By the process engineering team. What follows comes out of our own crimp validation programme for J1962 diagnostic harnesses and OBD2 cable assemblies, run over eleven months across 2023 and 2024.
A crimp that passes a 210 N pull test can still be sitting at 96 milliohms after five hundred thermal cycles, having started at 0.3. The tension test is measuring the wrong joint. It loads the barrel’s grip on the bundle, while the part that actually fails is the electrical interface a few tenths of a millimetre inside it, and almost no incoming lot in this industry checks that interface at all.
This is what we found when a fleet customer sent six harnesses back.
The return that started this
The email arrived on a Tuesday in March with a photograph attached. Six OBD-II pass-through harnesses, pulled out of a fleet of delivery vans. Someone on the customer’s side had stripped the insulation back about fifteen millimetres from each terminal, and under the jacket, in a place nobody had looked at in three years, the copper had gone green.
The fault itself was undramatic. The telematics units kept rebooting. Now and then the gateway would log an under-voltage event on the supply pin, clear it, and behave for another few weeks. Two vans had run three months clean and then thrown four faults in a single afternoon, in the middle of a heat wave.
The second attachment is what made it worth a lab ticket. It was the incoming inspection record for that lot: pull force, ten pieces, mean 236 N, minimum 221 N, against a print requirement of 210 N. Every piece with margin. The lot shipped with no nonconformance recorded anywhere.
A part that passes every test you give it and still dies in the field is not a manufacturing defect. It is a specification defect, and it belongs to whoever wrote the print.
Why the pull test was never going to answer this
Every crimped joint is two connections formed in the same instant, decaying on completely different schedules.
The mechanical one is the barrel wings folding around the bundle and crushing it. Done properly the crush is hard enough to break the oxide on both surfaces and cold-weld them together. That weld, plus the friction of a bundle compressed inside a channel, is what the tensile machine pulls against. It is a real joint, it is genuinely strong, and it is not where these failures come from.
The electrical one is smaller and fussier. Current does not enter the barrel everywhere the surfaces appear to touch. It enters at asperity contacts, the microscopic peaks where metal is genuinely in contact, a small percentage of the apparent area. Three things govern it: how hard those peaks are pressed together, whether oxide has formed on them, and whether they stay in the same place.
A pull test loads the first joint along the wire axis and reports on nothing else. The second can go from healthy to failed while the first still holds 95 percent of its original strength, which means there is no mechanical symptom to catch. Both parts feel tight in the hand. A 210 N crimp and a 96 milliohm crimp are indistinguishable until one of them is bolted into a vehicle.
What 85°C does that 20°C does not
Move the same crimp off a bench and into an enclosure that sits at 85°C for hours a day, and four things change. Only one of them is about heat the way most people mean it.
Tin at 85°C has stopped behaving like a solid
Tin melts at 232°C, which sounds like plenty of headroom. Metallurgy does not care about the gap in degrees; it cares about the ratio, in kelvin. At 85°C, or 358 K, against a melting point of 505 K, tin is running at 0.71 of its absolute melting temperature.
Past about 0.5, metals creep. They flow slowly and permanently under loads that would do nothing at room temperature. Tin is one of the worst offenders: soft, low yield strength, and at 85°C it stops holding its shape. The pressure pinning those asperities together does not disappear in one event. It bleeds away over months, and cooling the joint down does not put it back.
That is the whole reason terminals for this temperature band get a nickel underplate beneath the tin, or lose the tin entirely in favour of silver on the high-current pins. Nobody is worried about tin melting. They are worried about it flowing.
Expansion is reversible. Creep is not.
Brass runs at roughly 20.5 parts per million per kelvin. Copper runs at roughly 17. Across a one millimetre barrel bore and a sixty-five kelvin swing, that gap works out to about a quarter of a micron of relative movement between the barrel wall and the strand bundle.
Once, that is nothing.
