|
Loop Optimization:
How To Tune A Loop
The Right Approach Can Reduce Variability,
Cut Response Time, and Increase Robustness
By Michel Ruel, P.E.
Reprinted with permission from CONTROL Magazine, May 1999
Plant efficiency and consistent product quality depend on
proper loop performance, but PID tuning is only the last step.
This is the third in a three-part series on loop optimization. In
March, Part I discussed defining your objectives and
understanding the limitations of equipment. In April, Part II
described how to optimize loop characteristics.
Tuning control loops for
optimal performance is a noble endeavor, and modern loop-tuning
software tools make it look easy. But before tuning a loop,
it's critical to understand the importance of defining the
objectives, understanding the limitations of your equipment, and
dealing with loop characteristics.
Plant efficiency and consistent product quality depend on
proper loop performance, but PID tuning is only the last step.
This is the third in a three-part series on loop optimization. In
March, Part I discussed defining your objectives and
understanding the limitations of equipment. In April, Part II
described how to optimize loop characteristics.
As we discussed in Part I of this series, the EnTech study
estimated that some 30% of all loops oscillate due to
nonlinearities such as hysteresis, stiction, deadband, and
nonlinear process gain. Only 30% oscillate because of poor
controller tuning.
In Part II, we described a series of tests to find any
conditions that would compromise loop the results of loop tuning.
The tests answer the questions:
1. Process gain: Is the control valve sized properly?
2. Are hysteresis or stiction excessive?
3. Is the deadtime short enough?
4. Is there an excessive amount of noise in the loop?
5. How nonlinear is the loop?
6. Asymmetry: Does the loop respond differently in one
direction than in the other?
7. Is the loop optimally tuned?
Loop tuning should be performed only after answering these
questions and, if possible, correcting pre-existing conditions to
make the tuning more effective. Corrections might include valve
maintenance, filtering, linearization, repairing or maintaining a
sensor, or identifying and removing upstream cyclic upsets.
The last step is to identify the highest-gain,
largest-deadtime location in the loop and plan to tune for that
worst case.
Tune for the Worst
Figure 1 shows the worst case for the paper mill steam
pressure control loop example from Part II. Here, the PID
controller is part of a Measurex system. This DCS uses a parallel
algorithm. The software contains a database with more than 200
PID controllers. This database informs the software of the
algorithm, units, and special functions.
THE WORST CASE
Tune for the
highest-gain, largest-deadtime location in the loop. Shown is
the worst case for a paper mill steam pressure control loop
SEEK ROBUSTNESS
This robustness plot
shows the tradeoff between tight tuning
and stability. It can be used to quickly analyze the stability
(sensitivity or robustness) of a loop.
When the controller uses a series or an ideal algorithm, it is
easy to link the equivalent process deadtime to the integral time
and derivative time. However, with a parallel algorithm, the
numbers are sometimes quite different from what people expect.
The tuning parameters are chosen, in this case, to ensure
robustness. This loop must eliminate disturbances quickly, but
the valve has not been tested throughout its range, so a safety
factor of 2.6 is used. Also, the valve operates only over a small
part of its range and the behavior of this loop is greatly
dependent on the valve.
If the loop interacts with others, the parameters must be
chosen accordingly: in this case, the other loops are a lot
slower and the loop can be tuned fast. If another loop is at the
same speed, one of the two loops should be detuned to be sure the
speeds are different.
The selected tuning parameters are "Load Fastest,"
which gives fast recovery after a disturbance and enough
robustness. At the opposite, "Setpoint" tuning (also
named Lambda tuning) would be too slow (three times slower) and
not aggressive enough. Other applications could use settings
between these extremes.
The robustness plot (Figure 2) is an analysis tool. It shows
how sensitive (or robust) the loop is to process gain or process
deadtime changes. Robustness plots graphically show the tradeoff
between tight tuning and stability. Use the robustness plot to
quickly analyze the stability (sensitivity or robustness) of a
loop.
The plot has a region of stability and a region of
instability. The solid (red and blue) lines on the robustness
plot are the limits of stability. To the right and above the
solid lines (higher gain and delay ratios), the closed loop
process is unstable. To the left and below the solid lines, the
closed loop system is stable. The cross, where both ratios are 1,
shows the process gain and deadtime at the selected process
values.
Check It Out
To see how the new tuning parameters affect the loop, compare
variability before and after tuning. Collect process variable and
controller output data with the controller in automatic at normal
operating conditions (Figure 3).
CHECK RESULTS AT A CONSTANT SETPOINT
Process variable and
controller output data before (left) and after tuning
(right) shows significant improvements.
If possible, do a setpoint change to verify how the loop
reacts and compare old and new tuning parameters (Figure 4). In
this case, the response time is a lot smaller. Also, the cycling
is now eliminated, and the valve moves by an appropriate amount.
CHECK STEP CHANGE RESULTS
Comparison of before
(left) and after tuning (right) shows the
response time is a lot smaller, cycling has been eliminated, and
the valve moves by an appropriate amount.
You could also do a before and after comparison using power
spectral density (Figure 5). This one shows that cycling at
longer periods has been greatly reduced, but cycling from the
relief valve remains.
POWER SPECTRAL DENSITY
Cycling before (top) and
after tuning (bottom) shows great reduction, but
cycling from the relief valve remains.
Statistical analysis before and after (Figure 6) shows
variability is cut by half. The short oscillations remain but the
long oscillations are removed. This was easily visible on the DCS
trends.
STATISTICS
Statistical analysis
before (left) and after tuning (right) shows
variability cut by half.
Write the Report
It's a good idea to write a report and insert pictures to
help troubleshoot the loop in the future. The Multi-Loop Tuner
package from ExperTune, Hartland, Wis., used for this example,
has built-in tools to generate an automated report including
graphics, values computed, and analysis.
The report on this loop also noted that the loop would perform
better with a properly sized valve. A hidden cycling of five
seconds from a leaking relief valve will disappear after the
relief valve is replaced or repaired.
The new tuning parameters increased the loop performance:
• Three times more robust.
• Response time reduced 80%.
• Variability reduced by half.
After three weeks of operation, the paper mill steam pressure
control loop was performing very well and the operators no longer
complained about poor performance, cycling, and instability.
Michel Ruel, P.E., at TOP Control Inc., St. Romuald, Quebec,
has 22 years of plant experience at companies including Monsanto,
Domtar Paper, Dow Corning, and Shell Oil. Author of several
publications on instrumentation and control and frequent
university lecturer, Ruel is experienced in solving unusual
process control problems and a pioneer in implementing fuzzy
logic in process control. His e-mail address is mruel@topcontrol.com.
|
