A Weld That Will Not Hold: The Order to Check Things In
A weld that will not hold is usually blamed on the parameters, and the parameters are usually the last thing that changed. This is the order we work in when a customer sends back a part that failed: decide which family of fault it is, then measure the one number that separates the candidates.
Three families before any parameter
Almost every bad weld on an ultrasonic metal machine sits in one of three families: not enough energy, too much energy, or the two parts never met properly. They look different in the break, and each one has a different first measurement.
Welders get adjusted in circles because the three are treated as one problem. Turn the amplitude up and a contact fault gets worse. Turn it down and an energy fault gets worse. Decide the family first, then the knobs have somewhere to go.
| Family | What the break looks like | First number to look at |
|---|---|---|
| Not enough energy | Bright, unjoined interface; bond islands are small and scattered | Pull force, then amplitude against the day-one baseline |
| Too much energy | Thinning, deep sink marks, brittle fracture, cracks at the edge of the horn footprint | Thickness after welding and the surface profile |
| Parts did not meet | Good bond in the middle, nothing at one edge; or weld position drifted | Break pattern by location, then anvil support and horn flatness |
What to measure, in a fixed order
The order matters more than the list. Each step either closes a family or leaves one candidate standing, so nothing is measured twice and nothing gets adjusted on a hunch.
- Pull one part to failure. Where it breaks tells you more than the number. A break in the base metal means the interface held. A break at the interface means it did not. A break that starts at one edge and runs inward means the load was never even.
- Read the amplitude log against the baseline. We record amplitude on the first article and keep it. If today reads lower than day one, look at the stack and the horn before touching the settings.
- Measure impedance of the whole stack. A shift here means the transducer, booster or horn changed, not the process. New horn, cracked ceramic, loose stud, all show up in this number.
- Look at the horn face under magnification and check the anvil. Wear rings, aluminium pickup, a flat that is no longer flat. Then check that the workpiece is fully supported under the weld line. Overhang at the anvil edge produces a fault that no parameter can fix.
- Only now adjust, one thing at a time. Amplitude in steps of about ten percent, then weld time, then pressure. After each change, repeat the same pull test on the same material. Two changes at once and the result cannot be read.
Step four is where most investigations end. In our own service records the horn face and the anvil account for more repeat failures than the generator does, and both are visible without any instrument. If you want the parameters in a wider context, the guide to quality control and process comparison covers how amplitude, time and pressure trade against each other.
Eight defects, three suspects each
| What you see | Three suspects, most likely first | The measurement that rules one out |
|---|---|---|
| Interface not joined, bright metal | Amplitude below the working window; pressure not reaching the parts; oil or oxide on the surface | Amplitude against baseline, then a solvent-wiped sample welded again |
| Bonded in the middle, open at one edge | Workpiece overhanging the anvil; horn face not flat; part located off centre | Break pattern by position, then anvil coverage under the weld line |
| Over-welded, base metal thinned | Amplitude too high; time too long; pressure too high | Thickness before and after, plus sink depth at the horn footprint |
| Crack running along the weld | Amplitude too high; horn edge worn to a sharp lip; part not restrained | Horn face photo at magnification, then one amplitude step down |
| Crack running across the weld | Contraction stress during cooling; excessive pressure; hard temper material | Direction of the fracture surface and the pressure record |
| Deep marks, metal stuck to the horn | High amplitude with high pressure; rough horn face; heat building up over a run | Horn face condition, then surface temperature after a run of parts |
| Foil stack joined on the outside only | Total thickness beyond what the machine can do; oxide between layers; pressure too low to close the stack | Peel the stack apart layer by layer and look at each interface |
| Copper to aluminium joint goes brittle | Energy high enough to grow intermetallic; contaminated surface; time too long | Fracture appearance and colour, then weld time against the original setting |
Cold welds: never joined, or joined a little
Two different faults get called the same name. In the first, the interface never bonded at all and peels apart clean. In the second, small islands bonded and the rest did not, so the part holds a low load and then lets go.
A clean peel with no bond islands is an energy or a contact problem, not a parameter that drifted by itself. Check that the amplitude is still where it was on day one, that pressure actually arrives at the parts, and that the surface is clean. A soldering flux residue or a film of stamping oil will do it on its own.
