Industrial CT scanning: how it works, key parameters and standards

A radiograph tells you there is a pore in the casting. It rarely tells you how large the pore is, how deep it sits, or whether it will open up on the surface the cutter exposes during machining. Those answers decide whether the casting ships or stays in the warehouse. Industrial computed tomography (CT) exists to close that gap. From hundreds of projections acquired around the part, it reconstructs a volume in which the flaw has a measurable position, shape and size, and the internal geometry can be compared directly against the CAD model.

The same volume serves two different jobs, defect inspection and dimensional measurement, each with its own standards and its own questions. Both depend on how the volume is built, so that is the place to start.

From projections to volume: what an industrial CT scan reconstructs

A radiograph records the attenuation accumulated along every ray crossing the object, following the Beer-Lambert law, and everything the ray meets is superimposed on the same image. A pore two millimeters below the surface and one at the center of the section cast the same shadow.

CT resolves the ambiguity by rotating the part, or the source-detector pair around it, and acquiring hundreds or thousands of projections over a full turn. A reconstruction algorithm combines them into a volume of voxels, the volume elements in which each gray value represents the local attenuation of the material. The most common method is filtered back projection, in its FDK form for cone-beam geometry; iterative algorithms are slower and cope better with noise and incomplete data. The result is still an indirect measurement. ISO 15708-2:2025 says so in its section on the limitations of the method: absolute measurements, such as the size of an inhomogeneity or a wall thickness, have to be based on comparison with other measurement procedures. The German foundry guideline BDG P 203 says the same and recommends metallographic sectioning, which the scan helps place in the right spot.

The standards picture is more recent than many specifications suggest. The ISO 15708 series was reissued in full between October 2024 and June 2025: Part 1 covers vocabulary, Part 2 principles, equipment and samples, Part 3 operation and interpretation, Part 4 qualification of the system against the inspection task. On the American side, ASTM E1441, reapproved in 2026, is the general guide to the method. If your contracts still cite the 2017 or 2019 editions, the reissue does not change how the work is done; it changes the citation, which now points to withdrawn documents.

Cone beam or fan beam: two architectures, two examination practices

A cone-beam system pairs a cone-shaped beam with a flat-panel detector and captures the whole volume in one rotation. It is fast, and it is the architecture of almost every laboratory and shop-floor system. It pays for the speed with more scattered radiation and with the artifacts typical of large cone angles. Its examination practice is ASTM E3375, published in 2023.

A fan-beam system collimates the beam onto a line detector and reconstructs one slice at a time. It is slower, but collimation makes it far less sensitive to scatter, which is why it remains the preferred choice at high energies and on massive parts. ASTM E1570 covers this geometry. A specification for a flat-panel scanner that cites only E1570 is citing the practice for a different architecture.

On the source side, focal spot size matters. Minifocus tubes serve castings and large components, microfocus tubes, with spots of a few micrometers, handle fine detail at high magnification, and nanofocus tubes go below one micrometer for microelectronics and materials research. For the most massive components, linear accelerators take over, at energies in the megaelectronvolt range.

Voxel size, focal spot and energy: resolution is decided by the part

Voxel size comes from geometry. BDG P 203 gives the formula: detector pixel size multiplied by the ratio of source-to-object distance to source-to-detector distance. Moving the part toward the source magnifies it on the detector and shrinks the voxel, as long as the whole object stays in the field of view. With the object framed in full, voxel size is therefore roughly its width divided by the detector's pixel count, and on the same machine a large component will never get the voxel size of a small one.

A voxel, though, is a grid, and the grid alone does not tell you what you will see. The same guideline sets the theoretical minimum: under ideal conditions a detail is recognizable at two to three voxels per side. Reliable detection takes more than that. On aluminum-alloy castings, Yosifov and co-authors (NDT&E International, 2026) measured the size detected with 90% probability at 95% confidence. It came to 100 µm at a 20 µm voxel size, five voxels, and 235 µm at 35 µm, almost seven. That gives the rule we use across all our CT articles. To guarantee a flaw, divide its size by five, or by seven if the voxel is coarse, and the result is the maximum voxel size to write into the specification. The worked example on a die-cast housing is in the article on porosity in castings and die castings.

At high magnification the limit shifts to the focal spot, because geometric unsharpness grows with the spot size times the magnification minus one: a 5 µm voxel with a 30 µm focal spot does not show 5 µm details. Small spots tolerate little power, and power is what gets through the material in a reasonable time. That is where energy comes in. ISO 15708-2:2025 requires at least 10% of the radiation to pass through the sample at every projection angle. The attenuation coefficients in the NIST XCOM database let you check in advance whether a material reaches that at a given thickness. The range runs from 225 kV, enough for polymers and light alloys, to accelerators for massive steel. A copper or tin filter on the tube window hardens the spectrum and reduces artifacts at the cost of intensity, which has to be recovered with more current or more time.

Artifacts: what degrades the volume and how to contain it

The best known is beam hardening. A tube spectrum contains many energies, the low ones are absorbed first, and the beam leaves the material harder than it entered: the volume shows brighter edges, an apparently darker core (the cupping effect) and streaks between dense regions. Filtration, software linearization and, in difficult cases, dual-energy acquisition keep it in check.

