By my standards, the tuning does not look nice overall, especially on pitch, with actual/desired discrepancies easily reaching 4 degrees. This is consistent with the overall shakiness. But I have never dealt with large machines, so “nice” or “not nice” is to be taken with a grain of salt.
What is probably more objective is the PID contributions. And here I see an increase of a D-term in pitch with time, which may quite likely have something to do with the change of sound you report.
If I switch the D-term off, one can see not only that P-term also increases with time, but that it is about 2x to 3x time smaller than D, which normally indicates that there is too much noise coming through.
Too much noise is consistent with quite high vibration levels.
This all does not quite explain why CTUN.ThO decreases with time. I had several hypotheses, in the order of appearance:
- Battery was heating up over the course of the flight, which decreased its internal resistance. While one cannot eliminate this completely, there are battery temperature readings in the logs, and they show quite steady temperature, so this seems unlikely. However, estimated battery resistance does decrease with time.
- Voltage scaling was configured incorrectly. But this does not seem to be the case, and battery voltage reported is well within the limits of the scaler.
- Outdoor temperature and/or pressure changed over the course of the flight. Again, based on barometric readings, this did not seem to be the case.
- Something heated up (or got loose) on the machine and changed its frame resonances and responses.
- ESCs have some kind of voltage sag compensation that played tricks with Ardupilot’s assumptions.
For #4, unfortunately, the logs do not contain any information to be analysed via Filter Review, so I cannot test this. However, this may explain both why PIDP increases towards the end of the flight, if one of the resonances becomes more active somehow. (#5 provides an alternative explanation for this though).
For #5, I additionally plotted the consumed power (as a product of reported voltage and current), and this stays the same over time. This tells me that energetically there is nothing strange. (It could have been if resonance response changed over time, as hitting frame resonances and not compensating them in filters usually increases power consumption).
However, if ESCs indeed scale the inputs with voltage, then double-scaling can effectively result in both perceived throttle decrease AND increased control noise, because Ardupilot, as configured now, would scale PID outputs with voltage.
If ESCs are indeed “clever”, disable voltage scaling in Ardupilot. (Or, alternatively, disable this feature in the ESCs. I don’t know which would be better.)
