Industrial CT: When to Buy a Scanner and When to Use a Service

The die-cast batch has been sitting in the warehouse for four days. The CT report from the outside lab should arrive tomorrow, and the quality manager already knows how the next round will go: another shipment, another queue, more days of parts on hold. Computed tomography shows porosity and internal geometry in three dimensions, something 2D radiography cannot do. But when the same scene repeats every month, the question goes up to management in its most awkward form. Keep buying the service, or bring CT in-house?

Four pieces of information the company already has will decide it: what you need to know about the part (a dimension or a defect), how large it is and what it is made of, how many scans you need a year, and how quickly the result must get back to production. The price list comes later. Teams that work in the opposite order, starting from the spec sheet of the most capable machine, often end up paying for performance their parts will never use, or find out that the machine they chose falls short on the one part that matters.

CT for metrology or for defects: two machines under one name

Two trades share the CT label. The first is dimensional metrology: reconstructing the internal and external geometry of the part, comparing it with the CAD model, checking dimensions and tolerances on surfaces a probe cannot reach. The second is defect detection, meaning finding porosity, inclusions, cracks and disbonds and judging them against an acceptance criterion. They call for different calibrations, software and budgets.

For measurement, the reference point is acceptance testing of coordinate measuring machines. The part of the ISO 10360 series written for X-ray CT systems, ISO 10360-11, is still at draft stage (prEN ISO 10360-11). In the meantime industry applies VDI/VDE 2630 Part 1.3 of December 2011, which explains how to use ISO 10360 on CMMs with CT sensors. That test gives you the machine's maximum permissible error, and it has to be sensitive to the instrument's own geometry errors: a 2019 NIST study by Muralikrishnan and colleagues quantified how detector misalignments distort sphere-to-sphere distances measured on a cone-beam system. The uncertainty of one specific dimension, on that material and with that scan strategy, is a different quantity. It has to be determined case by case, for example with the procedure in VDI/VDE 2630 Part 2.1 of June 2015, and a quote for metrology work should address both.

Defect detection follows a different framework. The ISO 15708 series, reissued between late 2024 and 2025, covers vocabulary, principles and equipment, operation and interpretation, and system qualification. Under ASTM the examination practice depends on the architecture: E1570 applies to fan-beam systems, while for cone-beam systems with a flat-panel detector, the type most companies are evaluating, the reference is E3375, published in 2023. None of these standards says how much porosity a casting may contain. Customer and supplier set that limit on the drawing; for aluminum, magnesium and zinc castings the German foundry guideline BDG P 203 provides the syntax for writing it and leaves the values to agreement between the parties. Buying before settling this fork exposes you to two opposite mistakes: paying for metrological accuracy you will never use, or learning too late that the system cannot support a traceable measurement.

Voxel size and real resolution: the part dictates the machine

The figure everyone quotes is voxel size, the volume element of the reconstruction. BDG P 203 gives the formula: detector pixel size multiplied by the ratio of source-to-object distance to source-to-detector distance. In practice the voxel shrinks as the part moves closer to the source, as long as the whole part stays in the field of view. With a detector about two thousand pixels across, a casting 300 mm wide imaged in full cannot go below roughly 150 µm voxels (300 mm divided by 2,000 is 0.15 mm), while a 20 mm connector gets to around 10 µm. Same machine. Two different problems.

And the voxel is only the grid the volume is built on. The same guideline notes that under ideal conditions a detail becomes recognizable from two or three native voxels across, so 8 to 27 voxels in volume, and that stating the voxel size alone does not prove the resolution achieved: resolution measured with the part in the beam is usually worse than without it, because contrast drops. On the casting scanned at 150 µm, a 20 µm pore sits below the grid. It may still show up if contrast is high, through the partial-volume effect, but it cannot be measured. The tube focal spot also limits sharpness. A smaller spot gives crisper edges but tolerates less power, and power is exactly what you need to get through thick sections in reasonable time.

This is where material comes in. The NIST attenuation tables compiled by Hubbell and Seltzer in Standard Reference Database 126 show that the mass attenuation coefficient rises with atomic number and falls as photon energy increases, so thick steel, nickel alloys or copper need far higher energies than aluminum, polymers and composites. BDG P 203 matches tubes up to 225 kV with polymers and light alloys (steel only in limited thicknesses), tubes up to 450 kV with steel as well, and 1 to 15 MeV accelerators with steel and copper-alloy castings. Each step up in energy brings more shielding, more cost and usually a larger focal spot. No machine does everything. There is a right machine for a family of parts, defined by maximum envelope, material, thickness to penetrate and smallest detail to see, and starting there rather than from the catalog is what separates a well-aimed investment from an oversized one.

