Accurate 3D Printed Wind Tunnel With Live Force Data

Hello, I'm Alistair, a 1st (going into 2nd) year aerospace engineering student at the University of Surrey.

Project Goals (these are worked out in step 1).

This project aims to:

1. Achieve a minimum of 20m/s in the testing zone that is variable with a dial
2. Have live drag and lift/downforce data-plotting through Arduino IDE into MATLAB
3. Be portable, I should be able to break this wind tunnel down into individual components so it can travel around with me
4. Custom object fitting, I should be able to test ANYTHING that can fit in the wind tunnel.
5. Build the majority out of a 3D printer.

Why did I build a wind tunnel?

There are hundreds of engineering students at my university, a good portion of them are aerospace engineering students like myself, where the study of aerodynamics plays a big role in their degree. The problem is that while the University does have wind tunnels, it can be difficult to get access to them due to the sheer volume of students.

So I've designed, tested, manufactured and assembled my wind tunnel, which I'm allowing all engineering students at my university to use for free, accessible testing without having to go through the long, tedious process through the University.

This project aims to aid students in the upcoming module where we are tasked to design, test, make and evaluate our own remote-controlled RC aeroplane.

If you want to build a wind tunnel to these parameters, following my journey might help :). If you're looking for different objectives, you can use the information in this instructable to give you an understanding of what you might need.

FINAL EXTRA NOTES BEFORE STARTING

From placing 3rd in the world in the STEM Racing (Formerly known as F1 in Schools) competition, I knew the biggest challenge would be incorporating electronics since I've never done anything with it before, so I've used this project as motivation to learn. It worked with the last project with CAD, so why wouldn't it work here?

Funding this was also another difficult part; while it was not directly related to engineering, it was something else I had to learn for this project to be successful. If you see some 'RS' looking Instagram-style posts, that would be my social media post where I publicly announced my wind tunnel to gain access to the RS Student prize funds. If you are struggling to find funds for this project, try and market it to other companies' contests who could be potentially willing to invest.

ACCURACY NOTE

I am measuring the accuracy of the wind tunnel, so I go through my steps, but I might not have completed it yet, so if there is no mention of the percentage of accuracy at the end of the wind tunnel, it means I'm still testing/working it out! When that's done, I'll update this instructable. I have worked out the drag (Fx) measurement accuracy and precision

This is my first big solo engineering project, so I hope you enjoy reading through :).

Software:

CAD - Autodesk Fusion 360

FEA - Autodesk Fusion 360

CFD - Autodesk CFD 2026

Force balance - Arduino IDE & Matlab

3D printing utilities - Bambu Studio

Electronic Schematics - KiCAD

Electronics:

Soldering kit

Solder

Flux

HX711 Chips x2

TAL220 load cells x2

Arduino Uno R3

mini breadboard

Jumper wires

Single plug head

power cable (earth, live and neutral wires)

Other physical components

M4, M5 and M6 nuts and bolts

1300m^3/h 230V AC centrifugal fan (very expensive, so this was covered by RS)

Acrylic

Metal Meshing

Printer (I used a Bambulab X1C)

Fillament (very expensive, so this was covered by RS)

See the finances breakdown above :). It might be slightly inaccurate as I ended up buying a few more things down the line, but it's pretty accurate. I didn't spend more than £250 out of pocket

Base dimensions

I started off by identifying the design objectives and the motive for creating the wind tunnel. I decided a good size is one of the casings for my STEM Racing World finals car, so I based my testing dimensions (110mm x 110mm frontal area) on that and then worked out the other venturi dimensions with the wanted testing speed (~20m/s). After working out the volumetric flow rate and the headloss from the wind tunnel, a fun could be searched for.

Manufacturing choice

I also knew that to get the funding for this, the wind tunnel would have to look somewhat nice to be able to be marketable for funding, so I went with 3D printing as the manufacturing choice, which adds more assembly complexity, but looks better in the overall finish.

Without funding from a sponsor, I don't recommend 3D printing a wind tunnel, it's very long to manufacture (depending on the number of 3D printers you have, I only have 1), and it's very expensive in terms of filament.

