Specifying UQD and Blind-Mate Quick Disconnects for Liquid Cooling

A quick disconnect is the cheapest component in a liquid-cooled rack and the one most likely to take it down. Here is how to specify UQD and blind-mate couplings so that "OCP compliant" actually means something — and how to catch the ones that quietly are not.

Every liquid-cooling chain has a component whose failure cost is orders of magnitude larger than its purchase price. The GPU cold plate is expensive but rarely fails. The CDU is expensive and is specified carefully. The quick disconnect (QD) coupling is a few tens of dollars, gets bolted on at the end of the bill of materials, and is the single most common origin point for the leaks that reach the silicon.

This article is a working specification guide for UQD and UQDB couplings in direct-to-chip deployments. It covers what the OCP standard actually guarantees versus what it leaves to the vendor, why "compliant with UQD Rev 1.0" is not the same as "interchangeable," how to size a coupling so it does not silently consume your CDU's pressure budget, how to choose seal compounds for the coolant you actually run, and what the failure data says about where to spend money. It closes with a procurement checklist and a commissioning sequence.

Why the smallest part carries the largest risk

Start with the arithmetic, because it explains why this component deserves more engineering attention than it usually gets. A 16-tray liquid-cooled rack has two couplings per tray — one supply, one return. Add the manifold inlet and outlet and the CDU connections, and a single rack carries roughly 36 fluid break points. Scale that to a 100-rack hall and the facility contains on the order of 3,600 pressurised mechanical joints, each with an O-ring, a spring-loaded valve, and a latch.

Joints outnumber every other wetted component in the loop by an order of magnitude. That alone would make them the statistically dominant failure origin, and the reliability literature bears it out: cracked fittings, failed quick-disconnect seals, fatigued hoses, and incorrectly torqued connections are described as the most common category of liquid-cooling failure precisely because the joint count is so high.^[1]

What makes the consequence disproportionate is the failure taxonomy. A leak is not a plumbing event; it is an electrical event that happens to start with fluid. A weep — millilitres per day — is the most dangerous form because it often escapes detection entirely while corroding board traces over months. A drip is visible and localised, usually survivable if point detection exists. A spray is what turns one joint failure into a multi-server event.^[1]

Quantify the spray case and the case for engineering rigour becomes obvious. At a typical CDU secondary-loop flow around 20 LPM, a fully open fitting failure releases about 20 litres (5.3 gallons) in the one minute it takes an operator to reach an isolation valve, and roughly 100 litres (26 gallons) if detection takes five minutes. Detection-driven supply shutoff inside 60 seconds cuts the same failure to about a fifth of its released volume.^[2] The coupling choice and the detection architecture are therefore one decision, not two.

What the OCP specification actually guarantees

The Universal Quick Disconnect standard originated as an Intel initiative inside the Open Compute Project, with the UQD Specification Rev 1.0 standardising key mechanical dimensions and baseline performance so that products from different manufacturers would at least physically connect.^[3] That distinction — physical interoperability versus functional equivalence — is the entire story of this component category, and it is where most procurement mistakes begin.

What Rev 1.0 pins down. The envelope and the floor. Four nominal sizes (UQD02, UQD04, UQD06, UQD08), maximum working pressure, minimum burst pressure, minimum flow coefficient, minimum cycle life, and a maximum fluid loss per disconnect. These are thresholds a product must meet, not the performance it will deliver.

What Rev 1.0 leaves open. Sealing geometry, spring rates, seal compound and cure system, surface finish in the wetted path, internal valve design, and the actual flow coefficient above the minimum. Two couplings can both be "OCP UQD Rev 1.0 compliant" and behave differently in the same loop.

The table below sets the specification floor beside the actual published performance of a mainstream compliant coupling. The gap is the point.

Table 1 — OCP UQD Rev 1.0 requirements^[3][4] alongside actual published vendor performance^[5]. Specifying to the minimum column means accepting a coupling with less than half the flow capability of a compliant peer product.

The fluid-loss number nobody checks

Fluid loss per disconnect is the most under-examined figure on a UQD datasheet, because at 0.02–0.07 mL it looks like nothing. Run it out over the specified service life and the picture changes.

