CT porosity: same casting, two different numbers. What to write in the specification

A CT report can follow the foundry industry's reference guideline to the letter and still not say which algorithm produced the porosity figure it contains. BDG P 203, the document with which the German foundry association moved porosity assessment from the section plane to the volume, lists the laboratory, the procedure, the part, data quality and results among the minimum contents of a test report. The evaluation software, its version and the segmentation method appear in a different list, the laboratory's internal documentation, and handing that data to the customer is left to an agreement between the parties.

Buyers and suppliers of castings find out what this means when the same part goes to a second laboratory and comes back with a different percentage. Neither lab has made a mistake, and often neither can prove it: the two numbers come from choices the purchase specification never fixed. Knowing what those choices are is the cheapest way to avoid a dispute, because nearly all of them can be settled with a few extra lines in the supply specification.

The number comes from a threshold, and nobody prescribes the threshold

A CT scan returns a volume of grey values. To count pores, someone has to decide, voxel by voxel, what is material and what is void. That step is segmentation, and the guideline names it as the precondition for any quantitative analysis. Section 6.2 separates global thresholds, identical across the whole volume, from local ones, and names three global methods. The first is Iso-50%, described as the simplest and most widely used, which sets the threshold halfway between the air peak and the material peak in the grey-value histogram. The other two, Otsu and Kittler-Illingworth, derive the threshold statistically from the same distribution.

The guideline makes none of them mandatory. What it does define is who judges the result: an experienced examiner assesses visually whether segmentation has succeeded and, in doubtful cases, compares it with metallographic sections, bearing in mind the limits of that comparison. The criterion ends, in other words, in a visual judgement. That is a sensible choice, since no algorithm works on every casting, but it has a practical consequence: two competent examiners using two different software packages can settle on two different thresholds and both sign off work that was carried out correctly.

Where the halfway threshold stops working

The guideline itself shows a case in which the most common method fails. It is a stepped aluminium cylinder, cast and deliberately gassed, where the different wall thicknesses appear from light to dark because of beam hardening, the progressive loss of lower energies as X-rays pass through more metal. With a background that shifts from step to step, a single threshold cannot suit all of them, and on the upper steps Iso-50% segmentation of the porosity does not succeed. The document draws a general point from this: with strong artefacts a global threshold becomes difficult or unworkable, and local thresholds that adapt to each voxel's neighbourhood take over. It also notes that the cylinder is not a standardised test piece, so it should be read as an example and nothing more.

The second weak spot concerns good castings. A review published in 2021 in Measurement Science and Technology by Jaques, du Plessis and colleagues observes that the halfway threshold works when the histogram has two recognisable peaks and breaks down at low porosity, when the void peak can no longer be told apart from the rest. That describes most parts that reach inspection: few pores, small ones, close to the limit of what the voxel size allows.

There is a way to tie the threshold to something that does not depend on the operator. The guideline describes it in the same section: a reference body with a known internal volume is scanned together with the casting, and the correct global threshold is worked back from the measurement of that volume. The example in the appendix uses seven aluminium spheres inside a hollow cylinder of the same material. It is the only passage in the document that links the threshold value to an independent quantity, which makes it the first one worth calling up in a specification.

One study in four says how it counted the pores

The figures in that review show the problem is far from theoretical. Out of 53 scientific papers reporting CT porosity measurements, 13 give all the parameters needed to repeat the measurement: one in four. Otsu is the most used method, in 36% of the studies, with manual thresholding close behind at 34%. The authors group the sources of uncertainty into three families (acquisition, reconstruction and segmentation) and write that there is no consensus on the ideal thresholding method, and that existing metrology standards give no specific guidance on the uncertainty of segmenting 3D CT data.

The most cited experimental check is a 2019 round robin published in Additive Manufacturing: ten laboratories, the same three titanium alloy parts made by laser powder bed fusion, the same protocol supplied to everyone. The review sums up the outcome in two halves. On pore distribution, meaning where the pores are, agreement is good; on quantitative values the differences are substantial. The scope needs stating, because this is additively manufactured titanium and the results do not carry over unchanged to a light-alloy die casting. The mechanism, though, is the one at work on any casting: tomography says with confidence where the voids are, and with far less confidence how much volume they occupy, until the way of counting them is fixed.

Gap, distance to surface and shape factors: the parameters that give way first

Porosity percentage is only one of the parameters in a P 203 key. On machined and sealing surfaces, the distance between adjacent pores and the distance between a pore and the surface matter more, and for both the guideline states that the calculation rule must be agreed between supplier and customer. It adds two warnings worth reading before an order is signed. The conventional method for pore spacing measures the distance between the circles drawn around the pores and not between the pores themselves, to the point where the circles can overlap and return apparently negative distances; the document calls it insufficient. Pores close to the surface, or inside thin remaining walls, can also be assigned by mistake to the outside air, and so drop out of the count exactly where they do most harm.

