Computed Tomography More Than Just Surface Deep?

AuthorGerd Schwaderer Business Development Metrology – Volume Graphics GmbH


Oh dear, there is a new technology called Industrial Computer Tomography and they pretend it can measure.

Didn’t we have exactly the same situation about 20 years ago? There were those guys that used light for metrology. Few people believed in it, and yet many years later we find them in every company all the way down to the shop floor. White light scanning. There is norms for it, and very few people still doubt that you cannot measure with them any more. Great advantages and a few disadvantages made them state of the art, and thousands with metrology quality are sold every year from major companies worldwide. The technology even moved into the consumer area.

Now here we are, Industrial Computer Tomographs move into the focus of many companies. So why is that and how does that work, what’s the advantages?

How an Industrial CT works:

Color-coded results of a wall thickness analysis

A  CT uses an X-Ray Source and the materials absorption to create a 3 dimensional model of the product from a series of projection images. So far there is no major difference to a medical CT. Only that the specimen is moving, whilst the medical solution needs to move the X-ray source and detector.

CT’s with metrology quality exist only since about 10 years, they consist of the tube, a detector and a rotary table with CMM quality machine bed and movement axes. The CT is taking digital radiographic images 360 degrees around the specimen. Very clever mathematical methods reconstruct that stack of pictures to a 3D object, that represents density, geometry and material very accurately. You may know their name in medical, which is “DICOM” and pretty much the same with patient meta information.

In analogy to 2D picture pixels (think of your digital camera), this dataset consists of voxels, a giant Rubics Cube made from easily a billion elements. They enable you, being not just the surface of the object, to identify, qualify and quantify faults and features like Porosity and Inclusions, Orientation and Location of Fibers, Simulation of Flow through the Material (transport phenomena), and even Structural Mechanical Simulation of the effects of those defects directly on the Voxels.

One of the beauties of the method is certainly the ability to replace a CMM with a new technology, being able to look not just on the outside of the object, but also into the object. If it is “just” the surface and the dimensions, that are of interest, there is already huge advantages.

CT does not care about historic problems of CMMs

Interior Geometry, deep pockets and undercuts that are either impossible to reach or hard to measure. Certainly soft materials are another sweet spot, as nothing touches the object.

CT also does not care about classic issues of the newer optical scanners, both light or laser based. The model having an uncooperative surface which is reflective, shiny or translucent doesn’t phase a CT. Colors on the opposite side of the scanner light spectrum (blue light scanners hate orange objects) doesn’t matter either. Remember it is all about absorption. So no spray necessary, also no calibration artifacts in the view needed, and no stickers to be present in the scene.

Color-coded deviations from Nominal/Actual Comparison

Certainly classic functionality like 3D compare and wall thickness works seamlessly and at any point of the model, not just where points where acquired, dimensional control and GD&T measurements are as straight forward as using any other technology.

The voxel model enables to calculate measurement points at any given position in space any time, always ensuring that there is enough surface points to do the math with. This is a big difference to the discrete points that are taken by optical sensors or CMMs. The points taken are discrete information in space, XYZ coordinates. Once taken, that’s the given data-set, no way to go back or forth, any point used for measurement in between is a good guess. That’s one of the reasons why it’s a major advantage to do the measurements directly on the native VOXEL data, rather than converting the voxel data to a mesh and using a metrology solution based on that representation.

So how accurate, or better “how uncertain” are the results of such a Industrial Computer Tomograph, and how can I rely on the results?

Lets have a look at the reconstruction process. The typical metrology dataset consists of 2000*2000 single slice pictures representing the absorption of the part, each delivered with 16bit of greyscale information, delivering 16 Gigabytes of measurement voxel data. That data gets visualized in 3D, revealing the internal structures. But so far the external surface is not explicitly described in the data-set, and there is highly sophisticated surface determination algorithms available to extract the exterior skin of an object. As the X-Rays lose energy, when they travel through the object, material greyscale of identical composition result in different values. That’s why a so called ISO-50 calculation of one average grey-scale value, in order to calculate the materials surface is not sufficient for metrology. You will always need the local adaptive surface determination to compensate for this natural fact, but it will deliver a measurement uncertainty down to a 1/10th of the voxel size.

So how big is a voxel? That’s easy to calculate, if you have a 2k by 2k detector, and your part is 200mm by 200mm, your voxel is 0,1qmm. Using special calibrations, metrology grade CTs, a reliable environment, thorough texting has proven sub micron accuracy being achievable with cooperative materials and sizes of the specimen.

That brings us to the boundary conditions of a CT. What parts and what materials can be measured, and what size? Parts made of any type of plastics and polymers, also ceramics, but really often castings and light metals like aluminum are a great input, as well as composite materials. The closer to lead (which is the shielding material for CT’s and Superman!) the materials are ranked in the  periodic system, the harder it gets to acquire decent data.

CAD Import with PMI Information

But the more energy you need, the larger the focal spot of the X-Ray Source gets, reducing the possible resolution of the setup. With a CT you have the ability to scale parts up by positioning them closer or further away from the tube, remember those pictures of insects looking like T-Rex after acquisition. Also you can combine acquisitions, when the part doesn’t fit a single acquisition, only then you need a large CT or an open setup. But most of the time a single turntable setup or a helix scan is used these days.

Multi Material components, especially consisting of metal and another material, are more difficult to scan, especially for metrology purposes. Low energy doses needed for plastics do not penetrate the metal, and high doses will go straight through the plastics, so the contrast needed to extract the boundary of the different ingredients of the specimen gets more tricky. It is certainly possible, but not as seamless as with single material parts. One help is using automatically the assembly of a CAD file, to find the boundary of such combined parts. Multi-energy and multi-setup approaches can be very helpful.

And who makes sure this is metrology? Certainly organisations like the NIST, VDI, PTB and NPL are heavily involved in norming the uncertainty statements and methods for calibration. A quick look at VDI/VDE 2630 shows some of the ongoing efforts.

The best proof though is successful implementation by many companies worldwide, from R&D all the way to inline metrology with industrial computer tomography.

My personal recommendation? Take a part that you (think you) know very well, and approach a CT service provider. Compare the results, and be astonished about the data completeness and quality.

Where is may look like a large investment, don’t forget that the CT, once the parameters are set, runs by itself, and the metrology plans and templates can come right off a CAD file with PMI information. You might fit more than one part into the scanning chamber, the result is a 100% complete data-set with more information than you ever had. And the time needed is not hours anymore, it can be a few minutes, depending on the level of uncertainty your process lives with. And besides metrology, you can do structural stress analyses including the flaws, copy parts with reverse engineering, analyze fibers, porosity, material flow and create a true digital twin of your product.

For more information: