High-temperature materials in 3D printing for fluid-carrying components

Additive manufacturing arrived in test bench engineering long ago — as a fixture, as an adapter, as a flow straightener. But as soon as a printed part carries a medium under pressure, the question changes fundamentally. It is then no longer about freedom of form, but about leak tightness, reproducibility and the question of which property value you may put into a design.

This paper collects what can be substantiated on the subject. It is deliberately written to be uncomfortable: the interesting figures are not the ones used in advertising, but the ones one line further down in the same table. All values are traced back to material data sheets, standards or peer-reviewed publications; where a figure in common use could not be substantiated, it does not appear here.

01 · What “temperature-resistant” means on a data sheet

The heat deflection temperature — HDT to ASTM D648 or ISO 75 — is always quoted together with a load case. A bending bar is tested under a defined outer fibre stress; the temperature at which a specified deflection is reached is measured. Two load cases are common: 0.45 MPa and 1.8 MPa. The difference between the two is negligible for some materials and dramatic for others.

A high-temperature resin for stereolithography is listed with a heat deflection temperature of 238 °C. That value applies at 0.45 MPa, after post-curing and additional oven annealing at 160 °C for 180 minutes. The same material, the same specimen, measured at 1.8 MPa: 101 °C. Elongation at break is 2.3 %.^[16]

A second trap lies in where the figure comes from. Data sheets from the polymer producers quote injection moulding properties. Data sheets from the machine and filament manufacturers quote properties of printed specimens — and then additionally state the part orientation. Both sets of figures carry the same units and the same standard numbers, but mean different things. One PEEK filament data sheet explicitly points out that only density and diameter relate to the filament itself; the mechanical values are polymer properties, not properties of the printed part.^[9]

02 · The materials compared

The following overview lists only values that appear in the data sheet named in each case. Empty fields mean that the value is not stated there — not that it is zero.

Two rows deserve a second look. PAHT CF15 has a glass transition at 70 °C.^[14] A water-glycol circuit with an 80 °C flow runs this material above its glass transition — regardless of the fact that the HDT is 92 °C. And PA 12 from the multi jet fusion process, at 48 MPa in both directions, is the only polymer material in this list in which the build direction does not determine the strength.^[15] There it determines something else instead — the leak tightness, see section 04.

Metals

For permanently pressure-retaining components there is no way past metallic powder bed fusion. The data situation there is considerably more solid, because the data sheets state the as-built condition and the heat-treated condition separately.

03 · The Z direction: the most expensive property

Every component built up layer by layer is anisotropic. Within a layer there is continuous, fully melted material; between two layers there is an interface whose strength depends on the degree of fusion. In a pressurised pipe the hoop stress — the largest principal stress — lies exactly in that interface if the pipe is built standing up.

The figures are in the data sheets, you only have to put them side by side:

• ULTEM 9085: XZ 69.2 MPa against ZX 39.4 MPa. The Z strength is 57 % of the in-plane strength; elongation at break falls from 5.4 % to 1.9 %.^[1]
• ULTEM 1010: XZ 79.2 ± 4.9 MPa against ZX 28.2 ± 8.8 MPa. That is 36 % — with a standard deviation amounting to 31 % of the mean.^[2]
• PAHT CF15: XY 103.2 MPa against Z 18.2 MPa in the dry condition. That is 17.6 %.^[14]

The consequence for the engineering design is uncomfortable but unambiguous: for a pipe under internal pressure made from an FFF material, the lying build is the only defensible one. That moves the problem into the support structures and into the surface quality of the channel wall underneath — see section 05.

With PAHT CF15 a second effect comes on top, and it is regularly overlooked in test bench applications: conditioning at normal room climate for 72 hours alone lowers the XY tensile strength from 103.2 to 62.9 MPa and the modulus of elasticity from 8386 to 5052 MPa.^[14] In direct contact with water-glycol the moisture uptake is orders of magnitude higher; the conditioned values are therefore still the favourable limit.