Five hundred times is a mechanism. Each cycle nudges the tin at the contact points a little further under load, the creep does not unwind on the way back down, and the joint returns with slightly less pressure than it left with. Nothing reverses, which is what makes the effect cumulative rather than merely cyclical.
A few microns of movement is also exactly the amplitude band where fretting corrosion lives. Fresh metal gets exposed, oxidises, and the oxide sits precisely where the current has to cross.
We have no rig that measures contact normal force on a live crimp, so I am not going to quote a figure for the pressure lost. What we could measure was the downstream effect, resistance drift, and that was enough to rewrite our process.
Moisture arrives by breathing, not by dripping
Thermal cycling does something no static humidity soak can reproduce: it pumps. The air in the connector cavity expands during the hot dwell and gets drawn back in as the cavity cools. An unsealed cavity therefore breathes the interior air of the vehicle twice a day for years, and whatever meets a cold surface on the way down condenses there.
On the harnesses that came back to us, the corrosion had not started at the terminal. It started fifteen millimetres inside the insulation jacket, at the cut ends of the strands, where moisture ingress gathers in the capillary gap between strands and the jacket holds it there. From outside, nothing shows. We only found it by stripping jackets.
The joint heats itself
Three amps through 0.3 milliohms dissipates about 2.7 milliwatts. Nobody cares. The same three amps through 96 milliohms dissipates 0.86 watts inside a joint you could cover with a fingernail. Local temperature rises, oxide grows faster, resistance climbs, temperature rises again.
It is a loop with no brake in it, and it accounts for the shape of the field data. Faults arrived in clusters during hot spells instead of accumulating evenly.
The test we actually ran
We rebuilt the failure on the bench rather than reasoning about it, because reasoning about it had already produced one wrong answer.
The wire was 0.50 mm² tin-plated copper, between 20 and 21 AWG, called 20 AWG in most of the harness documentation we receive. Insulation was thin-wall XLPE. The termination was an open-barrel F-crimp onto a tin-plated brass receptacle, using the standard crimp terminal for this connector family. Two strand constructions went in, both nominally 0.50 mm²: sixteen strands of 0.20 mm, and seven strands of 0.30 mm.
The profile was minus 40 to plus 85°C, thirty minutes dwell, roughly 2 K per minute ramp, per IEC 60068-2-14 Nb, five hundred cycles. Groups carrying current did so at 3 A during the hot dwell, which is what an always-on accessory draws through a diagnostic harness supply pin. Groups carrying no current exist as controls, so that thermal effects and electrical effects could be separated. Half the groups were sealed with adhesive-lined heat shrink. The unsealed half went through 240 hours at 40°C and 93 percent relative humidity before cycling, to give the moisture somewhere to start.
Ten samples per cell. Pull force came from a separate set of samples from the resistance set, because a pull test destroys the joint and you cannot measure resistance on a crimp you have already torn apart. That point caused an argument in our own lab. We had been reusing samples across tests. It saves material and it invalidates the result.
Millivolt drop, in one sentence
You push a known current through the joint, measure the voltage across it with a four-wire setup, subtract the resistance of an equivalent length of bare conductor, and what remains is the crimp. Under half a milliohm for 0.50 mm² is a healthy joint. Ninety-six milliohms is a resistor.
In practice that means a 1 A source, a meter that resolves microvolts, and sense probes on the barrel and on the conductor fifty millimetres downstream. Two-wire readings from a handheld meter will not resolve any of this, and a good deal of incoming inspection is done exactly that way.
