Detect and ignore accelerometer if measured vibration is high

[digitalentity]

In some cases above certain throttle value accelerometer starts to read garbage rendering inertial modes unflyable:

Not sure where this effect comes from. Possible causes:

1. Electrical noise when above certain power level ESCs draw currect in spikes causing high-frequency noise.
2. Mechanical noise

In both cases I suspect resonance with MEMS accelerometer natural frequency. Datasheet doesn't specify it for accelerometers but for gyros it's in range 27-40kHz for typical gyros. If I'm correct adding mechanical damping that absorbs high-frequency vibrations and electrical filtering (adding a capacitor to power supply) will help.

In any case the resonance effect is unsafe for the craft and we should be able to detect it and act accordingly.

[bk79]

That"s the sacred Gral for any closed control loop...and very little can be
done...if the measure is noisy before the reading you can correct it...if
the measire come from the observer(sensor)itself nothing can be done and
you need to change it or an external same sensors which is not affected by
the noise. If the problem is resonance it have to be solved only by
hardware...a typical example of resonance is "tacoma bridge"(look for the
video on youtube) detecting to ignore it doesn't solve the issue and in a
very short time during the flight the aircraft dismatle by itself.

Il 21/ago/2016 03:43, "Konstantin Sharlaimov" notifications@github.com ha
scritto:

In some cases above certain throttle value accelerometer starts to read
garbage rendering inertial modes unflyable:
[image: image]
https://cloud.githubusercontent.com/assets/11059099/17834772/1264506a-6793-11e6-908f-214ebc1527e5.png
[image: image]
https://cloud.githubusercontent.com/assets/11059099/17834775/3363381c-6793-11e6-9391-e5ac5888979e.png

Not sure where this effect comes from. Possible causes:

1. Electrical noise when above certain power level ESCs draw currect in
   spikes causing high-frequency noise.
2. Mechanical noise

In both cases I suspect resonance with MEMS accelerometer natural
frequency. Datasheet doesn't specify it for accelerometers but for gyros
it's in range 27-40kHz for typical gyros. If I'm correct adding mechanical
damping that absorbs high-frequency vibrations and electrical filtering
(adding a capacitor to power supply) will help.

In any case the resonance effect is unsafe for the craft and we should be
able to detect it and act accordingly.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#486, or mute the thread
https://github.com/notifications/unsubscribe-auth/ALSWinNE1742i0OtIFwL-G7h_gRLE6ohks5qh603gaJpZM4JpNED
.

[digitalentity]

I know we can't solve it in software, but we can detect it and prevent issues with Alt/Pos hold by disabling inertial estimation and relying only on sensors that are less accurate but more reliable at the moment (baro/GPS).

This will degrade performance considerably but will keep the copter/pilot/surroundings safe.

[bk79]

Physical you can also add a diode in series that will also prevent surges
coming from esc...the problem is finding a diode small and of a lot of
amps...

Il 21/ago/2016 06:43, "Konstantin Sharlaimov" notifications@github.com ha
scritto:

I know we can't solve it in software, but we can detect it and prevent
issues with Alt/Pos hold by disabling inertial estimation and relying only
on sensors that are less accurate but more reliable at the moment
(baro/GPS).

This will degrade performance considerably but will keep the
copter/pilot/surroundings safe.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#486 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ALSWiuoR-L-f8Ln1cJW48sphB53vZ5Ayks5qh9dpgaJpZM4JpNED
.

[digitalentity]

I'm thinking about adding a lowpass filter to the line powering the FC

[digitalentity]

Hm... I wonder if this is the cause of problems with airplanes and Naze32 full boards. Baro weight is very small - good enough for stable accelerometer but insacely small for noisy one. GPS-only setups use GPS Z coord/vel to correct the inertial estimate and GPS weight is usually set much higher - enough to keep the estimated position/velocity within sane limits.

Airplanes also usually have much higher vibration levels than multirotors...

[bk79]

Cap will be 0hz lp and adding the diode in the line will make it better

Il 21/ago/2016 06:46, "Konstantin Sharlaimov" notifications@github.com ha
scritto:

I'm thinking about adding a lowpass filter to the line powering the FC

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#486 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ALSWinD1f6m5vvlD__U3vyQlJ8TK2l0oks5qh9gsgaJpZM4JpNED
.

[digitalentity]

ESCs can easily draw 20-30A each and 30A+ diodes are heavy and bulky and I suspect it's not electrical noise that is driving accelerometer crazy. I suspect it's a mechanical issue.

Going out to verify in about half an hour - added silicone pads to soft-mount my FC, will fly and do a few punch-outs to see what do I have in logs. No electrical change so experiment will be clean and correct.

[oleost]

"disabling inertial estimation and relying only on sensors that are less accurate but more reliable at the moment (baro/GPS)."

Good idea, but maybe we should test fly and see how it behaves without the inertial estimator?

[bk79]

Wilco1969 had it also with no props so i doubt it's mechanica or he should
had the reading with much more less throttle since the rotor spins at much
higher speed for the same throttle. If it is that, damping pads should
solve if the fc is not integrated in the structure..i suppose it's
electrical...an 1n4001series with 1000uf cap in parallel is small enough
for the fc filtering...about detecting the issue you already have the
answer..baro reading... if those big changes in altitude are there also
baro or gps or sonar should feel them... But a good idea is integrate acc
reading and derivate baro or whatever pos reading we take in consideration
so to compare them in terms of velocity
Il 21/ago/2016 06:59, "Konstantin Sharlaimov" notifications@github.com ha
scritto:

ESCs can easily draw 20-30A each and 30A+ diodes are heavy and bulky and
I suspect it's not electrical noise that is driving accelerometer crazy. I
suspect it's a mechanical issue.

Going out to verify in about half an hour - added silicone pads to
soft-mount my FC, will fly and do a few punch-outs to see what do I have in
logs. No electrical change so experiment will be clean and correct.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#486 (comment), or
mute the thread
https://github.com/notifications/unsubscribe-auth/ALSWiqSumPzSnl6eR3NzZ5RjLGwm04Zdks5qh9sUgaJpZM4JpNED
.

[digitalentity]

It's a mechanical issue. Added o-rings to my FC mount and hit 4.3g at full throttle without any garbage readings from accelerometer.

[wilco1967]

It's definitely mechanical.... If it where electrical, I would expect it also on the other motors...

What caught me by surprise was the better 'feeling' motor, actually being the one causing the resonance issue. No way I'm going to be able to remove the problem with balancing.. it is critical speed issue.
What about the ACC sensor range ? Is it large enough to never ever be overloaded with the highest peaks? Similar problem happened on MultiWii a few years ago.

Don't worry about the copter disintegrate on flight... It's very small, but enough to make the ACC go crazy... I'm just unlucky to hit the wrong frequency.

[wilco1967]

I think I will order an SPracing Evo board
That is not part of the frame,so can be mounted on some double sided foam tape.... The same copter with all its vibrations has been flying perfectly before using a Naze on foam.
Then F3FC provides a clean build, but a nightmare to remove vibrations....