By Greg McMillan and Stan Weiner
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 firstname.lastname@example.org.
Stan: There are a lot of diverse processes and applications out there. We all tend to think our loop performance problems are unique.
Greg: But a deeper understanding of concepts will guide you through the thousands of details to the source of the problem and the most effective solution.
Stan: We've picked Mark Coughran, a senior industry consultant in Global Industry Solutions at Emerson Process Management in Austin, Texas, and previously a research specialist in the Research and Test Department at Fisher Controls in Marshalltown, Iowa, to be our guide.
Greg: For you as a troubleshooter, what are the most common performance problems?
Mark: I am seeing integrator loop tuning and valve response problems in all of the process industries in plants all over the world.
Stan: Why are these problems so prevalent?
Mark: First, valve specifications generally do not have entries for valve performance requirements. The tuning for integrating loops also is different and counterintuitive. Finally, plants are being built where there is no expertise.
Greg: What valve response parameters are most critical for loop performance?
Mark: The key parameters are valve dead band from backlash and shaft windup, resolution limits from stiction, and the installed characteristic that is the result of the inherent valve characteristic, fluid properties and the process pressure at the valve inlet and outlet. Dead band and resolution are lost motion, and the slope of the installed characteristic is the valve gain. The control valve is the principal source of nonlinearity and limit cycling in control loops. Split-ranged valves and integrating processes are particularly susceptible to valve response problems.
Stan: Dead band, stiction and characteristic nonlinearity are generally greatest near the closed position. The transition between cooling and heating for temperature control and acid and base reagent for pH control requires operation near the closed position of all valves at the split-range point. I have seen a lot of loops cycling across the split-range point, wasting energy and reagent. So what I am hearing is we have a double hit. What processes have an integrating response?
Mark: Batch temperature, bioreactor process variables (dissolved oxygen, pH and temperature), some gas pressures, level, dissolved solids and impurity concentrations in processes with large recycle streams have integrating responses. Composition, temperature and pH loops in continuous processes with a large dominant process time constant have a near-integrating response. To shorten tuning tests dramatically, these extremely slow self-regulating processes can be tuned as integrating processes.
Greg: These are important loops in terms of process yield and efficiency, since analyzers provide direct measurements, and temperatures provide inferential measurements of process composition. How about an example of a typical tuning mistake?
Mark: A drum level loop was cycling with a controller gain of 1 and a reset time of 20 seconds. The settings that proved best ended up being a controller gain of 12 and a reset time of 2000.
Greg: This is consistent with my experience that loops with integrating processes are running with a significantly smaller than desirable controller gain and a reset time at least an order of magnitude too low. I have had success increasing the reset time by a factor of 10 as a quick fix until an auto tuner can give results. This is a move in the more stable direction. If the reset time is not greater than 300 seconds, and the controller gain is not greater than three, then the tuning settings are probably too low. More than two significant figures is wishful thinking.
Mark: If the controller gain is too low, you become more sensitive to reset action. The product of the controller gain and integral time must be greater than four divided by the integrating process gain (See Equation 10 in Adaptive Level Control or below).
Thus, you can reduce slow rolling oscillations and overshoot by increasing the controller gain. This is counterintuitive because we are taught that a lower gain and slower tuning provides a smoother over-damped response.
Some integrating processes, such as header pressure and jacket temperature control, have a lead time where the initial rate of change is much faster than the eventual rate of change of the process variable. The tuning rules for integrating processes are not well known, and the tuning rules to deal with a lead time are particularly rare. Most of the tuning rules in the literature are for self-regulating processes with a first order (single lag) plus dead time response approximation. The user doing trial-and-error tuning finds it especially difficult for integrating processes.
Stan: In an inverse response where the process variable first moves in the opposite direction of the final response, you have to back off on the controller gain. Drum level, column sump level and furnace pressure can have an inverse response. In modern plants, the degree of inverse response has decreased. Pre-heating feed water and air by flue gases for energy recovery has reduced the inverse response for boilers and furnaces. For column sump level, the inverse response is only seen when steam is manipulated to control level, not a preferred control scheme unless the bottoms flow is too small for level control.
Mark: I have seen a lot of loops where the filter time was too small or too large. If the filter time is too small, noise causes the valve to move, which self-inflicts a disturbance and wears out the packing. If the filter time is too large, the loop may oscillate, since integrating process performance and tuning is particularly sensitive to the additional lag.
Stan: What do you recommend projects do for these important loops?
Mark: I recommend that users request the supplier to provide the dead band, resolution, response time and installed characteristic for control valves. These lost-motion and response parameters should be the results of small steps in valve signal (e.g. 0.5%) per the current test methods established by the ISA-75.25.01-2000 (R2006). The user should plot the integrating process gain, which is the product of the control valve gain, process variable gain and measurement span gain, versus controller output. The control valve gain is the slope of the flow versus stroke plot of an installed characteristic curve. The process variable gain is the slope of a plot of the ramp rate of the process variable versus valve flow provided by the process engineer for various setpoints, and the measurement span gain is 100% divided by the process variable span used by the controller. The resulting integrating process gain must have dimensions of % PV per sec per % flow (units of 1/sec). The user should use tuning rules for integrating processes to estimate tuning settings, and decide if gain scheduling is needed for changes in the gain with operating point. Once the loop is commissioned, auto tuners and adaptive controllers should be used to identify the actual integrating process gain and dead time for the expected range of setpoints and load disturbances.
Greg: The installed characteristic and lost motion near the seat determine the real rangeability of the control valve as estimated by the equations on slide 21 of Deminar #2 posted April 22, 2010 on http://modelingandcontrol.com/. Deminar #2 and its posted review also provide a more in-depth discussion of dead band (backlash), resolution (stiction) and the resulting limit cycles.
Top Ten Things You Don't Want to Hear in a Project Definition Meeting
10. I don't want any smart instrumentation talking back to me.
9. Let's study each loop to see if the valve really needs a positioner.
8. Let's slap an actuator on our piping valves and use them for control valves.
7. We just need to make sure the control valve spec requires the tightest shutoff.
6. What is the big deal about process control? We just have to set the flow per the PFD.
5 Cascade control seems awfully complex.
4. The operators can tune the loops.
3. Let's do the project for half the money in half the time.
2. Let's go with packaged equipment and let the equipment supplier select and design the automation system.
1. Let's go out for bids and have the purchasing agent pick the best deal.