3D Mechanical Database Brings Together Components, Machine Learning

A team of mechanical engineers from Purdue University has undertaken the task of establishing a nearly 60,000-component, open-source, annotated database of 3D mechanical parts. The database aims to bring together applications in machine learning with the individual parts that enable those applications and more.

Though computer vision researchers often use machine learning principles to train computers to visually recognize distinct objects, few, Purdue reported in a press release, apply machine learning to mechanical parts — from gearboxes to clutches, and nuts to bolts.

After experimenting with the notion of visual searching for parts throughout the 2000s, teams led by Purdue’s Karthik Ramani determined that to complete such a data set, the engineers would need to prepare a number of unique samples large and varied enough to allow a computer to build up a large enough store of pertinent information.

“Deep learning is data hungry,” Ramani said. “It needs a lot of examples for the computer to learn what humans mean and how things relate to each other. That means we needed a lot of 3D models of parts, which also required an underlying engineering classification.”

Purdue researchers ultimately partnered with TraceParts, a French company, to gain access to a comprehensive database of 3D engineering parts. Qixing Huang, an assistant professor at the University of Texas, also joined the collaboration in an attempt to identify similar databases for 3D models. The team eventually compiled a database of 58,696 mechanical components.

The Ramani-led group at Purdue next built a hierarchical taxonomy of almost 70 classes, based on the International Classification for Standards, that the International Organization for Standardization maintains.

“Now when a computer sees a picture of a seal component, it will know that it fits in the category of dynamic seals and then, more specifically, under composite seals,” Ramani said.

A new database would help engineers and manufacturers apply machine learning to mechanical parts. Courtesy of Purdue University/Sangpil Kim.

The researchers debuted their open-source database at the 16th European Conference on Computer Vision in August.

“We see many real-world situations for this technology,” Ramani said. “Imagine you’re working maintenance in a factory, and you’re replacing a part of a machine. You can point a camera at the part, and the computer will recognize it, and instantly give you all the specifications of that part — what it’s called, what it connects to, and where they are physically stored in the factory. This could even happen through augmented reality glasses; you could have your company’s entire visual catalog instantly at your fingertips and learn how to fix things or order parts.”

The published data set is online. Research continues at Purdue’s Convergence Design Laboratory. It is partially supported by National Science Foundation grants.