Quality testing

When the wind tunnel is done, the round accuracy of the wind tunnel should be calculated. It will be the mean percentage of the 'real' value of the object. The 2 objects are a flat plate and an NACA0012 aerofoil at different angles of attack. This is testing Fx and Fy; once that's done, I'll test different speeds to see how that changes the accuracy of the results. My initial hypothesis is that the accuracy decreases when speed decreases.

At the end, I should have a graph of the accuracy of the wind tunnel at different wind speeds (Reynolds number is more professional, but I find it easier to visualise wind speed).

Initial manufacturing Tests

With the design objectives, base dimensions and manufacturing method being set, I then got to work on 3D modelling, incorporating variable parameters inside my constrained sketches, I could iteratively modify the design to my liking in an instant. At this point, I also started testing different joint mechanisms, what sort of sealant? expoy or suprtglue? I concluded that silicone sealant on its own, due to the vibration resistance and the clean cured finish. Along with this, a few personalised 3D printed nuts and bolts would be enough to hold it together with some 'legs' to support the weight.

To hold the separate pieces together during silicone sealant curing, I made my own PETG-CF nuts and bolts. I iterated the design so they could be printed on their side for a strong structural integrity, as the force would not act against the printed layers.

Initial CAD

Since the maximum area that can be printed is 256 x 256 (actually around ~210 x 210mm for boundary tolerances), the CAD model had to be split up into many pieces. By using the mass conservation formula, I am able to work out my inlet and diffuser dimensions. If I needed to change them over time (spoilers, I did), I could use the variable parameters that I embedded into the sketches for instant iteration. I opted for a maximum length of each part being 200mm. This is done by lofting the sketch (solid modelling) with 3D sketches and then splitting the bodies afterwards.

CFD SETUP

Before I get started explaining results, etc, ill quickly go through my CFD Set-up

Test material: ABS moulded

Fluid material: Air

incompressible

Meshing: automatic (I want to do some mesh refinement testing at some point, but I'm overloaded with other jobs)

Turbulence model: K-epsilon (I wasn't too sure what to put for this, so I left it on the default)

Inlet angle testing

I modelled an inlet surface with varying heights, ranging from ~340mm to 0mm. The idea is that I can use the 'planes' feature in Autodesk CFD to visually see all of the velocity vectors in that plane. The length of the inlet was set to 593mm, as that was the longest my printer could handle with 3 layers of prints (see the image with the inlet with different coloured sections); any more and the manufacturing would have taken ages (which it already did). So this was testing to see how large I could make the inlet, as I already knew from research that the larger the inlet, the better.

When the CFD solving was finished (see video attached), I could see that at too steep an angle, the flow wouldn't travel parallel to the inlet body. I picked a 'plane' that had good flow vectors and used the scale system in Autodesk CFD to then work out the height needed (around 200mm, so 18-degree average incline). I then iterated the design in Fusion and made it slightly shallower (now 15-degree average incline) to ensure that the flow stayed attached to the surface.

Finishing base CAD before manufacturing.

I finalised the inlet CAD so I could start printing. This was the 'FINAL' CAD as I didn't have the RS fan yet, so I didn't want to make a diffuser that I wasn't 100% sure was going to fit, as that would be a very expensive mistake.

3D Printing Entry Zone

I then started printing the entry zone, which consisted of a lot of 3D-printed pieces. To avoid confusion between the pieces (unfortunately, they weren't exactly the same due to millimetre differences in the bolt placements), I numbered the first 2 rows and drew a coloured line to ensure that the assembly was correct during the sealing process. After all parts were printed, I joined them together with the silicone sealant and left them to cure (a very stressful process, combining the big pieces). This was a very long process, as I waited 4 to 7 hours every time I applied the sealant.

To keep the wind tunnel portable, both the inlet and the diffuser would be split into 2; this also makes it easier to assemble.

Now we get to the parts where I learnt the most. The part that I had pretty much been procrastinating doing because I initially didn't know how to do it, but I was too far in the project to turn back... electronics and coding

What I did (lots of mistakes, don't recommend)

While I was waiting for the printing and curing of the inlet, I got started on the testing zone electronics. I have never done anything to do with electronics or Arduino before. So, incorporating 2 load cells with real-time data plotting was going to be the biggest challenge yet. I realised at this point that the HX711 amplifier schematics on KiCAD did not have the same 'pin layout' as the real ones I had in my hands. So I spent a few days with a multi-meter testing and working out 'what bit was what'. I learnt how to solder and then edited the code. I mounted the electronics onto the testing zone of the wind tunnel, and voila, it works. That was a long week.

