Greg McMillan and Stan Weiner bring their wits and more than 66 years of process control experience to bear on your questions, comments, and problems.
Write to them at [email protected].
Greg: In case our readers have been lying awake at night wondering what the answers to this years Puzzlers are, we here present the solutions that are every control engineers dream or nightmare.
Stan: How much information is in the process variable versus the output of a PID controller on disturbances, sensor drift and offset, and valve dead band and resolution?
Greg: We answered this puzzler in the July column titled Disturbing Remarks. Since a tuned PID controller transfers variability from the process variable to its output, and each mode of the PID adds information content on magnitude, duration and rate of change of the process variable, the story is generally in the controller output on the process, measurement and valve.
Stan: What are the implications of an integrating or runaway process response on control valve dead band, resolution and stroking requirements?
Greg: The implications are all bad and generally worse the higher the integrating and runaway process gain. Dead band will cause these valves to cycle, and if the valve response is too late or too slow, the process variable can ramp or accelerate off-scale.
Stan: What was the rampant valve problem fostered by a lack of position feedback?
Greg: The belief that a valve actually moved when a controller output changed was a widespread fantasy until the advent of smart positioners. Smart positioners are sometimes worse than dumb ones when the feedback is on the actuator shaft position rather than the trim position for some classic high-friction rotary valves. In these dastardly situations, where a cheap piping valve is posed as a throttling valve, the smart positioner is providing negative knowledge. (See www.chemicalprocessing.com/articles/2007/200.html.)
Stan: What are the major sources of changes in loop dead time?
Greg: The sources are everywhere, but one of the most insidious is from control valve dead band and resolution because step changes in the controller output from most manual tuning methods will not reveal this dead time. The controller should be in auto to identify it. This dead time is the dead band or resolution divided by the rate of change of the controller output, which is extremely variable. Dead times also come from secondary time constants, such as mixing, sensor and thermal lags, and the one thing you can say about time constants is that they are not constant. Dead time also comes from transportation delays, where the dead time is the volume divided by the total flow.
Stan: When does a steam rate feed-forward for a three-element boiler drum-level control cause problems?
Greg: I think we answered this and other feed-forward questions in prior columns. If the steam and feed-water flow is too noisy or not representative of the flow, which can happen at low rates, then it is best to go on straight level control directly to the valve. If the inverse response is so severe, the boiler may trip on level. In this case, the three-element feed-forward, which is based on a material balance, may make it worse, and an impulse feed-forward with the opposite sign is used instead during the inverse response to prevent a level trip.
Stan: What are the application considerations for nuclear magnetic resonance analyzers and near-infra-red analyzers?
Greg: In some ways, they are similar to neural networks in that you need a lot of lab samples and training data sets and if anything in the process changes, you may need to redevelop the models.
Stan: When does a proportional-integral-derivative (PID) controller perform better than a model-predictive controller (MPC)?
Greg: For a single loop where derivative action is essential, a PID may be best, but either one can be made to look great or lousy based on tuning. So pick the one you like best and tune it better, in the time-honored tradition of most papers touting a special algorithm. We also have an answer to this question from George Buckbee.
George: PID costs substantially less than MPC. It costs less to design, less to install and less to maintain. PID achieves very good results for single-loop control, with low cost, simplicity and relative ease of tuning. For individual control loops, where there is little process interaction, PID is the low-cost solution, and there is little benefit offered by MPC. With complex, interacting systems (such as distillation, for example), there can be substantial benefits to an MPC layer, and the added expense and complexity may be justified. However, keep in mind that MPC systems are often put in place on top of lower-level regulatory PID loops. You should be sure that the PID layer is functioning properly before investing in a layer of MPC on top. Most plants are indeed dynamic environments. Raw materials change. Products change. Production rates change. Seasons change. To be successful, a controller must be sufficiently robust to handle these changes. Both robust PID tuning and adaptive MPC models are capable of handling these scenarios. You should use control-loop monitoring tools to ensure that the controllers continue to perform well as these conditions change, regardless of the controller.
Stan: Now for the question of the hour: Why wasnt an engineer selected for the VP spot?
Greg: The answer is in these timely top ten lists from Randy Reiss, who is becoming my best resource for my delisted brain.
Top Ten Reasons Not to Pick a Control Engineer as a Candidate for Vice President of the United States
10. There is no model-predictive control for the vetting process
9. Attack dog label is often overlooked for the Ole Dogs in the control room.
8. Keeps asking the press corps, Where are the valves?
7. Campaign stops have to be scheduled around production.
6. Foreign policy?
Hasnt been seen since his visa expired last week.
5. Nobody would understand, Its the process gain, stupid
4. Fuzzy logic may be misinterpreted as scandalous.
3. Keeps checking if the campaign bus is properly terminated.
2. Only knows one speech, and its that one about the time constant being so-o-o-o-o important.
1. The dress code at the White House does not include steel-toed boots, a pocket protector and a coffee-stained shirt.
Top Ten Reasons to Pick a Control Engineer as a Candidate for Vice President of the United States
10. Was vetted by his last wife
That means she took the Vette, right?
9. Foreign policy experience Can you say outsourcing?
8. Increase production, reduce costs, properly handle waste
Its the economy stupid!
7. No personal life outside of the plant means no possibility of scandals.
6. Has inside connections at the oil and energy companies; Thats a requirement these days, right?
5. If he can figure out why Big Paul brought that monkey into the control room last Tuesday, he can decide any
split vote in the Senate.
4. One time spent six hours out in the plant with a maintenance guy who kept singing that How a Bill Becomes a
Law song from Saturday morning cartoons.
3. Knows how to handle dead time.
2. When that 3:00 a.m. phone call comes, hell already be awake, worrying about loop tuning
1. A natural diplomat; has a keen understanding how logic plays a tiny role in management decisions.
