Automotive Why is an FFT done after a noise signal is captured?

[k.udhay]

Presently there is a gear whine noise issue in a vehicle transmission. Our NVH team captured the noise signal using a microphone and then did an FFT of the signal and gave us a frequency plot. What exactly can I take out of it?

1. Can I assume every harmonic belongs to a noise generated by a gear having the same meshing frequency?
2. Then, what are those harmonics which don't match with any gear order?
   
3. An irrelevant question - Does excess PCD runout of a gear cause whine? How to prove if that is the major cause?

 

[Grinkle]

An FFT translates data from the time domain to the frequency domain. Seeing the data this way often gives insights into what is causing the data look like what it looks like. For instance, if I hear a hum on my stereo and an FFT of the hum shows a peak at some frequency that I happen to know matches the AC frequency of a nearby power line, I will suspect the power line to be the cause of the hum.

I don't know the answers to your questions, but my guesses -

1. If the only whine is coming from gears, maybe.
2. Noise that is not coming from gears. Maybe a shaft is vibrating, or the casing, or etc ...
3. I don't understand the question (my knowledge limitation, not your phrasing)

 

[k.udhay]

Thanks, Grinkle! The description you gave answered my first two questions, I think.

I have an extension question. I have heard that our human ear has got millions of hair that sense different frequency noise signals and that is sensed by the man as one single noise function. Now, can I say FFT is just the reversal of what happens in an ear? FFT decomposes one noise function as noise signals of multiple harmonics? Thanks.

An FFT translates data from the time domain to the frequency domain. Seeing the data this way often gives insights into what is causing the data look like what it looks like. For instance, if I hear a hum on my stereo and an FFT of the hum shows a peak at some frequency that I happen to know matches the AC frequency of a nearby power line, I will suspect the power line to be the cause of the hum.

I don't know the answers to your questions, but my guesses -

1. If the only whine is coming from gears, maybe.
2. Noise that is not coming from gears. Maybe a shaft is vibrating, or the casing, or etc ...
3. I don't understand the question (my knowledge limitation, not your phrasing)

T

 

[Grinkle]

I don't know how the brain does its signal processing but keep in mind that the signal reaching the ear is a native time-domain signal with all frequencies superimposed on each other - no combining is needed by the ear unless one considers the vibrating eardrum to be a combiner. You are correct in your thinking that an FFT does take a time domain signal that is made up of superimposed frequencies with differing amplitudes and phases and breaks the signal into buckets of frequencies, but I don't think its helpful to consider this the inverse of what an ear or a record needle does. Ears and record needles are time domain sensors that detect signals already in the time domain.

I suspect you want to develop your instinct for time domain vs frequency domain representations and that is motivating your question. Google Time Domain vs Frequency Domain or similar searches and read a few of the descriptions. They tend to be heavy on math and short on concepts but if you read enough of them you will be able to distill the concepts I think.

The usefulness of looking at signals in the frequency domain depends on how much an engineer knows about possible sources of noise. For instance, on a circuit board one may observe that the ground looks very noisy when checked with an oscilloscope probe. The noise seen on the scope in the time domain may look random to the eye and not give much clue about how to reduce the noise. Looking at the same noise in the frequency domain may still look random, or it may show a peak at a frequency that also happens to be a harmonic of one of the clocking signals on the board, which leads one to look at a bank of 10 gillion switches that are all toggling at once from this clock which leads one to see that these switches need a better connection to their power rails to quiet the ground plane. As an engineering tool, its just a different way of looking at a signal and it might provide different insights if one is practiced at thinking about frequency domain signals and what can cause them to look as they do.

 

[boneh3ad]

The brain essentially performs its own real-time Fourier transform in order to interpret sound signals. The ear canal is lined with cilia of many sizes and they et smaller as you go further inside. Larger hairs have a lower natural frequency and the smaller hairs have a higher natural frequency. Each size of hair is therefor going to vibrate according to how powerful the sound signal is in a small band around its natural frequency. Your nerves carry information about which hair sizes are vibrating the most to your brain, which interprets all the data as sound like music or a voice.

