3d scanned miata parts- interest check
 

I have been slowly building a library of 3d scans of miata parts for my own projects unrelated to any real car fab. However, it occurred to me that having a library of high quality and accurate 3d scans is something that all miata fabricators could find useful. So far I have been working on this slowly, and on my own schedule. Is there significant enough interest for people to back a patreon for parts scans?

Here is an example:
https://cimg8.ibsrv.net/gimg/www.mia...cae912576f.jpg


The ultimate goal would be having a fairly complete library of parts freely available for download. Imagine you want to design a part, you could iterate on the design and check fit without having to actually build anything. I think this would be hugely useful for engine swaps, designing aero parts, exhaust routing, turbo manifolds...

Your question is silly.

What do you use to scan these?

That was 141 photos with a nikon d800E and a sigma 50mm art processed in reality capture.

I am not familiar with that process. I will research. Thanks.

But yeah, 3D files of actual parts is a great thing.

I think tim needs to bring a miata into work. And plop the thing on the faro arm table :D

I have. I scanned my hacked up shifter hole and some random interior bits with it. But it's manual point cloud data, so like maybe 100 points, whatever I click. We've been eyeing a laser scanner for years but have yet to be impressed with the software used to get 3D files from the scans. Seems simple, should be simple and common, but year after year, multiple sales guys, demo guys, phone conferences, live demos...we've yet to get a solidworks file of something simple that same day let alone in 30-60 minutes which is how long it takes us to do something complicated (i.e. cylinder heads) manually.

Actually this year we also looked at a lesser expensive product that use visible light cameras instead of a laser and it produced better results, quicker.

Anyway...TL;DR this looks pretty sweet.

The first image I posted was 3.1 million triangles. Probably overkill for people to mock up fit off of. Here here is the same model chopped down to 14 thousand triangles which is much more manageable for most people's computers.
https://cimg0.ibsrv.net/gimg/www.mia...4c91c82a8c.png

Careful; if you get a system that works well it reduces your job security. ;-)

This is relevant to my interests.

I recently discovered Onshape, it's a free CAD website. It seems pretty slick to me, although I'm an utter noob when it comes to CAD. I'm sure real Solidworks users would find quite a few little features it doesn't have, but it's free and there are quite a few tutorial and videos to show how it works.

The problem I've been left with is how to find good models of parts to design against, so having a library of good accurate 3d models would be awesome.

I remember reading a paper about a research conducted by a few MIT students (grad?) that uses a cheap 3d scanner like the Kinect and dslr with a polarized filter to create really detailed scans.

I would love a library of 3d scanned parts!

I'm attaching a sample for people to take a look at. The trickiest part is establishing accurate scale. I think adding a large machinist ruler to the scans is going to be the most reliable method.

What kind of accuracy is this system capable of? I think an accurate scan of the trans bellhousing would go a long way towards making adapter plates to use a stronger trans.

Accuracy is hard to pin down to a specific number for two reasons
*accuracy of the surface varies depending on how clean and well resolved an area of the scan is.
*Scale and accuracy aren't quite the same thing, A scan can be super clean and accurate but be scaled wrong. For something as large as a subframe, a .5% scale error results in bolt holes being mis placed by several mm.

I ordered a 600mm machinist ruler that I'm going to include in my next scan. That should give me really good data to pull scale off of. I'm going to focus on making sure areas that have bolt holes are especially well resolved. Overall, I think I can get placement of bolt holes on parts accurate to about a half mm on large objects.

Any 3D scans are usefull to the community. We really need a central repository to dump them.

I use Agi photoscan for lots of parts and find its accurate enough for most development work. The key to using photos is to make sure objects have no specular highlights or reflections. Any reflections and you'll get points at the wrong depth. The photos also need to have as little noise as possible and be in focus.

I actually find using a mobile phone works fine in most cases with the scans are accurate to less than 1mm error once scaled. Plus its fast and easy.

Here is a scan of an old 1.6 dohc lotus head that got mashed up when it dropped a valve. I photographed a clean chamber and built a 3D model from it. This will be used to CNC the welded up damaged chamber.

The entire process took less than a few hours including reskinning the 3D model.

https://cimg3.ibsrv.net/gimg/www.mia...318b2cd946.jpg


https://cimg2.ibsrv.net/gimg/www.mia...54a87459a8.jpg

With photoscanning it's easy to get a bend or a warp in the 3D scan if you take a series of photos along the length. Ideally it's best to take photos that cover the entire object rather than lots of overlapping photos. You at least need some photos that cover the entire object that act as a baseline for the closer photos.