The photos of the physical force balance wiring are wrong!! i addd them because I thought they were funny looking back at them. the correct wiring is shown by the 2 HX711 chips.

What you should do (not as many mistakes, probably, and I do recommend)

I didn't see a schematic online. However, when typing this up right now, I just timed myself finding a correct schematic, and it took me 8.93 seconds... use the schematic. After testing everything, making sure it works (don't worry about a bit of noise in the readings, this will decrease when you solder), you can solder the wires into the chips.

Extra info

If you want a better explanation, you can go here to this setup:

https://sung.seas.upenn.edu/publications/wind-tunnel-force-balance/

Acrylic

I also had an acrylic window for pictures. I got it by printing a bracket, which I could superglue the cut acrylic sheet onto. I used a small hacksaw to cut the acrylic without it cracking,g and then I sanded it down to fit inside the bracket.

Arduino IDE

Using the link below, I was able to find a public IDE file that gets the Arduino to read the 2 load cells. I used that for my IDE code:

https://sung.seas.upenn.edu/publications/wind-tunnel-force-balance/

MATLAB Code

Using the aid of ChatGPT, I created an m script that splits the reading into 2 main phases: Acceleration phase and segment phase.

Since the load cells read the change in voltage, the fan would have to turn on after the load cells started sending data, so that's why there is an acceleration phase. For testing, it was adjusted to 10 seconds, but when I got the real fan, it changed to 25.

I also made it so that 2 separate graphs instead of 1, as that makes it easier to read in my opinion.

At the end of the 3 segments (or however many the person wants to add), MATLAB automatically saves all the data, both graphical and numerical, into a folder for easy transfer to the other students.

Load Cell Drift

To be completely honest, I looked into drift correction and tried to link up all of the factors (temperature, amplifier offset, etc) into 1 big equation to differentiate and try to find an 'optimal' tarring sequence, however, I don't know if this is an accurate method of finding the 'answer', even with the help of the internet. The graph attached shows my attempt to try and recreate the drift offset over time for a TAL220 and an HX711 amplifier. I decided on 25 seconds because it looked like the curve had flattened out and it was the rough intersection of the separate load cell drift equations, but I don't think that actually means anything.

Load cell drift is only a noticeable problem at low wind speeds. When the wind speed increases, it has a negligible effect. From my testing, anyway.

Thoughts

When the electronics and coding were done, I was very happy. The hardest part of the project was over. But not the scariest, that was still to come.

Writing this up after finishing everything, I relied on ChatGPT very much in this portion of the project, it's the area of engineering that I am currently quite weak in, so I should try to do more projects like this to challenge myself.

At this point, I pretty much had a working wind tunnel! The only issue is that it could only get windspeeds up to 3m/s with a big, low static pressure pushing fan at the base of the inlet (not ideal). But while sorting out funding from RS and going over the fan electronics plan, I decided to do a bit of accuracy testing without airflow and at 3m/s.

Force Balance Calibration

Without airflow, I calibrated the scales with a known weight of 50g (0.4905N), as the load cells measure the slight change in voltages through the deformation of the material. I have to convert these changes in millivolts to Newtons. To do this, I used a 'scaling factor' which, over the course of many tests on both Fx and Fy force balances, got both load cells to measure newtons to 0.5% accuracy.

I also found that the load cells interfere with each other slightly, around 0.1N

Airflow stabilisers functionality

To check if the airflow after the stabilisers was working, a dental floss was taped to both the testing stand and the back of the first stabiliser. Both showed that the airflow was laminar after the stabilisers and turbulent without them. The turbulence probably came from vortices off the pushing fan, and not the speed of the airflow, since the Reynolds number is 22,000, which is nowhere near the critical Reynolds number (500,000). The videos attached show hands without airflow stabilisers

Initial Shapes testing

I decided to test 2 shapes at 3m/s. The shapes test included 1 flat plate that pretty much blocked up the entire frontal area and a curved chevron that had a much smaller frontal area the the flat plate. I ended up being incorrect by 280% on the flat plate and so incredibly incorrect on the chevron that I stopped testing after the 3rd round out of 10. Load cells were reading that the chevron was experiencing lift and being pulled towards the direction of airflow?!

