ArduCopter Analytic Tune Tool

ArduCopter
,
1

This is a video I put together to start the conversation about using the Analytic Tune Tool to tune Multirotor Aircraft.

Helicopter roll and pitch here:

This was a tool created by Bill Geyer and he goes into detail in the following blog post.

I hope you find this usefull and let the discussion begin :slight_smile:

8 Likes
2

I’ve been playing with this tool some more and I’m very interested. I tried it a few weeks ago and I’m circling back to take another shot.

A log from a pitch test. I’ll admit it was windy and it was a bit of fight to keep it in place. Looking at the notes I may have hit the data saturation point, but it’s the data that I have today to try out.

I’ve come up with two possible tunes and I’d like some feedback on them (based on the tool).

  1. Following the video instructions I got this tune. Interestingly it’s comparable to what Autotune had given me for this quad. Lower P/I, but higher D. Similar angle P.
    Leonard Method.param (581 Bytes)

  2. After some chat’s with @bnsgeyer in another thread I used his approach to drop the _FLTD and came up with this. Much higher PIDs, but less filtering.
    Bill Method.param (578 Bytes)

This is a Holybro X500V2. It has been tuned before and generally flies well, but for this I reset the PIDs just to go through the process. This is less about the quad, and more about learning this method of tuning.

and for reference, here’s the values from autotune that it normally flies on:

ATC_ANG_PIT_P,25.86118
ATC_INPUT_TC,0.15
ATC_RAT_PIT_D,0.004628581
ATC_RAT_PIT_D_FF,0
ATC_RAT_PIT_FF,0
ATC_RAT_PIT_FLTD,21
ATC_RAT_PIT_FLTE,0
ATC_RAT_PIT_FLTT,21
ATC_RAT_PIT_I,0.1297368
ATC_RAT_PIT_P,0.1297368
3

The less filtering you have the faster you can make your pids work and the higher the gains can be.

The problem is we need to offset our filter settings against the level of noise. In your example your P and I gains are almost double and the D gains are almost 50% higher. So those are direct multipliers on the level of noise. On top of this you have removed all D filtering. The combined impact can be an unacceptable level of noise.

My normal approach is to find the minimum level of filtering I can tolerate then tune the aircraft and reevaluate the noise. If it is then too high I increase the level of filtering and tune again (I also reduce the PID’s to ensure the aircraft is stable with the increase filtering)

1 Like
4

In the “Leonard” method I left the filtering alone. Do the PIDs that I came up with there make sense?

Thanks for helping me understand this tool!

5

I had a quick look but I need to spend a little more time to see if I get the same results. I will get back to you.

1 Like
6

Since there’s two different discussions on this tool I’ll risk double posting this. I wrote a script for changing the SID_AXIS values in flight. Posted here: Using Analytic Tune Web Tool to tune your helicopter - #12 by Allister

2 Likes
7

Hi @Allister,

Here is what I would suggest as a stable tune. I went back to -6dB on the rate loop at 180 degrees.

ATC_ANG_PIT_P,18
ATC_RAT_PIT_FF,0
ATC_RAT_PIT_P,0.148
ATC_RAT_PIT_I,0.148
ATC_RAT_PIT_D,0.0046
ATC_RAT_PIT_D_FF,0
ATC_RAT_PIT_FLTT,21
ATC_RAT_PIT_FLTE,0
ATC_RAT_PIT_FLTD,21

We are still working out what the best process would be though so everything needs to be taken with a grain of salt.

1 Like
8

Thanks for taking the time to do that @Leonardthall!

I think I was pushing for -3dB based on my understanding of the video, but I may have messed that up. This tune you’ve provided looks very reasonable.

I’ll certainly play with this more and I’m looking forward to any guidance or process instructions that may come out.

9

@Allister As Leonard mentioned earlier, the noise in the control signals is a very important consideration when tuning. In looking at your data, I was very impressed with the highest frequency you were able to achieve with excellent coherence. In the plot below, you can see that you are seeing excellent coherence out to 22 hz and good coherence out to 33 hz. Most of my data of a multirotor using throttle based notches hardly gets above 8 hz.

image
image1920Ă—1080 225 KB

I looked at your INS notch filter settings and it appears you were filtering each motor based on the ESC RPM. So I believe this really helped with the clean data.
After looking at logs using this tool, I am beginning to equate the max frequency with good coherence with limitations due to noise. I have seen other logs where the data was noisy and the frequency sweep was unusable to tune because as you pushed up the rate D term, the drop off of the gain plot started to exceed the frequency where there was good coherence.
So in your case, I was satisfied in removing the D term filter and tuning with out it. I would recommend that if you do this, that you want to fly with the new tune and filter and look at the time history of the RATE.ROut, RATE.POut, and RATE.YOut signals with the fast attitude logging setting to ensure that the noise is less than 0.05 which is what I believe Leonard uses as his metric to judge low noise in the control signals

5 Likes
10

Thanks for making the connection between coherence and noise. I wasn’t clear where noise would show up in these charts.

11

Can you clarify what you mean by “going back to -6dB at 180 degrees”?

12

Hi @Joshua_Kordani ,

The 180 degree point is where the middle graph steps from -180 to 180 degrees. This is the point where the reaction is perfectly in the oppoiste direction to the input.

-6dB is the magnitude of the response, shown on the upper graph, and represents a response of half the input.

Did that make sense?

13

Do you mean you chose filter settings that would result in -6dB at -180deg?

14

Yes, PID settings that would result in -6dB at -180deg.

1 Like