principal and principle world of Greg Shinskey. His books Process Control Systems, Distillation Control, pH and pION Control in Process and Waste Streams, Feedback Controllers for the Process Industries, Controlling Multivariable Processes, and Energy Conservation through Control each have orders of magnitude more practical knowledge on process dynamics and relationships and how they affect control than any book I have written or seen.  My associate Terry Tolliver, the world’s foremost expert in distillation control feels the same way. Our deep knowledge of first principles all originates from Shinskey. I was so honored to be with Greg Shinskey and Bela Liptak as the first inductees into the Process Automation Hall of Fame.

As my career progressed I expanded my world of application into batch control, fermenter and bioreactor control, plug flow reactor control, extruder control and plastic sheet control. The worlds of gas and oil, pulp and paper, and food and beverage processes were and still are foreign to me.

Bill Bialkowski principally specializing in pulp and paper applications developed lambda tuning. In the same time frame academics developed and popularized Internal Model Control (IMC). The tuning parameter for both was the closed loop time constant (time to reach 63% of a setpoint change) for self-regulating processes (called lambda for lambda tuning and gamma for IMC tuning). Both methodologies concentrated on setpoint response and considered disturbances to be on the process output, which then had similar tuning requirements for load response as setpoint response. In one case there was a step change in the process variable (PV) and in the other case a step change in the setpoint (PV). For a structure of PID on Error, the tuning was essentially the same.

Shinskey’s complaint was that the emphasis should be on load disturbances at the process input that have to pass through the process time constants and consequently on being able to deal with processes with large time constants. If you used lambda or IMC tuning for such processes, the reset time could be an order of magnitude or more larger than what would provide the minimum integrated absolute error (IAE), the common criterion in the literature for load disturbances.  Furthermore, tuning tests should be made by emulating a load disturbance by momentarily putting the PID in manual and making a step change in PID output.  Shinskey pointed out the setpoint response would also be good for good load response tuning by simply adding a setpoint lead-lag. Alternatively, for today’s modern controller with many choices of PID structure, you can get the same results as a lag time on the setpoint equal to the reset time by the use of a “PD on PV, I on error” structure as noted in the ISA Mentor Program 1/26/2106 post “Equivalent Methods to Eliminate proportional Step and Derivative Kick”. Additionally, the use of a “Two Degrees of Freedom” PID structure can give similar performance to a setpoint lead-lag. 

I found that if you use lambda rather than lambda factor setting it equal to about ¾ the dead time and switching to lambda integrating process tuning rules when the process time constant becomes greater than 4 times the dead time, you end up with good load response tuning. Whether this is robust enough considering nonlinearities and unknowns and whether you want to move the PID output so aggressively is another story. Additionally if you simply add the requirement that rate time is not only the secondary time constant but also ½ the dead time, you end up with tuning similar to what you would get for minimization of IAE.  Also, the focus is switched from a setpoint response metric  (lambda being a closed loop time constant) to a load response metric (lambda being an arrest time - time for the PV to reach the maximum of its excursion to start its return to setpoint) in lambda integrating process tuning rules. If you are concerned about abrupt changes in the PID output from affecting other loops, you can put a setpoint rate limit on the analog output or secondary loop setpoint and turn on external reset feedback to provide directional move suppression. No detuning is needed to provide a smoother output change.

Furthermore, for small process time constants, which is the case in pulp inline and paper sheet line loops, whether the disturbance is on the input or output of the process doesn’t make much difference. Finally, for processes where the time constant is less than the dead time, putting a limit on lambda and thus reset time being about ½ the dead time, prevents the transition to an integral-only type of tuning.

For more on how the size of the process time constant has caused such different perspectives, see my 10/24/2013 Control Talk Blog “The Primary Source of Disagreement in Process Control Tips”.     

What I am offering is that you can be a fan of both Shinskey fan and Bialkowski taking advantage of the knowledge they respectively offer on the value of knowing chemical engineering principles and different tuning objectives if you just know how to modify tuning rules and use key PID options. For more on this perspective see my Control 10/16/2014 white paper “So Many Tuning Rules, so Little Time”.

This is probably much more background you have time for, so let’s cut to the chase.

Frist we need to explain that the term Final Resting Value (FRV) is what the PID output ends up after a load disturbance or setpoint change. The generalization that FRV overshoot should always be eliminated by some experts is a prime example of not understanding other worlds. Overshoot of FRV is necessary for a setpoint and load response for integrating and runaway processes. In fact the elimination of FRV overshoot leads to level, pressure and temperature trips in vessels and columns and thus a safety issue.