Two operational consequences follow. First, at spec-limit performance a UQD06 coupling loses a drop every time a technician touches it — and a UQD08 loses more than a drop. In a rack serviced quarterly across 32 trays, that is a measurable and continuous coolant inventory drain, not a rounding error. Second, and more important, that fluid does not disappear. It deposits on the break surface. Every dry-break design leaves a thin film on the mating faces, and that film is where the contamination and corrosion story begins.^[6]

This is why the strongest products publish fluid loss an order of magnitude below the specification ceiling — one mainstream compliant line publishes 0.004–0.010 mL per cycle, which is between 5× and 7× better than the requirement at every size.^[5] When you compare quotes, this is a line item worth reading.

Hand-mate versus blind-mate: the decision is made by physics, not preference

OCP originally maintained UQD (hand-mate) and UQDB (blind-mate) as separate specifications. The distinction is mechanical, and it resolves itself once you look at the rear of a modern accelerator tray.

Hand-mate UQD requires a technician to align the coupling, push it home, and confirm the latch by visual, tactile, or audible feedback — the OCP specification requires feedback of at least one of those kinds. It is a simpler, more robust mechanism with fewer parts, and it demands skill and line-of-sight access.

Blind-mate UQDB connects automatically as the tray slides onto its rails. A guide cone and a floating mount absorb installation tolerance; the mainstream radial misalignment tolerance is ±1 mm, with some suppliers advertising 0.8 mm and others larger float for Open Rack V3 class assemblies.^[7][8] No visibility, no tools, no manual alignment.

The practical rule: once tray density and rear-cable congestion make manual alignment unreliable, blind-mate stops being an option and becomes the only workable architecture. The higher part count and cost of a floating mechanism are repaid in connection success rate, which is the metric that determines whether a hot-swap procedure is safe to run at all.

The direction of travel is one unified specification. A draft OCP UQD Specification v2.0 consolidates UQD and UQDB into a single document and adds a hybrid configuration in which a UQD plug mates directly with a UQDB socket, with backward compatibility to V1 hardware. Supplier documentation already references Rev 2.0 designs.^[10][11] OCP's stated intent is genuine part interchangeability rather than mere dimensional compatibility, with working groups continuing on both UQD v2 and interoperability.^[12]

The interoperability problem, and the fix that finally worked

Dimensional compatibility did not deliver interchangeability, and the industry spent several years discovering that the hard way. Because OCP's early work concentrated on mechanical dimensions, manufacturers diverged in sealing performance, flow characteristics, and material compatibility. Products from different brands could physically connect, yet their reliability in service varied.^[3]

In August 2025, Intel established the Universal Quick Disconnect Interoperability Alliance with five founding certified partners, explicitly targeting cross-brand interchangeability.^[13][14] The engineering findings from that programme are more useful to a buyer than the announcement:

• Testing concentrated on the UQD04 size — the mainstream in-rack coupling.
• Over roughly a year, more than 250 sample groups were subjected to more than 1,500 test cycles.
• Root causes found and corrected were not dimensional. They were spring stiffness, compression spacing, and O-ring material differences — the exact variables Rev 1.0 left open.
• The single most instructive defect: a pull-ring that failed to self-return after approximately 300 connect/disconnect cycles, so the coupling would no longer latch and required manual intervention. The fix was increasing the pull-ring spring design force so it could overcome friction across all mating combinations.^[14]

Sit with that last finding. A coupling can pass every dimensional check, seal correctly, and meet the specification — and still become unlatachable at cycle 300, which is 6% of its rated 5,000-cycle life. The failure is invisible on a datasheet because "latch self-return force" is not a line item in Rev 1.0. It only appears when two different manufacturers' halves are mated repeatedly, which is exactly the condition a mixed-source fleet creates.

Pressure drop: where a cheap coupling becomes an expensive one

The flow coefficient is the parameter that connects a coupling decision to the CDU you already bought. C_v is not a marketing number — it determines pressure drop at your design flow, and every coupling in series adds to the pump head the CDU must supply.

The relationship is standard hydraulics. Volumetric flow through a valve or fitting scales with the square root of pressure drop divided by specific gravity:

Now put real numbers through it. A 25% propylene glycol loop (SG ≈ 1.02) flowing a representative UQD04-size circuit at 1.0 GPM:

One point eight psi on one coupling looks trivial. Multiply it across the loop. A 16-tray rack carries 32 tray couplings plus manifold and CDU ends — call it 36 break points, and treat the tray pairs as the dominant series path. Thirty-two couplings at 1.80 psi of avoidable loss is roughly 58 psi (4.0 bar) of additional pump head demanded from the CDU, before you account for any of it being recovered. Compare that against a typical in-rack TCS loop operating band in the low hundreds of kPa^[15] and the coupling decision is consuming a large fraction of the available pressure budget.