 

[S_Happens]

Presently there is a gear whine noise issue in a vehicle transmission. Our NVH team captured the noise signal using a microphone and then did an FFT of the signal and gave us a frequency plot. What exactly can I take out of it?

1. Can I assume every harmonic belongs to a noise generated by a gear having the same meshing frequency?
2. Then, what are those harmonics which don't match with any gear order?
   
3. An irrelevant question - Does excess PCD runout of a gear cause whine? How to prove if that is the major cause?

Clarification: Every spike that you see in the FFT is not necessarily a harmonic. A harmonic is a multiple of a specific frequency, which is to say that it is twice (2X or 2 times) or thrice (3X or 3 times) the frequency of the fundamental frequency (datum frequency) that it is being compared to. So, by definition the harmonics are at least related mathematically, if not due to the same phenomenon (coincidences do occur). If you knew what one actual gear mesh frequency should be, you could look for it on the FFT and then look for its harmonics at 2X, 3X, etc. Seeing only the fundamental frequency for the gear mesh tells you something different than seeing harmonics along with it. What these things mean will vary depending on the individual amplitudes. If that's not clear, then at least realize that the spikes you see on the FFT could be either related or unrelated.

1. No, you cannot assume that every frequency you see corresponds specifically to another gear mesh frequency. You can (and should) verify this by calculating the gear mesh frequencies and comparing them to what you see in the FFT. The frequencies that you see in the FFT could be caused by many things, because there are many components and independent dynamics inside that gearbox; bearings, seals, hydraulics, multiple shafts and gears, etc. Vibration analysis is a large and complicated field, that certainly won't be summarized in this response.

2. Beyond what I've already alluded to in 1. above, it is impossible to say without knowing the details of the system, whether the cause of the frequency is constant, and also the limitations of using a microphone. Vibration analysis can be done using three different measurements; position (proximity probes), velocity (velometer), and acceleration (accelerometer). In industry, an accelerometer or velometer would be used in specific orthogonal directions on the gearbox case. This would catch the vibration that is transmitted to the case, rather than potentially just the air around it. Proximity probes, which don't concern you for this, would measure a shaft position relative to the case.

3. Unfortunately my main lack of experience is in gears and gearboxes as I mainly deal with drivers and driven of industrial equipment (turbines, motors, pumps, & compressors). I could do some educated speculation, but I'll refrain for now. Basically it depends on how that runout ultimately affects the gear and its interaction with the corresponding gear. That sounds like a lame answer, but I don't know if the gears are component balanced or how everything is set up, so I don't know whether to start wondering about balance (probably not considering the size of the gearbox/gears and tight tolerances), misalignment up to root to tip interference, etc.

There is a lot of information out there, so I'm sure you could find something on PCD runout. If not, you might want to contact an industrial gear manufacturer such as Philadelphia Gear. They may or may not be willing to discuss it with you or share some information that they already have. If not, keep asking questions until they at least point you to a good reference book.

 

[k.udhay]

Thanks Grinkle. I will sure try to understand time and frequency domain approaches!

 

[k.udhay]

Hi S_Happens - Thanks for giving such a detailed answer. It means a lot!

Just for an assumption, if there are two gears and they are meshing at two different frequencies (not at a harmonic of the other one) and there is no other source of noise, will I see two fundamental frequencies and their harmonics? By this way, I will be able to find from the FFT that there are majorly two noise sources. Thanks.

 

[S_Happens]

I think the question you are trying to ask is just whether you can see different frequencies independently in the FFT, and the basic answer is yes. The whole point of the FFT is to allow you to take the time based signal and show it as the component frequencies. Within the limitations of the measuring/display equipment, any of multiple frequencies that exist due to the dynamics (or are necessary to approximate the time based signal) are shown. If there are no issues with the gearbox, you will probably just see the gear mesh frequencies without any harmonics (you will also see 1X of each gear/shaft). Harmonics of a measurable amount are typically due to other issues in the system such as preload, hard rub, etc.

