You think you'’ve got PID troubles?

A builder of turnkey, PLC-based batching and blending systems is finding it increasingly harder to control some loops with basic PID. Find out what the experts recommend in The Answer to Your Problems.



Can You Help Resolve Our PID Troubles?
We build turnkey, PLC-based batching and blending systems for the wet chemical producing industry. As customers’ requirements become more demanding, we're finding it increasingly harder to control some loops with basic PID. We’re aware of advanced options such as fuzzy logic and adaptive control, but mostly in the context of much larger operations. Does anyone have some experiences to share?

— From September 2005 CONTROL DESIGN


Stick with PID
It’s possible to control most processes using PID controllers, and PLCs are perfectly capable of handling most PID loops. However, the control engineer must be very careful when configuring the PID algorithm. In most PLCs, you’re faced with configuring about 40 parameters to properly configure a PID loop. It’s been my experience that more than 90% of PLC-based PID Loops are not properly configured. The most common errors include:

  • Selecting the wrong algorithm options;
  • Failing to match scan time with PID loop time;
  • Improper scaling; and
  • Lack of coordination between interacting loops such as cascade loops.

If you’re confident that the PID block is properly configured, the next step is to confirm proper tuning of the PID loop. This certainly will include performing some bump tests, and tuning, using a scientific approach. A good PID tuning package such as our PID Loop Optimizer will ensure the selection of optimal tuning parameters.

Some batch processes must handle non-linear measurements (pH, for example) or non-linear process responses. Again, these can be handled easily in a PLC with the proper design. Gather data from process bumps, define the non-linearity, and then build a characterizer to correct for it. Characterizers can be built easily in most PLCs by using a set of X-Y pairs. A sample characterizer is shown in Figure 1 below.



A characterizer can help a PLC correct for non-linear measurements.

Except for large, complex, interacting systems, such as distillation columns, it’s rare to find a loop that PID can’t handle. Solving these problems with PID is far less costly than the training, design and implementation cost of more complex technology such as fuzzy logic.

George Buckbee, product development director, ExperTune, Hartland, Wis.

You Can Do Better Than PID

WE RECOMMEND that you use Model-Free Adaptive (MFA) control. MFA is an adaptive control method that doesn’t require process models. Therefore, it’s easy to install, use, and maintain. Once installed, no controller parameter tuning is required.

Since you have to deal with batch changes and product changeovers, PIDs might need frequent tuning. If you use a fuzzy or model-based approach, you might still need to build rules and models. Also, maintaining these rules and models can be a headache.

Why is MFA better? MFA isn’t just one controller. There are a number of MFA controllers readily available. You can simply choose the one that fits your application, do some simple configuration, and launch it. These MFAs include:

  • SISO MFA Controller to replace PID and eliminate manual tuning,
  • Nonlinear MFA to control extremely nonlinear processes,
  • MFA pH controller to control pH processes,
  • Anti-delay MFA to control processes with large time delays,
  • Robust MFA to force the process variable to stay within defined bounds,
  • Feedforward MFA to deal with measurable disturbances, and
  • MIMO MFA to control multivariable processes.

A chapter on MFA is included in the newly published Instrument Engineers’ Handbook—Process Control and Optimization by Béla Lipták, CRC Press LLC, October 2005. You can buy the book from the ISA book store at or via

George Cheng, CTO, CyboSoft, Rancho Cordova, Calif.

Control the Gain; Control the Process

BASIC PID has the property that the gain settings for the P, I and D portions are constant, regardless of the magnitude of error or change in error.

In microphones, for example, we often have automatic gain control. For a large signal, the gain is small, and for a small signal, the gain is high. If a speaker talks loudly, he will not overdrive the microphone, and yet when he whispers, he can still be heard.

Similarly, many PID controllers work better if the gain is different for large signals and for small signals. For small signals, a smaller P gain often is used. Also, the D portion is often zero, especially for small signals. This then helps to stabilize and quiet the system near zero error.

Fuzzy control is one way to provide variable gains over the input range of the error signal. It also allows other variables to participate in the control. For example, a cruise control in an automobile also may take into consideration whether you’re driving uphill or downhill. An article I wrote in CONTROL DESIGN ["Warm and Fuzzy," Nov ’03, p.39] demonstrates the influence of a second variable.