Sizing the coupling to the circuit

Work backwards from flow, never forwards from a size label. The procedure:

1. Establish the design flow per circuit. Per-tray or per-cold-plate flow, in L/min or GPM, at the target temperature rise. Cold-plate reference flow rates are commonly quoted in the region of 1.25–2.0 L/min per kW of IT load.^[16]
2. Set an acceptable pressure-drop budget per coupling. Allocate it deliberately out of the CDU's total head, alongside the cold plate, hoses, and manifold.
3. Solve for the required C_v. C_v = Q / √(ΔP / SG).
4. Then select a size with margin — 20–30% above the computed requirement — and record the vendor's published C_v for that specific part number, not a family figure.
5. Verify the whole series path against the CDU head curve at the design flow, including the redundancy configuration you will actually operate.

Seal compound selection: match the chemistry, not the price

The O-ring is the only part of the coupling doing the actual sealing, and the most common cause of premature QD leakage is a seal compound that is chemically wrong for the coolant rather than a manufacturing defect. The failure mechanism is straightforward: a polarity mismatch causes fluid absorption, volumetric swelling, loss of tensile strength, and eventual leak.^[17]

Table 2 — Seal compound compatibility by coolant family. Ratings synthesised from liquid-cooling seal selection guidance^[17][18][19]. Always qualify the specific compound against the specific coolant formulation; a material class is not a compound specification.

Rules that matter in practice

• Direct-to-chip with water-glycol coolant: peroxide-cured EPDM. This is the mainstream choice and the one OCP-aligned UQD and BMQC products standardise on. Peroxide curing gives lower compression set than sulfur curing and avoids sulfur-induced contamination of the coolant. Target compression set below 30% after extended high-temperature exposure.^[18]
• Do not treat FKM as an upgrade. It is a different tool. Standard FKM grades can degrade in prolonged contact with hot water-glycol, and FKM compounds diverge widely in water-rich service — meaning FKM must be validated against your coolant data rather than assumed superior.^[19]
• Immersion and dielectric fluids: EPDM is disqualifying. EPDM absorbs non-polar hydrocarbon fluids and can swell by over 50% by volume, extruding the O-ring from its groove and causing the leak it was meant to prevent. Use FKM for mineral and synthetic hydrocarbon fluids.^[17]
• Fluorinated fluids are the hardest case. Their very low surface tension lets small molecules penetrate the elastomer network, extracting plasticisers and drifting dimensions. Standard FKM still swells; FFKM is the reliable choice, with modified high-fluorine FKM as a cost compromise only after immersion testing.^[19]
• "EPDM" is not a specification. Cure system, filler, hardness, and formulation all change compression set, swelling, and ageing behaviour. Once a compound is qualified, control it — do not accept an alternative EPDM formulation as automatically equivalent across thousands of installed couplings.^[19]

Materials, wetted path, and the corrosion question

Coupling bodies are typically stainless steel or aluminium alloy. The material discussion is less about the shell than about the wetted internals — springs, valve seats, and the O-ring groove surfaces.

• Wetted springs must be corrosion-resistant stainless. A spring is thin-section, highly stressed, and permanently in contact with coolant. It is the component most vulnerable to chloride-driven pitting, and its failure mode is a valve that does not reseat — which converts a dry-break coupling into a drip source.
• 316L stainless is preferred where chloride exposure is plausible. The added molybdenum improves resistance to chloride-induced corrosion relative to 303/304 grades, which is relevant because coolant chemistry degrades over time and inhibitor depletion changes the corrosion environment.^[8]
• Sealed, smooth wetted surfaces matter more than alloy choice. The OCP requirement is for sealed surfaces with no sharp edges or tool marks in wetted areas.^[4] A burr at the sealing face is a leak path and a particle generator simultaneously.
• Double O-ring sealing on blind-mate designs provides redundancy where the coupling is inaccessible for inspection — a sensible design response to the fact that you cannot see a blind-mate joint to inspect it.^[7]
• Helium leak testing is the quality gate to ask about. Some suppliers 100%-test every coupling by helium mass spectrometry rather than sampling.^[20] For a part with this failure consequence, a 100% test is worth paying for.

