UPDATE: On January 29, 2020, Physna announced that it had raised an additional $20 million of financing. Sequoia Capital led the series B investment. Drive Capital also contributed. This brings the company’s total funding to $29 million.
"This will allow us to significantly grow our team and accelerate advancements made both in Physna's enterprise software as well as our free consumer site - Thangs.com - which is quickly becoming the standard way to find parts and models." said CEO and co-founder Paul Powers who plans on using the additional funds to grow his team.
________________________________________
You may not know that you need a search engine for parts. You probably didn’t know that you needed a search engine for words, either—until Google came along.
Paul Powers, co-founder and CEO of Physna. (Picture courtesy of Physna.)
You will be free to move on with your brilliant design—your company’s next big hit. What engineer wouldn’t sell his soul for that?
You don’t have to, says Powers. Thangs is free.
A Brief History of Geometric Search
Design search engines have come—and gone. We made an extensive listing of search applications in 2015. Since then, we have looked at CurvSurf in 2016 and Model Search in 2018.
They don’t work, says Powers, dismissing the lot of them.
“I started this company because I tried every geometric search engine. Geometric search looks at models like a 2D projection. They more or less take snapshots. They understand edges, points and volumes from CAD data. They can’t work with 3D scan data; they can’t work with agnostic models; they can’t find parts within parts.”
And so, part searches continue to be manual and text-based, such as searches for standard parts like mechanical fasteners, electrical components or building hardware. But the bulk of the parts in everything are nonstandard, and a usable, Google-like geometric search for an arbitrary shape has remained elusive.
DNA Is Part of Physna’s DNA
The company’s name Physna is a portmanteau, from “physics” and “DNA.” Physna has secured $6.9 million from one investor, but $8.9M in total. It has paying customers—most of whom Powers was careful not to reveal.
Physna is entirely based on extracting what it calls the DNA of a part—its own creation of a unique identifier of a part’s geometry that recognizes meaningful features of the part and its geometry and their relationships to each other. This part DNA can be used to match a part with other parts whose DNA has already been cataloged.
Like biological DNA, Physna’s part DNA is a set of stored information that is unique to a shape. Like biological DNA that uses molecules called chromosomes to map the detail of the cell and the organism, Physna’s DNA does the same with an identifier made of digital data.
This is an approach unlike the many geometric search engines to date, which rely on pattern matching, such as comparing the profile of a part to a library of stored profiles. Such a technique records a “hit” if the profiles match, but in order for that to happen, the profiles have to be in identical orientations. But creating, storing and matching all the possible profiles is a Herculean task. Considering only the three major axes and rotating the part in 1-degree increments about each of them gives us 3603, or 46,656,000 profiles. If the orientation is arbitrary, that is, not along the major axes, which is what you would want for an automated search, the number of orientations is unbound.
That approach never made sense to Powers.
“You have the same DNA if you are upside down or standing up, correct?” he asks rhetorically.
Look to the Stars
That lack of a good geometry search was galling to Powers in his first job as a patent attorney who refused to believe that there was not a way to look for devices that were similar to others, a necessary part of his job.
Powers was no ordinary attorney, however. He was homeschooled, studied chemistry when he was 12, and gained admission to Harvard at the age of 16 to study astrophysics and astronomy. He was, by his own admission, a “weird, nerdy dude.” Law school was at the University of Heidelberg.
Looking at the stars during his undergrad days had given Powers ideas.
“The universe is a 3D, non-oriented place. On Earth, we think up, down, left, right, but the universe isn’t like that,” said Powers. “The universe is a giant ball, if you will, and there is no up, down, left or right, and you have to be able to understand your way around it regardless of orientation. The way to figure out where a star is, is not with constellations. You look for patterns in a different way, gravitational signatures, et cetera, and you create patterns out of that.”
As an attorney, Powers was sifting through one patent drawing after another, with an eye toward protecting the client’s IP. It seemed weird to him that there were algorithms to find copyright violations in words, even in music, but there was nothing for doing that with 3D models. While it is obvious to a human that if one were to “take Mickey Mouse’s head and put it on Donald Duck’s body it would be a Disney infringement. But that is not discernible by any algorithm.”
Since there was no search engine to solve this problem, Powers took it upon himself to make one. He founded Physna. He took his CTO and holed up in Saint Croix, where his girlfriend’s family had a home, and spent every day writing equations on a cheap whiteboard from Kmart.
Physna's co-founder, who came from a 3D printing world, was all about slicing a part into orderly layers, but Powers, coming from the world of astrophysics, knew the universe was anything but orderly.
“But algorithms that would reveal star patterns wouldn’t work for part searches, so we created our own way of analyzing how parts relate. We could figure out how all parts are related no matter where they come from or what file type they were. They don’t have to have any metadata or any kind of file structure. We could change the precision so the size of the part does not matter. We have found parts as big as the Empire State Building and as small as a virus.”
What follows is the rest of our interview with Paul Powers.
A geometry search engine is not a new concept. Tell us what makes Thangs different?