What the data showed
The crimp height ladder
Table 1. Crimp height ladder, 0.50 mm² sixteen-strand wire, n=10 per setting, 200 cycles −40/+85°C, unloaded and unsealed.
| Crimp height | Pull mean (N) | Pull minimum (N) | Strands intact | ΔR after 200 cycles | Note |
| 1.10 mm | 168 | 141 | 16/16 | +14.2 mΩ | Loose. No gas-tight seal. |
| 1.06 mm | 205 | 188 | 16/16 | +5.8 mΩ | Fails the 210 N print outright. |
| 1.02 mm | 238 | 221 | 16/16 | +0.9 mΩ | The setting we had been running. |
| 0.98 mm | 251 | 236 | 15/16 | +3.1 mΩ | Highest pull force. One strand nicked. |
| 0.94 mm | 244 | 219 | 12/16 | +9.4 mΩ | Four strands severed. Pull still passes. |
| 0.90 mm | 226 | 198 | 9/16 | +18.0 mΩ | Seven strands gone. |
The two right-hand columns contradict each other, and that contradiction is the entire failure mode.
Pull force clears 210 N on the mean at four of the six settings. Three of those four clear it on every single piece. Two of those three are quietly wrecking the conductor.
Going 0.04 mm deeper than nominal bought thirteen newtons and nicked one strand. Going another 0.04 mm deeper lost seven newtons and three more strands. The pull curve had already turned over while the damage kept stacking up. That is the trap built into this design: pull force only tells you that you have gone too far once you are well past the point where the joint starts to run hot.
The drawing made it worse. A crimp height of 0.98 mm sat inside the plus-or-minus 0.03 mm tolerance. On paper it was a conforming part. It also gave the highest pull force of anything we tested and three and a half times the resistance drift of nominal.
Five hundred cycles, sealed and unsealed
Then we ran the full cycling matrix
Table 2. Five hundred cycles −40/+85°C, n=10 per group. Group E is a deliberate control: the under-crimped setting that fails the pull print, carried through the full matrix to see whether a bad mechanical joint is also a bad electrical one. Group F is the unloaded control.
| Group | Crimp height | Sealed | Loaded at 3 A | Pull before | Pull after | Retention | R before | R after |
| A | 1.02 | Yes | Yes | 238 N | 229 N | 96% | 0.31 mΩ | 0.40 mΩ |
| B | 0.98 | Yes | Yes | 251 N | 241 N | 96% | 0.29 mΩ | 0.58 mΩ |
| C | 1.02 | No | Yes | 236 N | 224 N | 95% | 0.32 mΩ | 41 mΩ |
| D | 0.98 | No | Yes | 249 N | 238 N | 96% | 0.30 mΩ | 96 mΩ |
| E | 1.06 | Yes | Yes | 205 N | 194 N | 95% | 0.36 mΩ | 1.4 mΩ |
| F | 1.02 | Yes | No | 237 N | 230 N | 97% | 0.31 mΩ | 0.34 mΩ |
Every group held between 94 and 97 percent of its pull force. Group D included, and group D finished the test at 96 milliohms and 238 newtons.
If pull force retention is your acceptance criterion, group D is a good part. It would have cleared incoming inspection, first article and annual requalification without a flag. On the vehicle it would have been carrying 0.29 volts of drop and 0.86 watts of self-heating, which is precisely the fault signature the fleet reported.
We cut group D open afterwards. The barrel-to-wire weld was still intact and still strong. The corrosion was sitting between the strands, inside the bundle, in the fifteen millimetres the jacket had been hiding.
Two results in that table caught us out. Group E, the under-crimped control, was the worst part in the study mechanically and slightly better than nominal electrically, because it was loose enough that no strand had been cut, and its failure mode is a slow one that declares itself early. Groups B and D, the ones with the cleanest paperwork, are the ones that fail late and without warning.
Why the strand construction decided the outcome
The seven-strand wire came through cycling consistently better than the sixteen-strand wire at the same nominal area. We re-ran it before believing it.
The mechanism is not subtle once you have sectioned a few joints. Coarser strands offer fewer, larger contact patches to the barrel wall, and a large patch is harder to shift by a quarter of a micron of differential expansion. Fine-strand wire brings more surface, more capillary gaps to wick moisture, and thinner strands, so a larger share of them get nicked by the crimp punch at any given crimp height. The 0.20 mm strands were the ones vanishing from the cross-sections at 0.94 mm.
