This quad has over 2 hours of hard flights on it and today, while flying with the owner, she came down hard at the end of the day. Battery pack voltage looked OK even though we were getting towards the end. Vibrations look good also although we did suffer a few prop strikes throughout the session due to a nervous pilot. Anyways, I noticed that baro alt was true and shows a descent while alt and dalt shoot up vertically. Any help is much appreciated!
[quote=“jschall”]IMU1 gets severely offset from IMU2.
This is a sensor failure… The MPU6k is messed up. I suggest getting a replacement board from 3DR.[/quote]
Thanks, that does look like the case after checking IMU2. It’s odd though that the code would ignore the baro in this situation. I wonder if EKF would have helped in this situation?
Anyways, yesterday I saw when I booted the Pixhawk up that the horizon was way off reality in Mission Planner. I was confused and cycled the power. The HUD horizon was OK and I moved on. But now after this IMU failure it makes me think that the IMU was bad or going bad yesterday. Thanks for your help, much appreciated!!
Just thinking; EKF helps sort out wrong values in a flow of data based on just previous correct events, right(?) In your case, wouldn’t EKF just filter the bad IMU data for a short while, then softly adapt to errorus data flow since the correct “average” actually turns bad over time? Or is a EKF the key to sort out and disregard a corrupt IMU??
[quote=“Flyhard”]Just thinking; EKF helps sort out wrong values in a flow of data based on just previous correct events, right(?) In your case, wouldn’t EKF just filter the bad IMU data for a short while, then softly adapt to errorus data flow since the correct “average” actually turns bad over time? Or is a EKF the key to sort out and disregard a corrupt IMU??
Sent from my iPhone using Tapatalk[/quote]
EKF has a fairly hack-ish IMU weighting algorithm which is updated on every GPS fix. It does not do well with hard failures because it does not update fast enough, but it does help out quite a bit with minor aliasing.
Now, hard failures of the mpu6k often involve much larger offsets (~6g) than you got. I would imagine that your failure wouldn’t have messed up the state vector that much before the ekf could deweight the mpu6k.