Turbo Speed by Sound

I know nothing beats a dedicated, purpose-built sensor for any specific application. However, in tuning of of VNT/VGT turbochargers, I'm slightly surprised to not see any audio engineers here on the forums. We have all sorts of other engineers present. Anyhow, I digress. It dawned on me last night to open up a spectrogram app while I was coming home from work and look at my engine and turbo sound. Since I have an open exhaust, I have a whistle from the turbo that is readily available and I figured the FFT would pick it up; well, it did:

Click to embiggen:

Note: That's the weakest sample I had, but the only one I paused the recording to take a screen shot.

That's a shift from 2nd to 3rd gear with a peak frequency about 13.5 kHz. With 9 blades on the turbine and converting from times-per-second to RPM, this gives a wheel speed of 90K RPM. For the GTB1756 specifically, if the 'ole 2056 map is close enough with a rated speed of 190K RPM, this puts the aural frequency at 28.5 kHz. Yes, that is outside of the human hearing range, but a standard 96 kHz, 24-bit pickup would capture that as long as the mic isn't low-passed to filter out ultrasonic frequencies.

Usually weird side-application stuff like this will pull a few hits on Google, but I have nearly nothing on this specific application. For my own curiosity, I thought I'd share. Probably going to use a mic with specs stated above to capture some other samples and see what Audacity will spit out. Granted, this isn't a real time application with a turbo speed sensor, but the temporal resolution is hard to beat.

it would be interesting to see how the sound correlates with an actual turbo tachometer

Huh? I have no clue what you are posting about.

KERMA said:

it would be interesting to see how the sound correlates with an actual turbo tachometer

Click to expand...

Yeah definitely. Engineering curiosity aside, what benefit would this knowledge have on tuning the engine over current methods?

Fascinating! Yep, it would work very well and accurately provided that you know the number of turbine blades and all the stuff you said.

Science, for the win!

Evidently if you have appropriate gear you can use this same idea for all kinds of automotive noises... starting with is it a valve train noise (1/2 RPM) or a crankshaft noise (RPM), or...

Starting back in Formula1, competing teams sat track side with microphones to analyse their competitors' cars as they whizzed by. In addition to determining the engine RPM, by analysing the FFT signatures more closely, one could even figure out firing orders, vee bank angles (when there were a lot more engine design degrees of freedom in F1), exhaust header designs, etc.

DC, why not just epoxy the mic to the turbo housing? Strongest signal, I bet.
This is shades of the vibration tachometer I used to tune 2-stroke outboards.

TDIMeister said:

Starting back in Formula1, competing teams sat track side with microphones to analyse their competitors' cars as they whizzed by. In addition to determining the engine RPM, by analysing the FFT signatures more closely, one could even figure out firing orders, vee bank angles (when there were a lot more engine design degrees of freedom in F1), exhaust header designs, etc.

Click to expand...

Wow that's pretty neat, never knew that! Some of those rivalries were intense back then.

Sent from my SM-N910V using Tapatalk

That's a really elegant solution! We use simple $15 web cameras to do similar things at work, but I never thought of doing it for turbo speed. I like flee's idea of putting it right in the housing/pipe.

I wonder if there's a different number of blades on the compressor vs turbine if it confuses things at all?

There's a tremendous amount of data in those spectral graphs - once you get out of the time domain and into the frequency domain, a whole another world of interpretation and understanding opens up.

Cool stuff!

Interesting stuff. I observed some time ago that whatever frequency my turbo makes steady state cruising on the highway is not at all soothing to my poor beagle.

True. Interesting stuff. I guess I'll just listen on this stuff................

Ol'Rattler said:

True. Interesting stuff. I guess I'll just listen on this stuff................

Click to expand...

Measuring turbo shaft speed is something typically only done by turbo manufacturers or OEM's in an R&D setting. Not many enthusiasts like us will spend the time/money to buy the sensors, data acquisition, software (thousands of $$), do the machining to the turbo, etc to get a turbo shaft speed measurement. We look at compressor maps and guess, which is usually close enough.

Digital Corpus came up with a simple and very cost effective way to get a pretty accurate turbo shaft speed measurement using a generic microphone and knowing the number of blades on the turbo.

Most "digital" microphones (web cam's, cell phones, voice recorders, etc) use a 96khz/24 Bit A/D's so they can record sounds really fast (96,000 times per second) and have a huge range quiet to loud (2^24 or 16,777,216 discrete levels of "loudness").

Very similar to clipping a card to trip on the spokes of a bicycle, you can gauge how fast the bicycle is moving by how fast you hear the card flap, you can do the same thing with a microphone and some software.