Both wires are 0.50 mm². Both are called 20 AWG. Both meet the same conductor standard. Specify cross-section and nothing else, and you have specified about half of what decides whether the joint lives.
We now put strand count on every drawing we quote against, and a supplier change in strand construction is treated as a change requiring requalification, not a drop-in substitution. On one European programme that cost us a four-week delay in 2019 and saved the customer a recall.
Step by step: making a 210 N crimp survive 85°C
This is the sequence we now run, in order, on any high-temperature harness that will live above 70°C. None of it needs equipment most harness shops do not already own.
- Add a resistance criterion at time zero, and write it into the drawing. Acceptance is a four-wire measurement across the joint at a defined current, with the conductor contribution subtracted. We use 0.5 milliohms for 0.50 mm² and scale from there by cross-section. This belongs in inspection, not in a laboratory report. With a Kelvin clip fixture it takes eleven seconds per joint.
- Establish the crimp height window from the electrical curve, not the mechanical one. Run the ladder. Ten pieces at each of five heights spanning nominal plus or minus 0.06 mm. Plot pull force and resistance drift at two hundred cycles. The setting you want is the centre of the flat part of the resistance curve, even where it costs you fifteen newtons of pull force. Then tighten the drawing tolerance to whatever that flat region will bear. Ours went from plus or minus 0.03 to plus or minus 0.02.
- Confirm the setting with a cross-section, every shift. Look for barrel fill between roughly 70 and 85 percent, no voids, no strands outside the crimp, and no strand sectioned by the punch edge. A pull test cannot see a cut strand. A crimp cross-section can. Photograph it and file the image with the lot record.
- Specify the plating for the temperature, not for the price. Tin over nickel for terminals that see 85°C and moisture. Silver on high-current pins where the contact resistance budget is tight. Bare tin on its own is fine in a dry, sealed, current-limited cavity and nowhere else. If a supplier quotes you tin and cannot tell you what underplate sits under it, that is your answer.
- Seal the transition, not just the connector. Adhesive-lined heat shrink over the crimp and at least twenty millimetres of jacket. The adhesive is the whole point. Plain tubing keeps splashes out and lets vapour in. Every one of the returned harnesses failed in a cavity where the pigtail entered an unsealed housing.
- Qualify with current flowing. A thermal cycle with no load tests expansion. A thermal cycle at rated current tests expansion plus creep plus self-heating plus oxide growth, which is closer to what the part actually does. Group F drifted about a third as much as group A with everything else held constant.
- Freeze the wire construction. Strand count, plating, insulation compound and supplier go on the drawing. Any change goes through requalification. This is the control most often traded away in a cost-down, and it is the one that decides the outcome.
- Change the control plan, not just the part. Crimp height every two hours with a digital micrometer, logged. Pull force per lot, minimum five pieces. Resistance per lot, minimum ten joints. Cross-section per shift. Capability study on crimp height against a target Cpk of 1.67. When we ran ours after retuning, the applicator was drifting 0.015 mm across a shift as the press warmed up. Inside tolerance, and enough to walk us along that spike.
Five mistakes we see in incoming inspection
- Testing pull force and calling the joint qualified. It is the only crimp test most incoming labs run, and the table above shows it moving by one percentage point between a joint that lasts twenty years and one that corrodes to 96 milliohms in three. It catches a badly set applicator. It does not catch a joint that has to survive temperature.
- Cycling without current. Close to universal. The chamber is full, the power supplies are somewhere else, and getting live conductors into a chamber takes planning. The result is a qualification that passes and a field that fails.
- Reusing samples across tests. Pull the sample after you have measured it, never before, and never on a piece that has already been sectioned. Test sequence order changes the answer.
- Qualifying with a hand tool and producing with an applicator. The two do not produce the same crimp height distribution, and the operator spread of a hand tool hides exactly the sensitivity you are trying to characterise.
- Treating a conforming crimp height as a conforming crimp. The tolerance on our drawing was satisfied at 0.98 mm and the joint was still marginal. Meeting a tolerance and being adequate to the duty cycle are different questions, and only the second one belongs to you.
