I am measuring BLDC motor vibrations with ADXL345. I use 2 propellers, one perfectly balanced and other bad, which vibrates a lot. The problem is I get similar results from accelerometer with both props. I tried ODRs from 200 to 3200 and results look similar with both props (1000-5000 RPM). I get real time data from MCU which reads raw data from accelerometer, then my program draws a real time chart (amplitude vs time, no processing just raw accelerometer data), so I visually check the results while trying different things.

What can be the problem here? I want to be able to detect vibrations on the motor.

  • $\begingroup$ Where is the accelerometer, and how is it fixed to the motor? The basic fact here is that the accelerometer measures the acceleration of itself. If your propeller has a relatively flexible shaft and/or blades, and there is some play in the bearings, the vibrations might not be getting to the place where you are trying to measure them. $\endgroup$
    – alephzero
    Commented Feb 8, 2017 at 2:44
  • $\begingroup$ it is fixed with screws to the motor holder (hard connection). When I mount it on a flexible tabe I get a little bit better results. $\endgroup$ Commented Feb 8, 2017 at 7:03
  • $\begingroup$ Also, I can feel the difference between the props by hand, bad one vibrates a lot. $\endgroup$ Commented Feb 8, 2017 at 7:07
  • $\begingroup$ "When I mount it on a flexible tabe I get a little bit better results" - that makes sense. You will only measure an acceleration at a point where the structure is moving. You might get better results mounting the accelerometer as close as possible to the motor bearings. Alternatively, try using a strain gauge (or several) instead of the accelerometer. Strain gauges should be smaller, and easy to attach anywhere with a suitable adhesive (e.g. epoxy). $\endgroup$
    – alephzero
    Commented Feb 8, 2017 at 10:38
  • $\begingroup$ An ideal way to measure the prop vibration would be to use something like a scanning laser Doppler vibrometer to make non-contact measurements on the props while they are actually running - but unless you can get access to that sort of test gear, the cost would be much to high for a "one-off" experiment. $\endgroup$
    – alephzero
    Commented Feb 8, 2017 at 10:44

2 Answers 2


Typically, motor vibration analyses and monitoring are done in the frequency domain, taking the time data and processing it into the frequency domain via subsequent FFT's.

Roughly speaking, the approach would go as follows:

  • Determine the frequencies of interest. In this case, you'll probably be interested primarily in the rotational frequency and harmonics, so ~16.7Hz - ~83Hz. This is where you would anticipate finding peak amplitudes in the frequency domian.

  • Generate a baseline frequency/amplitude profile. Typically this would be done with the ideal case, for you, this is probably the balanced prop. and identify the frequencies and amplitudes that correspond to the frequencies of interest.

  • Once that's done, the effect of any change in vibration will show up as a shift in the amplitude and frequency of the corresponding peak.

  • $\begingroup$ Interesting approach. Will try it. I wanted to make something like this d39f523dxuzndv.cloudfront.net/wp-content/themes/freefly/images/… where I can visualize the vibrations and remember max numbers. $\endgroup$ Commented Feb 8, 2017 at 15:28
  • $\begingroup$ Smartphone apps also use accelerometer to detect vibrations and draw such charts. When I secure my smartphone to the motor holder I can detect the vibrations. $\endgroup$ Commented Feb 8, 2017 at 15:33
  • $\begingroup$ There's a lot of signal processing going on behind the scenes in those apps $\endgroup$
    – DLS3141
    Commented Feb 8, 2017 at 17:29
  • $\begingroup$ Can you point some key things that need to be done in processing? $\endgroup$ Commented Feb 8, 2017 at 17:33
  • $\begingroup$ @RomanSimonyan there are entire graduate courses on signal processing. In the case of the screen you showed, I'd say that there is some kind of narrow bandpass filter and some curve smoothing. it's hard to say $\endgroup$
    – DLS3141
    Commented Feb 8, 2017 at 21:05

Not 100% sure what might be the problem in this case, but there are several reasons what could go wrong. I personally had a tough time not logging the right CAN data, so you should make sure you are doing it correctly. Then it can be the actual domain frequency... There might be a problem in hardware or software you are using or the actual connection between those two.

My guess would be the issue comes within the ECU and doesnt transmission the information properly.

  • $\begingroup$ The OP just has an electrical motor and a propeller. What makes you think there is an ECU or a CAN bus? $\endgroup$ Commented Oct 26, 2021 at 16:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.