04 · Leak tightness is not a material property

The central finding of this paper comes from work by the Karlsruhe Institute of Technology on thin-walled fluidic components made by laser powder bed fusion. Its title is at the same time its key statement: Gas-tight means zero defects.^[23]

The work distinguishes two thresholds — gas-tight below 10⁻⁸ l·mbar·s⁻¹, liquid-tight below 10⁻⁴ l·mbar·s⁻¹ — and finds for 316L:

• Reproducibly gas-tight from a wall thickness of about 300 µm. At 200 to 216 µm the reproducibility fell markedly.
• Gas-tight specimens required a line energy above about 320 J·m⁻¹.
• The overlap of the melt tracks of two layers must be above 60 to 70 % , otherwise lack-of-fusion defects occur.
• At 30° inclination to the horizontal, gas-tight specimens were still produced, but with lower probability.
• The best compromise parameter set delivered 78 % leak-free specimens — 18 out of 23. The thinnest-walled set delivered 48 %.

For polymer processes the threshold lies elsewhere, but it is just as sharp. A study of PA12 from the multi jet fusion process tested walls from 0.4 to 0.7 mm over one hour at 20 ± 2 °C.^[24] The result is not a curve but a cliff — and it hangs on the build direction.

The authors' scanning electron microscopy confirms the finding in the microstructure: specimens built vertically show markedly more pores. Their conclusion is that the orientation has “a great impact on the leak values, with the vertical orientation being the most critical”.^[24]

Even where the strength looks good, the leak tightness need not. Laser-sintered PEEK reaches 88.7 MPa tensile strength against 100 MPa in injection moulding — with a total porosity of 4.36 % and partially sintered particles on the fracture surfaces.^[20] For strength, distributed porosity is largely uncritical; for leak tightness it is the opposite.

The question of reproducibility

The practically most important figure on process scatter comes from a parameter study of PEEK in the FFF process. Across 16 parameter combinations the porosity varied between 0.12 % and 10.8 %, the crystallinity between 20 % and 27 %, and tensile strength and modulus of elasticity by up to 27 %.^[22] The same material, the same machine, a factor of around 90 in porosity — from the parameter set alone.

In addition: a review of PEEK in FFF quotes pore volume fractions of 6.7 % for XY printing and 7.6 % for Z printing, with the individual pores in Z being markedly larger. It also shows why a heated build chamber is not a comfort feature for these materials: only at a chamber temperature of 200 °C did the modulus of elasticity exceed that of the injection moulded material.^[21]

05 · What happens inside the channel

The diameter you do not get

A printed channel does not have the diameter shown on the drawing. A non-destructive study of L-PBF channels with a 1.5 mm nominal size found a geometric mean diameter of 1.31 mm — 190 µm less — and a functional equivalent diameter of 1.27 mm. The mean topography height of the channel wall was 23 to 24 µm, largely independent of the channel diameter.^[26]

Roughness and the limits of Moody and Colebrook

The question of whether AM channels can be calculated with the classical pipe friction approaches has been investigated experimentally. Work on semicircular channels made by selective laser melting, with relative roughnesses between 0.0023 and 0.0044, finds:^[25]

• The transition from laminar to turbulent flow already occurs at Reynolds numbers around 1400 to 1800 instead of around 2300.
• For the smoother 4 mm channels the Blasius approach was usable; for the rougher 2 mm channels it was out by over 40 % .
• Filleting the channel corners raised the friction factor by an average of about 9 %.

A second study uses the Colebrook-White relation in the reverse direction — to back-calculate the actual hydraulic diameter from flow measurements — and thereby achieves a deviation of under ±3 % against computed tomography.^[26] That is the practicable route: the printed channel has flow put through it and is characterised from that, instead of being calculated from the drawing.