I've also used a white chalking spray to cover shiny objects in a matt finish, then splatter paint or use a marker pen to create identifiable marks over the object.

For scaling, a metal ruler isn't the best as it's too reflective. You are better of printing a scale on an A4 or A3 sheet that can be seen from the main photos. My idea is to make up a scale card that can be placed next to or under smaller objects. The card will have a 'L' shaped scale for easy alignment as well as a couple of other glyphs to double check the square. If I make one and test it I'll post it up, but the main thing is that the scale card needs to be inflexible and large enough to be accurate.

Interesting, I recently had a BP4W head shipped to my uni's workshop and have pretty much unlimited time with the Faro arm. And flowbench for that matter :)

Front upright and LBJ. I'm reprocessing with scale information and then I will be able to make some conclusive statements about dimensional accuracy.
https://cimg5.ibsrv.net/gimg/www.mia...3bb67455b9.jpg

He bought a scale with a matte finish made specifically for this use, pretty cheap on amazon.

I remember trying the Kinect thing in college, but if memory serves me correct, the dimensions were so inaccurate that they could not be used. We ended up using a local company's Faro arm instead.

Yeah, it has a satin finish which isn't quite ideal. I ran a strip of painters tape down it so the scan could have something easy to grab onto. The tick marks are placed right up to the edge of the painters tape so I can make accurate distances.

I think this is a very good result for accuracy. I'm using cheap harbor freight calipers and my method for measuring on the model is not super precise. If I had better calipers I could take 4-5 measurements and try and find a trend for what % difference between model and real, then apply that scale factor. Either way, this is damn close:
https://cimg2.ibsrv.net/gimg/www.mia...7240a01a12.png

The differences are 0.10015% and 0.065822%. Assuming a .1% error over a 600mm span the error would be .6mm.

Error is completely application depended, making a set of brake ducts with this scan and the accuracy is overkill. Try to use this scan to make an adapter plate to locate a transmission bolt pattern and .060" error is enough you'll never get it bolted together let alone concentric enough to the output shaft to run for any length of time.

Error is completely application depended, making a set of brake ducts with this scan and the accuracy is overkill. Try to use this scan to make an adapter plate to locate a transmission bolt pattern and .0024" error is enough possibly have issues with the output shaft being too far out of center.

Correct. The usefulness of the scans is application dependent. The transmission bell housing bolts are probably closer to 400mm apart than 600 so the error in the case would be lower. Still too large. For really critical dimensions that scans can be adjusted to fit real world measurements.

What I'm imagining is not just individual parts scans but most of the car in an assembly that can be loaded piece by piece. This would allow people to design for part fit on the specific mounting locations and also for interference for within the whole assembly/car. Also things like body modifications and aero templates generally rely on tracing cardboard templates and iterative fitting.

Here is the paper I read, the results looks pretty good. Sadly I don't have access to MatLab to test out their code

Polarized 3D - Achuta Kadambi - MIT Media Lab

I've used KScan3d before to scan some stuff and it came out awful.

Okay, another lesson learned. Process the mesh at the highest resolution possible then chop the result down. The cleaner mesh here is actually lower triangle count after reduction. The mesh I posted before (the lower quality one in this gif) took about 45 minutes to process. The new one had to run overnight:

https://cimg8.ibsrv.net/gimg/www.mia...42bc717e6a.gif

Organic objects like body panels, interior pieces, some suspension components, etc are a great fit for this, but for more technically built pieces, the deck of the engine block for instance, I think would be better modeled with actual blueprints/dimensioned sketches.

Example: suspension arms would have to be cleaned up and bushing holes would have to be simplified and dimensioned, as well as smoothing out ALL that casting flash.
The reason for this is because your object doesn't have any planar faces or cylinders to base an origin reference from. This is something that is needed when assembling components together in a program like Solidworks.
My point is that these parts would need further modification to be used in the intended computer application.

For example, imagine wanting to machine a billet aluminum UCA from the exact dimensions of the cast original using these models.
The operations would go as follows (for Solidworks at least, since it's what I use)

simplify polygons
assuming an edge is not generated somewhere automatically, spend a bunch of time setting the edge on the flattest face you can in order to get the correct origin of part.
create planes and sketches outlining and dimension all circular and flat faces
extrude cut all the bushing holes and dimension them.
"shave" away all flashing from the cast part
FEA test the part in whatever material you will make it from
send to CNC

I think this is worth pursuing, but I fear it would be time intensive for people wanting to modify and fabricate parts based off them. My only real concern for the usability of these scans are in getting the surfaces acquired.
If it takes me hours just to pin down a flat surface it wont be worth my time since I could just take measurements from the part with a ruler and tools in the same amount of time.