If a bicycle wheel has 32 spokes and is spinning at 40 RPM for example: 40 revs per minute / 60 seconds per minute leaves you with 0.66666 revolutions per second. With 32 spokes per revolution and 0.66666 revolutions per second you end up with the card flapping at 21.33333 Hz which you could measure with a microphone and see with the software.

On a turbo charger like the one in Digital Corpus's example, he recorded a frequency of 13,500Hz (13.5kHz) and his exhaust wheel has 9 blades on it. That's it's like a bicycle with 9 spokes spinning really fast . 13,500 "flaps" per second x 60 seconds/minute = 810,000 flaps per minute. Since there are 9 blades which will create 9 flaps per revolution, we divide by 9 to get 90,000 revolutions per minute.

The screen shot that is shown in the first post (and below) shows time across the bottom, frequency up and down and the color shows the intensity of the frequency. The light blue "line" shows the turbo frequency described above. That specific frequency shows up as a different color because it's louder than the other frequencies (engine speed, road noise, etc). Blue is quieter, red is louder. By being able to record what this frequency is and knowing how many blades are on the turbo, we can get a good estimate of turbo shaft speed.

I bumped over from Blackman to Blackman-Harris windowing in hopes to isolate frequencies a little bit better without having larger samples and loosing temporal resolution.

I like the idea of affixing a mic, though you'll get a pickup of wind noise which can be close to the same bandwidth of what we're trying to measure. The cabin or a mic near a regular exhaust's tip might do the deed just fine too though playing around with it would be the only other way to try. I do fancy recording exhaust over intake because we're dealing with lower frequencies at that end, but the heat and environment is difficult to place a mic just right.

The garage was occupied by other car maintenance so I haven't had time to swap my slipping clutch. but from the other testing I've done, the amplitude pickup in the car of the whistle is working well. If the mic is close enough to a [laptop] computer you log with, then you can hear the keystroke of starting and ending the log.

A little fancy work with Audacity, a screen capture, and Engauge Digitizer, and you'll be able to adjust your tune more precisely. If you have a very fast RTD or thermocouple, you can effectively re-create a turbo's compressor map.

Another sample from driving last night. It shows about 109K RPM:
Click to embiggen:


If you look closely, I've relatively sure that what looks to be steady of the steady-state after the dip shows my injector imbalance since I think I discern seeing 4 bands in there...

I'm going to play around with the aforementioned tools and see how much I can extract. Depending on the noise floor, I don't see why this wouldn't be horribly inaccurate, thus one could nail down the speed to within a couple hundred Hz, generally speaking. More samples means refining that too.

I don't think that the 4 injection events would come through in the turbo rotational frequency, especially at the same-ish frequency. Combustion frequency will be in the 20-40Hz range.

I can't think of what would cause that banding off hand, but I'd be surprised if it was something timed with the engine (combustion, valve events, etc)

Well FUB, tell me what you think it is here. I can zoom in a pull greater temporal resolution, but I adjusted the spectrogram towards frequency bias. Here is a quick proof of concept of internal laptop mic (which is crap), spectrogram in Audacity, import a screen shot into Engauge, and then graphed data in Excel/Libreoffice.

Audacity with a 16384 window size using Blackman-Harris windowing. The gain was set at 36 dB with a range of 60 dB and a Freq. gain of 6 dB due to the mic's quality.

Click to embiggen:


Axes setup and points fitted to convert to a dataset. I exported @ 1/10th of a second resolution.

Click to embiggen:


And into a spreadsheet program to graph after some small arithmetic:

Click to embiggen:


But back to the initial spectrogram. Given that the exhaust gasses are the medium creating the tone that is observed, either the length of the manifold to the end of the downpipe creates a different tone due to the effective horn length, there are flow variations between the 9 blades broken into regular quarters of the wheel, or there is different flow speeds/amounts coming from each cylinder. One of those, or something else either of us haven't thought of, methinks.

Edit: If you want to have a look at the audio files, they are listed below in your compressible format of choice:
Speed-Test.zip (591 MiB)
Speed-Test.tar.bz2 (434 MiB)
Speed-Test.tar.lzr (379 MiB)

I wonder if a microphone on the turbo inlet would be easier/better? There might be less combustion 'noise' to deal with and depending on how/what microphone you use, it could "see/hear" the wheel pretty good.

You've already blown way past my knowledge in terms of setting filters, methods, etc. I've always struggled to really wrap my brain around some of these concepts.

Regardless of that, I think it's pretty obvious you've demonstrated that there's data behind the method and it seems to fit the theory. Some of this is drawing circles around darts on the wall, but if it's right, then it's right .