Here are some of the many loop objectives:

  • Minimum PV peak error in load response to prevent:

–        Compressor surge, SIS activation, relief, undesirable reactions, poor cell health

  • Minimum PV integrated error in load or setpoint response to minimize:

–        total amount of off-spec product to enable closer operation to optimum setpoint

  • Minimum PV overshoot of SP in setpoint response to prevent:

–        Compressor surge, SIS activation, relief, undesirable reactions, poor cell health

  • Minimum Out overshoot of FRV in setpoint response to prevent:

–        Interaction with heat integration and recycle loops in hydrocarbon gas unit ops

  • Minimum PV time to reach SP in setpoint response to minimize:

–        Batch cycle time, startup time, transition time to new products and operating rates

  • Minimum split range point crossings to prevent:

–        Wasted energy-reactants-reagents, poor cell health (high osmotic pressure)

  • Maximum absorption of variability in level control to prevent:

–        Passing of changes in input flows to output flows upsetting downstream unit ops

Here are some of the different worlds:

  • Hydrocarbon processes and other gas unit operations with plug flow, heat integration & recycle streams (e.g. crackers, furnaces, reformers)

–        Fast self-regulating responses, interactions and complex secondary responses with sensitivity to SP and FRV overshoot, split range crossings and utility interactions.

  • Chemical batch and continuous processes with vessels and columns

–        Important loops tend to have slow near or true integrating and runaway responses with minimizing peak and integrated errors and rise time as key objectives.

  • Utility systems (e.g., boilers, steam headers, chillers, compressors)

–        Important loops tend to have fast near or true integrating responses with minimizing peak and integrated errors and interactions as key objectives.

  • Pulp, paper, food and polymer inline, extrusion and sheet processes

–        Fast self-regulating responses and interactions with propagation of variability into product (little to no attenuation of oscillations by back mixed volumes) with extreme sensitive to variability and resonance. Loops (particularly sheets) can be dead time dominant due to transportation delays unless there are heat transfer lags.

  • Biological vessels (e.g., fermenters and bioreactors)

–        Most important loops tend have slow near or true integrating responses with extreme sensitivity to SP and FRV overshoot, split range crossings and utility interactions. Load disturbances originating from cells are incredibly slow making load response a nonissue. 

Maybe I will write an article and addendum like I did for “Valve Response - Truth or Consequences” on the different worlds. At least for now this blog is a starting point.

I hope I have given you a way of seeing what world you are in and expanding your horizon and understanding of how it can all come together. A more unified profession is a stronger profession.

What world do you live in and where would you like to go? 

For a concise presentation of the concepts and details on PID control see my ISA book Good Tuning: A Pocket Guide, 4th Edition. If you are up for a more comprehensive view, see my Momentum Press book Tuning and Control Loop Performance - 4th Edition

Show Comments
Hide Comments

Join the discussion

We welcome your thoughtful comments.
All comments will display your user name.

Want to participate in the discussion?

Register for free

Log in for complete access.


  • Thanks Greg. It is interesting to get a more diverse view-point as most people stay in particular industries. You mentioned Entech, Bill and pulp and paper. I've been in business for 24 years and before that worked for EnTech. I was there around the time we were developing the EnTech Valve Specification. EnTech was a promoter of Lambda tuning, not really a developer. I've tuned perhaps 1000 loops all using the Lambda method. I have been surprised in recent years how a method that is a defacto standard for our industry for 25 years is not also the standard in other industries. I have seen published articles written by knowledgeable people who seem to be new to Lambda tuning and have some misunderstandings that may confuse some readers (e.g. that Lambda tuning is not well suited to integrating processes like levels or that performance was poor when in fact, the incorrect Lambda tuning rule was applied). I agree that slow performance is an issue with the Lambda method for processes with large time constants. There are few of these in the P&P industry as most are fast self-regulators, integrators or near-integrators. It is understood in our industry that such loops are tuned as near-integrators. Almost all steam pressure controls are tuned this way and Lambda tuning courses are usually split into first-order, integrating and near-integrating methods. Occasionally the near-integrating method is not practical. With very large time constants, I will sometimes replace the PID with a Modified Smith Predictor which would typically be superior to PID tuned by any method. But the Lambda method for me has been extremely helpful. I have just returned from tuning a type of process I've never even seen before and achieved reductions in variation of 90 to 95 %. Typically, the tuning parameters by the Lambda method, compared to what was present at the time (probably by trial-and-error) results in much less variation, faster control, better load attenuation, better setpoint response, more stability (handling changes in process gain) and less unnecessary valve movement (noise propagation). Lambda stresses stability and wide ranges of operation and of process gain are common in P&P. The advantages of a method where the time constant in auto is known and chosen based on calculation before you enter them into the controller, whose intial tuning is usually the final tuning, that allows comparison to cut-off periods, speeds of interacting loops, speeds of master and slave loops, Bode plots and frequency spectra are endless. The other odd thing is that other methods used are rarely discussed in detail and I have never seen the equations (for gain and integral) to use such other methods. Could you enlighten us with the equations for the minimum IAE method? G. Givens, Givens Control Eng'g Inc.


RSS feed for comments on this page | RSS feed for all comments