Each rotor fires once per rev, so the engine has one torque pulse every 180 degrees of E-shaft rotation.

If the sine wave might be combustion related, can you move the sensor array (both trigger and vibe sensor) 90 degrees & compare timing of the peaks to the original configuration? If it's combustion events, then sine wave will 'move' 90* earlier or later related to the trigger point. If it's actual imbalance, the peak should follow the sensor array.

Filter frequencies are typically spec'd at the -3dB 'knee' point. So if the filter was constructed based on frequency 'x', then rolloff will start significantly sooner than the spec freq. How much sooner depends on what 'order' the filter is. 1st order (6 dB/octave) will begin to roll off *much* earlier, removing more in-band and *not* filter out as much out-of-band as a 3rd order filter (18 dB/octave), which has a much steeper slope. If you want to avoid affecting in-band levels, the cutoff must be significantly out of band. See the 'Butterworth' filter graph on Wikipedia. https://en.wikipedia.org/wiki/Low-pass_filter

Have you thought about going 'old school' with balancing? Rig one of your LED landing lights to trigger off the peak pulse of the ADXL335, put a 'witness mark' on the face of the flywheel, and the strobe will show you the rotational position of the flywheel when the ADXL335 feels the peak out-of-balance.

Or, you could let an A&P do it. Or get on with flying your time off, and then fly it up here; I've got our EAA chapter's Dynavibe in my hangar...

Charlie

On 3/18/2021 3:38 PM, Finn Lassen finn.lassen@verizon.net wrote:

LED and photo transistor in same housing. I put it inside  a 1/2" CPVC tube so I can't read the part number.

In an ideal world, a mass rotating around a center point should produce a sine curve. But what if you have more than one mass? You should be able to use vector math to find the resulting single equivalent mass.

However, looks like a combustion engine is not that simple. I'm now fairly confident that the near-sine curve I see on the scope is mainly the combustion event. Someone need to refresh my memory: how many times does combustion occur per flywheel rotation? Looks to me like one. Rotor 1, then next rotation rotor 2, then next rotation the next surface of rotor 1, etc. Is that right?

To make matters worse, you also have other sources of vibration: alternator, water pump, etc. Their vibrations (different frequencies) are added to the basic one period per flywheel sine curve. If I'm right, the combustion event won't be a simple sine curve either: a pulse resolves into multiple frequencies (spectrum).

When you look at a scope the signal is not stable: any little misfire will jar it up or down. Makes is really hard to read amplitude on scope.

In theory, a real effective low pass filter should remove all other frequencies than the one you're interested in but doesn't seem to work all that well in practice. So, back to FFT on my Arduino Nano. Basically using FFT to produce a frequency spectrum.

My Arduino program takes 128 samples with a sampling frequency of 300 Hz. Sampling starts at the trigger pulse (may or may not be a good idea). This is repeated 8 times and averages calculated. Then the FFT transforms the 128 sample values into 64 frequency "bins", the magnitude of each  are then plotted.

Using Ernest's method, I took 5 readings (no nut, then a 1/4" lock nut zip-tied to each bolt nut in turn), running engine at or just below 3,000 RPM (50Hz). (Varied RPM slightly using mixture knob to ensure got result in the 50 Hz bin only) (Basically using the FFT as a band-pass filter.)
Result was that the 4.4g weight worsened the amplitude at bolt 1, 2, 3 and 4.
No nut: 5-10 (varying between 5 and 10 on successive plots)
Bolt 1: 20-22
Bolt 2: 18-20
Bolt 3: 30
Bolt 4: 20-22

So if adding a 4.4g weight increases vibration at least by a factor 2 no matter where it's added, I decided that's good enough, removed it and called it quits. Onto prop balancing. (Except, should probably remove the PSRU and replace the oil seal now that the prop is off ...)

Now, my circuitry doesn't match theory. The LP filter op-amp should have had a 3x amplification and 3db frequency at 120 Hz. A following op-amp should have 100x amplification. Hooked up a generator to the input instead of the ADXL335 and measured input and output. Filter amp has no amplification and the filter seems to cut in way lower than 120Hz. The following op-amp only seemed to have 80 to 90x amplification. I should probably use low-noise op-amps as Charlie suggested instead of the LM358. The LM358 runs on single-supply voltage (5V) but that requires caps between filter output and amplifier input. I guess I should make a schematic and submit for critique. Got the theory from my electronic tech apprenticeship papers back in 70's.

As for measuring phase or angle on the scope, sort of can be done. But again, this is not an electric motor!

Also the FFT gives both amplitude and phase info for each frequency "bin", but I have not been able to relate that to the trigger pulse. Hmmm, now I think I get it. Supposed to have two channels being sampled simultaneously: one the trigger pulse, two the accelerometer. Not sure I want to spend more time on it at the moment. The "nut" methods appear a lot simpler (although require more runs).

Finn