Artifacts, positioning and fixtures: what a demo does not show

A CT reconstruction carries artifacts no spec sheet lists. The most common one comes from the fact that the tube emits a spectrum of energies: the soft component is absorbed just below the surface, the beam hardens along its path, and the reconstructed volume shows brighter edges and an apparently darker core, the effect operators call cupping. Metal filters on the tube window reduce it, and for aluminum castings BDG P 203 recommends copper up to 50 mm of wall thickness and tin up to 100 mm, in sheets 0.5 to 1 mm thick. The filter also cuts intensity, which has to be offset with more tube current or longer exposure. The second family is scattered radiation, stronger in cone-beam systems with flat-panel detectors. It is reduced by moving the part away from the detector, with collimators, or with software correction.

How the part sits on the rotary table matters just as much. The same guideline notes that the severity of many artifacts depends on orientation relative to the beam, and that several scans in different positions can clean up the region of interest. That is why parts are tilted, so that no flat face stays parallel to the beam through the whole rotation, and mounted on low-density supports such as foam or polymer that add no material to the beam path. The part must also stay perfectly still. A motion artifact cannot be corrected afterward. For series inspection, multi-part fixtures load several samples into one volume and cut the time per part; the price is a wider field of view, which means a coarser voxel, and the arithmetic above applies here too.

CT scan time: the short part of the cycle

Tomography takes time. The system acquires hundreds or thousands of projections while the part rotates, often averaging several frames to reduce noise, and only then reconstructs the volume. Depending on resolution, material and required quality, a single acquisition runs from a few minutes to several hours. Then comes the work purchase comparisons tend to forget: reconstruction, segmentation of voids or surface extraction, comparison with CAD, the report. In many applications these steps take as long as the scan, sometimes longer.

Teams inspecting many identical parts use dedicated fixtures and optimized protocols and can get down to a few minutes per part, while development labs accept long cycles in exchange for extreme resolution. These are two different kinds of productivity, and each should be budgeted for what it is, ignoring the peak figure in the brochure. During a trial, time scan, reconstruction and analysis separately. The total tells you whether the system will keep pace with production or turn into an elegant bottleneck.

CT inspection service or in-house system: where the numbers shift

An outside lab needs no capital, brings trained expertise and takes care of calibration, maintenance and software updates. In return you pay per scan, you wait, you ship parts, and your part data leaves the company. For low volumes, constantly changing parts, prototypes and first-article inspection it remains almost always the sounder choice.

An in-house system changes the math in three situations: recurring scans on the same family of parts, critical turnaround because CT sits inside the production flow, or product-data confidentiality that outweighs cost. The investment, however, goes beyond the machine. There is the shielded cabinet or bunker, whose weight has to be checked against the floor. There are radiation-protection obligations, which in Italy derive from Legislative Decree 101 of 31 July 2020 implementing Directive 2013/59/Euratom and have to be defined case by case with a qualified radiation protection expert. Then come operator training, the service contract, analysis software licenses and floor space. Taken together, these items can amount to a significant share of the system itself.

Back to the batch on hold. To see which side of the line they are on, the quality manager can run two annual calculations with data already at hand. The first adds up lab invoices, shipping, days of parts on hold and engineering hours spent waiting for a report; the second covers depreciation of a system suited to the part family, staff, maintenance, licenses and radiation protection. As long as the first stays well below the second, the service wins without argument. When the two converge, a proper sizing exercise is worth doing, and often the most prudent route is a few months of targeted service on the critical part family, because the scan parameters developed there remain valid afterward. The exit condition works both ways. If volumes stay low and variable, the service has won; if they stabilize and turnaround becomes critical, the purchase has the numbers to back it.

2D radiography, ultrasonics or a CMM: when CT is overkill

On some problems CT adds cost and time without adding information. A crack with a known orientation, parallel to the beam, shows up sooner and far more cheaply on 2D digital radiography, and the same goes for much casting porosity when all you need to know is whether it is there. A dimension on an accessible surface is closed out by contact measurement, usually with lower uncertainty. Residual wall thickness comes from ultrasonic testing, in real time and without reconstruction. CT earns its keep where the value lies in the inaccessible interior and the complete 3D geometry: internal channels, undercuts, closed assemblies, porosity that has to be located relative to a surface still to be machined. The review of industrial applications published by De Chiffre and colleagues in CIRP Annals in 2014 documents this across cases ranging from mechanical parts to electronics and inhomogeneous materials.

What to prepare before requesting an evaluation

A useful evaluation starts from a few concrete inputs: drawing or CAD model, material, maximum thickness to penetrate, envelope. Then the detail of interest, written as a number, meaning the minimum defect size to detect or the dimensions and tolerances to verify. For porosity you also need the counting rule, that is, the reference volume and the size below which a pore is ignored, which the customer is expected to set on the drawing. Annual volumes, frequency, acceptable turnaround and the quality system the data will live in complete the picture.

With these inputs the choice between service and purchase is made on numbers, and the configuration is sized to the part instead of chasing the most powerful machine. Configurations and fields of use are collected on the industrial computed tomography page. If your volumes sit in the zone where the two calculations converge, the next step is an evaluation on real data, with a neutral comparison of both routes before committing budget.