The following discussion is part of an occasional series, "Ask the Automation Pros," authored by Greg McMillan, industry consultant, author of numerous process control books, and 2010 ISA Life Achievement Award recipient. Program administrators will collect submitted questions and solicits responses from automation professionals. Past Q&A videos are available on the ISA YouTube channel. View the playlist here. You can read all posts from this series here.
Looking for additional career guidance, or to offer support to those new to automation? Sign up for the ISA Mentor Program.
Here is a question I have based on some experience:
I ask because I had an experience of upgrading a feedwater valve on our recovery boiler and I literally took print outs of the before and after Cv curves, eyeballed the slopes in the operating range of the old trim, and expected operating range of the new trim, then changed the gain as a ratio of the two.
I expected to have to retune the valve after or during startup, but it started up at 2 a.m. and seemed perfect so we haven’t touched it in two-three years. Is this legitimate?
If there are no process changes, the installed valve characteristics are linear, the valve actuator type is the same, the positioner tuning is similar and the valve precision and response time are the same, the tuning can simplify to a ratio of the slopes. An auto tuner should be used to confirm there are no changes in the process or valve response other than valve gain. Testing per ISA-TR75.25.02 can detail if there any changes in resolution or response time or nonlinearities in valve response.
Going to a valve with greater capacity and tighter shutoff is often a disaster because of the dramatic increase in installed valve characteristic nonlinearity, backlash, shaft windup, and stiction. These are on-off valve posing as control valves. If the control valve is the greatest source of deadtime in the loop, the valve response time plays a big role in loop tuning and performance. See the Control article “Is your Control Valve an Imposter” to avoid these and many other valve problems that even more intelligent tuning cannot completely fix.
I agree with Greg. However, this agreement holds true only under the condition that the tuning parameters were appropriate. Unlike a temperature loop or a level loop, a flow loop involves a relatively straightforward process. Essentially, the process model primarily consists of the valve itself. Therefore, if there are variations in the valve dynamics, it may be necessary to adjust the tuning parameters accordingly.
Michel Ruel is right in that the tuning needs to be verified that it meets performance objectives (e.g., minimum peak error and minimum integrated absolute error for load disturbances) with sufficient robustness for changes in process and equipment conditions over the entire operating range realizing there is a tradeoff between robustness and performance. The tuning before the valve was changed may not have been the best. Scheduling of tuning settings can be used to help deal with identified changes in process operating conditions, most notably production rate. A setpoint lead-lag or Two-Degrees of Freedom PID (proportional–integral–derivative) structure can be used to achieve the desired setpoint response using the tuning identified for the best load response.
Michel and Greg make some very good points. Updating the tuning based on Cv alone may be a good first estimate of the required tuning for a simple loop, such as a flow loop. However, you may be ignoring some other important aspects of robust and stable tuning. For instance, a new valve may also bring a new actuator, positioner, booster or type of valve trim. Not that it would happen at your plant, but the valve could be installed backwards!
The real question in my mind is “Why not take the time to bump and tune the loop?” I would prefer to extend the work from 2 a.m. to 3 a.m., rather than risk being called back out of bed to go to the plant at 5 a.m. because there was a tuning problem.
As I (re)read the original inquiry/premise, the expression “better lucky than good” keeps coming to mind. To be clear, it’s not that the approach is fundamentally wrong, but there’s too little information to distinguish whether it was a good approach or if they just got lucky. What remains unknown is why the valve was being replaced: was it a “replacement in kind” (that merely had a slightly different Cv); were there any piping modifications and/or a pump upgrade/replacement – all of which potentially affects the system’s hydraulic characteristics (the main factor affecting flow controllers); what size and frequency of disturbances was the system subjected to; etc.?
This particular application sounds like it was effectively a “stand-alone system” – that is , one that was only remotely connected to or interacting with other processes. Thus, it operated at/near some nominal “design point.” For such processes, the approach seems warranted. Where it might fail to achieve the intent is if there are unexpected/unanticipated nonlinear effects, which result from changes in hydraulic behavior, as indicated above (i.e., pipeline differential pressure (DP), pump curve, valve Cv curve, etc.) that would affect the process gain as seen by the flow controller.
But back to the “why,” the quote from Simon Sinek’s book, “Start with Why” comes to mind: “…if we’re starting with the wrong questions, if we don’t understand the cause, then even right answers will always steer us wrong…eventually.”