I believe it was incredibly inaccurate on both models because:

1. The Blockage ratio to too large; if I want to fix that issue, I need to redesign the wind tunnel entirely. I can use a blockage correction formula to reduce the effects, but it's still not great to rely on.

2. Low windspeed, if the windspeed is too low, the change in voltage is so small for the load cells and amplifiers to record accurately, the data gets lost in the 'noise' in the system.

Environment Testing

I also tested how the testing zone environment affected the data. I used the flat plate again, but this time removed the top cover. Doing this increases the standard deviation by about 400%, which proves that the wind tunnel was measuring precise results. If you compare the 2 graphs, you can see the difference.

Considering each second is about 100 data samples, I got organised VERY quickly to not lose any of this data.

Why a Centrifugal fan

After emailing a lot of companies (there are so many fans out there), I eventually settled on this high-power, efficient centrifugal fan! While most people tend to go for axial, I went for a centrifugal one as there isn't enough room in my year 2 student accommodation to have an axial fan suck the air backwards, as this would suck it directly into a wall, which would significantly reduce the flow rate. However, having the airflow move perpendicular around the back of the fan is much better for space efficiency.

Financing - Winning the Student Grant

Centrifugal fans are more expensive when compared to axial, so at this point, I was re-considering my options while looking at a £500 fan. I came across a student fund from RS for people to pitch their projects with the hopes of landing some funding. So I filled out and sent a form, and I ended up winning! which was surprising considering how bad my writing was after reading it back! I tend to type how I talk, which isn't great for essay writing.

Fan speed controller issue

Because of the Student grant I won from RS, they shipped this to me for free! At this point, I had realised that I had the incorrect fan speed controller; I needed a PWM (Pulse Width Modulator) controller for full variability. Luckily, the centrifugal fan that I bought came from a company that made speed controllers just for their fans, so it all worked out quite nicely in the end.

Bracket

The centrifugal fan only had one connection, and that was the back. The entire black component spins, which caught me off guard as I at first believed that it was just a casing.

Over 3-4 days, I designed, tested, manufactured and assembled my own custom bracket for it, which was really fun. Because of the danger that the fan poses if it were to escape the bracket, I ended up using metal bolts and PETG-CF for the main stress points. I also used Fusion 360's FEA to help with the bracket design. It wasn't too complicated, so I decided to do the simulation in Fusion and not transfer it over to ANSYS.

To ensure that I could model around the fan accurately, I 3D scanned it and referred to the technical drawings provided.

The main design issue was that on the fan's side, the PLA plastic around the connection point would buckle and bend, which created some friction between the metal inlet and the rotating fan, which can NOT happen at high RPMS. To fix this, I extruded the connection point in line with the back of the fan bracket and put bolts aiming outward so that the applied stress was even throughout the design. I also added 5mm more tolerance to the inlet ring, but that wasn't as complicated.

Bracket - FEA - Results

More than double the expected force was applied to the connection screws (50N)

You can see the FEA diagrams attached. MK1 had a safety factor of 5.25, which was okay, but the displacement and stress were all located at the back face of the fan holding zone, which meant that it would bend and intersect with the metal inlet ring. It did everything as simulated in the physical setup.

MK3 moved the stress over onto the extruded PETG-CF cylinder, which stopped the back fan board from bending. The safety factor was 15, which was perfect. Everything simulated in Fusion 360 was accurate to the physical setup

Setting up the electronics

With the bracket made, the fan electronics can be set up and make the fan operational! Getting the singular plug and the power cable, I connected the cables to their respective zones (making sure to solder beforehand for a nice connection). Then I connected the fan's main power source and the PWM speed controller. Then, when everything was double-checked, I plugged it into the wall and turned it on for the first time!

Initial test

Everything worked, which was a good sign; however, the recorded windspeed at the inlet was 14.4m/s, which was 5m/s lower than the theoretically predicted calculated flowrate value (~19m/s). However, after further calculation,n 14.14m/s still theoretically reaches 20m/s at the testing tunnel, but the margin for assembly error is reduced. The entire fan was able to be fully controlled from ~1m/s to 14.4m/s, so that was another tick off the checklist. I was also very happy that nothing exploded. The attached video is the first time I turned on the fan.

