CG1211-CtTalk
CG1211-CtTalk
CG1211-CtTalk
CG1211-CtTalk
CG1211-CtTalk

Ruel Rules for Use of PID, MPC and FLC

Nov. 12, 2012
When Do You Need to Move Beyond PID Control?
About the Authors
Greg McMillan and Stan Weiner bring their wits and more than 80 years of process control experience to bear on your questions, comments and problems. Write to them at [email protected]. Follow McMillan's Control Talk Blog.

Check out ControlGlobal.com on Google+

Greg: In the Control Talk columns "Show Me the Money–Part 1 and 2" (Nov and Dec 2009), Michel Ruel demonstrated to us how to improve the performance of the basic process control system by fixing valves, tuning loops and improving sequences. In some cases, the capability of the PID was used to perform advanced regulatory control (ARC). A key part of the success was the calculation of benefits and working with the operators to improve PID actions and sequences for start-up and abnormal situations. 

Stan: When do you need to move beyond PID control?

Michel: Given that process models can be identified, model predictive control (MPC) is advisable if the interactions between controlled variables can't be sufficiently reduced by detuning or decoupling. Detuning where one PID (hopefully the least important PID) is made about five times slower than the other PID can handle weak interactions. A PID decoupler eliminates detuning and the consequential deterioration of loop performance. The decoupler can also deal with stronger interactions. A simple decoupler uses the output of one PID as the feed-forward for the other PID and vice versa. For interactions between more than two controlled variables or for more than one constraint, MPC is advisable.

Greg: The advent of adaptive tuners has recently automated the identification of process models and scheduling of tuning. MPC process model identification was an automated and essential feature from the beginning of MPC technology.

Stan: Oil, gas, petrochemicals and commodity chemicals are produced by large continuous processes with well-known process models where a 0.1% increase in process efficiency or capacity provides enormous benefits. MPC with the integrated LP for process optimization is the obvious solution. What industries don't have process models and why?

Michel: I have found that many processes in the mining industry can't be modeled. The process interrelationships and dynamics in the processing of ores are not defined due to the predominance of missing measurements and unknown effects. PID loops are often in manual, not only for the usual reasons of valve and measurement problems, but also because process dynamics between a controlled and manipulated variable radically change, including even the sign of the process action (reverse or direct) based on complex multivariable effects that can't be quantified.

Stan: What process did you recently tackle where process models could not be identified?

Michel: Last year, we worked in a nickel plant on a semi-autogenous grinding (SAG) mill. The SAG mill process uses steel balls and large rocks for grinding. The load consisted of ore, steel balls and water. In addition to stabilizing the process and increasing production rates, the control system must protect the lining by ensuring the rocks fall on other rocks and not the lining. The controlled variables were density (inferential measurement), power, weight (bearing pressure) and recirculation flow. The manipulated variables were speed, feed and water flow. The disturbance variables were ore size, ore hardness and crusher opening. A camera provided the mean size, the percent below 4 inches and below 1 inch. The detailed population of ore sizes was not known.

Greg: What solution did you use to eliminate manual control when process models are not possible?

Michel: Since the operators could control the process manually, rule-based models would work. Since process performance varied significantly from shift to shift, we worked closely with operations to find the best operator logic, and put it in the form of simple linguistic rules with relative grading via a fuzzy logic control program.
Fuzzification of the controller inputs consisted of rating the measurements as to members of Lo-Lo, Lo, OK, Hi and Hi-Hi sets. The membership can be crisp where there the shapes are rectangles with no overlap.

More traditionally the membership consists of symmetrical triangles whose sides intersect at the middle of the side, providing equal overlapping. For our application, we used some special shapes and weights. Decisions were made with "If-Then" rules between each graded process input and a process output. An example of a rule is: "If the power is high and the weight is low, then the speed is medium and the feed is low." The rules of the best performing operators were nominated and closely reviewed by the process engineer (metallurgist). Defuzzification consisted of establishing a relative grading of the change in the controller outputs for each rule by a membership set. The resulting increment or decrement in each output is a velocity algorithm that inherently eliminates windup and the bumpless transition from manual to automatic. Weights for each rule and shapes for memberships were determined from a design of experiments (DOE). The resulting fuzzy logic controller (FLC) was commissioned in the advisory mode.

Stan: What were your controller inputs?

Michel: We used the controlled variables and rates of change of controlled variables as FLC inputs. A second order Butterworth filter was used to effectively reduce noise in the rate of change calculations.

Greg: An FLC I designed was used on a large waste pH neutralization system to minimize reagent use. The FLC worked quite well for decades, but when the control engineer who implemented the FLC in the DCS left the plant, the control engineer who inherited the system did not know how to maintain or improve the FLC. The process engineers never could figure out what the FLC was doing, and just knew the reagent cost was less when the FLC was in automatic. What did you do to help FLC analysis, troubleshooting and performance assessment?

Michel: We could see what rules were firing when and how often. It turned out that 20% of the 500 rules were doing most of the work. We also had online metrics of process performance. We paid particular attention to interfaces. Metallurgists had to be able to modify targets, constraints, production goals and limits. The control engineers must be able to easily adjust rules and tweaked weights.

Stan: How long did it take to commission and what were the benefits?

Michel: The FLC was on advisory control for three days. Operators could see the FLC was anticipating their actions. During the next four days, the FLC was on automatic during the day, and the shapes and ranges were modified. On the eighth day the FLC was used continuously and has been operational ever since. Every week metallurgists validate rules, make slight adjustments, and work with control engineers to make slight adjustments. A production record was achieved in the first week. The average use of energy per ton has decreased by 8%, and the tonnage per day has increased by 14%.

Greg: Fuzzy logic has uses beyond control of mining processes.

Top 10 Reasons to Use Fuzzy Logic Control on Your Children

10. Model-predictive control did not work.
9. Decoupling all your kids' interactions requires too much dynamic compensation.
8. There too many states of nonlinearity for an adaptive tuner.
7. You can write your own rules.
6. You can claim to be an expert.
5. You can add and subtract rules at will.
4. Your children will be baffled.
3. It gives you a warm fuzzy feeling.
2. You can throw away your child psychology books.
1. Your children won't move back in after graduation.

About the Author

Greg McMillan | Columnist

Greg K. McMillan captures the wisdom of talented leaders in process control and adds his perspective based on more than 50 years of experience, cartoons by Ted Williams and Top 10 lists.