James comes from a culture of internal model control favored at Eastman as the result of ties to the University of Tennessee. He became a consultant with Emerson in a group formed by the acquisition of EnTech Control, where Lambda tuning, like IMC tuning, "quantifies the tuning response" by its setpoint response and a corresponding, well-defined load regulation response. Lambda is a closed-loop time constant for self-regulating processes and an arrest time for integrating processes.
I am a product of Shinskey's approach of maximizing disturbance rejection, and the transfer of variability from the process variable to the manipulated variable, except for isolated cases, such as surge tank level control.
Stan: Before software for tuning became prevalent, plants often used a trial-and-error method. I principally have seen reset times too small because of human impatience. Besides the obvious problem of inconsistent and poor results from human limitations in understanding the future effects of tuning settings, what are some of the other downsides of the ad-hoc approach?
James: By having sound loop-tuning methodology, you can detect if something has changed in the process rather than just re-tuning the loop over and over. For example, I was asked to retune a temperature loop for a polypropylene reactor. Even though the process operating conditions had not changed, I spent the time to re-test the process response. The process analysis software showed that the process model was significantly different that when I previously tested it. However, since the process was at the same conditions as before, and I had confidence in my process analysis system, I questioned the operating personnel about what might be different. They did some investigation and found that there was a ppm level of moisture in the reactor catalyst, which severely changed the reaction rate. They replaced the catalyst and the process preformed as desired using the tuning that I had initially installed! A year later I got a call to retune the same loop. Before I spent the time to re-test the process, I asked them to check the catalyst for moisture. Sure enough, they found the same problem had reoccurred. They replaced the catalyst, I stayed on the lake fishing, and we both were happy!
Stan: There is a lesson here that tuning should not be used to cover up field problems. You can make almost any problem seem less by detuning the PID to the point there is no response by the controller as noted by Michel Ruel in the November 2012 Control Talk, "Ruel Rules for the Use of PID, FLC and MPC" What should our focus be?
James: Tuning should help the PID do the real job of the loop; that is, to increase process efficiency and production. Sometimes there is a need to coordinate loops and sometimes to absorb variability. Though the "closed-loop setpoint response" is used as the "response knob" for Lambda tuning, the load regulation response is also very important and very well-defined for Lambda tuning. In the presence of dead time (an enemy of loop performance!), all tuning can amplify variability in the process. In other words, the PID controller can actually increase variability! And the more aggressive the tuning is, the higher the variability amplification will be. We use Lambda tuning techniques to intelligently attenuate variability or transfer it to the least detrimental point in the process.
Greg: Alternately, you can retain disturbance rejection tuning, enable the external-reset option, and use a AO or PID setpoint filter or rate limit to slow down the transfer of variability as described in the Nov. 28, 2012, Control Talk blog, "Is It Better to Have a Controller with More or Less Knobs." Rate limits provide the move suppression (penalty on move) found to be so important in model-predictive control.
James: The most frequently poorly tuned loop is a level loop with severe consequences of initiating a variability that pervades a whole train of equipment. Level loops most often have an "integrating" response, and good tuning methods for these processes don't seem to be taught and used as commonly as for self-regulating processes. A 10,000-gal surge tank level loop not tuned to absorb variability can make that tank effectively a mere 100-gal tank in terms of its ability to absorb variability! These loops need a gain setting that will only react enough to keep the level within tank alarm limits for the expected load change. The computation of Lambda equal to twice the allowable error divided by the product of the integrating process gain and maximum change in controller output required will effectively maximize the absorption of flow variability into the surge tank. The integrating process gain is inversely proportional to residence time of the portion of the vessel measured by the level transmitter. The Lambda tuning rule makes sure the reset time is large enough to avoid any tendency to oscillate (i.e., critically damped) for the PID gain setting.