Mavic 2 Pro out of control, crashed into lake... help?

[BudWalker]

Sorry for the delay - I was on a transatlantic flight with a broken internet connection.

This is a strange one. The problem began very suddenly at around 321 seconds, when both IMUs recorded what looks like an unphysically large negative y axis (to the left) acceleration, and applied a fairly extreme right roll to attempt to correct it. Then everything went south, figuratively speaking. This needs some more in depth analysis.

View attachment 76207

I'm not entirely satisfied with this but another possibility is that the Y axis acceleration was real and was caused by an object striking the AC from the right and above. The CW roll wasn't commanded by the FC it was the result of the impact from above on the right side of the AC. In addition to the Y axis acceleration the Z axis acceleration increases almost to 0.0 indicating the strike was from above.

The strike appears to have been on the right front prop. Motor speed, commanded and current are all consistent with the right front prop being the strike point.

Also, the CCW yawing indicates a propulsion issue with the right front.

The effects of the strike appear to be transitory as the AC recovers normal attitude in about 6 secs.

I don't know why there is an altitude gain after this.

 

[BudWalker]

Only smoking gun I can see is motors powering up at 320.6 despite no throttle input, then big elevation change, then this in the log:

324.130 : 17931 [L-FDI][CTRL]: fault on , height_ctrl_fail
324.150 : 17932 [L-RC]craft ctrl failed!!!

..so could be it realized by monitoring its commands vs what the IMU sees that undemanded throttle was happening?

The motor speed increases were the result of the FC commanding higher speeds. See post #41 above.

 

[BudWalker]

..... Otherwise it must be an invisible UFO applying tractor beam trying to kidnap it and failed.

I think you're right. See post #41

 

[BudWalker]

There have been a few confirmed examples of Mavic 2s that don't request recalibration based on either time or distance. Those not requesting at 30 day intervals appear to be because the last-calibrated date is set far in the future. I haven't seen the DAT files for aircraft not requesting based on distance - you should take a look at the event stream and see what it reports.

My personal theory is that those of us that have to recalibrate have different magnetometers than those who don't have to recalibrate. For what ever reason DJI has determined that the problematic magnetometers are more likely to change with time and require recalibration. The distance thing is just thrown in for good measure.

 

[Gryphon962]

The motor speed increases were the result of the FC commanding higher speeds. See post #41 above.

Excellent plots. I agree this looks like an impact. However, I would expect the impacted rotor speed to go down sharply then recover normal rpm. In this case the left front rrpm seems to go up, which is odd unless a piece of blade, or a whole blade, got sliced off. Speculation - would losing one blade from the front left cause the large lateral acceleration we see?

 

[ff22]

Each of the three or four times I've traveled more than 25 or 30 miles I have been required to do the Calibration Dance. Both arriving at a destination and then on return home.

 

[codaroommedia]

I'm not convinced that this was a compass problem.

View attachment 76208

So... what does this graph show us? Sorry for my inexperience with this. It’s clear something’s haywire here... I’m just curious what it is, in layman’s terms. Thanks!

 

[gnirtS]

Its every 30 miles for me (50km) and the .dat logs even show it triggering this.

Although the error is magnetic interference if you look at the actual compass values on the sensor status page you see the noise level as low as it normally is as opposed to genuine interference.

The manual states 30 miles and the drone actively enforces this on mine. As soon as it gets a GPS lock it'll trigger a magnetic warning if its over 30 miles and write an event saying distance from last calibration point in the .dat.

 

[RayOZ]

Excellent plots. I agree this looks like an impact. However, I would expect the impacted rotor speed to go down sharply then recover normal rpm. In this case the left front rrpm seems to go up, which is odd unless a piece of blade, or a whole blade, got sliced off. Speculation - would losing one blade from the front left cause the large lateral acceleration we see?

But usually, any strike to the prop would result in a 'motor obstructed' warning. There wasn't any of that in the log.

 

[codaroommedia]

But usually, any strike to the prop would result in a 'motor obstructed' warning. There wasn't any of that in the log.

 

[codaroommedia]

Also, I had clear line of sight to the drone, as did the people on the boat. We were over water, nothing to obstruct the drone, and no birds/anything in the area at the time.