How to confirm the fix actually held
Do not confirm it with a pull test and a visual. Confirm it with a sequence, on production samples off the actual applicator, with the acceptance limits written down before you start.
- Joint resistance at time zero — four-wire at rated current, 0.5 milliohms or less for 0.50 mm².
- Resistance drift — after 500 loaded cycles at −40/+85°C, no more than 20 percent rise, and no joint above 1 milliohm.
- Continuity through cycling — logged during dwell, no interruption longer than 100 microseconds.
- Pull force after cycling — destructive on separate samples, 210 N minimum on ten pieces.
- Cross-section before and after — barrel fill 70 to 85 percent, strand count unchanged.
- Intermetallic thickness — sectioned and measured, trended rather than pass or fail at this temperature.
- Moisture — 240 hours at 40°C and 93 percent relative humidity then cycling, no rise beyond the drift limit.
- Corrosion — 480 hours neutral salt spray to ISO 9227, no base metal corrosion at the crimp.
On intermetallic growth, one note so the case is not overstated. At 85°C the copper-tin layer does thicken. Ours went from roughly half a micron to just under a micron across the hot-dwell hours of the cycling run. At this temperature it is not the villain. It is the clock. It matters over a twenty-year under-dash life, and it matters enormously at 125°C. At 85°C the tin creep and the moisture arrive first.
Where salt spray and PPAP fit, and where they stop helping
Every supplier audit questionnaire we get from Europe or North America asks for salt spray and PPAP, nearly always in that order, which is backwards. Here is what each one actually buys.
Neutral salt spray, 480 hours to ISO 9227, runs on our line against finished assemblies with the crimp intact. It is good for plating porosity, for the quality of the nickel underplate, and for whether a housing material will chalk and craze. It is poor for this failure mode, because salt spray is isothermal and static. It cannot reproduce the breathing that draws moisture into the strand bundle, and it will pass a crimp that dies on the vehicle. We run it, we keep the reports, and we do not let anyone present a salt spray pass as a thermal cycling qualification.
PPAP is the other half. A Level 3 submission, meaning control plan, PFMEA, MSA on the crimp height gauge, capability studies, dimensional results on the cavity and barrel, material certifications and the PSW, is how a customer knows that the process which produced the qualification samples is the process that will produce their order. That is the whole argument in an automotive wire harness programme. Ask for the capability study on crimp height specifically. If Cpk sits below 1.33 against a 0.02 mm tolerance, the process is not holding the window the thermal data requires, whatever the first article said.
The reason a 0.02 mm crimp tolerance survives three shifts here is unglamorous. Terminal reels live in a dehumidified store, because tin oxide grows in a damp warehouse and a re-oxidised terminal does not crimp like a fresh one. The applicator is checked against a master at every shift change, not once a week. The crimp height gauge carries a calibration record an auditor can pull without asking us for it. ISO 9001, IATF 16949, ISO 14001, RoHS and REACH are on our certification records. They are the entry ticket, not the argument.
The harnesses this applies to
We build the J1962 sixteen-pin pigtails and OBD-II pass-through harnesses that this study came out of: 0.50 mm² and 0.75 mm² conductors, open-barrel crimps, overmoulded or assembled housings, with or without the always-on supply tap. The same crimp discipline goes into our OBD-II extension leads, diagnostic Y-cables, OBD2 cable assemblies and the custom OBD-II cable assemblies we build to drawings rather than to a catalogue.
Most of what leaves here is drawn, not catalogued. A first order usually arrives as a PDF with a housing part number, a pinout table and a length in millimetres. We send back a crimp specification, a cross-section photograph, and one question, usually about whether pin 16 is switched or permanent, because the answer changes the plating call. Overmould tooling is the only long-lead item and the only thing we quote separately.
There is no price list because there is no standard part. Two customers can order the same housing and need different plating, a different crimp height and a different sealed transition, and the quote follows the drawing.
