Crimp pull test: how many, how often, and what to do on a fail

Pull at 25.4 mm/min with the insulation grip released, and record the force and how the crimp let go, not just pass or fail. Take the sample count your customer's document sets. If nobody sets one, our suggestion is at least five at set-up, Molex's floor, and at least three at each later check, NASA's number. Pull at the start and end of every run so the run is bracketed, and set the interval between by dividing the quarantine you can afford by your output rate: 1,000 crimps at 1,200 an hour means a check every 50 minutes. When one fails, stop that tool, hold everything back to at least the last pass, and do not retest your way out.

Expect less from the test than its reputation suggests. Scored only as pass or fail, three clean pulls show at 95% confidence that fewer than 63.2% of your crimps are weak.

What a handful of pulls can catch

The chance that k pulls catch a process making a fraction p of weak crimps is 1 - (1 - p)^k.

The sum assumes independent random picks and that every weak crimp fails the pull. Both flatter the test, since a check usually takes pieces made together. Read the other way, zero failures rules out only so much at 95% confidence: 63.2% weak after 3 pulls, 45.1% after 5, 25.9% after 10, 11.3% after 25, 5.8% after 50 and 3.0% after 100. You would need 299 clean pulls to rule out 1%.

So, scored pass or fail, a pull test finds a broken set-up. It will not find a 1% problem at any sample you are likely to pull. A NASA team reached a similar view, reviewing 780 logged pulls that all passed and 48 crimp problem reports. Only 11 reports were relevant, and two involved a pull test failure. One was a temporary test-rig crimp that should not have been reported. The other was an oversized contact, so in that case the pull did find a wrong part. The authors still conclude that pull testing is a process control and not a way of finding defects.¹ These reports cover what reached flight hardware work, so they cannot show how much a pull test would catch earlier. It is also aerospace data on machined contacts, and not a benchmark for your line.

How to run it

Published methods agree on the pull and differ on how many.

NASA-STD-8739.4A is public,² but it is a workmanship standard for NASA work, and a commercial customer is unlikely to specify it unless the drawing says so. Its Table 12-1 can serve as a reference point, not as a limit your customer has agreed to. Yours may call up IPC/WHMA-A-620, a UL standard, DIN EN 60352-2 or its own numbers. Take the class, minimum force and sample count from your licensed copy. Those tables are not reproduced here.

A working method, mostly from Molex:³

1. Make the sample on the production tool, from the current reel and wire lot, at the nominal crimp height. Inspect it first for bell mouth, brush and cut strands.
2. Take the insulation grip out of the load, by stripping long enough that it grips nothing or by opening it.³⁴ You are testing the conductor crimp. The insulation crimp gets its own bend test.
3. Use about 150 mm of wire. Knot the free end if the insulation slips.
4. Clamp the terminal by its contact end and never across the conductor crimp, and hold it square. NASA notes that a break at the barrel entrance caused by a contact held out of square does not count as a preferred break.²
5. Pull at 25.4 mm/min. Molex says a faster rate rarely changes much on most applications, but you should verify it against data taken at that speed.
6. On a double crimp, pull each wire on its own. TE requires it, and Molex warns that pulling both together reads higher.⁴³
7. Record the force, how it let go, the tool, both lots and the time. Compare the lowest reading with the minimum, not the average.

Molex also accepts hanging fixed weights on the wire for at least a minute.³ That gives you a pass or a fail and throws away the force, which is the part worth having.

Record the force, because pass or fail says almost nothing

NASA's paper says in its abstract that averages were at least 182% of the minimum, but its own table runs from 126% to 300% of NASA's minimum across 18 contact and wire pairs.¹ If your margin looks anything like the upper part of that range, a drifting tool has a long way to fall before any pull fails. Take an illustration with invented round numbers: minimum 100 N, process mean 180 N, standard deviation 10 N, and a chart on the mean of each set of 3 pulls with a lower limit of 162.7 N.

