When do I Use PID, MPC, and FLC for Basic Control - Tips?
For basic control, when do I use Fuzzy Logic Control (FLC) and Model Predictive Control (MPC) instead of PID control? The automation and integration of FLC and MPC into distributed control systems makes these a more viable choice. Surprisingly, there may be less decisions and expertise required than for PID control. Also, the use of more technically advanced solutions can help promote the recognition of the individual and the profession. Here are some factors.
This is the 5th in a continuing series of practical and useful questions on PID tuning raised by Brian Hrankowsky a knowledgeable process control specialist in the pharmaceutical industry. Brian is not representing his company in these posts. The questions will be addressed in much greater detail in the long overdue 4th edition of my book Tuning and Control Loop Performance being published by Momentum Press in June 2013.
I see an increased use of MPC for basic control applications that were previously domain of the PID. Why is this the case considering that the PID has been shown to provide optimal rejection of unmeasured disturbances and the PID has incredible flexibility and power? The details in the 2 part series Nov-Dec 2009 "Show me the Money", the 4 part series May - August 2011 starting with the Control Talk column "35 Years of Extraordinary Innovations" and my ISA Automation Week 2012 presentation "The Effective Use of Key PID Features" would seem to make PID the obvious solution fro basic control. The following is my take on the main reasons why MPC is gaining favor, when is it best to stay with PID and how to maximize PID effectiveness.
Why is PID Control is being replaced with MPC?
•(1) PID performance depends heavily on tuning. A poorly tuned controller can perform as badly as a loop with poor dynamics
•(2) PID tuning rules and/or factors particularly for reset time and rate time for maximum unmeasured disturbance rejection change with the dead time to time constant ratio for self-regulating processes and the relative speed of the ramp rate, dead time, and secondary time constant for integrating and runaway processes.
•(3) There may be more interest in the response to measured disturbances for feed forward control or in setpoint changes for transitions, startups, and batch operation.
•(4) The unmeasured disturbances may be so slow, that even slow tuning can catch up and suppress the consequences (e.g. peak error negligible).
•(5) Process objectives, such as the minimization of transfer of variability from the process variable to the manipulated variable to absorb variability (e.g. surge tank level), promote coordination (e.g. blending) or to minimize interactions may be more important than disturbance rejection
•(6) The MPC dependence on steady state process knowledge is more in sync with the chemical engineering curriculum than the PID dependence upon dynamics.
•(7) PID expertise is not extensively documented, organized, or automated.
•(8) The understanding of the power of external reset is largely missing in action as discussed in the Automation World Jan-Feb column "User View"
•(9) The suppression of cycles from slow and sticky valves and from the violation of the cascade rule by the use of external reset is not recognized
•(10) The enabling of setpoint rate limits to provide directional move suppression to reduce interaction and promote optimization and coordination by the use of external reset is a new concept
•(11) The connection of the BKCAL signals in the configuration for external reset is not well documented particularly for split range and override control
•(12) You don't get much as much recognition for a PID solution
When do you need to stay with PID?
•(1) PID gain needs to be greater than 10
•(2) PID rate time needs to be greater than 1 minute
•(3) PID execution time is justifiable less than the minimum MPC execution time
•(4) PID needs scheduling of tuning settings (e.g. adaptive control)
•(5) PID directly manipulates a nonlinear valve (e.g. flow control)
•(6) Open loop unstable processes (e.g. runaway reactions)
•(7) Process variables can go off-scale within seconds of loop being place in manual (e.g. incinerator and electrical furnace pressure)
•(8) A momentum balance is at play (e.g. compressor surge)
So when do you go with FLC? If a dynamic model process model is not possible due to uncertainty in process relationships and measurements, the PID cannot be tuned and the MPC has no model. This is a clear case for FLC as noted in the Nov 2012 Control Talk column "Ruel Rules for Use of PID, MPC and FLC"
The setpoint response of a FLC for lag dominant loop is excellent if the FLC has an auto tuner that computes the fuzzy factors. The providers of temperature equipment have offered FLC temperature controllers that have done well by meeting these criteria. Publications promoting FLC largely ignore PID techniques such as tuning scheduling, setpoint feed forward, and Bang-Bang Control that can be used to provide as good or better or better setpoint response. In the defense of FLC advocates, the FLC offers a practical advantage in that the FLC may be an easy to implement solution out of the box requiring less expertise than MPC for lag dominant processes. FLC robustness is not a concern because the ultimate gain for these loops is far beyond the normal range considered by users.
How can I Maximize the Effectiveness of PID Solutions?
•(1) Get anything presented or written by James Beall (e.g. "Tuning to Meet Process Objectives"), Mark Coughran (e.g. "What's Going on with Loop Performance"), Michel Ruel and Jacques Smuts (e.g. "How can we provide Mechanical and Process Engineering Regulatory Control Guidance Tips - Part 4?"), myself, and of course Greg Shinskey.
•(2) Gain access to a good auto tuner or adaptive tuner and online loop metrics. Use lambda integrator tuning rules for lag dominant self-regulating, integrating, and runaway processes. The Ziegler Nichols reaction curve method gives similar results if the rate time is set equal to the secondary time constant and lambda (arrest time) is set equal to the total loop dead time.
•(3) Learn how to use external reset and to test/verify control loop configurations
•(4) Understand the source of process dynamics as seen in "Appendix F - First Principle Process Gains, Deadtimes, and Time Constants" of the ISA book 101 Tips for a Successful Automation Career
Next week we have the logical continuation "When do I Use PID, MPC, and FLC for Advanced Regulatory Control?