Islands scattered across part of the weld, with clean metal elsewhere, point at geometry. The horn was not parallel to the part, or the part was not flat, or the anvil did not support the line. A quick test: weld the same material at a different position on the horn face. If only one position fails, the fault travels with the tooling. If every position fails the same way, it is energy.
When the pull force falls off gradually over weeks rather than overnight, the horn is the first suspect. Face wear changes how the energy enters the part long before anything looks broken, which is what the guide to horn wear is about.
Cracks: where they start tells you why
A crack along the length of the weld usually means the energy was applied too hard at a line rather than over an area. High amplitude, or a horn edge that has worn into a lip, concentrates stress at the boundary of the footprint. Look at the face first, then come down one amplitude step and weld a fresh sample.
A crack across the weld is more often mechanical than acoustic. The joint contracts as it cools while the fixture holds it rigid, or the pressure was high enough to emboss the part. If the crack appears only on parts that were clamped hard, the fixture is the place to look.
A crack that starts exactly at the edge of the anvil is the clearest case of all. The material under the horn was vibrating while the material beyond the anvil edge was not, and the boundary took the difference. Moving the part or extending the anvil fixes it; no setting will.
Two cases that fool people
A stack of thin foils can look fully welded and still delaminate in service. The outer layers bond first because they see the horn and the anvil directly, and the middle sees whatever pressure got through. Pull testing the stack as a whole hides this. Peel it apart: if the middle layers come away clean, the weld is a shell.
Copper to aluminium is the other one. The joint can pass a pull test and still be wrong, because the interface has grown a brittle intermetallic layer that holds load in tension and cracks when bent. Add a bend or a peel to the acceptance check on these joints. A pull force alone will keep signing off parts that fail in the field.
Repair or replace
A worn or chipped horn face can be refaced, up to the point where the overall length no longer tunes. After refacing, amplitude and frequency both change, so the first article has to be re-run and the baseline re-recorded. A horn that has been refaced twice is usually cheaper to replace than to keep chasing.
A transducer or booster whose impedance has moved is a replacement, not an adjustment. So is a stack that will not hold frequency at temperature. On the generator side, an over-current or over-temperature trip is a symptom: check the cable, the connector and the ground before opening the generator, because a bad joint in the cable produces the same trip as a failing board.
Whatever was changed, the close-out is the same: measure amplitude, measure impedance, weld five parts and pull all five. Without those numbers the repair is a guess that happened to work that afternoon.
Questions we get
The pull force passes, but the customer reports cracks after assembly.
A pull test loads the joint in one direction. Cracks after assembly usually come from bending, vibration or thermal cycling, and a joint with a thin brittle interface can pass tension and fail in bending. Add a bend test to the check for copper to aluminium and for very thin foil stacks, and look at the fracture surface rather than only the number.
The same settings worked yesterday and fail today.
Settings do not drift, hardware does. In order: horn face condition, anvil position, air pressure at the machine rather than at the compressor, then impedance of the stack. If all four are unchanged, check that the material lot is the same, because hardness and oxide thickness vary between deliveries and both change the working window.
Can we just raise the amplitude to cure a cold weld?
Sometimes, and it buys a bigger problem. If the fault is contact or geometry, more amplitude drives the horn harder into a part that was never properly supported, which wears the face and moves the failure to cracking. Raise amplitude only after the interface has been shown to be clean, flat and fully supported.
How often should the baseline be re-recorded?
After any change to the stack: new horn, refaced horn, new transducer, new generator. Otherwise once a quarter is enough to catch slow drift. The point of the baseline is comparison, and a baseline from a different hardware setup compares nothing. Our published amplitude window data shows how much the pull force moves across the usable window on a 70 mm² copper to aluminium joint.
Send us the break
If you have a part that failed and you are not sure which family it belongs to, send it. A photograph of the fracture surface at magnification, the amplitude reading from the day it was welded, and the settings are usually enough for us to name the cause before anything is adjusted. If you would rather work through it on a call, our engineers answer within 24 hours.