Problem with Dutch Roll?


I have a rudder-elevator model that has a pretty low CG - about 20 cm below the root chord. I should have a total of over 15 hours of mission time on this plane. This was with APM:Plane 3.0(.2/.3). I never had the time to fine-tune the autopilot. The flight performance was pretty acceptable with the minimally adjusted parameters.
I upgraded to 3.1.0 a few days ago. Wanted to do some additional tuning. I started with roll tuning. After increasing P and I a bit the plane started to oscillate in a motion very similar to Dutch Roll. I tried decreasing the parameters and finally reset to the previous parameters. Next I tried autotune - the plane became uncontrollable very quickly - it overshot the bank angles and exhibited Dutch Roll like behaviour when flying straight. I reset the modified parameters again and landed.
Then two days later I went flying again. I thought the plane should fly fine like it did before with the “good parameters”. Nope, the plane started to oscillate and wanted to go into a death spiral as soon as I switched to FBWA. I tried decreasing the roll P parameter, but it didn’t seem to have any effect. I had to switch to manual many time to save the plane. The plane flew perfectly well in manual. No sight of Dutch Roll or any tendency for corkscrewing itself into the ground.

I looked at the logs for both days, but I’m not at all sure what to start with. Do you have any ideas what to look for when examining the logs?

How about sharing your tlogs and/or data flash logs with the rest of us so that we can help you troubleshoot your problem.

Sorry about that. Here are the logs --> … N-ewa?dl=0

I think I also discovered a possible bug in the code during testing yesterday. When the plane went into a death spiral and pitched down over 90 deg then it tried to recover by pitching down further, but that only made the spiral more stable. Taking over manually the recovery was pretty simple - I just needed to pitch up and level the wings.
I added yesterday’s log (9.bin) to the dropbox folder --> … N-ewa?dl=0

After another flying session today it seems that the plane is truly untunable. The basic mechanism how the autopilot looses control over the plane is following:
[ol]the plane is in FBWA,
the plane overshoots the desired bank angle and
tries to correct it by overshooting in the other direction,
the oscillations grow in amplitude and the plane pitches down and starts spiraling towards the ground.[/ol]
This happens every once in a while. The plane seems unresponsive to the commands of the pilot.
Increasing the roll I-parameter seemed to alleviate the problem quite a bit, but still the autopilot lost control over the plane at least once.
I have no idea what to do next.