Araqev Tackles 3D Printing Quality Control
Purdue University company leverages machine learning to create closed-loop, smart controls to enhance 3D print quality.
May 2, 2022
Despite all the technology milestones and ensuing enthusiasm behind advanced 3D printing, quality control remains an-ongoing issue that hampers more widespread adoption of additive manufacturing as a mainstay production method.
Araqev, a startup affiliated with Purdue University, is now aiming to solve that challenge by taking an algorithmic approach to 3D printing quality control. The company has developed software that leverages machine learning models to learn from past 3D prints—both successful and flops—to iterate and modify designs in support of fewer shape deviations and ultimately, 3D printed part failures. Shape deviation is a consistent problem across the spectrum of 3D printing technologies, both metals- and plastics-based, contends Araqev CEO and President Arman Sabbaghi.
“The quality control issue is ubiquitous across all types of 3D printing processes due to the inherent physics involved with rapid heating and cooling going on,” explains Sabbaghi, also associate professor in Purdue’s Department of Statistics in the College of Science. “We really believe that when it comes to 3D printing, the future depends on leveraging data from past builds, even scrap, to ensure better prints. That way, we’re not printing so many iterations to get the accurate and quality product we want.”
Based on its research models, Araqev estimates that quality-control issues surrounding additive manufacturing are costing companies nearly $2 billion in global losses annually.
While a market for 3D printing quality control software is emerging, Sabbaghi says the options are still pretty limited while explaining that Araqev is taking a different approach. Unlike existing software that employs physics-based modeling to solve 3D printing quality control issues, Araqev’s concept is to apply machine learning and transfer learning algorithms to the problem to take advantage of the existing large data sets that can provide insights into quality issues and guidance on how to avoid them. “Physics-based models have fundamental limitations for this—they are built under certain context and settings that may not be applicable to the specific 3D printer being used,” he explains.
The way Araqev’s software works is users 3D scan past-printed products, both usable and scrap, and upload that point cloud data to the platform, along with a nominal shape design they are looking to 3D print. The platform runs algorithms against those inputs to derive modifications to the nominal designs, known as compensation plans, so when the modified designs are printed, they exhibit fewer shape deviations, Sabbaghi says. The software works as a closed-loop process so if an engineer isn’t satisfied with the build after the initial go-around, it can work with the data to further iterate and improve the compensation plan. “This can eliminate reprinting significantly, improve throughput, and improve overall quality in one or two design iterations,” he adds.
In one test case, the Araqev platform was able to reduce shape inaccuracies on metal 3D printing output by 30% to 60%, depending on the geometry, in two iterations, with three training shapes and one or two test shapes for a specific geometry. The algorithms also enable the transfer of knowledge encoded via the machine learning models across different 3D printer technologies, materials, and shapes so organizations don’t have to start from scratch for each analysis.
“When 3D printing, the hardware is continuously changing and the software needs to adapt to those changes—that’s where transfer learning comes in,” he says.
Araqev is working to establish partnerships with 3D printing manufacturers to commercialize the software, which is currently in test mode.
About the Author
Beth Stackpole is a contributing editor to Digital Engineering. Send e-mail about this article to DE-Editors@digitaleng.news.Follow DE