Three things. Thangs is more powerful. It uses AI. It can predict things about those parts. Second, it’s free and easily accessible. You don’t have to have a PLM system, a CAD system—anything. You can go right now to thangs.com and upload anything you want, and it’ll find everything that’s like it. Not just things that look like it. And third, it’ll find things that are inside of other things.
So, let’s say you’re a mechanical watch designer. You upload a mechanical watch that you designed. It’ll find the gears and the components that go inside of it, and it’ll list those for you and say, “Here you are. This is what you need to build that watch.” Not just a similar watch, and you don’t have to pay for that. It’s totally free.
And how did you come up with Thangs?
[Powers laughs.] We had to have some names to call this thing. On the way to work, I thought of a few names. Thangs was one of them. I never thought it would go over, but everyone loved it. Plus, the domain was available. It wasn’t cheap, but available.
Thangs is free to use. How are you planning on making money?
We’re not trying to make money off of the student or the hobbyist. We have enterprise customers that pay for access. If you’re a user of Thangs, you have two options. You can make a model public or private. It’s free either way. If its private, no one can find it. If it’s public, anyone can find it. So, think of GitHub. If you want to display your work to the world, then you would make it public. If you’re just working with somebody else in your class or in your company, you’d make it private.
Who are your customers?
I can’t name a lot of our big customers, but almost all of them are Fortune 500 companies. We have a few government agencies. We made a large 3D part repository for the Department of Defense, and their people—not engineers—use it to find a part in the supply chain and also to identify something in the field. Again, you don’t have to be an engineer to use Thangs.
If I am designing an assembly of parts, I can have Thangs find similar parts and where I might buy them?
[Powers cranks up a demo.]
Let’s say I’m looking for a valve. Not a big deal. Thangs will find valves in the Thangs database and you can download them. It will find things outside of Thangs and you can buy them and download them.
I can use Thangs to find parts without having to thumb through catalogs and without part numbers?
Absolutely. I’ll upload a part here from McMaster-Carr. Part number 118… whatever. I won’t use the part number. I can search with the part geometry.
It’s processing.… It’s a little slow today. We added over 500 users in the past two hours. I got a text message from our CTO five minutes before this meeting started saying that he’s worried that the site’s going to break down because it has more people on it than we expected. It’s not a problem. We use Google’s cloud service. We can scale up quickly.
There… it found the parts from McMaster. Now it’s looking on Thangs to see if it finds anything.
Can you show how it can find parts within parts?
Yes, it works on full parts and also parts in parts. Let’s just go with this one. You see it found another part that from the outside does not look similar—but wait. We take the found part apart and there are similar parts inside it. A traditional shape search would not have found them.
I can use the component finder to find all the different components that exist worldwide that go inside of this assembly. I can set the correlation 100 percent or 95 percent for something that is close. You see it found the same parts in Grainger, Parker, Schaeffler, et cetera. They are all the same parts. And some are similar or have some geometry that is similar.
Think of it as an autofill. Let’s say you are modeling a valve. You model a part—a tiny part—and give it to Thangs. It will find the valve.
Can Thangs find models from 3D scan data?
Yes, even bad scan data with a lot of missing data. I can click on scan match and find different types of 3D scans. Then I can find the CAD model for it.
If it exists in the public database, you can take the world’s ugliest scan, throw it up in Thangs, and in theory, Thangs will find it.
What’s in the Future for Thangs?
With 3D part matching in hand, Physna wants to extend its search to 2D input. Like you taking a picture of a part and Physna being able to find the part and tell you where you could buy it.
We’re not there yet, but it’s coming. We have it working, but it’s not released yet, because we want to make sure it works for everything. A process by which we take a chair, for example, extract the relevant data. Right now, we go through a proprietary 2D to 3D machine learning process. We find not only a chair. We find the exact one that correlates to that chair. We find the 3D model that’s perfect for that 2D picture.
Or let’s say you find a bolt that fell out of your car. You take a picture and ask Thangs to identify it. You need to know where it goes. Thangs will be able to tell you where it goes—and how to install it. That’s where we’re heading. Anytime you don’t know what something is, take a picture and we’ll figure it out.
Think of it like IKEA instructions on steroids.
But a photo cannot determine what is hidden from view. How would Thangs get around that?
Because of the way Thangs works, Physna, will over time, using machine learning, be the first to figure it out.
With a 2D picture, the orientation is critical. I can transfer that DNA into a 3D model. Thangs will match it against every possible orientation of every 3D model and quickly eliminate 99.9 percent of these 3D models. Of the few that remain, does any orientation correlate with the 2D picture? Let’s say 10 do. Thangs will find which one correlates the best.
Is there a plan for Physna beyond CAD?
When we started with patents, people asked if we could improve engineering efficiency. People also asked about use cases in healthcare, like identifying tumors at a much earlier stage. And inspection of cell towers automatically. We must have had hundreds of suggestions, most of them from the Fortune 500. We kept going down these different rabbit holes, but one day we decided we had better focus on one thing. We knew the software could do it, but we didn’t have the manpower to build it out. Our goal with Thangs now is to be the springboard, a platform, and supply an API so people can license the technology and build their specific applications.