At Sandia National Laboratories, a new inspection workflow is taking shape that could help catch tiny defects earlier in the manufacturing process for ceramic components.
“We manufacture ceramic components for nuclear deterrence applications,” said Process Engineer Jesse Adamczyk, who is leading the project. “We realize there’s a big opportunity here.”
Teams from across the Labs are installing new optical and acoustic imaging systems and building an AI-assisted review tool designed to speed inspections while keeping people firmly in the loop.
“We do manual inspections of all our parts. It is extremely time-consuming,” Adamczyk said. “These parts go into various weapon systems.”
The project begins by scanning ceramic billets, the starter pieces that are later manufactured into finished components, using high-throughput imaging systems that create detailed digital records of each billet.
“It’s pricey to get billets to their final component,” Adamczyk said. “If we can identify defects at the billet level, we don’t put all that work into manufacturing the final component.”
The earlier inspections will save time and money.
Right now, inspectors rely heavily on manual microscopes for inspecting final components. It takes one to two years to fully train an operator on the manual inspection process, which is time-consuming and challenging on the eyes.
The new approach for final components is designed to shift that work to a digital workflow in which images can be reviewed at a workstation.
“Right now, an operator looks through a manual microscope for defects. They’re subtle, so they can be hard to find,” Adamczyk said. “We’re setting up software — an AI augmentation interface — where operators can do anomaly detection from their desktops and have AI highlight defects for them.”
Adamczyk emphasized that inspections will not rely solely on AI.
“Operators will double-check to make sure the AI is highlighting real defects, and if there’s a defect AI misses, the operator will catch it,” he said. “AI augmentation is going to be more effective than manual visual inspection and more effective than just letting the AI run loose.”
Adamczyk said this is a big shift, but operators are embracing it to help meet demand.
Here is an exclusive Tech Briefs interview, edited for length and clarity, with Adamczyk.
Tech Briefs: What was the biggest technical challenge you faced while developing this AI inspection workflow?
Adamczyk: That's a really good question, and the answer might be a little surprising. We have put together a web app to help display the images for the AI anomaly detection. And, honestly, the biggest part was getting that web app up and deployed. So, even though we had AI coding assistants build that web app, deployment is still a really challenging and human-focused area still. Basically, getting the web app up onto the server and then making all the connections; that turned out to be the biggest challenge.
Tech Briefs: Oh, wow. That is surprising. Now, can you please explain in simple terms how everything works?
Adamczyk: Let me give a little bit of background. Our existing process has our inspection operators looking at these tiny parts under a microscope, and they have to measure these really small defects using an eyepiece reticle — it’s all manual microscopes, and it's really hard.
Our new AI augmented process is going to take those same parts and they're going to run through these high-speed imaging systems that are designed for semiconductor manufacturing. Then those images that come off of the systems will have AI anomaly detection run on them. That AI will find the defects for us; all those defects will be highlighted in images and presented to the inspectors in our web app.
Then, the inspectors will go in and verify that the AI made the correct determination on the parts there. So, they're going to be looking at screens and looking at computer monitors instead of looking through microscopes; it’ll be a lot easier.
Tech Briefs: The article says, “This is a big shift, but operators are embracing it to help meet demand.” My question is: Was that unanimous? Was there any dissent or pushback at all?
Adamczyk: There has not really been any pushback. I think, at first, they were like, ‘Well, what's going on here?’ But as soon as they started trialing out the software and saw that, ‘Oh, this isn't here to take my job. This is just to make my job easier.’ Then they were all onboard. ‘This is going to be way better.’ ‘We’re not going to have to strain our eyes staring through this microscope.’ ‘Yeah, this is clearly the future.’ So, we’ve had a lot of buy-in from the inspection operators.
Tech Briefs: The article also says that it's scheduled to be up and running by early fall. Is that timeline still on track?
Adamczyk: Yeah, we’re working hard at this every single day. We have a big team working to move things along. Every day we’re making progress and meeting our milestones on this. I think by fall we’ll have a good initial process going on our production floor.
Tech Briefs: Those are all the questions I have; is there anything else you'd like to add that I didn't touch upon?
Adamczyk: Maybe just one thing. A lot of the stuff we're doing with AI, we’re being very pragmatic about using AI on our production floor. We’re really just trying to make people’s jobs easier here, and we’re doing it through pretty common-sense means — things that would be in line with lean manufacturing. We’re not doing anything crazy, leading edge, or anything like that. And I think we can have a pretty big impact with even those relatively low-cost, common-sense tools.