Colour coding — red for warm return, blue for cold supply, with matching O-ring and sleeve identification — is not cosmetic. It is the control that prevents cross-connection during a hot swap when a technician is working by feel in a congested rear rack.^[4][6]

Where the failures actually come from

Four mechanisms account for most quick-disconnect leakage in service. Recognising them tells you which specification lines to weight and which installation practices to enforce.

1. Particulate damage and foreign-object embedding at the sealing face

Metal swarf, weld spatter, and debris left in piping during fabrication travel with the fluid and act as micro-cutting tools against the O-ring, or hold the sealing faces apart. Industry analysis attributes the majority of liquid-cooling system stoppage events to scratched or contaminated sealing faces at quick-disconnect points. The defence is commissioning cleanliness and side-stream filtration, not a tougher O-ring — no elastomer resists an embedded metal particle.

2. Dynamic wear and thermal cycling

Every hot swap slides the seal against the metal valve core. Materials with inadequate abrasion resistance shed rubber fragments, which both clog narrow passages and accelerate valve sticking. Layered on top is thermal cycling: repeated temperature excursions and pressure pulses drive stress relaxation and micro-cracking in the elastomer. Both effects are cumulative, which is why cycle life is specified and why exceeding it has consequences.

3. Cross-brand tolerance mismatch and false locking

When couplings from different manufacturers are mated, small deviations in thread form and sealing face cone angle can produce a joint that feels and looks seated but is not fully compressed — a "false lock." The result is a gap that opens under vibration. This is precisely the failure class the Intel interoperability programme was created to eliminate, and it is why cross-mating test reports are the right evidence standard.

4. Installation error and incomplete seating

The simplest and most preventable. A partially engaged coupling can appear connected while the latch is not home, and it will leak under pressure. The specified control is locking-mechanism feedback — visual, tactile, or audible — and the specified discipline is verifying it before the loop is pressurised. On blind-mate interfaces, where no operator can inspect the joint, this control has to be designed into the guide geometry and confirmed by a pressure-hold test.

Detection: point sensors, not perimeter ropes

Coupling specification and leak detection are one architecture. A perimeter rope around the CDU tells you that fluid reached the floor — after it has travelled. Point sensing at the couplings, manifold junctions, and supply and return valves tells you which joint failed, in time to isolate it.

The economics favour point-level detection strongly. At 20 LPM, detection-driven shutoff within 60 seconds holds a full-flow failure to about 20 litres. Without it, the same failure runs to roughly 100 litres in the five minutes it takes to locate and reach the isolation valve.^[2] That five-fold containment difference is materially cheaper than building enough secondary containment to survive the slow-detection case.

Two additional controls belong in the same layer. First, treat the facility water system, the CDU secondary loop, and the in-rack loop as separately isolatable zones with valves at each boundary, so a fault in one does not escalate into the others. Second — and this connects directly to a different failure mode — the CDU heat exchanger must hold the technology-loop supply temperature above the hall dew point. Every ASHRAE liquid-cooling class permits supply water cold enough to be below room dew point in many halls, so where that interlock is missing, condensation produces the same electrical outcome as a leak with no leak to find. We cover the calculation in how to calculate dew point in a data center.

Commissioning and service sequence

A coupling that is correctly specified can still be destroyed by the commissioning procedure. The bracketed checks below close the gap between specification and installation.

1. Inspect every coupling on receipt. Seal faces free of tool marks, scratches, and burrs; colour coding present and correct; no debris in the bore. Reject, do not clean and install.
2. Confirm the part number, not the family. Match the delivered P/N to the C_v, seal compound, size, and termination you specified. Mixing terminations between plug and socket is a common receiving error.
3. Flush the loop before couplings are installed where possible. Use a temporary flush path. Driving commissioning debris through new couplings — and through the CDU pump and heat exchanger — is a documented way to damage equipment in its first week of service.^[1]
4. Pressure-hold test each zone independently. Zone-by-zone verification localises any leak to a known boundary instead of forcing a whole-hall investigation.
5. Verify latch feedback on 100% of hand-mate couplings — visual, tactile, or audible — before energising the loop. For blind-mate interfaces, verify engagement by pressure hold, since visual confirmation is unavailable by design.
6. Witness-cycle a sample and re-test. Cycle representative couplings through a fraction of their rated life, then repeat the pressure-hold test. This detects the latch-return class of defect that only surfaces after hundreds of cycles.^[14]
7. Record the baseline. Log coupling P/N, size, seal compound, and installation zone per rack. Without a parts baseline, mixed-vendor failure analysis is not possible and warranty claims cannot be substantiated.
8. Validate point sensors in the current cycle. A failed leak sensor reads clean. Coverage counts only for sensors verified active, and sensor validation belongs in the routine, not the incident response.^[1]
9. Set a service-life counter. Track cycles per coupling against the 5,000-cycle rating and the 10-year service life. A coupling in a hot-swap test rack will exhaust its rating long before the calendar life expires.^[4]
10. Pre-stage spares to the specified revision. Keep the correct size, C_v, and seal compound in stock, and note the compatibility caveat if you hold both V1 and V2 hardware.^[10]