Another useful article that I wrote for CONTROL DESIGN ["Fuzzy or Algebraic?" Feb ’05, p.39] describes the application of variable gain, especially to a PI controller (D portion is zero). It uses the property that the gain functions for P and I are monotonic. In the example of the microphone, if a speaker whispers the gain is high, and if he speaks loud the gain is low. How about if he yells? Certainly the gain would be even smaller (and not larger). That is monotonic. Using this property, relatively simple algebraic gain control might improve the overall loop control.

Ernst Dummermuth, independent consultant, Chesterland, Ohio

Advanced Tools Can Work Small

MOST PLANT loops don’t require adaptive control. For fast changing processes, adaptive control might not work well because it might be too slow to respond to changes. Gain scheduling is a better solution. Adaptive tuning software, however, can be very useful in a plant environment when dealing with a significant number of loops. Adaptive tuning software can monitor many loops at the same time.

ased on observed process responses, it will give recommended PID parameters automatically. The adaptive tuning software will converge to a new set of PID parameters as needed. This effort could be the source of data to determine the gain scheduling requirements.  Plant management can use their resources on other endeavors, while the software does the tuning.

As a side note, model predictive control (MPC) or advanced process control (APC) solutions are not limited to applications with many loops. As these technologies have been developed for the PC, you can optimize very small applications using MPC. A coordinated control application has only three loops to control. A long dead-time issue can be easily resolved with an internal model controller, and this could be just one loop. With PC-based supervisory process control there is no minimum size for MPC.

Matt Petras, technical support engineer, and Paul Botzman, marketing/sales, ControlSoft Inc., Cleveland, Ohio

Above All, Know Your Process

I'VE PROVIDED control engineering services to the pulp & paper industry for 13 years now. You're right that there are problems that are hard to control with PID. We sometimes use a Modified Smith Control for critical processes with lots of dead time and a large time constant. We use our own controller, which allows for some model uncertainty, but it's important that you model the process gain, dead time and time constant, etc., properly and adapt it to process conditions.

Having said that, we find that most of the problems with PID control are not that PID is unsuitable. We find that tuning probably is the worst problem: too high a gain, too little integral action, very slow control, yet acting too much on noise, sometimes over-damping of transmitters.

Use a mathematical method of tuning. In the paper industry, we use the Lambda tuning technique. Other suitable methods exist. Avoid methods such as Ziegler-Nichols, which are intolerant of changes in process gain. Another problem is poor valves. We very often find new valves have excessive backlash and stiction. Do bump tests to find the stiction and backlash, multiply by the process gain (in % of span) and the answer should be less than 2% for non-critical loops and less than 1% for critical loops or those close to the final product. Unfortunately, valve brand (manufacturer) matters a lot. Make sure your level controls do not cycle as this tends to drive the entire process to cycle.

Going further than just getting a PID loop to work well is the next step. Rather than choosing a more complex controller than PID, consider adding the intelligence that you already have. For example, consider the concentration or consistency measured in a line after a pump. You are controlling the concentration by adding something or diluting it and you know that the process gain will vary inversely with the nominal flow of the diluted liquid. For consistency control in a paper mill where water is added at the pipe inlet to a stock pump, you know the process gain is inversely proportional to the stock flow. Why not add that logic to adapt the controller gain? Sometimes we also adapt to the square root of the dilution pressure. We often see our adaptation logic (gain scheduling) use a max-to-min gain ratio of four or more, so it does make a big difference.

Likewise, you might know that the dead time and time constant of some processes are inversely proportional to the nominal flow. Adapt the tuning accordingly if they vary a lot. Use feed-forward where appropriate. Don't allow a master loop to windup when its output goes outside the setpoint range of the slave loop. Consider midranging control for valves with a large process gain. Avoid output ramp rate limiters except where you are going in or out of "tracking" for an interlock situation. And for ratio controls or any calculations that result in a setpoint, don't just blindly use the PV and propagate that noise to the loop receiving the setpoint.

Concentrate on the nuts and bolts of practical process control. Understand the control logic and the real control objectives before you even start bump tests.

R. Glenn Givens, PE, president, Innovention Industries, Burlington, Ontario