Where the Standard Stops: Setting a Pull-Force Limit for a Joint
A standard tells you how much the wire itself should hold. It does not tell you how much the joint should hold, and for copper to aluminium it does not tell you anything at all. This is how we set the number that goes on the acceptance sheet when no table covers the joint.
The standard covers same-metal splices, and stops there
The table most harness plants work from lists conductor sizes from 0.22 mm² to 150 mm², with separate columns for copper and for aluminium: a tensile value, a minimum strip strength, and a weld width range. It is a solid reference for what one wire of a given size can take.
Three things it does not cover. Copper joined to aluminium is not in it. Busbar and foil stacks are not in it. Transition joints between dissimilar conductors are not in it. And the number it prints is the breaking load of the wire, not a requirement on the weld.
That last point causes most of the confusion we see. A buyer takes the 2400 N in the 70 mm² copper row, writes it on a drawing as the weld requirement, and then argues with the supplier about a joint that already exceeds the wire. The number was never about the joint.
Why a good joint tests at about twice the wire strength
An ultrasonic splice is a lap weld. The horn presses two conductors together over an area, and the joint carries load in shear across that area, not across the wire cross-section. The bonded area is typically several times the conductor area, so the joint can hold more than the wire.
That is the expected result, not an overclaim. It also gives the real acceptance criterion: where the break happens. If the sample breaks in the base metal away from the weld, the joint is stronger than the wire and the part passes. If it breaks at the interface, the joint is weaker than the wire, and no number on the certificate changes that.
Measured against the standard rows, the joints we build land at roughly twice the wire value, and the ratio is steadier on aluminium than on copper. The table below is the same comparison across the sizes we publish.
| Conductor | Standard wire value | Our joint pull-off | Ratio |
|---|---|---|---|
| 6 mm² copper | 435 N | >1200 N | 2.76 |
| 25 mm² copper | 1350 N | >3100 N | 2.30 |
| 50 mm² copper | 2200 N | >5000 N | 2.27 |
| 70 mm² copper | 2400 N | >6000 N | 2.50 |
| 120 mm² copper | 3400 N | >6800 N | 2.00 |
| 35 mm² aluminium | 1200 N | >2400 N | 2.00 |
| 70 mm² aluminium | 1880 N | >3900 N | 2.07 |
| 120 mm² aluminium | 2200 N | >4500 N | 2.05 |
How to read a welding-dimension table
Once the joint is understood as a shear area, the welding dimensions stop being arbitrary. Weld width sets the bonded area together with weld length, and the height records how far the stack was compressed. Our own table, for the sections we quote, looks like this.
| Conductor | Weld length mm | Weld width mm | Weld height mm | Joint pull-off N |
|---|---|---|---|---|
| 6 mm² copper | 11 | 4 | 1.6 | >1200 |
| 25 mm² copper | 11 | 11 | 2.4 | >3100 |
| 35 mm² copper | 14 | 12 | 3.1 | >4000 |
| 50 mm² copper | 14 | 14 | 3.8 | >5000 |
| 70 mm² copper | 14 | 18 | 4.1 | >6000 |
| 95 mm² copper | 14 | 20 | 5.1 | >6000 |
| 120 mm² copper | 14 | 22 | 5.8 | >6800 |
| 35 mm² aluminium | 14 | 13 | 2.4 | >2400 |
| 50 mm² aluminium | 14 | 16 | 2.8 | >3200 |
| 70 mm² aluminium | 14 | 19 | 3.2 | >3900 |
| 95 mm² aluminium | 14 | 21 | 4.0 | >4100 |
| 120 mm² aluminium | 14 | 22 | 4.8 | >4500 |
Length stops at 14 mm while width keeps growing, from 4 mm to 22 mm. That is deliberate. A longer weld spreads the same acoustic energy over more material and the interface gets less of it per unit area, so the useful direction is wider, not longer. Width follows the horn face and the conductor bundle; height is the outcome of pressure and stack thickness, and a height far off the figure usually means the parts were not closed before the energy arrived.
Four steps to set a line where no standard covers you
This is the procedure we use for copper to aluminium, for busbar, and for any joint the standards skip. It takes about two days and it removes the argument that otherwise returns every time a batch is questioned.
- Build a baseline, not a sample. One material, one horn, one parameter set, thirty consecutive parts. Record pull force and fracture location for every part, and photograph the fracture surface of at least five.
- Look at the distribution, not the mean. The acceptance limit cannot sit at the average, because half the parts would fail by definition. Take the minimum, the standard deviation, and the spread between the first and last part of the run, which is where drift shows up.