Procurement checklist

Ten questions that separate a defensible coupling specification from a compliance claim.

The bottom line

The quick disconnect is where a liquid-cooling specification either holds together or quietly comes apart. Four things determine whether it does.

Specify performance, not compliance. "OCP UQD Rev 1.0 compliant" sets a floor that leaves C_v, sealing geometry, and spring behaviour to the vendor. Write the flow requirement, the pressure-drop budget, and the fluid-loss ceiling you actually need, and hold suppliers to the specific part number rather than the product family.

Sizing flows from the circuit, not the label. Compute the required C_v from design flow and an allocated pressure budget, then verify the whole series path against the CDU head curve. The difference between the specification minimum and a good compliant part reached roughly 58 psi of avoidable loss across one rack's worth of couplings in our worked estimate — enough to invalidate a CDU selection at commissioning.

Match the seal to the coolant, and control the compound. Peroxide-cured EPDM for water and water-glycol; FKM for hydrocarbon dielectric fluids; FFKM for fluorinated fluids. Then treat the qualified compound as a controlled item, because "EPDM" alone is not a material specification.

Test interchangeability if you intend to mix. Interchangeability is a tested property of a mating pair. The Intel alliance's most instructive finding — a latch failing to self-return at around 300 of a rated 5,000 cycles — is invisible to dimensional compliance and to any single-vendor certificate.

Get those four right and you have removed the most common origin of the leak that reaches your silicon. Get them wrong and you will find out at the worst possible moment: mid-hot-swap, at rack density, in a hall where a spray is not a plumbing problem but an electrical one.

Building a liquid-cooled rack or hall and need couplings, manifolds, or CDU plant specified to your actual loop conditions? Request a quote with your flow rates, coolant, and target densities, and we will work the hydraulic and material selection with you. Our rack-mount versus cabinet CDU comparison is a useful companion when the coupling decision pushes back on your distribution architecture.