Can you plot that vs. boost level for quick correlation?


Better use of a smartphone than most young people today, taking lewd photos of themselves...

Enabled said:

Can you plot that vs. boost level for quick correlation?


Better use of a smartphone than most young people today, taking lewd photos of themselves...

Click to expand...

I can, but I wasn't logging at the time. After the clutch replacement, I was planning on a scatter plot with that.

https://www.youtube.com/watch?v=H6AgNy0nUq8

Turbo rpm sensor with 0-5V output for datalogging....

majesty78 said:

https://www.youtube.com/watch?v=H6AgNy0nUq8
Turbo rpm sensor with 0-5V output for datalogging....

Click to expand...

End to end cost out of curiosity? Including how much it'd be for a machine shop to adapt to a turbo that doesn't have a factory place to mount the eddy current sensor.

The Sensor Kit is $400-$500 depending on where you buy.

http://www.atpturbo.com/mm5/merchant.mvc?Screen=PROD&Product_Code=ATP-ACS-020&Category_Code=BCS

Machining will probably be similar cost? Not terribly complicated machining, but hard to fixture.

https://turbobygarrett.com/turbobyg...accessories/Garrett_Speed_Sensor_Drawings.pdf

DC, you are such a nerd, love it,

Fix_Until_Broke said:

The Sensor Kit is $400-$500 depending on where you buy.

http://www.atpturbo.com/mm5/merchant.mvc?Screen=PROD&Product_Code=ATP-ACS-020&Category_Code=BCS

Machining will probably be similar cost? Not terribly complicated machining, but hard to fixture.

https://turbobygarrett.com/turbobyg...accessories/Garrett_Speed_Sensor_Drawings.pdf

Click to expand...

Know how much the micro-epsilon DZ140 unit costs?

turbovan+tdi said:

DC, you are such a nerd, love it,

Click to expand...

Have been, and will always remain a nerd. My coworkers warn new hires to not ask me technical questions. I'm no engineer, but that has never stopped me

To make all of this mean something more than individual datapoints though, I need to get that long, drawn out process of calculating VE for my engine and camshaft. From there I'll be able to plot RPM lines. I think I still have some of my fast response RTDs and if I can rig one up before the intercooler, I can get compressor outlet temps. Outlet air temp vs ambient air temp + turbo wheel speed + engine VE per RPM + MAP vs barometric temp and I can partially re-create a compressor map, theoretically.

Not specifically, but I don't think I've ever seen anything else that micro epsilon sells that's under $1k.

Digital Corpus said:

From there I'll be able to plot RPM lines. I think I still have some of my fast response RTDs and if I can rig one up before the intercooler, I can get compressor outlet temps. Outlet air temp vs ambient air temp + turbo wheel speed + engine VE per RPM + MAP vs barometric temp and I can partially re-create a compressor map, theoretically.

Click to expand...

Thats what I have in my car, p1-p4 + t1-t4 + turbo rpm.

I must say the rpm behaviour is very, very interesting when you see it first time as you will very quickly experience that turbo rpm has no direct proportional relation to pressure ratio or boost.....

I use acam picoturn in my car, and in customers cars, just drill a 5mm hole. Downside is sensor is 250 euro a piece.

Easier is take a sx00v or holset vgt. They usually come with a speed sensor from factory with 1 pulse/rev. Simple lm2917 converts to 5v

Took some more time looking closer as my flywheel gets resurfaced and noticed something. Here is an image for reference:
Click to embiggen:


Looking closely there are at least 7, maybe 8 little "waves" close to the primary tone. They are all evenly spaced, independent of windowing mode used (thus not artifacts of processing) so I feel that this next expressed opinion is fairly accurate. Whenever you speak or blow into a spinning fan, you hear a reflection and once per blade. Given that there s a base tone that is stronger than the others, there is some overlap, but the other "waves" are actually the immediate tones from the other blades from the turbo. The stronger the gas pressure, EMP, the more pronounced this is, but not only can you determine the speed of the turbo charger, but you can determine how many blades are on the turbine, which is 9 in my case.

I seriously cannot wait to look at a high sample rate and high quality audio capture of my turbo @ +25 psi.

Maybe all the more reason to listen on the compressor side? Not sure if the intake pulses will pass upstream through the compressor or not?

Thinking about this a bit more...why would the frequency of the waves be different if it's the number of blades causing them? It's not like one blade is at a different frequency than the others? If it was a combination of exhaust pulses, the spacing would vary with engine RPM I'd think?

Sorry to keep raining on your parade, I wish I knew more about it to know what causes these data signatures, but all I can seem to come up with are things that don't seem to fit