ORNL and RTX Develop Machine Learning In Situ Quality Control for Metal 3D Printing

October 31, 2023 by Matt Kremenetsky 3D Printing 3D Printing Research 3D Scanning 3D Software Aerospace 3D Printing North America

Share this Article

Oak Ridge National Laboratory (ORNL) and defense/aerospace giant RTX (formed by the merger of Raytheon and United Technologies) have collaborated to develop a software solution utilizing machine learning (ML) for in-situ quality control of metal parts produced with powder bed fusion (PBF) additive manufacturing (AM). The results of the team’s research were published at the end of September 2023 in the academic journal Additive Manufacturing.

The process underlying the solution involves leveraging data collected with both a near-infrared camera and added visible-light camera during the printing process, along with inspection of parts made after with CT scans. Next, the combination of both datasets is used to teach an algorithm to identify flaws in-situ during subsequent prints. After each print, the software is also trained by human feedback.

According to the researchers, the biggest leap forward achieved by their work is the foundation that has been laid for quantifying the reliability of ML-driven quality control — an indispensable prerequisite for this sort of technique to ultimately reach scale.

In a press release about the collaboration between ORNL and RTX software solution for in-situ quality control of 3D printed parts, ORNL researcher Luke Scime said, “We can detect flaw sizes of about half a millimeter — about the thickness of a business card — 90% of the time. We’re the first to put a number value on the level of confidence possible for in situ (in process) flaw detection.”
As ORNL researcher Zackary Snow explained, “For regular manufacturing we know what [the flaws] are and where and how to find them. (Operators) know the probability that they can detect flaws of a certain size, so they know how often to inspect to get a representative sample. Not having a number makes it hard to qualify and certify parts. It’s one of the biggest hurdles in [AM].”

As Snow summed up, the basic idea behind the researchers’ activities is to get to the point where manufacturers utilizing PBF can have “CT-level confidence without CT”.

This work by ORNL and RTX is a perfect example of how crucial the centralization and standardization of data has become, in order for the AM sector to propel itself into the phase of widespread commercialization. Currently, ML-driven in situ quality control may be the most critical element in this industry-wide project, given the potential it holds for saving time, money, and labor in what is probably the costliest and most time-intensive area of the sector, post-processing.

Moreover, ORNL and RTX are exactly the entities that will need to work together on this front in order to lead the rest of the sector forward. The ORNL/RTX study reinforces the prioritization of the objectives set forth, for instance, in June 2023 by ASTM International, with that body’s publication of its Strategic Guide: Additive Manufacturing In-Situ Technology Readiness Report.

Most specifically, increased confidence in the automation of quality control for 3D printed parts could have an accelerative effect in efforts to realize a goal I wrote about last week, the comprehensive digital traceability of 3D printed parts. Ultimately, after all, the ability to trace those parts will only be meaningful to the extent that their quality/certification can be automatically and reliably traced. If industry stakeholders can continue to unite for this purpose, in particular, it should provide a major boost to everyone hoping to deploy AM primarily as a means to keep fake parts out of strategically vital supply chains.

Images courtesy of ORNL