On shape factors such as sphericity and compactness the note is sharper still. They depend heavily on resolution, because a pore described by a handful of voxels has a shape that is largely the shape of the grid, and they should be used only comparatively. A sphericity limit written on a drawing without the voxel size it applies to is therefore a requirement a second laboratory cannot verify. The link between voxel size and minimum detectable defect is covered in our article on porosity in castings and die castings, which is best read alongside this one.

What reaches the test report, and what stays in the lab

Section 8.1 of the guideline lists the minimum contents of the report: laboratory and examiner, procedure, part identification, date and time, date of the last system stability check, CT data quality, image comparison with and without segmentation, results. Data quality has two specific entries. Spatial resolution must be determined on the part, not read off the machine's data sheet; contrast is expressed through the contrast-to-noise ratio, a quality factor defined by the document, or the image quality indicators of the ISO 19232 series.

Section 8.2 covers internal documentation instead: voltage, current, prefilter, voxel size, number of projections, reconstruction parameters, evaluation software with version and segmentation algorithm. Its stated purpose is reproducibility on the same type of CT system, and handing it to the customer requires an agreement. This is where the fact that opened this text comes from. A compliant report can contain the number and the segmented image without the data another laboratory would need to redo the count. Anyone who wants to be able to compare two reports has to ask for that data beforehand, in writing.

Method standards qualify the system and stop there

In negotiations the matter is sometimes closed with a reference to a standard: the system is qualified, so the number is reliable. The standards say something narrower. ASTM E1695 measures CT system performance through the modulation transfer function, derived from the edge of a homogeneous cylinder, and a contrast discrimination function; its significance and use section states that it describes overall system performance and is not intended to predict the detectability of specific features or flaws in a specific application. ASTM E3375, the 2023 practice for cone beam CT examination, sets minimum requirements, calls for a system performance baseline and its monitoring, and refers several points to agreement between purchaser and supplier.

On the ISO side, the 15708 series was reissued in 2025, with part 2 on principles, equipment and samples and part 3 on operation and interpretation. Anyone whose specification cites the standard generically, without part and edition, would do well to update it. None of these documents contains the criterion by which a pore is counted or a casting rejected. They qualify the instrument and the way it is used; the criterion remains a matter for the specification, as it does in dimensional metrology with CT, where the maximum permissible error is declared by the manufacturer and the acceptance test has to be chosen at quotation stage.

The second line of the specification

The porosity key on the drawing says how much void is allowed and in which zone. A second line is needed, in the supply specification or the control plan, saying how it is counted. The list below is an example we compiled from the sections cited above and from the six items the 2021 review asks to be reported every time; it should be adapted to the part and agreed with the laboratory.

• Segmentation method: global or local, with the name of the algorithm, the software and its version; the threshold value when it is global.
• Threshold anchoring: a known-volume reference body scanned with the part, or comparison with a metallographic section on an agreed sample.
• Voxel: maximum size allowed for the scan and minimum pore size included in the count.
• Calculation rules: how pore spacing and distance to surface are measured, and how pores open to the outside are handled.
• Shape parameters: allowed only with the voxel size they apply to, and only for comparison between parts scanned the same way.
• Report contents beyond the minimum: histogram, a slice with pores highlighted, description of the procedure (manual, semi-automatic, automatic), uncertainty estimate, scan and reconstruction parameters.

A serious laboratory accepts these requests without difficulty, because they describe data it already holds. They have a cost, and the guideline says so openly when it warns that specifying porosity under its rules brings testing obligations whose cost must be weighed. For a structural casting with no sealing surfaces and porosity well away from the limits, a specification this detailed is probably out of proportion, and a comparison scan against a reference part is enough. It becomes necessary when the measured value approaches the limit, when more than one laboratory is involved, or when the part has already been the subject of a dispute.

What tomography will not tell you anyway

Two limits remain that no specification removes, and the guideline states both. Area porosity and volume porosity as a rule do not match and are not comparable: anyone moving from section-based inspection to tomography cannot expect the same number, and a limit born for micrographs cannot be carried over to the volume by changing the unit alone. Tomography also provides no information on whether a casting is leak-tight. It can show that a pore sits a few tenths of a millimetre from the machined surface, but whether that casting leaks is settled by the leak test.

Within these boundaries the practical rule is easy to state. A porosity number is worth as much as the description of how it was obtained, and a report without that description can be used to decide on that part but not to compare it with another report. If you already have one in hand, perhaps two that disagree, that is the place to start: send us the report and the drawing with the porosity key and we will check which data is missing to make it comparable and what to ask the laboratory, before any talk of a new scan or an in-house industrial CT system.