Mesh Vs stabilisers flow control

I added a thin metal mesh sheet in front of the inlet as a 'last defence' in case a foreign object made its way past the second airflow stabiliser. However, after testing the speed with and without the mesh grid, I was very surprised to see the results...

The inlet with no cover was reading around 13-14m/s, and the same result was obtained with my printed airflow stabilisation filter. But with the thin mesh grid, it was barely hitting 10m/s. I was genuinely confused at this. The frontal surface area of the meshing grid was so much smaller than the airflow stabiliser, surely it would not have that big of an impact! let alone be worse than a 3D printed version.

So while the final few pieces of the diffuser were printing, I iterated the attachment to include my own 3D printed airflow stabilisation filter to have the best of both worlds. This solved the low airflow issue.

After some research, the reason why it was occurring is that the airflow doesn't 'see' the mesh grid as a grid, just a circle (put onto a 2D plane). whereas the airflow sees the stabilisation filter as a long rectangle. It doesn't seem intuitive in my opinion, but I can't really argue against data. the same

Rough assembly test

Although the whole assembly wasn't together, I did do a quick test with half of the diffuser components. My anemometer was reading around 19 - 20m/s, which meant that the design was working as intended. I expect it to increase slightly when parts are properly sealed with a smooth surface. I also tested the meshing grid on top of the rough diffuser assembly, and the results were pretty much in line with the original test.

Other Things to note:

Vibrations: While I couldn't sense any vibrations, I was and am still very cautious, as vibrations can tear things apart. To avoid this, all screws have had 2 nuts attached to them to ensure nothing catastrophic can happen.

Angular momentum: the 3000RPM fan was moving so much air that the bracket on the floor was rotating slightly at the high RPMs, because of this, the 'leg' that the fan will be mounted on needs to be very stable laterally. This was done by increasing the width of the stand and allowing the stand to be screwed on (with the steel screws) to the fan bracket

Essentially the same steps with the inlet assembly and manufacture, but the silicone sealant was along the entire edge of the wind tunnel, not just a rough boundary. The biggest concern with this is that it should be as airtight as possible, so the silicone sealant had to be very thorough.

After the diffuser was manufactured and assembled, the legs were printed, and everything was put together!

For all the engineering students to use it, they must know how their object attaches to the wind tunnel. Since some of the students are relatively new to Fusion 360, I want to make it as easy as possible for them to iterate on the testing block to fit their design. So I sent them 2 files.

Basic iterable setup: a simple Fusion file with 1 variable parameter that only changes the height of the aerofoil component. It also has a pre-built functioning locking mechanism which works with the parameters.

Advanced iterable setup: For other students who want to play around with different sweeps and dimensions with a bit more freedom, I modelled this more advanced setup with even more parameters to allow for extra freedom. I also left the locking mechanism up to them if they want to work one out for themselves; if not, they can use the standard mechanism in the basic iterable setup.

The issue

Initially, the wind tunnel was only reaching speeds of 18.8 m/s at maximum, with no object being tested, which was really annoying considering it was designed to reach 20 m/s. The only part of the wind tunnel that would have an effect was the attachment hole for the force balance. When I covered the hole and tested again, it did hit 20m/s (yippee).

The solution

The biggest potential issue with plugging the hole would be that it would interfere with the force balance and skew the results. Thankfully, the force balance was only measuring 2 axes, so the width of the hole could be reduced from 20mm to 6mm. I chose to print with TPU for additional flexibility and tightness so that it wouldn't fall out between tests.

CFD

I used Autodesk CFD to see what the difference would be by adding the 'plug'. The first 4 CFD images show the testing section with the original big hole, and the second batch of CFD pictures shows the small hole CFD Simulation.

Both simulations show very similar results, but as the opening is wider for the larger hole, more airflow can exit the testing section. You can see this with the Iso surface (both iso surface values were set to 595.5 to keep it even), on the larger hole test, it is MUCH larger than the smaller hole test.

I also added some screenshots of traces and velocity planes, but they tell us what we already know.

Did it work?

Yes, the maximum tested speed jumped up to 20.3m/s, so it was a big success!

Was it necessary to use CFD for this? No.