The full measurement set behind this article, six crimp heights and six groups at ten pieces each, is available as a CSV. Ask for the 85°C crimp study through the contact page and we will send it with the test method written out.
Questions we get asked about this
Does 210 N meet the usual automotive crimp standards for 0.50 mm²?
It sits above what most crimp standards require at this gauge. Those numbers generally land lower, in the 110 to 160 N band depending on strand class. A 210 N print is a house specification, usually inherited from a legacy drawing. Meeting it proves your applicator is not badly out of adjustment, and not much more.
Why did the joint pass at 20°C and fail at 85°C?
Two different properties were being measured. Pull force reflects the cold-welded barrel-to-bundle joint, which is stable. Resistance reflects the asperity contacts, which depend on contact pressure held by tin that creeps at 0.71 of its melting temperature. The second degraded. The first did not.
How do I run a millivolt drop test without a lab?
Four wires, not two. Source one amp through the joint, measure the voltage across it with a meter that resolves microvolts, and put the sense probes on the barrel and on the conductor fifty millimetres away. Subtract the resistance of an equal length of bare conductor. A two-wire handheld reading will not resolve any of it.
How many thermal cycles is enough?
Five hundred cycles of minus 40 to plus 85°C with current during the dwell separated good crimps from marginal ones reliably in our data. For a twenty-year life, follow it with an isothermal soak of one thousand hours at 85°C, unpowered, to catch the creep-only mechanism on its own.
Should I specify gold plating?
Not at 85°C in a wet cavity. Gold is excellent at low signal levels and does not oxidise, but it is thin and soft, and it does not stop moisture reaching the copper underneath. Tin over nickel, or silver on the power pins, is the better answer here. Gold earns its price in dry, low-current, high-mating-cycle applications.
Can I just crimp harder to get more margin?
No, and the ladder data shows why. The first 0.04 mm past nominal bought thirteen newtons and cost a strand. The next 0.04 mm lost seven newtons and cost three more. Pull force rises, conductive cross-section falls, and the joint gets stronger and hotter simultaneously.
Does the wire strand count really matter at the same cross-section?
It shifted our results by roughly a factor of two on resistance drift with everything else held constant. Sixteen strands of 0.20 mm and seven strands of 0.30 mm are both 0.50 mm² and both get called 20 AWG. Put the strand count on the drawing.
What is the difference between thermal shock and thermal cycling?
Thermal shock moves between extremes quickly, to test for mechanical cracking from differential expansion. Thermal cycling dwells at each extreme, which is what lets creep and moisture absorption happen. For crimp durability you want cycling with dwell, not shock.
How do I inspect a crimp cross-section properly?
Pot the sample in epoxy, grind to the barrel centreline, polish to one micron, and etch lightly if you need grain structure. Look at barrel fill, voids, strand deformation, and whether the punch has sectioned a strand. Photograph at 50x and keep it with the lot. A section is the only inspection that shows you a cut strand.
What should I ask a cable supplier for before awarding a harness programme?
The crimp height window with the electrical data behind it, a cross-section from the production applicator, the resistance acceptance criterion, a thermal cycling report with current applied, the salt spray report, and a PPAP file with a capability study on crimp height. If any of those is missing, the pull force certificate is doing more work than it can bear.
If you are specifying a harness that runs hot
Send the drawing and tell us the duty cycle: ambient temperature, continuous current per pin, whether the cavity is sealed, and the wire construction if it is already fixed. We will tell you where the crimp window should sit and whether the plating you have specified will hold it.
If the crimp window on your print is the one that fails at temperature, we will say so before tooling rather than after. That conversation has saved two programmes here from a redesign, and cost us one order.
WhatsApp, talk to Linda directly: https://api.whatsapp.com/send/?phone=8617307168662&text=Need+Help%3F+Chat+linda+WhatsAPP&type=phone_number&app_absent=0
Contact page, for drawings, specifications and the raw data from this study: https://obd-cable.com/contact/
