A new deep learning framework developed at the Department of Energy’s Oak Ridge National Laboratory (ORNL) speeds up the inspection of additively manufactured metal parts using X-ray computed tomography (CT) while increasing the accuracy of results. The lower costs of time, labour, maintenance and energy are expected to accelerate the expansion of additive manufacturing or 3D printing.
“The speed of examination reduces costs significantly,” ORNL Senior Researcher Amir Zibari said. “And the quality is higher, so post-processing analysis becomes simpler. “
The framework has already been incorporated into software used by commercial partner ZEISS within its hardware at ORNL’s Department of Energy Manufacturing Demonstration Facility, where companies are working to fine-tune their 3D printing methods.
ORNL researchers previously developed a technique that could analyze the quality of a part as it was being printed. Adding a high level of imaging accuracy after printing provides an additional level of confidence in additive manufacturing with the potential for increased production.
“Using this, we can examine every part that comes out of 3D printing machines,” said Pradeep Bhatad, Director of Business Development at ZEISS Additive Manufacturing. “Computed tomography is currently limited to prototypes. But this one tool can push further industrialization towards industrialization.”
A CT X-ray is important to certify the integrity of a 3D-printed part without damaging it. An object placed inside a cabinet is slowly rotated and scanned at every angle by a powerful X-ray. Computer algorithms use the stack generated from 2D projections to create a 3D image that shows the density of an object’s internal structure. CT X-rays can be used to detect defects, analyze failures, or certify that a product matches the intended composition and quality.
However, X-ray CT is not widely used in additive manufacturing because current methods of scanning and analysis are time-consuming and imprecise. Metals can completely absorb low-energy X-rays in the X-ray beam, resulting in an image inaccuracy that can be doubled if the object has a complex shape. The resulting defects in the image can obscure the cracks or pores that the scan aims to detect. A trained technician can correct these issues during analysis, but the process is time-consuming and labor-intensive.
Training a supervised deep learning network on CT usually requires many expensive measurements. Since metal parts pose additional challenges, obtaining proper training data can be challenging. Zibari’s approach provides a leap forward by generating realistic training data without the need for extensive experiments to collect it.
The Generative Adversarial Network, or GAN, method is used to create a realistic looking dataset to train a neural network, making use of physics-based simulations and computer-aided design. “GAN is a class of machine learning that uses neural networks that compete with each other as in a game. It has rarely been used for practical applications like thisZibari said.
“Because the X-ray CT frame needs scans at lower angles to achieve accuracy, it reduced imaging time by six times.From about an hour to 10 minutes or less, Zibari said. Working quickly with very few viewing angles can add significant “noise” to the 3D image. But the ORNL algorithm taught on training data corrects this, even boosting small flaw detection by a factor of four or more.
The framework developed by the Ziabari team will allow manufacturers to quickly adjust their chassis, even while changing designs or materials. With this approach, sample analysis can be completed in one day instead of six to eight weeks.” Bhatad said.
“If I can inspect the entire part quickly in a very cost-effective manner, then we have 100% confidence,” He said. “We partner with ORNL to make CT an accessible and reliable industrial inspection tool. “
ORNL researchers evaluated the performance of the new frame on hundreds of samples printed with different scanning parameters, using complex and dense materials. “These results have been good, and ongoing trials in MDF are working to verify that this technique is equally effective with any type of metal alloy.”Bhatad said.
This is important, because the approach developed by Ziabari’s team can make it much easier to certify parts made of new metal alloys. “People don’t use new materials because they don’t know the best printing parameters,” Zibari said. “Now, if you can characterize these materials very quickly and improve the parameters, that will help move these new materials into additive manufacturing.”
In fact, Zibari said, the technology can be applied in many areas, including defense, automobile manufacturing, aviation, and electronic printing, as well as the non-destructive evaluation of electric vehicle batteries.
for more information: www.ornl.gov