If you post a file, that control arm for instance, I could play around with it and report back :)

I hear what you are saying with finding reference surfaces on a dense polygon model but the usage example doesn't make a lot of sense to me. I can't see anyone looking to CNC a copy of an OEM steel piece in aluminum by directly cleaning up scan data.

I can align the model's orientation and placement in the scan software. Generally the scans should be treated like physical objects to take measurements from. Again, I think the advantage here would be having a complete assembly to work with. It would drastically reduce the number of oops moments where a prototype interferes with another part in an unforeseen way and it allows for designing in areas where there is limited space to route a part.

I got one of these einscan-s-3d-scanner at work for scanning architectural elements (primarily stone) so we can recreate them with a variety of technologies. Never thought of using it for scanning OEM parts is a pretty neat idea. It does take some time to scan the object in, especially if you have to rig it up to scan it (better if the object rotates at a set distance from the scanner as opposed to moving the scanner around the object). would really love to scan the whole car so you can play with custom aero bits without having the fabricate. So much technology now with simulating aerodynamics is so cool!

Ok maybe the control arm was a bad example, but my concern still stands about what kind of process and programs this assembling you are thinking of will take place in.
I fear these scans are going to be way too large to play with in any CAD program unless they are cleaned up or simplified to the point where touch up will indeed be required.
It would be nice to post a test scan of something. Anything really. A pipe with threads for example.

My PC is on the top end of what people use/have access to. I would like to see if this is even possible to load something this complex and successfully edit it in the way I am trying to convey.

My shitty work computer can open the stl without issues.

Xenon processor, 16 gigs of ram, and a crappy Quadro card.

My dinosaur of a home computer opened the file up without issues.
i5 4790k, 16gigs of ram, gtx970


Cleaning up the scans would help with load times if you're trying to clean it up in a cad program. Load times and editing in MeshLab (or a similar program) is a lot faster.

I don't know how this thing works, but could you use it to finally get a real model of the bottom of the miata? This could then be used for cfd work, if it's that easy. In my head it sure is simple.

It depends on the CFD software. I don't know a whole lot about CFD but i have seen some CFD software that is intended to be used on polygon models.

Does anyone in NC have junk transmission they want to donate?

I use an artec spider 3D scanner at work along with polyworks to create a functional part model.
We, as a community, should definitely build a library!

Just as an update- I am still scanning parts and I plan to post the first set of scans soon. Also, I wanted to share the primary purpose of the scans and hopefully get some help tracking down cheap parts for me to continue. I am working towards building a VR game where you can completely disassemble an NA (gameplay would probably involve diagnosing and fixing). Imagine walking around a small car with a room scale VR setup, flashlight in one hand, and multi-tool in the other. My first goal (hopefully within a month) is the completed front suspension corner sitting on the front sub frame. After that I will move onto the rear suspension, then all the drive train parts. I expect I can have the roller skate finished in about 6 months.
I have been buying some used parts but this is getting expensive. If anyone has junk stock parts they are willing to part with please let me know. I am currently looking for:

Stock shock + spring (or bilsteins)- I only need one front and one rear
Front subframe
5 speed transmission
Rear Subframe
Rear upright and hub
A 1.8 diff with housing.

If you are local I can scan a part in a weekend and return it thoroughly cleaned.



First few game ready assets:
https://cimg3.ibsrv.net/gimg/www.mia...7e6f44a7ec.jpg
https://cimg5.ibsrv.net/gimg/www.mia...23f23be5ff.jpg
https://cimg9.ibsrv.net/gimg/www.mia...40009b66b4.jpg



Some of the newer scans. These are a mix of scan data an manual modelling. The hyper clean/machined surfaces and things like the axle nut and wheel stud are remodeled:

https://cimg1.ibsrv.net/gimg/www.mia...414fb8c995.png

https://cimg5.ibsrv.net/gimg/www.mia...985fef21f8.jpg


https://cimg6.ibsrv.net/gimg/www.mia...d27c2a231a.png

https://cimg8.ibsrv.net/gimg/www.mia...9f0f6a36f5.png
https://cimg9.ibsrv.net/gimg/www.mia...74f06a6b8c.jpg

Ill send you a rear upright and hub if no one local does.

How much time did it take from start to finish to make 3d model like this?