Self-supporting cross-sections instead of support structures

Support structures in internal channels cannot be removed after the build. The solution lies in the shape of the cross-section: surfaces with less than 30° inclination to the build plate cannot be manufactured; a diamond-shaped cross-section with a 45° profile angle and a teardrop cross-section with 50° are self-supporting. The price in pressure drop at Re = 8000 against the circular channel is 22.5 % for the diamond and 8.6 % for the teardrop. At the same time the circular channel shows the largest shape deviation in the overhang — around 300 µm — and the highest downskin roughness, with Ra values between 5.4 and 40.5 µm depending on position.^[27]

Residual powder

In powder bed processes, powder remains in internal channels. A study of channels from 1 mm diameter with helix and U-bend geometries shows: hand-held ultrasonic polishing clears straight channels in under ten seconds, but fails at sharp U-bends. The most effective method for laser powder bed fusion was the combination of an ultrasonic bath and vacuum boiling.^[28]

What is decisive for acceptance is the detection limit: a balance detects about 1 mm³ of powder — that is around 100,000 particles. X-ray computed tomography detects powder quantities an order of magnitude below that.^[28]

06 · Chemical resistance and the role of annealing

For PEEK a differentiated resistance matrix from the polymer producer is available. It rates on three levels — no, slight, strong interaction — and across several temperatures:^[7]

For printed materials the picture immediately thins out. The chemical resistance comparison of a large FDM supplier rates ULTEM 9085 as “limited” against coolant, engine oil and petroleum-based fuels in each case, while ULTEM 1010 and the PEKK grades do better.^[4] That contradicts the widespread assumption that PEI is generally chemically resistant. More notable is the reservation the same supplier states: the ratings are relative and “not based on any specific testing”.^[4] The table is therefore a preselection document, not a document of proof.

What annealing achieves

The most informative substantiated test comes from a study of printed PEKK to ASTM D543, 168 hours immersion, followed by tensile testing to ASTM D638, in both orientations and in each case stress-free as well as under 1 % strain:^[5]

Dichloromethane makes the unannealed material swell by 30.6 % in thickness and destroys it mechanically. After annealing, the change in mass in the same medium falls to 0.4 to 0.6 % and the change in strength to −15 to +1 %; all the other media stay between −19 and +19 %. The manufacturer summarises that annealing removed the chemical sensitivity.^[5]

For fuel contact the finding remains unsatisfactory nonetheless: there is no solid data base other than your own test to ASTM D543, with your own part orientation and your own annealing condition. Material data sheets are of no further help here.

07 · Post-processing and its limits

The common process for smoothing internal channels is abrasive flow machining. A study of additively manufactured titanium quantifies the benefit and the side effect:^[29]

• In a 10 mm through-hole, Ra fell over 15 cycles from 10.66 to 6.32 µm, that is by 40.7 %; the standard deviation of Ra fell by 64.2 %.
• In 4 mm channels with bends the improvement was 20.4 % (straight) to 38.5 % (with three bends).
• Side effect: The 4 mm channels widened by 6 % in radius after 15 cycles; slot widths grew by up to 0.216 mm.
• Limit: Walls below 0.4 mm were already permanently deformed after five cycles.

08 · Standards, pressure equipment and approval

The framework of standards

The terminology is governed by ISO/ASTM 52900.^[30] For the purchase of AM parts — that is, for what a test bench builder usually needs — ISO/ASTM 52901 is the relevant standard.^[31] It specifies which information has to be agreed between purchaser and supplier. ISO/ASTM 52920, by contrast, concerns the process and the production site, not the individual part.^[32]

For the mechanical evaluation of metallic AM materials, ASTM F3122 exists as a guide.^[34] For a continuous qualification path from material through process to part, DNV-ST-B203 in edition 2025-11 is the most complete publicly obtainable basis.^[41] Freely accessible and, despite their spaceflight background, usable for part classification are NASA-STD-6030^[35] and MSFC-STD-3716.^[36]

The finding on pressure retention

Here lies the result that is most often reported incorrectly. ASME PTB-13-2021 is a criteria document, not a code: it describes which points a future standard for pressure-retaining metallic components made by powder bed fusion would have to address.^[37] It also limits its criteria to low-temperature applications — defined as a design temperature at least 25 °C below the temperature above which time-dependent material properties become governing.^[38]