Scattered radiation works less visibly. It lowers contrast and skews gray values, especially in cone-beam systems on dense materials, and it is contained by moving the object away from the detector, with collimators, with software corrections or by switching to fan-beam geometry. Ring artifacts come from detector pixels with non-uniform response and are handled with flat-field calibration and small detector shifts during acquisition. Away from the central plane of a cone-beam geometry, FDK reconstruction is an approximation; the remedies are a smaller cone angle, helical trajectories or iterative algorithms. Very dense inserts and parts wider than the field of view, finally, produce streaks and wrong values, managed by reorienting the object and raising the energy.

None of these countermeasures removes the artifact; each brings it under control. ISO 15708-2:2025 lists artifacts among the limitations of the method and expects the operator to recognize the common ones. For anyone buying a scan the consequence is practical: achievable quality is judged on the real component, not on a demo sample chosen because it scans well.

Porosity, internal dimensions, assemblies: what the volume delivers

The application that most often justifies a CT scan is porosity in castings, aluminum high-pressure die castings above all, and in additive-manufactured parts. For every pore the volume returns size, position and distance from the surface that will be machined, and the porosity percentage is calculated by functional zone. No standard says how much porosity is acceptable, though: the limit is written by whoever signs the drawing, as the article on porosity in castings and die castings with CT explains. And if the specification does not fix how pores are counted, two labs can return different numbers on the same casting, the case told in CT porosity: same casting, two different numbers.

Next to defect inspection sits dimensional measurement. From the surface extracted from the volume, the scanned component is compared with the CAD model in a color map, and dimensions and tolerances are measured even on channels and cavities no stylus can reach. Accuracy has to be demonstrated, however. Acceptance testing of CT systems as coordinate measuring machines currently runs through VDI/VDE 2630 Part 1.3, because ISO 10360-11, the standard dedicated to these systems, is still a draft. The uncertainty of each dimension, meanwhile, is determined case by case under Part 2.1. The subject is developed in the article on CT metrology of internal geometry.

When the drawing is missing, the same volume becomes an STL mesh, the basis of reverse engineering with industrial CT. On assembled products it shows where seals, clips, connectors and adhesive beads ended up without taking anything apart. On polymers and composites it reveals voids, fiber orientation and warpage against the mold, within the limits described in the article on CT scanning of plastic and composite parts. In electronics, microfocus and nanofocus systems look for voids in ball-grid-array solder joints and defects in wire bonds. The review by De Chiffre and co-authors in CIRP Annals (2014) documents these uses on cases ranging from mechanical engineering to heterogeneous materials.

CT, 2D radiography or ultrasonic testing: a table for choosing

The three techniques answer different questions, so read the table row by row, starting from the information you need.

When inspection has to fit inside a line's cycle time, the answer is often industrial radioscopy; the comparison between radiographic and ultrasonic methods is in the article on industrial radiography vs ultrasonic testing.

The spec sheet describes the machine without your part in the beam

The mistakes people make buying a CT system, or a scanning campaign, look like six and come down to one: the machine gets judged empty. Spec-sheet resolution is measured on a small sample the manufacturer picked, at maximum magnification. The resolution that matters is the one you get with your own part in the beam, and it is worse. BDG P 203 says so directly: stating the voxel size does not prove the resolution achieved, which has to be checked with a reference body placed on the component, at the wall thickness you care about. ASTM E1695, the test method for CT system performance, makes the same point from the other side. It measures spatial resolution and contrast sensitivity on a uniform cylinder, and its scope says it is not meant to predict whether a specific flaw will be detected in a specific application.

The other mistakes grow from the same root. Energy gets undersized because it is specified for today's typical workpiece, and one thicker casting in next year's mix is enough to fall below 10% transmission. Dimensional measurement goes wrong because the maximum permissible error stated in the quotation belongs to the machine, not the task. The uncertainty of one feature in one material is worked out under VDI/VDE 2630 Part 2.1:2015-06, which means repeated measurements on a calibrated workpiece. Cycle time goes wrong because the demo times the scan, while reconstruction and analysis often take longer. Software, training and service drop out of the comparison for the same reason: none of them is on the spec sheet.

The fix costs one shipment. Send the supplier three of your own parts (the thickest, the most complex, and one a customer has already rejected). Ask for more than images in return: the voxel size used, the resolution measured on the workpiece, and separate times for scan, reconstruction and analysis. A supplier who will run that test is telling you something. So is one who won't. Whether to buy the system or stay with an outside service is a separate calculation, with its own numbers, covered in the article on industrial CT: buying a system or using a service.

What to send us for an assessment

To tell you which voxel size is achievable and how to set up the scan, we need a few precise details:

• drawing or CAD model of the part, with overall size and weight;
• material or alloy and maximum thickness to penetrate;
• goal: defect inspection, dimensional measurement, reverse engineering or checking an assembly;
• smallest detail to detect, for example the pore diameter, or the dimensions and tolerances to verify;
• acceptance criterion or customer specification, if there is one;
• number of parts, frequency and the turnaround you need;
• any CT report you have already received, if you want it compared.

You can send them by email to info@dipitech.com or on WhatsApp, with drawings or photos attached.

An industrial CT scan is judged by the verdict it delivers on your part, and that verdict depends on voxel size, energy and counting rules chosen before the scan. If you are weighing a CT analysis on a casting, an internal dimension or an assembly, the next step is a scan of a real component. The contract industrial CT scanning service exists for exactly that, while custom industrial CT systems come into the picture once scans recur every week. Configurations and fields of use are collected on the industrial computed tomography page.