- Set the limit from the spread. With a process capability target of 1.33, the lower limit sits four standard deviations below the mean. If the thirty parts do not support that, the process is not ready for a limit yet, and the answer is more trials rather than a softer number.
- Write the fracture criterion next to the number. A pull force with no fracture requirement can be met by a joint that is already cracking. State that the break must occur in the base metal, and for copper to aluminium add a bend or a peel, because the interface can pass tension and fail in bending.
If you already have a quality system, the same logic sits behind the process comparison in our quality control guide; the difference here is that the limit is derived from your own distribution rather than copied from a table.
Strand loss and unbonded layers: two things one number hides
A stranded conductor can pass a pull test with several strands never joined. The load is carried by the strands that did bond, so the number looks acceptable while the joint has less copper than the drawing says. Current capacity and fatigue life both suffer, and neither shows up in a tensile test.
Foil stacks have the mirror problem. The outer layers see the horn and the anvil directly and bond first; the middle sees whatever pressure reached it. A stack tested as a whole can pass while the centre is still loose.
Neither can be settled with a percentage we invent. What we do instead is prescribe the check rather than the threshold: peel the splice apart and count the unbonded strands, and for foil stacks slice or peel layer by layer and look at each interface. The acceptable count depends on the current the joint carries and on the vibration environment, so it belongs in the drawing, agreed between the harness maker and the end customer. A single percentage applied to every joint hides more than it controls.
A 70 mm² copper to aluminium example
Copper to aluminium is the case the standards leave open, so it is the one worth walking through. On a 70 mm² joint we publish two figures: 5800 N with a fresh horn, and 5200 N after eight months on the same machine. Same material, same parameters, worn face.
Read against the tables that is consistent. The standard row for 70 mm² copper is 2400 N, our same-metal joint is above 6000 N, and the dissimilar joint sits lower than the same-metal one because the interface is not a single metallurgy. The joint is still more than twice the wire value, which is what matters for the break to land in the base metal.
The line we would write for that joint is not 6000 N. It is 5200 N as the floor, with 5800 N as the reference for a new horn, and a fracture requirement that the break occurs in the copper. A limit copied from the same-metal row would have been 2400 N, which any badly worn horn still passes.
The amplitude window matters here too. Our published measurements on a 70 mm² copper to aluminium joint show how pull force and contact resistance move between 35, 45 and 55 micrometres, so the limit has to be tied to the amplitude the joint is actually welded at.
What goes on the acceptance sheet
Four lines, no more. Conductor and material, including whether the joint is same-metal or dissimilar. Pull-force floor for that section. Fracture requirement, meaning where the break must occur and what the surface must look like. Inspection frequency with the record that proves it was done.
Frequency in practice: first article, first part of every shift, after any horn change, and after any material lot change. Record the amplitude and the impedance alongside the pull force, because a joint that fails later is usually explained by one of those two numbers, not by the pull test that passed.
Questions we get
The standard says 2400 N and you quote 6000 N. Is that inflated?
They measure different things. The standard value is the breaking load of the wire itself; 6000 N is the load our joint carries in shear before it fails, and at that point the wire has already broken. If you want one number to compare against the standard, use the standard’s: the wire breaks at 2400 N either way. The 6000 N describes how much margin the joint has over the wire.
Is a higher pull force always better?
No. Above a point, more force means the weld was driven harder than the material wanted, which shows up as thinning, deep sink marks and cracks at the edge of the footprint. On aluminium it also accelerates intermetallic growth at the interface. The target is a joint that reliably breaks in the base metal, not the highest number the press can produce.
Our customer asks for a report against a standard that has no row for our joint.
Use the standard’s method and your own limits. Quote the standard for the conductor sizes it does cover, state that the dissimilar joint is outside its scope, and attach the baseline, the distribution and the fracture photographs that produced your limit. In practice this is accepted more readily than a copied number, because it can be audited.
How often should the limit be re-derived?
Whenever the hardware changes (new horn, refaced horn, new transducer, new generator) or the material lot changes. Otherwise once a year as a review. A limit derived from a worn horn and applied to a new one is either too loose or impossible to meet, and in both cases the operator stops trusting it. See the horn wear guide for how fast the face moves.
Send us the joint
If you are quoting a copper to aluminium or busbar joint and the drawing has no pull-force value, send the section, the material and the target current. We will weld a trial run, send the distribution and the fracture photographs, and propose a limit you can defend to your customer. Engineers reply within 24 hours.