Shape-based Feedforward Tuning:

Velocity Pattern — Adjusting Kvff

Let's start off by saying that Velocity feedforward tuning is very different in the PID and PIV algorithms. If you are using the PIV algorithm, refer to the PIV tuning section(s) and skip this velocity feedforward section.

Adjusting Kvff is very similar to Kaff. The main difference is that you will be looking for patterns that look like the velocity plot, not the acceleration plot.

The first plot is with the acceleration feedforward set to 100,000 as determined in the acceleration feedforward page. Notice that the position error scaling has been changed from 500 peak-to-peak to 70 peak-to-peak to better see the error patterns.

	10,000 count move Trapezoidal move:       5e5 accel, decel, vel	Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error
	Kvff = 0 Kaff = 100,000 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 28 counts

When we zoom in on the error plot, we can see that there is a trend in the position error (white) that is similar in shape to the velocity plot (green). Since the position error and velocity plot are similar in sign (both have a positive peak), we will add Kvff. If the traces were opposite in polarity, we would reduce Kvff. Notice that the procedure that follows is virtually identical to the Kaff procedure.

	10,000 count move Trapezoidal move:       5e5 accel, decel, vel	Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error
	Kvff = 10 Kaff = 100,000 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 25 counts

	10,000 count move Trapezoidal move:       5e5 accel, decel, vel	Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error
	Kvff = 20 Kaff = 100,000 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 22 counts

	10,000 count move Trapezoidal move:       5e5 accel, decel, vel	Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error
	Kvff = 40 Kaff = 100,000 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 16 counts

Notice that there is not a noticable trend similar to the velocity plot at Kvfff = 40. There is a shift vertically that is constant for the entire motion in addition to some ringing on the transitions in acceleration and some Kaff error. At this point, we would call Kvff tuned. Remember that we are trying to get the velocity plot shape out of the position error plot, not going for absolute minimum position error peak.

After adjusting Kvff, you can see a small amount of position error pattern that looks like the acceleration pattern. It is common to first tune Kaff as best as possible and then Kvff, only to find that some small error in your Kaff tuning that you can only see after tuning Kvff. Just go back and fine tune your original Kaff tuning now that you can see the pattern more clearly.

Previous | Next