.

# Shape-based Feedforward Tuning:

## Acceleration Pattern — Adjusting Kaff

 General Rule of Kaff: If the position error plot and the commanded acceleration plot have a similar shape, add acceleration feedforward (Kaff). If they have a similar, but opposite shape, reduce Kaff.

Notice that in the plot below, the position error plot (white) has a similar shape to the commanded acceleration (yellow), therefore we will add Kaff.

 10,000 count move Trapezoidal move:       5e5 accel, decel, vel Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error Kaff = 0 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 187 counts

On this stage, adding a Kaff of 1 makes no noticeable difference:

 10,000 count move  Trapezoidal move:       5e5 accel, decel, vel Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error Kaff = 1 Kp = 100 Ki = 0 Kd = 2,000

After doubling Kaff each move, we eventually get the following plot when Kaff = 50,000.

 10,000 count move Trapezoidal move:       5e5 accel, decel, vel Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error Kaff = 50,000 Kp = 100 Ki = 0 Kd = 2000

It looks like the position error was roughly reduced by half when we changed Kaff = 50,000. That means that if change the Kaff = 100,000, the position error should be fairly close to zero.

 10,000 count move Trapezoidal move:       5e5 accel, decel, vel Green - Commanded Velocity              Yellow - Commanded Acceleration                    White - Position Error Kaff = 100,0000 Kp = 100 Ki = 0 Kd = 2,000 Peak Position Error = 28 counts

The resulting plot shows a position error that is very close to zero and only has a peak position error of 28 counts—an impressive 85% reduction.

NOTE: This reduction comes with no reduction in stability.

As you can see, adding Kaff (acceleration feedforward) can dramatically reduce position error. The stage is actually more stable because P, I, and D need to deal with smaller disturbances, while Kaff is doing most of the work.

Obviously, for the example above, we found a good value for Kaff rather quickly. To see some additional plots with different Kaff values (for the same commanded move) to show that the shape pattern of Kaff is fairly simple and consistent, click here.

 Legal Notice  |  Tech Email  |  Feedback        Copyright © 2001-2021 Motion Engineering