Note 5491: Manual Tuning using the Velocity

Application Note #5491
Manual tuning using the velocity zone method
General information
There have been many methods developed over the years for tuning the loop constants of
a servo axis, ranging from various auto-tuning methods to step-by-step manual tuning.
All of these methods offer users various benefits, but all have the same goal; to simplify
the process of tuning a PID filter while giving the user optimum performance. One
manual method of tuning which works very well when combined with the GalilTools
scope is what we call a velocity zone tuning method. This application note will guide the
user through this approach, ending in a well tuned servo axis.
In traditional PID loop tuning, a position step response is plotted and the KD and KP
term are increased up to a point just short of resonance or instability. Typically KD is
increased first, followed by the KP term. However changes to KP may affect KD and
vice versa, so many iterations are required before arriving at the final value. This
approach results in a well tuned servo axis, but relies on a good deal of experience on the
part of the user, who must be able to read the positional responses and determine which
values to raise or lower.
The velocity zone tuning approaches tuning somewhat differently. The main feature of
this tuning method is in isolating the velocity tuning parameters from the position
parameters, and tuning those first. Specifically, the KD term which provides the
derivative gain in the system as well as the FV term which applies a feedforward velocity
gain are tuned first. These parameters are tuned based on the commanded velocity
profile of the system. Once these values have been selected, the position error of the step
response is used to select the appropriate KP (proportional gain) and FA (feedforward
acceleration) terms. Finally, the position error information is again used to tune the KI
(integral gain) and IL (integral gain limit) terms.
The benefit of this method is not that it necessarily yields better tuning results than
traditional PID tuning. Rather, this method is easier for a basic user to follow along as a
step-by-step manual tuning approach. That step-by-step approach is described below.
Step-by-step approach to velocity zone tuning
1. Setup GalilTools
The key information to display when tuning a servo axis is commanded/actual
encoder velocity, encoder position error and servo torque. Figure 1 shows how
the scope is set up for initial tuning. Values for “Scale” are selected based on
move length, and may need to be adjusted during tuning. Set the trigger for
“Repeat” and click on “Start” to get the scope ready to collect data.
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
Figure 1: GalilTools setup
2. DMC Code
This tuning method requires the motor to perform a short step move back and
forth. The distance of this step move should be sufficient to break through static
friction and fully accelerate an inertial mass. In addition, there should be some
amount of steady state velocity in the profile (ie. No triangular profile). The
velocity, acceleration and deceleration are selected based on maximums required
for the application. Below is a sample program for this particular system being
tuned. This program is then downloaded to the controller. Upon execution, this
code will continually step the motor back and forth by the PA amount. The code
may be stopped at any time via the “AB” or “ST” commands.
AC 7500000
DC 7500000
SP 50000
PA 10000
WT 250
PA 0
WT 250
3. Set initial parameters
Assuming the default gains are currently being used on the axis (KD64;KP6;KI0),
set the KP term to 0 and leave KI at 0. Set KD to 10 to give a minimal derivative
gain which will be acceptable for most systems.
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
4. Perform step response
With the DMC code downloaded to the controller, the step move is initiated by
typing in the “XQ” command into GalilTools. The motor will make a small
velocity move, trying to match the commanded velocity profile. For most
systems, the KD of 10 will be insufficient to match the profile, so KD will need to
be raised significantly (detailed in step 5). After the move has completed the
scope will show the commanded velocity profile in addition to a trace for actual
velocity profile, as shown in Figure 2:
Figure 2: Step response with minimal KD
5. Tune KD term
The goal of this step in the tuning is to continually increase the KD term until the
actual velocity profile nearly matches the commanded velocity profile (Figure 3).
If the initial KD10 results in an actual velocity profile that is <50% in magnitude
of the commanded velocity profile, the first steps can be large (ie. Increase from
KD10 to KD50 and repeat). Otherwise steps of 5 or 10 are sufficient. As the
actual velocity profile gets closer to the magnitude of the commanded velocity
profile, the steps of the KD value should decrease to narrow in on a value. One of
three things will indicate that the proper KD value has been selected:
1. The actual velocity profile is within ~5% of the magnitude of the commanded
velocity profile.
2. Increasing KD no longer changes the shape or magnitude of the actual
velocity profile.
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
3. Resonance in the system prohibits KD from increasing further. In this case,
proceed to step 6 below for help in tuning PL term.
Figure 3: Step response with ideal KD
6. If necessary, tune PL term
In certain systems, mechanical resonance can be magnified by the high frequency
component of the KD term. This can turn minor resonances into actual position
oscillations. The Galil motion controller has a low pass filter which is used to
remove these resonances, thus allowing the user to increase the KD term for
sufficient damping. This resonance can be heard as a high frequency buzzing, or
can be seen on the actual velocity profile. If this situation occurs, the low pass
pole filter is added, starting at a low value of PL0.1 and continually increasing
while checking the step response. If the high frequency noise is removed, return
to step 5 and increase KD. If the high frequency persists, raise to PL0.2 and
check the step response. Continue this process until PL has removed the
resonance, and KD has been raised to a sufficient value
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
7. Tune FV term
While working with the velocity response from the system, it is convenient to also
tune the feedforward velocity gain of the system. As seen in the plots of
commanded and actual velocity in GalilTools, it is typically not possible to raise
KD high enough to match the two plots 100%. The FV term is added such that the
two plots match each other completely. Similar to the process for KD and PL,
start FV with a low value of FV5. The actual velocity trace should increase.