Things change! The design and tuning of a control loop may have been brilliant for the application that motivated the original project, and may have successfully adapted to years, even decades, of changes in the way the greater process works. Getting some new maintenance activity or an equipment change working properly (hopefully optimally) usually benefits from reviewing and testing all the fundamentals involved – much as you would with the original project. It can be useful and time-saving to review any questions or unusual features – I have Greg Shinskey’s book "Process Control Systems" a short step away.
Even when there are no intended changes in the original design it can be beneficial and time-saving to repeat the tuning process. This can expose what were thought to be benign changes over the years that may have produced unexpected issues. Perhaps a measurement location was moved, perhaps with some piping work. Depending on exactly what happened this might impact timing or bias on the data available at the controller. It is easier to deal with such things as part of the current project while everyone has focus and stimulated memory of how and why some change was made. Evaluating performance and optimizing adjustments is a thorough way of assessing how this new project’s design, equipment and tuning are doing toward meeting requirements.
Thanks, Greg! I appreciate all the experience and wisdom from this group.
A little background, the feedwater valve that was replaced was getting an upgraded cage to give it higher capacity. The positioner, valve body, piping, pumping, etc. were all identical. Also, the flow loop had been tuned with the help of Andrew Waite from Emerson using the lambda tuning methods and was the inner loop of a cascade scheme. I had reasonable confidence in the initial tuning and that the only change was the Cv curve.
The maximum PID gain is inversely proportional to the open loop gain that is the product of the manipulated variable gain, process gain, and measurement gain. For a self-regulating process with the PID working in % of input and output scales, the resulting open loop gain is dimensionless. If the valve has a linear installed characteristic and the PID process variable *PV) scale is the measurement transmitter calibration span, the manipulated variable gain is valve capacity divided by 100% and the measurement gain is 100% divided by calibration span. For a flow loop, the process gain is 1 and the open loop gain simplifies to simply the ratio of the valve capacity to calibration span.
If the PID works in engineering units, which is rare and undesirable, for a 0 to 100% output scale and a PV scale that matches calibration span, our flow controller open loop gain becomes the valve capacity in the same engineering units as PV divided by 100%. For a flow loop range and valve capacity much greater than 100%, the PID gain must be incredibly small. For example, if a typical flow PID gain of 0.25 is used and there is a PV change of 100 pph on a 1000 pph scale, the PID output would change 25% causing a change in valve flow of 250 pph instead of the desired 25 pph causing a severe overshoot instead of the desired gradual approach to setpoint via integral action for a flow loop. A PID working in engineering units causes all sorts of complications, including the devasting effects of changing the engineering units (e.g., from kpph to pph). Also, any similarity of tuning settings for loops with similar dynamics goes out the window.
Some additional questions:
1. Did the flow transmitter range change, too (i.e., get recalibrated to a larger span)?
2. Is your control system tuning based on range-based (% signals) or engineering units?
3. In what direction did the tuning change?
Here is the background to my questions:
1. If the transmitter was recalibrated for a higher flow range, there may have been little/no required change to the tuning, particularly if…
2. If it is a range-based control system, the relative change in valve capacity v. transmitter range may have canceled each other out; if it is an engineering unit-based system, (although I’m not personally aware of any), it could be different.
3. Assuming there was no change in transmitter calibration, my guess is that the controller gain (Kc) went down in proportion to the increase in flow capacity. This follows from the assumption that a change in output (valve position) resulted in a larger change in flow (compared to the previous valve), which means that the process gain (Kp) increased (from the controller’s perspective). As Kp and Kc are inversely proportional, a larger Kp requires a smaller Kc.
On a related topic, do you have sufficient long-term history that covers all/most of the valve’s range (0-100%) to examine the relationship between flow vs. valve position, thus allowing assessment of possible nonlinear behavior? If so, a “before and after” comparison would be a good follow up to perform, as well as to assess the need for gain scheduling or similar types of valve characterizations for the new valve.
The flow range did not change
The controller gain was change from 0.094 to 0.08001 due to: “CV/%OP is going to change at the operating point from 1.19 to 1.36, which is an estimated 14.3% increase in the process gain.” (From my notes)
This makes sense because the valve trim was intended to give more flow earlier in the stroke to prevent running the valve over 70% open.