This is all part of vibration analysis, how we are able to diagnose issues in rotating equipment through vibration measurement. It has led to predictive based maintenance, where we can mostly watch failures as they occur and take the equipment offline just prior to failure, or better prepare for the repair so that it costs less money and may or may not take less time.

 

[k.udhay]

Thanks, S_Happens! You have precisely amswered my question!

 

[David Macclain]

FFT is usually used to identify periodic components. You should know the theoretical model of your problem firstly. Then, according to your concern, you could use FFT to identify periodic components.

 

[Rx7man]

I'll throw in my two cents.

In an automotive gearbox (I'm going with a typical manual 5 speed for this example), you have the input shaft driving the countershaft in all gears except 4th (direct drive),.. In all the gears except 4th, the power goes through 2 sets of gears.. first the input-countershaft gear, let's say at a ratio of 30:60 teeth, and then from the countershaft to the output shaft, I'll take an example of 5th gear at 80:30 (real world examples will probably use prime or odd numbers of teeth for wear characteristics).
For an example, let's take an input shaft speed from a 4 cylinder engine at 4000 RPM, which has 8000 power pulses per minute, as a frequency,that is a frequency of 133hz.. you may see that on your graph.
Further down the line, we have the input-countershaft gearset, the input shaft is at 4000 RPM x 30 teeth = 120,000 tooth meshes/minute = 2000 hz... you'll probably see a small peak there, the worse condition those teeth are in, the higher the peak.
Next is the countershaft-outputshaft gearset... from the input ratio we know the countershaft is turning at 1/2 engine speed or 2000 RPM, multiply by the 80 teeth on the 5 gear drive gear is 160,000 tooth meshes/minute, which works out to 2666 hz,... you'll probably see another peak there. Again, with the amplitude corresponding with how bad the gears are. 2.6 khz would certainly be called a whine.

The same can be done with the differential gears. a 4.10 rear end is 10:41, figuring the driveshaft speed to be 5333 RPM, that's 53,333 tooth meshes per minute or 89 hz.. This is a low hum at this vehicle speed.

Now the wheel bearings and differential carrier bearings turn much slower.. 1300 RPM in this case, and as a frequency that's 21hz.. This would be a growl or grumble, This is also the wheel speed, so an out-of-balance wheel would show the same frequency.

By changing gears, you can select which components are under heavy load and which freewheel.. the freewheeling ones will not be producing much noise, while the loaded ones will, if there's a problem with them.

Bearings can cause noise as well, but they're a little harder to calculate frequencies for mathematically.. If you have a 1 ball with a flat spot, you can hear that and see it as a distinct frequency, but when you have many of them, and they aren't all hitting the flat spot at the same time, the frequency may be much higher and less defined... A flat spot on a race will be a function of how many times per revolution a ball passes over it (you'd need to use the planetary gear equations that I can't remember off the top of my head)Converting from time to frequency is just an alternate form of visualizing it, which is more intuitive... it's like graphing something on logarithmic axes instead of linear to better see the relations

 

[k.udhay]

Brilliant and practical explanation @Rx7 man! Thanks a lot!

 

[Rx7man]

Thank you :)

 

[xxChrisxx]

OP. Google 'LMS transmission error' For information on gears.

Also youtube 'Mobius institute' for the basics videos on using FFT for analysis.

Also what test condition that was recorded? Steady state or a run up? Also was it a spectrum or a colourmap you were given.

Also more to the point, why isn't your NVH guy analysing this.

 

[OldYat47]

In rolling element bearings, bearing ball pass, ball damage, inner race damage, outer race damage, etc., etc. Then there's foot mounting bolts, cover looseness, etc., etc. Just the FFT will not pinpoint the source of all those peaks. You have to look at the whole package.