 

[paulatkin73]

Also, I had clear line of sight to the drone, as did the people on the boat. We were over water, nothing to obstruct the drone, and no birds/anything in the area at the time.

are you absolutely positive about 'no birds' part - nothing at all, even if small, collided with that motor? i know you stated it before, but it really looks similar to a hit into that motor with possibly a destroyed prop and subsequent increased current given to compensate. but, if it is not what happened - then not. odd data...

as dji replaces it - then perhaps indeed it was a glitch.

 

[codaroommedia]

Positive. I was only 600-700 ft away, and was watching the drone when it started to freak out. Could a malfunctioning prop have caused this on its own? Even the sudden rise in elevation/48+ mph move?

are you absolutely positive about 'no birds' part - nothing at all, even if small, collided with that motor? i know you stated it before, but it really looks similar to a hit into that motor with possibly a destroyed prop and subsequent increased current given to compensate. but, if it is not what happened - then not. odd data...

as dji replaces it - then perhaps indeed it was a glitch.

 

[BudWalker]

Excellent plots. I agree this looks like an impact. However, I would expect the impacted rotor speed to go down sharply then recover normal rpm. In this case the left front rrpm seems to go up, which is odd unless a piece of blade, or a whole blade, got sliced off. Speculation - would losing one blade from the front left cause the large lateral acceleration we see?

Typically, if all or part of a prop is missing the AC will start rotating because the missing prop is no longer providing angular momentum. A propulsion issue in the rightFront and leftBack will cause the CCW rotation seen here. (BTW, it was the right Front not the leftFront). The P3 would just rotate and corkscrew into the ground but the Mavics also start tumbling.

I'm going to speculate that it was just a small piece that went missing. Although there was CCW rotation it was much slower than is typically seen. Also, the rightFront commanded only went to 92% where usually it'll go to 100%. And, the M2 was able to recover in the sense that it didn't spin and tumble it's way down.

Let me re-iterate that I'm not entirely satisfied with this analysis. It's tough with just the tablet .DAT.

 

[sar104]

Typically, if all or part of a prop is missing the AC will start rotating because the missing prop is no longer providing angular momentum. A propulsion issue in the rightFront and leftBack will cause the CCW rotation seen here. (BTW, it was the right Front not the leftFront). The P3 would just rotate and corkscrew into the ground but the Mavics also start tumbling.

I'm going to speculate that it was just a small piece that went missing. Although there was CCW rotation it was much slower than is typically seen. Also, the rightFront commanded only went to 92% where usually it'll go to 100%. And, the M2 was able to recover in the sense that it didn't spin and tumble it's way down.

Let me re-iterate that I'm not entirely satisfied with this analysis. It's tough with just the tablet .DAT.

That doesn't explain the accelerometer data though - does it? We are looking at around -40 m/s² (4 g) laterally.

 

[paulatkin73]

That doesn't explain the accelerometer data though - does it? We are looking at around -40 m/s² (4 g) laterally.

even if it was a hit - it does not properly explain how and why for exactly one sec with no throttle input it went up drawing crazy current - from 5m 22.8s till 5m 23.8s, then drops down ignoring given throttle up. it rotated and spinned prior to that climb, but stopped spinning during the climb. odd.
if you know structure of the xls file - does it have any metric for level of vibrations?

 

[BudWalker]

even if it was a hit - it does not properly explain how and why for exactly one sec with no throttle input it went up drawing crazy current - from 5m 22.8s till 5m 23.8s, then drops down ignoring given throttle up. it rotated and spinned prior to that climb, but stopped spinning during the climb. odd.
if you know structure of the xls file - does it have any metric for level of vibrations?

Wouldn't the accelerometer data should show vibrations.

 

[BudWalker]

That doesn't explain the accelerometer data though - does it? We are looking at around -40 m/s² (4 g) laterally.

I was speculating that it was struck from the right and above causing the Y axis and Z axis accelerations as well as the CW roll. Prop damage just indicates the strike point.

 

[paulatkin73]

Wouldn't the accelerometer data should show vibrations.

what they show is difficult to interpret. i also added current graph there to show how it pumped up power after that initial smack.
also, it remained 9.4A or so current for some time, then 7A until very end - 9.4A is a loiter level, yet, it was falling fast.

 

[BudWalker]

what they show is difficult to interpret.
View attachment 76288

You're right. The M2 .DAT has a higher sampling rate which would make it easier to do some filtering and see any vibrations.