References

1. Progressive Robot, "Liquid Cooling Failures in AI Data Centers: Causes, Risks, Prevention" — failure-mode taxonomy, joint-count dominance of QD failures, weep/drip/spray classification, point-versus-perimeter detection, commissioning flush damage, microchannel blockage thresholds. progressiverobot.com
2. US Made Supply, "Data Center Liquid Cooling Spill Response" — spill scenarios by volume, 20 LPM reference flow, five-minute containment arithmetic, five-layer containment design, ASHRAE TC 9.9 and NFPA 75-2024 framing. usmadesupply.com
3. Guchen Electronics, "UQD Blind-Mate Connectors: Driving Liquid Cooling Standardization" — UQD standard origin under OCP/Intel, Rev 1.0 scope limited to mechanical dimensions, divergence in sealing and flow between brands, OCP requirement table, ±1 mm UQDB radial tolerance, Intel alliance formation and scope. guchen-connector.com
4. UQD-USA, "UQD Series Products" — OCP Rev 1.0 performance requirements by size (C_v, fluid loss, pressure, temperature, cycles, service and shelf life), material and termination specification, validation test methods, IEC 62368-1 Clause G.15. uqd-usa.com
5. Amphenol, "Liquid Cooling Systems — Reliable Quick Disconnect Connectors, UQD Product Introduction" — published C_v and fluid loss by size for a compliant product line, wetted materials, seal options, mating force at zero pressure. farnell.com
6. Connector-Sensor, "Universal Quick Disconnect Couplings for AI & Data Center Liquid Cooling" — UQD specification interpretation, C_v guidance, termination options, colour coding convention, V1 versus V2 consolidation, application map. connector-sensor.com
7. CEJN, "UQDB couplings" — UQDB blind-mate technical data, ±1 mm radial misalignment tolerance, double O-ring redundancy, ISO 11926-3 ORB standard, IEC 62368-1 G.15.2.3 approval, EPDM seal, OCP contribution scope. cejn.com
8. Guchen Electronics, "UQDB Series Blind-Mate Floating Coupling" — floating compensation design intent, 316L stainless selection rationale for chloride resistance, zero-drip flat-face interface, nominal diameter range. guchen-connector.com
9. RapidAccu, "UQDB Blind-Mate Quick Disconnect" — clarification that the 1 mm figure is radial (lateral), not axial, that axial alignment is governed by insertion stroke and guide-cone length, and that latch verification is required before energising. rapidaccu.com
10. Connector-Sensor, UQD/V1-versus-V2 specification comparison — OCP UQD Specification v2.0 consolidating UQD and UQDB into one document, hybrid UQD plug to UQDB socket configuration, backward compatibility with the caveat that OCP recommends end-user assessment before mixing generations. connector-sensor.com
11. Amphenol Industrial, "UQD/UQDB REV2.0 | Liquid Cooling Quick Disconnects" (IAN-142) — supplier design conforming to OCP UQD/UQDB Specification Rev 2.0, hybrid configuration support, sizes 02–08, accidental-disconnection failure rationale. amphenol-industrial.com
12. Sector Signals, "OCP's Data Center Liquid Cooling Standardization: From Cold Plates and CDU to UQD" — UQD specification revision to v2.0, OCP Cooling Environments scope across five areas, Cold Plate Sub-Project participation, Meta ORv3 blind-mate EPDM hose evaluation (January 2026), vendor lock-in rationale. sector-signals.net
13. IT Home (IT之家), "Intel launches Universal Quick Disconnect Interoperability Alliance" (August 2025) — alliance formation, five founding certified partners, cross-brand compatibility objective, 250+ sample groups and 1,500+ test cycles, UQD design unification, TCO rationale. ithome.com
14. IT Times (IT时报), "Intel Universal Quick Disconnect Interoperability Alliance established" — UQD04 failure-mode analysis scope, pull-ring self-return defect at approximately 300 cycles and the spring-force design remedy, root causes traced to spring stiffness, compression spacing and O-ring material differences, rack lifecycle and server refresh mismatch. view.inews.qq.com
15. Sector Signals, "As liquid cooling becomes a baseline for AI servers" — in-rack TCS loop pressure operating band (140–450 kPa), ASHRAE coolant classes W1–W5, IEC 62368-1 leak testing and sensor requirements. sector-signals.net
16. Industry hydraulic standardisation analysis, AEE2026 / ODCC+OCP dual-track framework — cold-plate reference flow rate basis of 1.25–2.0 L/min per kW, ORv3 manifold branch flow balance tolerance within ±3%, UQD v2.0 dimensional and hydraulic requirements for DN04/DN06/DN08. Directional engineering figures. toutiao.com
17. Best Seal, "AI Data Center Liquid Cooling: EPDM & FKM Seal Compatibility With Synthetic Coolants" — polarity-matching principle, peroxide-cured EPDM for water-glycol with swelling under 3% by volume, EPDM swelling over 50% in hydrocarbon dielectric fluid, FKM requirement for immersion, extractables and cold-plate clogging. xmbestseal.com
18. Fannteer, "Data Center Liquid Cooling Seals: EPDM vs FKM Material Selection Guide" — EPDM as standard for D2C water-glycol loops, peroxide curing and compression set below 30%, helium leak testing requirement for UQD and BMQC O-rings, FKM limitations in hot water-glycol, seal cleanliness and low-extractable qualification. fannteer.com.tw
19. Titan Fluid, "EPDM vs. FKM Seals for Dry Break Couplings in Water-Glycol Cooling Systems?" — EPDM as first choice for dedicated water-glycol loops, EPDM incompatibility with hydrocarbon contamination, FKM as a technical option rather than a premium upgrade, compound-level qualification over material-class assumptions. Also: Hua Le Seals, "Which liquid-cooling O-ring for which coolant" — FFKM recommendation for fluorinated fluids with volume change under 5%, prohibition of sulphur-cured NBR and standard FKM in hot water-glycol loops. titanfluid.com
20. Danfoss, "Universal Quick Disconnect (UQD) fact sheet" — OCP UQD rev 1.0 compliance statement, 100% helium leak testing of all parts, wetted materials SS303/SS304, EPDM seal, colour identification convention, rated flow, fluid loss and C_v by size, ORB ISO 11926-3 termination. cnmec.biz