PX4 crash on EKF2

Hello guys!

We just had a terrible crash this Wednesday, where we lost the frame and a lot of expensive sensors (in total around 8k eur). Please help me to find out the source so it would never happen again!

Frame: Copter X8
Firmware: e1e15add01bf9292bd636581037a8673b376af15, EKF2 (I’ve been flying it for few weeks and it showed no issues in performance, that’s why it’s a bit outdated.
Conditions: Moderate wind around 7 m/s. Up in the mountains.
UAVenture log: http://logs.uaventure.com/view/N93kLksfGwbzkkXKNEVip3
Link to log, other files on SDcard: https://drive.google.com/folderview?id=0B4m7lGzrnKoub0U0c09JekoxU3M&usp=sharing

Unfortunately, it seems that Pixhawk hanged, so the log stops just before copter falls. And it wasn’t just falling but flipping at least 5 time around longitudinal axis before it hits the ground. So almost nothing survived.
I know it’s really hard to examine the problem without the log, but could you please try to give me at least some hints, I would appreciative any help.

The last message I got from msgs_date_time.txt is:
2016_05_18_11_35_03: Operation timeout
But it seems just like mavlink communication error.

Thank you in advance and God bless you! :slight_smile:

It does seem like a hang; I’ve had this kind of crash before when testing a local change which corrupted RAM.
I’m currently chasing a hardfault which occurs in ekf_att_pos_estimator https://github.com/PX4/Firmware/issues/4530, but haven’t done any testing with EKF2. Unfortunately, this fault is hard to diagnose since it occurs very infrequently.

@Ilia_Sheremet Sorry to hear about your crash. In general I can only advise against running software that is not the default yet on any very expensive or very risky airframe or mission.

What is interesting is that your attitude tracking deteriorates quite a bit before the log stops: http://logs.uaventure.com/view/N93kLksfGwbzkkXKNEVip3#AS_Roll_PLOT

It seemed to have mostly tracked ok (not great, the system needs tuning) the earlier part of the flight. You can also see that one of your motors has reached its minimum: http://logs.uaventure.com/view/N93kLksfGwbzkkXKNEVip3#Actuators_PLOT

What is notable is that you don’t have really a large yaw error at this point in the flight. I don’t want to speculate, but the sort of attitude runaway combined with motor spread has hinted at an electrical problem in the past:

@Mark_Whitehorn You can’t really draw a conclusion based on the log - it could be a hard fault or power loss. The issue you were facing was with a completely different estimator.

We’re digging into that with highest priority.

@Mark_Whitehorn Thank you for your reply, but as Lorenz mentioned, its seems like a problem of another estimator.

@LorenzMeier Thank you for your help! Is there a way to better log the copter’s flight in order to get more information in case the same fail happen again?

[quote=“LorenzMeier, post:3, topic:443”]
your motors has reached its minimum
[/quote] But isn’t it normal behavior during hitting wind gusts? We had quite a lot of it that day.

[quote=“LorenzMeier, post:3, topic:443”]
t. I don’t want to speculate, but the sort of attitude runaway combined with motor spread has hinted at an electrical problem in the past:
[/quote] Could you specify which kind of electrical problems it hints at?
I forgot to specify at once, but the battery was still plugged in after the crash and all motors were beeping as they don’t get the signal from Pixhawk. The board was powered through 3dr power module and servo rail (with Zener diod) and we had never had problems with power previously.

@LorenzMeier True, a loss of avionics power could be the cause. If that were the case, one would expect to find a bad power cable or a dead power supply post-crash. I’ve had one of those crashes also, now that you mention it; the ground lead in a work-hardened power cable (it was often connected/disconnected while debugging on the bench) broke in flight, cutting power to the FC, but not the motors. @Ilia_Sheremet In that case the broken ground lead was fairly obvious and clearly not the result of crash damage.