At 150 N the process has lost 37.5% of its margin and the pass or fail record still looks perfect. The pulls cost the same either way. NASA's authors make the same point for high volume work: recorded results can show die wear, tool setting changes and poor tool use.¹

If you report capability, read Molex's warning first. Pull force has only a minimum, so Cpk needs an invented upper limit, and high readings that widen the spread can lower Cpk even when the mean and the lowest reading have both gone up.³ Watch the mean and the lowest reading as well.

Set the interval from what you can afford to quarantine

If a check fails, everything made since the last pass is suspect. NASA's authors say those crimps need to be evaluated for risk or scrapped.¹ So the interval sets how much you would have to hold.

The rates are planning figures. 300 an hour is a round assumption for a hand tool or manual press. 1,200 is half the 2,400 an hour quoted as the maximum for one Molex bench press.⁵ 3,000 is a round planning figure for a fully automatic machine. Use your own.

Turn the table round to choose. Decide how many crimps you could re-inspect or scrap on a bad day, and divide by your rate. For 1,000 crimps that is every 200 minutes at 300 an hour, 50 minutes at 1,200 and 20 minutes at 3,000. Count in what you would actually have to open: 1,000 crimps sitting in a tray is a different problem from 1,000 crimps already taped into harnesses. This is the nominal quarantine, which assumes the last pass really was clean. The next section and the last table show when it was not.

Three rules go with the interval.

Bracket the run. A pull at the start says nothing about what came after it, so a run stays open until a pull at its end has passed. NASA requires both, and lets one check stand in for the next if the tools have not changed in between.²

Reset at changes. Pull after a new terminal reel, wire lot, tool or setting, and after a jam. Molex lists all of those as causes of a process shift, and warns that a tool loosened after the first hundred crimps can go unnoticed for thousands.³

Buy for it. NASA points out that connectors sold without enough spare contacts mean buying extra, with lot traceability between the test contacts and the build contacts.² On volume lines the pulls are cheap: 3 pulls every 2 hours, with the start and end checks, is 15 pulls a shift, 0.16% of output at 1,200 an hour. At 300 an hour it is 0.62%, and on expensive contacts that can decide the plan.

More pulls or more often?

A pass is not proof, so a partial fault can run through several checks. Assume the fault starts just after a pass, at 1,200 crimps an hour with hourly checks of 3 pulls. Before the first catch you can expect to make this many:

These are expected values. The last column hardly moves for low rates, because weak crimps made before detection come to about rate × interval ÷ pulls per check: 1,200 ÷ 3 = 400 here, and 240 with 5 pulls per check.

So more pulls per hour reduce escapes, and how you split them does not. One pull every 20 minutes uses the same 3 pulls an hour as 3 on the hour, and at 5% weak it makes about the same crimps before detection, 8,000 against 8,414. It cuts the nominal quarantine from 1,200 crimps to 400, but that is not 800 crimps saved. One clean pull is weak evidence: going back far enough for 95% confidence takes 59 checks and 23,600 crimps at 5% weak, against 20 checks and 24,000 with 3 pulls. The gain comes with a gross fault: at 90% weak it is 800 crimps against 1,200. The price is a noisier force record, so sets of 3 remain the sensible default.

Whichever you choose, the gaps belong to a cheaper check. NASA's authors say that if pull testing is reduced, other process controls should increase.¹ Crimp height is the obvious one: Molex says to check it every 250 to 500 parts.³ For the sums behind a formal plan, see sampling plans.

How it let go tells you where to look

NASA's figure shows four ways a crimp can separate: break at the crimp, break outside the contact, fray break and pull-out. It prefers the second and third, and accepts all four above the minimum force. When choosing a tool setting and the strengths tie, it takes the setting with more wire breaks than pull-outs. Below the minimum, wire pull-out, a wire break at the crimp and contact rupture are all failures.² Molex's troubleshooting table gives causes for the two common low-force modes.³

Two more readings are inference, not published rules. If the mode changes from breaks to pull-outs while force stays above the minimum, treat it as an early warning that compression is falling. And do not mistake the wire stretching for slip in the joint, which Wiring Harness News warns about.⁶

