Go Explore: Autodesk Reinvents Search

Reinvent the wheel no more, says Autodesk. The company has just launched a technology that will scour your company’s existing parts to see if the part you are designing may already exist.

Autodesk calls this technology Design Graph. If it lives up to its billing, the name will hardly do it justice.

Design Graph will let you sit at your desk and explore designs anywhere in your company—as long as they are in an A360 database. Design Graph is able to search for parts based both on shape similarities as well as textual clues, often entered as part metadata.

It couldn't be simpler. Design Graph, a search engine that adds shape-based searching. (Image courtesy of Autodesk.)

So What’s New?

Search engines are text-based, à la Google, or they are shape-based, like the ones we’ve covered here. Both can be misleading. For example, a shape-only search may find all screws but be unable to tell if they are made of stainless steel. Search functions that look for a chair by geometric similarity only will miss odd shapes such as the Adirondack chair or even bars tools and bean bag chairs.

Meanwhile, searching by text only relies on the proper text being added to the part and can be fooled by variations in file name, product descriptions and drawing notes. It is only when you combine text search with geometric search that magic happens, according to Autodesk.

Where Can I Get It?

Design Graph will exist as a web application and will also pop up as a search tool within existing applications. It will appear first in Autodesk’s A360, then in Fusion 360 and maybe eventually in desktop applications such as Inventor.

Mike Haley, senior director of Machine Intelligence at Autodesk. (Image courtesy of LinkedIn.)

“We hope to have it run in the background during a design,” said Mike Haley, who for the past four months has been the senior director of machine intelligence at Autodesk’s San Francisco office and has been working with the R&D team in Singapore. “Wouldn’t it be nice to have it suggest an existing part as the part you are designing is actually taking shape on your screen?”

I would imagine that a program that is always checking for similar parts as I am creating would slow my design process to a crawl. Comparing a part taking shape against all possible similar parts would not be a trivial amount of computing.

Luckily, we have infinite computing, according to Haley, using Autodesk’s term for bringing a vast amount of computing resources to bear on an as-needed basis. For example, a user that may sporadically run a series of complicated analyses can “rent” an array of servers rather than buy a supercomputer. Design Graph offloads the calculations and searching for matching parts onto its servers so your workstation doesn’t get bogged down.

It also helps with part searches if part libraries are stored centrally and cloud storage lends itself to this.

How Does It Work?

Design Graph was four years in the making. It is a result of machine learning research being done by Autodesk, much of it in its Singapore research facility.

Design Graph appears to be remarkably easy to use, with a one-line interface closely resembling a search box. You type in “chair” and see all the different chair and chair-like products.

For the first release all models and parts have to be in A360 for Design Graph to work with them, so it cannot currently index parts in other cloud applications. However, Design Graph supports over 50 different 3D design file formats so it can handle files from SOLIDWORKS, Creo and in STEP and IGES formats. Furthermore, Design Graph can work with mesh formats such as STL, OBJ, FBX, DWF, etc. Design Graph creates a "deeply indexed" database that enables quick exploration of large amounts of design data.

Blazing speed. Design Graph finds page after page of screws—and a few non-screws—in less than a second from a database of two million parts. (Image courtesy of Autodesk.)

In an online demo, I could see results for a search for screws come up really fast. In less than a second, Design Graph was able to evaluate a database of two million parts to generate page after page of results. However, some of the pages did seem to have non-relevant results such as non-screw parts.

Is there a way to teach Design Graph which parts are irrelevant to a given search, such as incorporating user feedback? This would help it perform smarter searches.

This will require patience, Autodesk seemed to suggest.

“That is something we would have in future release,” said Haley, who seemed quite accustomed to impatient editors that appear to be unappreciative of what may be a game-changing technology. “This is only the first release.”

For more information, visit Design Graph on the Autodesk site.