more sophisticated vibration analysis with full IMU logging

[Randy Mackay]

 

     We’ve recently introduce full rate IMU logging into ardupilot which means that we can now log the accel and gyro data at their maximum rate (1khz for the MPU6k, 1.6khz for the lsm303d) to the dataflash.  This is  leading to some in-depth understanding of the vibration of our vehicles by the likes of Leonard, Jon and Paul.

 

     For example, attached is a picture from Leonard produced in matlab of my IRIS’s vibration profile.  So the frequency of the vibration is on the left-hand scale, the flight time is along the bottom, and the brightness captures the strength of the vibration at that freq/time.  So you can see strong vibration bands around 100hz, 200hz, 320hz and above.  The thought is:

·         100hz (or 6000rpm) is caused by the rotation of the motor (i.e. slightly bent motor shaft, etc)

·         200hz is “blade passage frequency” where the blade passes over the arm of the vehicle (two sides of the blade so it happens at twice the frequency of the motor).  It might also be the prop getting caught by the wind as it comes around to the front of the copter (aka blade flapping?).

 

     To help make this kind of analysis more readily available to all the MP’s got a “FFT” button now on the secret Ctrl-F menu (see second attachment).

 

     The full rate IMU logging can be enabled by setting the LOG_BITMASK to “All+FullIMU” but beware, you need a decent SD card to keep up with the amount of data it writes and the log files are massive.  You can look forward to 10minutes of downloading from just a 3min flight.

 

     If you have more questions about this, I hope Paul, Jon and Leonard and answer your questions!

-Randy

[Randy Mackay]

 

     I forgot to mention that to complement this, we’ve got some in-flight vibration analysis that’s now in master and *may* be included in AC3.3.

 

     The vibration formula comes from Leonard/Paul and is the standard deviation of noise on the accelerometers in  m/s/s.  You can see the exact formula in the commit below:

             https://github.com/diydrones/ardupilot/commit/0db7acc6281c0c47544fb8706f0f99933e1b4b0f

 

     We also capture whether any of the accelerometers were clipped (i.e. went beyond their 16G range).  These values are all stored in a new VIBE dataflash message and are sent down to the ground station in a new MAVLink message.

            https://github.com/diydrones/ardupilot/commit/e677a100c625bce1385a8a047c0a7f515adc4097

 

     Once we get the GCS developers to build out support for this new message we could have live info to the Pilot when their vibration levels are high so that hopefully they can bring the copter down and sort out the problem before it leads to nasty flight behaviour.

 

     Attached is a picture of the vibration levels captured in-flight from my IRIS.  So vibes are very commonly up to 5m/s/s (i.e. 0.5G) with short peaks as high as 40m/s (i.e. 4G).  There’s no clipping however.   This is a totally average IRIS that’s flying well so it’s probably a good reference of a healthy copter.

 

-Randy

[Robert Lefebvre]

This is a HUGE development!  Thanks to all!

Any chance of getting FFT running live on the Pixhawk to get live feedback of vibration frequency and intensity?  It must be technically possible, as I've seen this running on even really cheap heli FBL controllers ($50).

FYI, the 200Hz thing in forward flight, the root problem is not blade flapping, but disymmetry of lift. The blade sweeping forward has a higher effective angle of attack than the retreating blade, so produces more lift.  This produces a torsion.  In helicopters, this is alleviated by "flap to equality" where the advancing blade rises, while the retreating blade falls.  Multirotors have fixed propellers, so they can't flap.

The effect of this has been visible already with the existing logging, if you just compare vibration levels to GPS speed.  But this should make it more clear, as I expect you'll see the 200hz-ish band will be the one that gets stronger with forward flight.

--
You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[David Pawlak]

Very Cool

[pritam....@gmail.com]

awesome already. It will be even more if we can hard mount another IMU on the frame and have the option of running similar monitoring on that as well. That way we will have two points one frame and other autopilot. Can easily detect bad vibration mount, loose foam which tend to happen in case of hard landings. And if you have dome can go unnoticed.