The lower control arm is about 3.5 days working normal 9-5 hours.

A thats to much work, i have a lot of MX-5 parts in 3D but all made by myself in 3D with 3Dsmax. A lower control arm would only take me about 2 hours, but yes it would not be accurate.

But it doesnt matter for me because i only use it for 3D animation movies as seen on the IL Motorsport youtube channel



EDIT: But it does look very cool even without textures on it

Most of the work is not the scan- that 3.5 days number is for the final in game asset. The scans themselves take about 30 minutes to shoot and then process for 4-8 hours depending on how many photos.

Maybe i should try it out sometimes

So you only looking for NA parts? I have stock NB suspension, could get you a set of NA too.

A stock NA or NB shock would be great. PM me.

In unreal engine:
https://cimg3.ibsrv.net/gimg/www.mia...f96a6a1cde.jpg

Two things- I have a development blog for my project and I started posting scan files:
Free 3d scans | Digital Mistake

Holy shit, man. How did I not find this thread earlier?

I want to have your babies.

This is one of the coolest things I've ever seen.

Nice work... I'm building a full space frame chassis at the moment and I loaded the design into unreal to display 1:1 on a HTC Vive. It's awesome to be able to see the car and chassis in actual scale, walk around it and even sit in the car and check clearances. I might download these and drop them into my model if it's ok.

Yeah, feel free to use it for your space frame chassis.
I still have a few more scans I can post, most of them probably aren't that useful. I don't think many people have a use for brakepad or rotor scans. I don't have a front upper control arm scan since I built that from scratch.

Almost done with the gameplay mechanics for the prototype:

Hey everyone, I am hoping to get some feedback on some ideas for my game. Most of us enjoy taking things apart so I think this is a good group to poll for opinions. Right now in my game every part down to bolts and cotter pins are pulled off of the car and are their own physics actors (meaning you can drop it, pick it up, or lose it). I have two directions I can go with for handling lots of small parts. One option is to continue this way, force the player to manage all of these parts, and provide them with really good organizational tools to keep track of them. I am thinking there would be interface clusters where you can attach objects to portable nodes- You would be able to place and orient pieces in those nodes however you wanted and they would stay in the same place relative to the nodes. Meaning you could lay them out in space with some notation to help remember how things go back together, and then carry that configuration around the shop. I could also implement some kind of loss prevention system, like if a component falls through the world moving it back to the 0,0,0 coordinate, or allow duplicate part ordering for lost parts.

On the other end of the spectrum, when something is removed from the car, I could blink it out of existence into an invisible inventory. When you want to reassemble a car part you would hover your hands over the area where it goes, and ghost copy would appear showing you the part- Click a button and a real copy is installed. You wouldn't be able to lose parts and the player isn't responsible for keeping track of anything.

As someone who has always liked taking things apart, I think option 1 is much more interesting. Taking apart something complicated for the first time is always kind of exciting because I know I need to keep track of how it all goes back together. Reassembly usually involves things like counting fasteners (and now as an adult checking photos or notes).

Would it be possible to have an option for either one? The detailed, more difficult option for those who enjoy that level of minutia, and a simpler option for the less detailed oriented?

This is the last set of art tasks for my prototype milestone. Front suspension fasteners:
https://cimg5.ibsrv.net/gimg/www.mia...f967f86558.jpg

Long bolts need more washers. Fantastic job!

I reuse the large washer for both sides of the long bolt.

This is all of the parts I have so far. 4k screenshot from game turned into a wallpaper:
http://digitalmistakegames.com/wrench4kwallpaper.jpg

I take take it all apart and put it together in game. Video incoming soon:

I am sure you have seen, my summer car.

Yeah, I haven't actually played it yet. I watched a little bit of it on youtube but I have been waiting to play My Summer Car and the Mechanic Simulator games until I have my game play systems worked out more. I want to come to my own conclusions about how it should work before seeing what decisions other developers made.

I hope there is an audience for this. I want the game to be focused on motorsports prep and include a bunch of aftermarket parts. The mechanic sim games and my summer car really simplify the cars and they don't go very deep into how to diagnose or repair beyond click on the stuff that is broken. The trade off is that my game will have less variety in the cars but with more depth in the way they are recreated. I want my game to be usable as a shop manual with enough optional help to guide newbies through the process so they aren't totally lost.

Video of the protoype:

This is amazing. Mad cats.

Thanks Dan. In ~6 months I will be done with all the OEM type parts and I will be looking to add aftermarket parts. Hoping to have 949 stuff in there.