What a pull test cannot show

The crimp you pull is not the one you ship, so the result is about the process. Beyond that, NASA's authors list conditions a pull test could not detect, including strands not captured in the barrel, damaged insulation and traceability errors, and say it is not meant to find malformed contacts, incorrect contacts or loose strands, even though one of their two failures was a wrong contact.¹ A sample with a full-strength joint can also hide a resistance problem. Molex describes resistance rising when pressure is too low to break the oxides and when over-crimping reduces the conductor's area,³ and a force reading measures neither. Where the customer cares about resistance, that takes its own measurement or a microsection. A harness also needs an electrical test and a terminal seating check, because none of this covers either.

When a pull fails

1. Stop that combination of tool, terminal and wire. Tag the tool and do not adjust it yet, because its state is evidence.
2. Keep the failed sample and write down the force, how it let go, the time, the tool, both lots and the operator. NASA requires the pull strength and the break or release condition for failures.²
3. Check the test before the product: tester calibration, clamping, insulation grip released, speed. A result is void only if you can name the test fault, and you record that too. Pulling more until enough pass is not containment. A process making 20% weak crimps passes a fresh set of 3 pulls 51.2% of the time.
4. Hold everything back to the last pass, including crimps already in sub-assemblies and finished harnesses. NASA's wording is that hardware produced prior to the failed test is suspect and must be evaluated before use.²
5. Decide whether the last pass is a safe boundary, using the table below. On our reading, if every pull in the set fails well below the minimum, the fault is probably large and recent, and the last pass will usually hold. That assumes a sudden fault and not gradual wear, so look at the force record first. One weak pull among good ones suggests a partial fault that may have passed earlier checks, so go back to the last change of tool, reel or lot, or to the last jam.
6. Find the cause from the failure mode, the crimp height chart, the stripping tool and the tooling.
7. Evaluate the suspects. Check height and appearance on all of them where that is practical, since neither destroys the part. Then pull samples from each box or hour to find where the fault began. As an illustration, 8 slices of 1,200 crimps at 10 pulls each is 80 pulls. A clean slice is not proof, though: 10 pulls catch a 20% fault 89% of the time and a 5% fault only 40% of the time. What you cannot clear, scrap or re-terminate if the drawing allows it.
8. Restart only after the cause is fixed and a fresh sample is clean. NASA's rule is zero failures out of three.² Your own number comes from your control plan or the customer's standard.
9. If any suspect crimps have shipped, your quality agreement decides who you tell and when. The recall record shows what can follow.

Notes

1. A. Constante, C. Fitzgerald and A. Boutte, Analysis of Pull Force Test Results for Crimped Connections, NASA Goddard Space Flight Center, conference paper for IPC APEX EXPO 2023, public use permitted. Read in the March 2022 text. Its data are machined contacts for space flight; stamped and formed contacts were excluded. Its counts differ between sections, 84 and 12 reports in the methods and 48 and 11 in the results, and both versions end with two pull test failures. It also quotes IPC requirements, which are not repeated here. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

2. NASA, NASA-STD-8739.4A with Change 4, 2022, clauses 6.5.3 and 12.3.5, Table 12-1 and Figure 12-1. Change 4 corrected captions only. A US government workmanship standard for NASA hardware, and not the document a commercial customer normally specifies. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

3. Molex, Quality Crimping Handbook, order no. 63800-0029, revision C, 2009, copy hosted by DigiKey. Written for Molex open barrel terminals and tooling. Its minimum force table is not repeated here. Use your licensed standard. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

4. TE Connectivity, Application Specification 114-18022, Rev K, 2011, copy hosted by Farnell. This is Rev K, the copy we could open. Check TE for a later revision. ↩ ↩²

5. Electronic Specifier, TM-4000 universal crimp press from Molex, trade press item on a maker's maximum rate. ↩

6. Wiring Harness News, Tensile Testing of Crimped Wire Terminals. Trade magazine article. ↩