Very Cool

--

You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Regards,

pritam

[Tom Pittenger]

The PixHawk 2 has one "rigid" IMU and two vibrationaly isolated IMUs

[jolyboy]

Amazing. Can't wait to see the results from a heli. Should be able to see everything, motor, main rotor, blades, tail, gears etc.

Thank you for adding this.

[pritam....@gmail.com]

cool, lets see if I can afford it .

[Luis Vale Gonçalves]

Hi Randy

I saw those new developments were only on Master. Are you planning to have them also on Copter 3.3 branch ?

brgds

Luis

[Philip Rowse]

Just for this reason!

Phil 

[Randy Mackay]

Luis,

 

     I’m not totally sure yet but probably.  I’m thinking about rebasing the next release candidate on master because the pde->cpp and submodule change has made merging between the branches quite difficult.

 

    The release is taking longer than expected because we’ve hit some unexpected problems including:

·         Tradheli is dropping in altitude when switching modes (Rob)

·         Altitude run-aways during high vibrations (Paul/Leonard)

·         Need to reset alt when taking off withouh GPS lock (Randy)

 

-Randy

--

[Andy Piper]

+1

This will make everyone's lives easier.

[Jesus Alvarez]

Really great!

It is going to be one of the best weapons against vibration problems.

Many thanks for the work

[Jon Farhat]

Wow.  Thank you for doing this!

j.

[Paul Riseborough]

I'll have a chat to Michael about cropping the FFT display in mission planner, so that we do not show the 'mirrored' data to the right of the Nyquist frequency.

[Andrew Zaborowski]

As far as I understand the vibration mount thing is to filter out
vibrations at frequencies higher than the sampling frequency which
could affect the gyros. 100Hz-200Hz generally are ok, as long as
there's no sensor saturation they can be dealt with in software.

It is still a great inspection tool for finding and fixing the sources
of vibration.

Cheers

On 12 June 2015 at 22:36, pritam....@gmail.com

[Tom Pittenger]

It's the accelerometers that can't handle vibration, not gyro.

[Andrew Zaborowski]

The effect on MEMS gyros is documented, the bias voltage can change as
a result. That's why vibration mounts are used even where there are
no accelerometers.

I'd be curious to learn what effect there is on accelerometers and
what frequencies are in question.

Vibration is obviously bad (inefficiency, fatigue) but I don't think
below sampling rate / 2 it affects the IMU particularly.

Cheers

[Jolyon Saunders]

Am I correct in my assumption that the primary issues here are sensor clipping below the Nyquist frequency, and all other vibes higher than the Nyquist frequency?

So would that make this addition mostly useful for detection of accelerometer clipping below the Nyquist rate, and finding causes of vibes below Nyquist frequency?

You received this message because you are subscribed to a topic in the Google Groups "drones-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/drones-discuss/VfjH0HIF50k/unsubscribe.
To unsubscribe from this group and all its topics, send an email to drones-discus...@googlegroups.com.

[john...@gmail.com]

Clipping and > Nyquist are critical vibrations that have an instant impact on performance. But other vibrations are also bad. The sensors are outputting approximations of real world events with varying degree of signal and sampeling errors in every part of the chain (sensor acquisition, ADC etc.). As such, aggressive movements like vibrations will have a much higher error ratio (S/N) then say a slow copter/plane movement, making it harder for the EKF/DCM to make an accurate estimation.

[C Wong]

Very nice information guys. Excellent data collection.

Of course, some frames will have unique vibration signatures, maybe even different results than what is extracted/generalized from this analysis.

Definitely would be interesting to compare this will high speed video/smoke tests. I may have a phantom camera one can borrow (we just dissolved our entire studio tech dept unfortunately--so no more drone-film R&D :( ).

[Andrew Chapman]

I can run my X8 coax motor mixing with this on too, you definitely hear the noise signature changing with the mix and you can hear quieter/smoother points, it will be interesting to see how that correlates

--

[Karl Easterly]

This is a great addition!!  Couple this with high speed data loggin the diagnostics are going to get better and better...

Thanks much!