Thread: Air flow calibration measurements

This is a technical post about measuring air speed/flow, which I know won't interest many woodies but it does highlight the many difficulties involved in making reliable air flow measurements.
It involves a cross comparison of 3 different air flow meters. This is just part one of a multipart post which will evolve in this thread over the next few weeks

I recently purchased a Testo hot wire anemometer (air speed meter) and was interested to see how well it agrees with my 2 other hot wire anemometers.
Lappa has posted in this forum on using a Testo to measure airflows in his DC system.

The Testo is a very handy product since it can be left operating in a remote location in a ducting system while recording of data takes place on a tablet or mobile phone communicating with the probe via Blue tooth.
It comes with a one point calibration certificate which states at 8.0 m/s it was measuring 8.2 m/s and has an absolute uncertainty of +/- 8.8%.
Knowledge of this absolute uncertainty is critical in comparing readings obtained by different sensors.

The two other meters I am comparing the Testo to are

1) Kurz meter.
This is an old analog meter with a zero to 30 m/s range and comes with a calibration certificate stating it has been compared to a NIST standard and has an uncertainty of +/-4%, but given how old it is I really doubt it's that good.
It jumps around quite a bit while measurements are made and you will see how I got around this problem below.

2) TSI meter
This is a much newer digital meter with a zero to 20 m/s range and comes with a 10 point calibration certificate traceable to NIST standards. It does have a short term integration feature which smooths the readings. The quoted uncertainty is +/- 5% but the calibration curve shows that the accuracy is > +/-2.5% for all readings above 60 FPM. Because of the quality of this calibration I decide to reference the Testo and the Kurz against the TSI meter.

Simple comparative tests have been done before by me between the TSI and the Kurz many years ago when I convinced myself they agreed within spec.

A real world test would be to use each of the meters to perform a full air flow (CFM) test in the same duct at the same air speed. Each measurement involves the systematic measurement of the air speed (m/s of FPM) at different radial points across duct and then performing calculations to get the flow (CFM). I will eventually do this but for the moment I wanted to see if I could get a comparison of just the air speed using another method.

To do this I needed to setup a test duct and locate the sensor in the same relative point in the air stream so the 3 probes measure the same air speed. This is not as easy as it sounds as the air inside a normal duct varies speed with distance from the walls of the duct and also (especially at higher speeds) preferentially weaves and wanders around in the the duct. If the sensors are placed serially along the duct they cannot be too close together as they will interfere with each other. If serial placement was used the would need to be placed meters apart and given the test duct for one sensor needs to be around 10x as long as the duct is wide, this would require a very long test duct.

The other arrangement is with all 3 sensors at the same position along a duct but distributed evenly (120º apart) at the same radius so the sensor elements are all the same distance from the wall of the duct and this is what I did. .

Probes.jpg

A complication arises in that the Testo probe has a 9.1mm diameter while the TSI and Kurz are 1/4" diameter probes.
This alone could change things significantly so I turned up some 9.1 mm diameter Al sleeves that were placed over the TSI and Kurz sensor shafts but I had to leave the ends of the sensor uncovered others I would disturb the calibration of the TSI and Kurz. This is major unknown effect #1.

The duct itself consists of two lengths of 9" (240mm) diameter PVC stormwater pipe.
The reason I used two pieces was because the sensors had to be located in the middle of the length of the duct and I needed to be able to reach in to make sure they were the same distance away from the wall of the duct. This I was able to do with a tolerance of <1%. I also had to be able to see if the sensor heads were oriented correctly. Being able to break the duct and get up close to the sensors was the only way this was easy to do.

Despite the great pains taken to get the probes at the same distance from the walls of the duct (113 mm) this is no guarantee they will experience the same air speed - see below,

Here is what the test duct looks like
It's around 4m long with the probes placed at the 2m mark.
For accuracy the test duct length should be ~10x the duct diameter. Even better would to have a length 10x the duct diameter before the sensors and another length 10X the duct diameter after the sensors.
This would require ~2400 mm before the seniors and 2400 mm after the sensors. My setup has 2000 mm before and 2000 mm after the sensors, so technically I am a little short but I am deliberately working at slow air speeds for this preliminary test so it should be OK, although it remains as a minor unknown. A good test is how the air speed varies in time which I will discuss later.

