Copter: prevent bad Z-vibrations causing rapid climb.

[AndKe]

Here's latest victim of AccZ vibrations, http://diydrones.com/forum/topics/fly-away-please-help-me-figure-out-why

please note that IMU2 provided good data.
I know that this is a user/assembly error, but still, there should be some handling of this scenario - after all, we can still detect an altitude gain that contradicts desired altitude/(input)

[rmackay9]

AC3.2 will include mixing of the two accelerometers on the Pixhawk (AC3.1.3 only uses the MPU6k). We believe that luckily because one accelerometer runs at 800hz and the other at 1000hz that aliasing will be much less likely.

It has not been thoroughly tested yet but in one test we saw one accelerometer which freaked out due to aliasing but the other did not.

[rmackay9]

While testing in the simulator I found that the EKF (fast CPU boards only like Pixhawk) does not have this problem. With a bias of 2m/s/s the vehicle only climbed 2m. On the APM2 the vehicle climbed 82m.

[R-Lefebvre]

I wonder if this can be closed now? Better filtering of the accels. EKF. Bad sensor rejection... all coming with 3.3.

[rmackay9]

Yes, the increase in the accel range from 8G to 16G should fix this. We've never actually tested this in a copter but I think we are fairly confident.

[vmayoral]

After these changes we actually started seeing the copters climbing. Refer to #2026

[rmackay9]

I'm closing this issue after all the work we've done to improve the vibration handling. We can open a new issue when/if we see it happen with AC3.3 (or higher) versions.