X-ray tomography and coordinate metrology benchmark for internal cavities in additively manufactured alloys
This paper introduces a comprehensive public benchmark dataset and evaluation framework that validates X-ray computed tomography for measuring internal cavity wall thicknesses in additively manufactured Ti-6Al-4V and Inconel 625 alloys by comparing CT-derived results against ground-truth coordinate measuring machine data across various surface conditions.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine trying to measure the thickness of a wall inside a sealed, twisted metal tube. You cannot see the inside, you cannot touch it with a ruler, and you cannot shine a light through it. This is the daily reality for engineers who build parts using additive manufacturing, a process often called 3D printing, where metal is fused layer by layer to create complex shapes. These parts often contain hollow channels or internal cavities that are vital for their function, yet they remain invisible to traditional inspection tools. To check if these hidden walls are the right thickness, scientists use X-ray computed tomography, a technique that takes thousands of X-ray images from every angle and stitches them together to create a digital 3D model of the object's interior. It is like taking a CT scan of a human body, but for a machine part. The challenge, however, is that the computer model is not perfect. The X-rays can be distorted by the density of the metal, and the software that turns the blurry X-ray data into a sharp surface has to guess where the edge of the material actually lies. If that guess is wrong, the measurement of the wall thickness is wrong, potentially leading to parts that fail under pressure.
To solve this puzzle, a team of researchers from Canada has released a new, open standard for testing how well these X-ray measurements actually work. They created a specific metal object, shaped like an hourglass with a hollow center, and manufactured it in two different metals: a common titanium alloy and a high-strength nickel alloy known as Inconel. They made each object twice, once with a rough surface straight from the printer and once after chemically smoothing it out. The clever part of their experiment was how they measured the truth. They first scanned the assembled object with X-rays to see the inside. Then, they took the object apart. Because the two halves were now separated, they could use a high-precision mechanical probe, a device with a tiny spherical tip that physically touches the surface, to measure the exact thickness of the walls at hundreds of specific points. This physical measurement served as the "ground truth," the reliable reference point against which the X-ray guesses could be judged.
The researchers then compared the X-ray measurements against the physical probe measurements to see where the digital models succeeded and where they failed. They found that the accuracy of the X-ray measurement depended heavily on the condition of the metal surface and the specific settings used to process the data. When the metal surfaces were rough, the X-ray software tended to underestimate the wall thickness, while the physical probe tended to overestimate it because the probe tip would ride over the tiny peaks of the rough surface rather than sinking into the valleys. After the parts were chemically milled to smooth them out, the gap between the X-ray results and the physical measurements narrowed significantly, showing that surface texture plays a major role in measurement error.
Perhaps most importantly, the study revealed that there is no single "magic setting" for the software that makes the X-ray measurements perfect for every situation. The team tested different ways to define the edge of the metal in the digital model, and they found that the best setting for the rough titanium parts was different from the best setting for the smooth nickel parts. In some cases, the error in the wall thickness measurement was as large as 0.16 millimeters, a significant amount for precision engineering. They also discovered that in areas where the metal was very dense, the X-rays struggled to penetrate, creating "blind spots" where the software could not clearly see the shape of the object. In these difficult areas, they tested a new approach using artificial intelligence to guess the shape, which performed better than the standard methods, though it still had limitations.
This work does not claim to have solved the problem of measuring internal cavities once and for all. Instead, it provides a clear, shared benchmark that other scientists and engineers can use to test their own methods. By releasing all the raw data—the X-ray images, the computer models, the physical measurements, and the specific instructions on how they were aligned—the researchers have created a common playing field. This allows anyone developing new software to see exactly how their tool performs against a known standard, rather than just claiming it works on their own private tests. The study confirms that while X-ray tomography is a powerful tool for seeing the invisible, its accuracy is a complex mix of material properties, surface texture, and the specific choices made by the software. The path forward is not a single perfect solution, but a better understanding of how to adjust our tools to the specific challenges of each unique metal part.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.