|
Plant
efficiency and consistent product quality depend on proper loop
performance, but PID tuning is only the last step.
Before
tuning a PID controller, it is wise to perform a series of tests
on the loop to find any conditions that would compromise its
performance, and correct those conditions if possible to make
the tuning more effective.
Factors that require consideration are as follows:
1.
Process gain: Is the control valve sized properly?
2. Is hysteresis or stiction excessive?
3. Is the dead time short enough?
4. Is there an excessive amount of noise in the loop?
5. How non-linear is the loop?
6. Asymmetry: Does the loop respond differently in one direction
than in the other?
7. Is the loop optimally tuned?
These
questions can be answered through a series of tests.
The Tests
|
|
|
Fig 1
|
 |
Fig
1 shows a typical set of tests to establish if your hardware is
correctly specified and performing satisfactorily.
Data
is initially collected with the loop in auto as found. From this
you can establish the variability in the loop, and whether the
loop is cycling and at what frequency. It is also a good idea to
make a setpoint change in auto to see what speed of response you
get. This response can then be compared to the response after
the optimisation work has been completed.
Place
the loop in manual and compare the variability to when it was in
auto. Has the cycling in the loop stopped? If the answer is yes,
then the problem lies within the closed loop. If the answer is
no, then the cycling is being generated from another source
further down the process. Does the PV measurement signal have
excessive noise? A small filter may be required, especially if
derivative is to be used.
In
order to establish the linearity of the loop, a number of open loop steps require to be taken throughout the normal
operating range. Any tuning must be done on the worst case
scenario i.e the step with the largest process gain and longest
dead-time (On fig 2 this would be the final step). This will
ensure the loop is stable throughout its full operating range.
If the loop has a greater than 3:1 ratio
in process gain then other measures such as characterisation, or
gain scheduling must be considered.
|
Linearity Check
|
|
Fig 2
|
 |
Reverse
the direction of the valve to establish whether it has
hysteresis. Excessive hysteresis will cause poor regulatory
control and cycling. Hysteresis for a valve with a positioner
fitted should be <1% and without a positioner <3%. Causes
of hysteresis include gland packing too tight, poor air supply,
loose linkages, positioner faulty, wrongly calibrated I/P, and
the actuator too small.
Does
the process act symmetrically i.e the same in both directions.
Temperature loops usually heat up quicker than cooling down.
That being the case, do you tune on the heating dynamics or the
cooling dynamics of the process?
|
Hysteresis Check
|
|
Fig 3
|
 |
To
determine if your valve is suffering from stiction increase the
controller output in small increments of 0.2%. If the PV
measurement does not alter, and then changes significantly in
one big step, the valve is showing signs of stiction. If this is
excessive then cycling will occur and the problem needs
rectifying (you should be looking for < 1% stiction) .
|
Stiction Check
|
|
Fig 4
|
 |
To
determine your optimum tuning parameters it is necessary to know
your system dynamics. The ratio of deadtime to the process time
constant is the critical factor in the speed of response. If the
dead time is larger than the time constant, the process will be
very difficult to control with traditional PID and other
alternatives may have to be considered.
Think
carefully what kind of response is required. On some loops
overshoot may not be acceptable so applying Ziegler &
Nichols quarter amplitude damping tuning would not be
appropriate. Most loops are set on a fixed setpoint, and are
required to respond to load changes. This being the case the
tuning should be set to respond to load changes and not setpoint
changes.
When
the loop has been tuned to your satisfaction, then place it in
auto and compare its performance to that previously.
|