Up till now, using X-ray computed tomography (CT) can take a long time and be inefficient because metals can absorb lower-energy X-rays, creating image inaccuracies that can be increased if the object has a complex shape. The resulting flaws in the image can obscure cracks or pores the scan is intended to reveal, Oak Ridge said.
Researchers at the scientific institute have developed a deep learning framework – a software package that can help computers learn like a human – using a generative adversarial network (GAN) method that can synthetically create realistic-looking data using physics-based simulations and computer-aided design.
This means that fewer x-ray scans are required to obtain an accurate image, with any flaws corrected by an algorithm, the researchers say.
“The scan speed reduces costs significantly,” said ORNL lead researcher Amir Ziabari. “And the quality is higher, so the post-processing analysis becomes much simpler.”
The framework is being incorporated into software used by commercial partner Zeiss in its 3d printers.
“With this, we can inspect every single part coming out of 3D-printing machines,” said Pradeep, ZEISS business development manager. “Currently CT is limited to prototyping. But this one tool can propel additive manufacturing toward industrialization.”
According to Bhattad, the company is seeing if the technique is also effective with other metal alloys.
