Mavic 2 Zoom crash / fly away

[Tenko]

Hi i am after some help please. Yesterday i was working internally at a factory site. I had flown the length of the factory and was bringing the drone manually back to land, when it appeared that the drone took off at a high speed without any button pressing my end, i believe that i hit the reverse button left and right, but nothing seemed to take affect on the drone. Luckily the factory was empty, but it crashed into a internal wall smashing the drone to bits. There was a lot of metal on the inside of the factory, so whether that screwed things up? but the fact it took off at a rate of speed, in what appeared to be of its own doing is a bit concerning, could someone check the .DAT for me please, as im going to go through the insurance, but if it was a dji issue then i will go through them. If you require any more info please drop a message in.

Thanks in advance

 

[boblui]

The fly away was due to the active IMU ( IMU1 ) suddenly made an erroneous adjustment of 156 degrees to the yaw value at 560 sec into the flight. As the result, the craft could not sense the direction correctly and ended up in crash. This phenomenon has been seen before but the reason has not yet been identified. The last time I saw it was on a Mini 2 a few weeks ago.



All related sensor outputs including the gyro and the compass ( active and standby units ) did NOT indicate any yaw change so the cause was not magnetic interference or malfunction of sensors but some strange thinking of the active IMU that was not agreed by the standby one.

 

[slup]

...This phenomenon has been seen before but the reason has not yet been identified.

The trigger for this is most probably that the FC in just that moment didn't have a reliable GPS health (2 of 5 possible) & lost the visual position from the VPS sensors ... can't say why this makes the FC act like it does though.



Exactly there the disagreement between IMUYaw1 & magYaw beacame first approx. - 30 degrees but then turned to +160 degrees.

We see from comparing all available Yaw data (VIO, IMU1, IMU0, magYaw) with the gyro data that the used IMU1 yaw value probably was wrong ... as @boblui points out. But despite this the FC trust the IMU1 to be the correct one & disregard the others ... at 564sec the DAT event log indicate a FUSION error, but pinpoint the magnetic heading to be the culprit.


Earlier in the flight it was a lot of periods where both the GPS health was below 3 & VPS inputs was missing ... so it's unknown why this happened just this time though.

But the result was that the FC thought that it was almost turned the opposite way compared the direction the AC actually was pointing. So the result of all your stick inputs became to the opposite direction for the AC ... this together with a VERY bad positional hold ... if any at all to be honest. All this must have been very confusing & making you use large stick inputs that just made everything worse.

This opposite reaction to your stick inputs can be seen here ... first elevator & pitch with the X speed (Yaw graph just there for a reference).



Then here aileron & roll with Y speed ...



So with the fusion error active & with no auto breaking due to no means of positional data ... you actually drove it into the wall by your self.

Weather this was a pilot or AC error ... well, it is a grey zone here. It was a very risky flight over all, ATTI mode could be expected. But with that said ... a fusion error couldn't be planned for in any way.

I would try to point this towards DJI ... think it's a good chance that they will take this as a warranty case.

 

[Tenko]

Thank you very much for your quick replies, i will indeed go to DJI.

 

[eaglesurfing]

Welcome to the forum and enjoy.

I hope you do get this resolved and I am always thankful for the experts on this forum.

 

[BudWalker]

This incident occurred only when both navHealth dropped to 2 and the vision system data was not valid. This graph shows navHealth (ctrl_level) and the VIO velocity data. Invalid VIO data is indicated by the gaps.


At time 558.5 secs navHealth was 2 and both VIO:velE and IMU_ATTI(1):velE were separating from the GPS:velE. I suspect the yaw adjustment was an attempt to make IMU_ATTI(1):velE track GPS:velE. Seems to have started to work before the impact.


I'm left wondering why the FC didn't switch to ATTI instead of continuing in GPS_ATTI.

 

[slup]

...I'm left wondering why the GC didn't switch to ATTI instead of continuing in GPS_ATTI.

I guess that ATTI wasn't far away there ... but it seems that the FC's on the newer AC's is really reluctant to fall back to ATTI.

 

[BudWalker]

I guess that ATTI wasn't far away there ... but it seems that the FC's on the newer AC's is really reluctant to fall back to ATTI.

ATTI hesitancy started with the MP roll out and has gotten more prevalent since then. I think it's in response to the P3 series always switching to ATTI - quite often with no apparent cause.

 

[Tenko]

Very impressed with the analysis guys, how would you suggest approaching dji?

 

[BudWalker]

Very impressed with the analysis guys, how would you suggest approaching dji?

I suspect you won't get very far with DJI. I'm not offering an opinion about pilot vs DJI fault. But, I think that another Mavic 2 would behave the same for the same set of conditions.

 

[slup]

I suspect you won't get very far with DJI...

I would try in this case ... @Tenko, go here ---> DJI Repair Services - DJI & push "Online repair request" & take it from there, DJI will tell you what they need, you most probably need to share the logs with them & then send the AC in.

 

[boblui]

Very impressed with the analysis guys, how would you suggest approaching dji?

If the drone is still under warranty, it does not harm to make a claim. The log clearly showed that the craft accelerated forward by itself and maintained the large pitch angle even after the elevator stick was released so I think its the drone’s fault. The complication is that during that last 4 seconds, full forward elevator input was applied for 60% of the time so DJI may consider the movement of the craft to be commanded by the pilot and its not easy to argue on that.

 

[Tenko]

Thanks again, I’ve engaged DJI, and it’s in its way to the Netherlands. Will keep the post updated

 

[slup]

Let us know the outcome ... good luck