Was it cool to include? yes

SHAPES & SETUP

I decided to test 3 simple shapes with known drag coefficients:

A Flat plate with a drag coefficient of 1.28

A Sphere with a drag coefficient of 0.47

A Cube with a drag coefficient of 1.05

Because blockage is a big deciding factor that I have not yet tested, I decided that if the blockage rate is 20% or higher, the blockage equation would be used.

I measure the speed the wind tunnel reaches with the object inside, and that is used for calculations (with objects in the wind tunnel, it doesn't always reach 20m/s). I then do 10 tests on each shape and get the average of all 3 segments if there are no visual glitches or errors inside the segments. If there is a very visible error, that segment is discarded. in the tests, there were no errors found.

Results

I spent a while on this because I couldn't quite believe the results. I triple checked the tests and the maths in the Excel spreadsheet, but I calculated the average Fx accuracy to be around 2% off the correct value?!

This result does sound freakishly close to the real result, so I am still a little uncertain, even though I have LOADS of data to back up my calculations.

I'm also finding that the precision of the results is even more impressive than the accuracy at around 1%!

This seems too good to be true, especially for a desktop wind tunnel with this amount of blockage

Fy (lift & downforce) is going to be a bit harder to test; there are fewer 'well-known' shapes with set lift coefficients, so I'll need to get started on that. I also need to test the accuracy of the results at different speeds and blockages. I bet the accuracy decreases as the speed decreases and at higher blockage rates.

Hypothesis from 3m/s pushing fan:

While I had tested the force balance without airflow and with 3m/s (using a sketchy, not air-tight pushing fan), I concluded that it was going to be around 5% out from the 'correct' value. My hypothesis was that it would increase in accuracy with higher wind speeds, as it would be easier for the load cells to read the forces and thus the fluctuations would have a smaller overall impact.

From my accuracy testing, my hypothesis is so far holding up to be correct.

Design objectives summary

Top windspeed is 20m/s or above AND variable?: YES, Max speed recorded was 20.5m/s

Live vertical and horizontal force data plotting?: YES

Is it portable?: YES (it breaks down to fit in a little Kia Picanto)

Changeable object fitting?: YES

Made entirely from a 3D printer: MOSTLY, apart from electronics, safety and visuals (acrylic sheet) 3D printing was the main manufacturing method.

Extra objectives I wanted to try and achieve, but only stated them mid-development

Is the airflow laminar at top speed? Reasonably, see the video attached

Final thoughts:

Overall, I would consider this project a MAJOR success, as 5/5 of the design objectives were hit. After further testing, one of the biggest issues with this design is that the frontal testing area is too small, so the majority of objects that are tested end up having a really high blockage ratio. While this can be sorted by using an equation, it still isn't a great thing to rely on heavily. for the time being, i am saying that for blockages upward of 20% should use the equation to maintain the accuracy of the wind tunnel.

Engineering students in my year have the 3D models for the testing connector. I think a good portion are looking forward to using it!

I am also very happy I managed ot include FEA, and CFD in the development for this design!

DIEMNSIONS

Wind tunnel length: 1.96 Meters

Wind tunnel height and width: 0.53 Meters (inlet)

==================================================

WIND TUNNEL STATS (I can't believe these results):

Fx: 2% off the 'correct value'

Precision (as a standard deviation): 1%

==================================================

Now, 3D printing a wind tunnel is NOT the best manufacturing method; it is quite expensive.e However, the point of using a 3D printer in this project was to make me feel more comfortable about building projects that take more than just 1 print. If I couldn't use my printer, I would use plywood to bend into shape using 3D printed Jigs to help maintain some level of accuracy.

Now that the wind tunnel is complete, I aim to help the ~200 engineering students in my year and the local STEM Racing team with free, accessible, and easy testing.

If I do end up being a winner for the Autodesk First-time Author award, I will use the gift voucher towards my next upcoming projects so I can continue pushing myself and learning these cool engineering skills!

Also, I can't believe I did this. I just designed, tested, manufactured and assembled my very own wind tunnel. I randomly grin throughout the day now. I can't lie, this project has fixed my sleep schedule. Setting alarms and waking up early to start new prints has been overall beneficial for my sleep!

Thank you for looking through this project! If you have any questions, feel free to reach out!