Continually raise FV until the two plots match (Figure 4).
Figure 4: Ideal velocity profiles with FV and KD
8. Tune KP term
At this point, the velocity terms are tuned properly and the positional terms can be
tuned. For safety, the position error of the system should be zeroed at this point to
prevent jumping of the motor when adding the initial KP. Stop the step response
with “ST”, and issue “MO” followed by “SH”. This will zero position error and
allow the user to assign an initial KP value of 5. Run the step response with
“XQ”, and evaluate the position error trace. On most systems, the KP value of 5
will be insufficient for proper operation, so there will be excessive position error
(ie. > 100 counts of error). Similar to the velocity terms, begin increasing the KP
value in increments based on the amount of position error. If the position error is
greater than 100, continue to increase KP in increments of 5 or 10. As the
position error decreases, decrease the KP increments. The signs that the KP term
is getting too high for the system are as follows. If these symptoms occur, back
off from the KP value and move on to the next step.
1. Position error trace shows unstable motor behavior.
2. Position error peaks at acceleration and deceleration begin to increase, or are
no longer decreasing.
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
3. Motor is getting audibly noisy, which can be seen as noise on the position
error trace.
Figure 5: Position error with KP5, Scale 100 counts/div
Figure 6: Position error with ideal KP80, Scale 20 counts/div
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
9. Tune FA term
The scope now shows that position accuracy during and after the move have
improved significantly, and should be minimized for this system. However, due
to the high KP term there is most likely a large position error spike at both the
acceleration and deceleration phases of the profile (as seen in Figure 6). The
feedforward acceleration gain (FA) on the Galil controller is implemented to help
remove that spike. Similar to other filter parameters, start with a low FA of 0.5
and begin increasing. As FA increases, the positive and negative spikes at the
acceleration and deceleration will begin to diminish. If the spikes ever invert
from their original direction, this is a sign that FA has become too high.
Figure 7: Position error with FA
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
10. Tune KI and IL
The final step in tuning the servo motor is to increase the KI term to remove any
steady state error. Due to friction and inertia in systems, motors when driven to a
position may not settle on the exact commanded position (ie. TE ≠ 0). The KI
term is used to overcome this friction and zero out the position error. Figure 7
above shows that during the rest between moves, the 2-3 counts of position error
is never removed from the system. This will be even greater on systems with
significant friction. The IL allows the user to set a limit on the integral term, as
well as selectively “freeze” it out during motion. For this tuning, we will assume
the goal is high speed step and settle. In those cases, freezing the integrator
during motion is ideal. This is first accomplished by setting IL to a negative
value. Start with “IL-9.9” to allow full torque from the integrator. Next, set a
minimal KI value of 0.5 and view the effect on the position error. Continue to
increase KI to minimize steady state position error, which is the 250msec between
consecutive moves, to as close to zero as possible. If the position error begins to
show oscillations, decrease IL until the oscillations are removed.
Figure 8: Position error with KI and negative IL
Additional Notes
The recommended GalilTools scope setup used for this tuning has a fourth trace
enabled for Torque. This value represents the +/-10V output from the Galil to the
amplifier, with +10V being full positive torque and -10V being full negative
torque. This should always be monitored for saturation. If this value ever
approaches the maximum +/-10V, this indicates that the system is being pushed
too hard. Either decrease the profile parameters (SP, AC and DC) or change the
amplifier gain in the system for more torque.
As mentioned in the KI tuning section, this system was tuned for high speed step
and settle. In this case, the critical number is how quickly the motor moves and
settles into position. Position error during motion is less critical. If a system needs
to be tuned for accurate position during motion, it is advised to keep the KI term
active at all times, which is done by keeping a positive value for IL (ie. “IL9.9”).
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
In that case, tuning the KI term is similar, starting with a low value for KI and
increasing until position error is minimized for both motion and settling. If the
position error begins to oscillate, or KI can't be raised high enough to reduce the
position error, the IL value should be reduced.
The motion controller also has a notch filter which can be used to remove system
resonances. A notch filter is used in cases where a machine has a resonance of a
certain frequency which prohibits proper tuning of the system. In those cases, the
notch filter is used to remove the resonance and tuning is performed as above. For
additional information on tuning the Galil notch filter, please see application note
2431 on the Galil website (link:
Galil Motion Control, Inc. • 270 Technology Way • Rocklin, CA 95765 USA • 800-377-6329 • Ph: 916-626-0101 • Fax: 916-626-0102 •
Download PDF
Similar pages