Testduct2.jpg

The piece of duct with the probes in it can be disconnected and rotated to see if there is a positional effect. This is necessary to check because the flow can differ inside the duct due to back reflections from junctions etc down the line. This involves lengthy systematic testing that I will do next.

Another view showing the pot plant reducer that attaches to the ducting for my lathe
In the middle you can see an MDF ring which is where the sensors attach to from the outside.
Testduct.jpg


Here is a close up of the MDF ring with the Testo probe (Orange and black) on top and the Kurz probe at the 4 o'clock position.
The TSI is at the 8 o'clock position on the other side of the duct.
The small screw in the middle of the Al strip straddling the probe, locks the probe in place.
sensorplacement.jpg

Now you have 3 meters jumping around like 3 frogs in a sock so how the hell do you record each reading at exactly the same time.
Ideally you plug all 3 into a computer and even the old Kurz has an analog signal out cable that was used to connect it to a computer but I have lost the calibration data for this and I did not want to go through that calibration again so I had to find another way.
I've done this a few times before and all I do is get a digital camera and take a photo of all 3 of the displays.
I just click the camera 10-15 times at the same air speed, open or close the gate valve to the lathe to vary the air speed and repeat the measurements.
The air speeds I used did not matter as long as a range of speeds was used during the test.

Then I review these on the computer and type them into a spreadsheet - it sounds long winded but its actually quite fast and it is a good chance to get up close and personal with the data rather than blindly using what comes out of the computer.
It's also a chance to reject wild outliers and start to make sense of the data

Testductprobes.jpg

I took 10 sets of readings between 1 and 9 m/s (with 2 sets at ~6 m/s)
The 10-15 data points for each set are averaged and age results for each meter compared to each other.
The "error bars" are a measure of the overall tolerance/uncertainty in the comparison for each point.
These are the simple sum of the standard statistical error for the TSI meter measurement and the standard statistical error for the other meter .
These should really be added in quadrature (square root of the sum of the squares but I was being conservative in my uncertainty assessment.

The air speed variation in time differs significantly between the 3 meters.
The Kurz reports instantaneous speed with about a 1/4 second response rate so that had the greatest variability, about 50% more than the variability of the Testo.
The TSI reported the least variability between readings, (about half that of the Testo) but that could be because the TSI does have a weak smoothing function built in.
This should not affect the final results.

The results are plotted below as % differences of the Testo and Kurz from the TSI readings.


Screen Shot 2017-04-22 at 10.37.49 PM.png

What does it all mean?
A simple conclusion is, if the error bars for a point on the graph cross the zero vertical axis the meter agrees within spec with the TSI meter for that air speed.
For the Testo it appears to consistently read on the low side compared to the TSI meter, with all bar one of the readings below that of the TSI.
However, 6 out of 10 of the Testo readings (1, 2.7, 3, 5 and two 6 m/s) are within uncertainty which shows they are not far from agreement.
The average deviation for the Testo from the TDI is -8.5% (the "minus" means its reading too low) and given the absolute uncertainty of the Testo is +/- 8.8%, from that point of view it could be said the two meters agree within spec.

The Kurz has an average deviation of -6.5% which while slightly closer than the Testo to the TSI is outside the Kurz quoted spec of 4%, but given how old this meter is, this does not surprise me.

Overall, given the gotchas, I am amazed these results got anywhere near agreeing to this degree, and there is much more testing to do especially at higher speeds.
The kind of test described in this post cannot be done at higher air speed because to achieve this will require the use of 6" duct (to get the air speeds up) and the smaller cut will put the probes closer together which is likely to create a mess of the air flow.
The only way to do sensor comparisons at these speeds is to perform full airflow measurements.

One possible outcome of this experiment is that if the Testo is indeed reading on the low side it could be that Lappa's system is better than he thinks.
And who's to say my Testo is the same as Lappa's in the first place?
Conversely the TSI could also be reading too high and I could have a slightly worse system than I think I do.

Now let me reiterate that the above conclusions could still be in doubt because I have not really assessed the chances of non-radial variation of air speed within the duct. This could easily result in the deviations observed.
To test this I will be rotating the section of duct that holds the sensors into a different positions with different sensors at the top each time and repeating these measurements.
Even this might not show up everything but it will be a start.
The real test will be a full volume measurement for which I will have to make up shorter Al sleeves for the TSI and Kurz meters.

What it does show is how different meters can give different results and how messy in general air flow measurements are.
If this doesn't put you off air flow measurements I think by the end of this thread you may convinced.