ArduCopter Crash - Help with Analysis?

We recently had a custom quad running ArduCopter 3.6.9 crash. We looked at the logs, but aren’t entirely convinced of what happened. Would anyone be able to shed light on this?

Copter is a FoxTech Hover 1 with a RadioLink Pixhawk. We are flying custom batteries (6S3P Li-Ion), with a 800 g payload.

The aircraft was flying at approximately 5 m/s westward along a west facing slope in an auto mission, leg starting approximately 5-7 m above the top of the slope. There is a small ridge running north where the drone crashed. The gully formed between the ridge where the drone crashed and the west facing slope slopes to the north. Aircraft was found on the north facing slope of the ridge.

The aircraft was found with the front right motor (No 1 motor) completely detached from the airframe. The motor bolts to a square arm with two bolts from the underside that also mount the landing leg. Neither the landing leg nor motor bolts were recovered. The distal bolt hole does not show evidence of thread damage, however, the proximal bolt hole has thread damage on the first two to three threads. The motor cables are routed through the arm, which is a CF square tube with aluminum tubing at the front and end of the arms where bolts run through. The motor cables were damaged where they entered the CF tube, with insulation cut all the way to the conductors in multiple locations.

The propeller on the front right motor has only a couple nicks on the leading edge, however one of the props has about 5 cm of trailing edge missing. These are the FoxTech folding props original to the aircraft.

The crash occurs at approximately 940 seconds from boot. I’ve linked the dataflash log here:

A quick summary of what we noticed in the log - the No 1 motor is pegged to 100% throttle near the start of the leg (927 s from boot). This is confirmed with power draw increasing at the same time from 200 W to 400 W, which is what we would expect for a motor going all out. The aircraft hits a patch of turbulence at 937 s, but recovers to what appears to be stable flight. At 940 s, the controller begins to spin down the No 2 motor, reaching idle by 940.5 s. At this point, we see the yaw of the aircraft begin to diverge from its setpoint, yawing left. At 940.55 s, there is a jump in x and z axis acceleration, with z axis acceleration reaching +0.5 G, and x axis acceleration reaching +0.5 G. We see the controller begin to spin down the No 1 motor as well at 940.6 s, and the roll and pitch begin to diverge from their setpoints. Immediately after this, power draw increases to 600 W by 940.7 s The copter rolls right beyond 30 degrees at 940.8 s. Ground impact is recorded at 942 s.

Our current suspicion is that the No 1 motor somehow detatched from the aircraft, presumably at 940.55 s. This is primarily because we did not recover the leg and motor bolts, and there is no evidence that the motor bolts sheared off while still threaded. However, we don’t have a good explanation for what caused the motor to detach, given that the copter was otherwise flying, and survived some turbulence, and then recovered for at least one second. One possibility that we don’t have the background knowledge to determine is whether or not the attitude estimation is accurate - we know that this aircraft has had vibration issues, but we assumed that they weren’t enough to affect the stability of the aircraft.

I’ve also attached a PDF of a Jupyter Notebook with our analysis so far.ArduCopter Crash Analyzer.pdf (329.6 KB)

Just having looked at the physical aspects as you have, I don’t quite agree with the summation of motor 1 detaching at 940.
Failure could have been sudden or protracted, and we see a peak in M1’s PWM demand 931, which would indicate the start of the failure, pointing to a protracted failure type.
If the peak in M1 was in fact a product of turbulence it would be expected to be seen in at leat one other motor reaction, even though it might be to a lesser degree.
But looking at the IMU acc data there are no vibration abnormalities at this 931 M1 demand for max PWM.
In fact the Acc reaction starts when the M1 motor demand drops, presuming the copter is already in free fall.
So a motor coming loose seems unlikely.
If the motor came off suddenly then why the prolonged demand for full throttle between 931 and 941 by which time it seems out of control.
There is a big drop in M2 output right from 931 when M1 is demanding more.
This would indicate a loss in yaw reaction that the controller was compensating for.

So the finger is still pointing at M1, but I think electrical more than mechanical??

Your tune looks quite busy in roll/pitch, was this a reactive copter to fly?

Just my 2c worth.

In terms of tuning, this was okay as an aircraft. I feel like I’ve definitely flown quads that were more reactive, but it wasn’t sloppy. I would characterize it as more floaty - we were trying for a long endurance platform.

It certainly could be an electrical failure. Closer inspection of the motor shows one of the wires inside the motor broken, but no way of telling when that failure occurred. Loss of a phase could explain this. But could also be after crash - the motor wasn’t attached to the aircraft, and the wire break is just after where the motor wires enter the motor case. But that still begs the question - where did those bolts and landing leg go?