It would be very useful if the readers could be better informed than the recommendation listed in the last "Control Update" under the title: "Cascade Control Recommendation Tips" in Greg McMillan's "Control Talk" blog. Number 8 is "Do not use setpoint filters on lower loops. Use lambda tuning for coordination of flow loops (e.g., maintaining stoichiometry for inline blending and reaction control). Set the lambda (closed-loop time constant) of each loop to match the lambda of the slowest loop.' I find it very saddening that even after someone of the stature of Greg Shinskey gets his paper, "A Case against Lambda Tuning" (http://bit.ly/1m8qygm) published in your May 2012 magazine, and that over the years, several people from industry and academy, including one of the inventors of the IMC tuning, have acknowledged the inherent deficiency of the method to cope with the most frequent dynamic behavior found in process control, your magazine keeps promoting it through this type of recommendation.
Summa Control Solutions
Greg McMillan replies:
Let me first say I think Greg Shinskey has provided an understanding of the effect of process and equipment design on dynamics and PID control in his books that goes far beyond what anyone else has done.
I shared Shinskey's concern about lambda tuning for over a decade. The major concern centered on a reset time that was much larger than what is needed to return to setpoint after a load disturbance to self-regulating processes with a large time constant to dead time ratio. Many of these loops are very important in that they involve concentration, pH and temperature control of large, continuous, liquid unit operations such as distillation, neutralization and reaction. A second concern was the lack of derivative action for these processes.
I realized about two years ago that if I switch from lambda tuning rules for self-regulating processes to lambda tuning rules for integrating processes with derivative action, when the time constant to dead time ratio became greater than three and I used a lambda equal to half the dead time, the problem that Shinskey raised about the recovery from load disturbances largely goes away. In fact, the tuning settings converge to those of the Ziegler-Nichols reaction curve method and many other tuning methods based on minimizing the peak and integrated error for disturbances at the process input (load disturbances).
In the lambda integrating process tuning rules, the lambda becomes the arrest time for the load response, which is more appropriate for these processes. In the lambda tuning rules for self-regulating processes, lambda was the closed-loop time constant for a setpoint response. The rate time is set equal to the secondary time constant in a second order plus dead time approximation. If the secondary time constant is not identified, I use one-half of the dead time as an inferred secondary time constant.
To provide more robustness and less tendency for the loop to oscillate, I have since used the rule of thumb to switch from lambda tuning rules for self-regulating processes to lambda tuning rules for integrating processes when the time constant to dead time ratio is greater than four, and use a lambda equal to the dead time. To deal with unknown nonlinearities that cannot be corrected by adaptive tuning and for interaction that cannot be corrected by decoupling, and to reduce resonance from disturbance oscillations with a period close to the ultimate period of the loop, I have increased lambda to be three dead times.
For more of this discussion, go to the comments section on article "Cascade Control Recommendation Tips."