ASME Code Case 3020, frequently cited as evidence, according to the account of Argonne National Laboratory does not concern laser powder bed fusion but the procedure qualification for metallic gas-shielded arc weld overlay, approved in May 2021.^[38] According to the same source, no formal ASME qualification exists for L-PBF in 316L; the development of a corresponding Code Case is waiting for additional fatigue tests on machined against as-printed specimens.^[38]

Pressure Equipment Directive

For the European market, Directive 2014/68/EU is the governing text.^[39] For test bench engineering, Article 4(3) is the decisive sentence: pressure equipment and assemblies that do not reach the category thresholds must be “designed and manufactured in accordance with the sound engineering practice of a Member State”.^[39] No conformity assessment by a notified body applies there — the manufacturer demonstrates fitness for purpose, and an additively manufactured part is admissible if that demonstration holds.

Above the thresholds it becomes laborious. Because AM materials are not listed in any harmonised material standard, the only routes left are a European approval for materials under Article 15 or a particular material appraisal by the notified body.^[39] One test house describes, for the second route, a certification programme based on the analogous application of EN 13445-4 section 3.1 and reports a pilot audit for laser powder bed fusion at a pump manufacturer.^[40] For one-off items in test bench engineering, neither is as a rule economically feasible.

Media contact and drinking water

Substantiated approvals for individual AM materials do exist — but not for the application most frequently asked about in plant engineering. Demonstrated are, among others: ULTEM 1010 in a certified version with NSF 51 as well as ISO 10993 or USP Class VI^[3]; a PEEK filament as USP Class VI compliant and tested to ISO 10993-1^[9]; PA 12 for multi jet fusion with USP Class I–VI.^[15]

For drinking water the picture is different. The German assessment basis KTW-BWGL is available in its 6th amendment of 2025-11-20^[42], in the United Kingdom material approval runs through WRAS on the basis of BS 6920^[43]. We were unable to substantiate any additively manufactured material with KTW-BWGL conformity, WRAS approval or NSF/ANSI 61 certification. The NSF entries in the polymer data sheets refer to injection moulding grades of particular base resins^{[11] [12]}, not to their additive processing form.

09 · Where we use it — and where we do not

From the above a fairly clear division follows. It is our conclusion, not that of the sources cited.

Not suitable

1. Continuous pressure with the build direction carrying the load. ULTEM 1010 in Z: 28.2 MPa with ±31 % scatter.^[2] There is no safety factor that absorbs such a distribution.
2. Thin walls below the respective threshold. Below 300 µm for metallic powder bed fusion^[23], below 0.6 mm for vertically built MJF PA12^[24] there is no gentle transition, but a cliff.
3. Sustained load at temperature. Creep data for printed polymers at operating temperature could not be substantiated. For PEEK, low creep at 23 °C and around 60 MPa over 1000 hours is documented^[8]; beyond that, for printed material with 0.12 to 10.8 % porosity^[22] there is no public basis for design.
4. Fuel carrying without your own testing. The manufacturers' tables are explicitly not based on testing.^[4]
5. Drinking water carrying. No substantiated AM material with KTW-BWGL, WRAS or NSF/ANSI 61.
6. Pressure-retaining parts above the PED category thresholds — there the effort for a material approval or a particular material appraisal is not feasible for one-off items.^[39]

Sensible

1. Flow straighteners and manifolds with complex internal geometry below the category thresholds, with a teardrop cross-section instead of a circular one — 8.6 % additional pressure drop against a considerable reduction in the number of parts.^[27]
2. Hot air and gas lines without appreciable internal pressure; there ULTEM 1010 plays out its 212 °C at 1.8 MPa.^[2]
3. Adapters and test fixtureswhere the leak test is part of the acceptance anyway.
4. Metallic parts from 316L or IN718 where the geometry cannot be produced conventionally — with a hundred per cent leak test, documented powder removal and the wall thickness as a test dimension in its own right.

10 · Sources

All sources were accessed in August 2026. Material data sheets are tied to revisions; before a design, the current version should be obtained from the manufacturer. Values from injection moulding data sheets are marked as such throughout this paper and do not apply to printed material.