Most welding faults in thermoplastic composite parts trace back to a routine borrowed from unreinforced plastics: heat both faces until they look wet, press them together, hold, release. On a neat polymer that routine works often enough to become a rule. On a fibre-reinforced laminate it fails quietly, and the failure rarely sits at the joint face you inspected. It sits a fraction of a millimetre underneath.
A 2026 study on laser-assisted winding of PPS tape shows the size of the effect. With a 320 °C laser source and a winding speed of 1.3 m/min, lifting tape tension from 50 N to 150 N raised the mean fatigue life of NOL ring specimens by a factor of 2.68. A single surface hot-pressing pass added after winding raised it by 3.39, cut the scatter of results by 82.3% and reduced surface roughness by roughly 30% (Composites Part B, vol. 314, 2026). Same tape, same fibre, same machine. Only the temperature and pressure history at the interface changed.
The same holds for a welded joint, and the consequence is practical: a weld is not a finishing step appended to a moulding route. It is a second consolidation, with its own pressure, temperature and cooling requirements, and it deserves its own process card.
Where Plastic Welding Logic Breaks Down
Four habits carried over from plastics do most of the damage.
- Heat until the surface looks molten. On a composite, a glossy surface says the resin-rich skin has softened. It says nothing about whether heat or pressure got through to the fibre bed behind it.
- Assume more heat gives a stronger joint. Past the point where the interface is fully diffused, extra energy buys thermal degradation, higher residual stress and a wider heat-affected zone. Not strength.
- Expect to fill a gap with added material. There is no filler rod. If the two faying surfaces do not conform, you are welding air.
- Expect weld strength to match the laminate. A weld competes with the interlaminar properties of the laminate, not with its in-plane tensile strength. Design to the wrong baseline and the joint ends up over-built and still delaminating.
Each of those habits assumes the interface behaves like bulk polymer. In a composite it behaves like a laminate edge: thin, resin-rich, and constrained on both sides by fibres that will never melt.
What Actually Happens at a Composite Interface
Two things have to happen before a weld carries load: intimate contact and chain diffusion. Intimate contact means the two surfaces wet each other completely, closing the microscopic gaps left by surface texture and warp. That takes pressure high enough to overcome the local viscosity of the resin, and time for the resin to flow into the valleys. Chain diffusion means polymer molecules cross the original interface and entangle on the far side. Its depth grows with temperature and with dwell time above the melting point.
Fibres interfere with both. A moulded thermoplastic surface normally carries a resin-rich layer a few tens of micrometres thick, and that layer is the only material available to bridge the joint. Machine it away, or sand through it, and the interface becomes two dry fibre beds pressed together with almost no resin between them. Surface preparation on composites is therefore less forgiving than on plastics. A peel-ply surface welds well, a sanded-through surface does not, and the two look identical under a lamp.
Cooling decides what the joint is worth six months later. Semi-crystalline matrices such as PPS and PA6 develop crystallinity at a rate set by the cooling curve. Crystallinity governs modulus, chemical resistance and dimensional stability, so a joint that is strong on the day of manufacture can behave brittly in service if it was cooled too fast. Residual stress adds to this. The surrounding laminate constrains contraction, and the weld line collects the result.
Voids are the quiet defect. Air trapped between the faying surfaces does not vanish under moderate pressure; it flattens into a disc. A disc of that kind survives a static pull test and then starts a fatigue crack at the edge of the weld, which is why fatigue data and static data so often disagree on the same joint.
Four Methods, Four Different Limiting Factors
Method choice is usually settled by geometry and access, not by strength. Each method delivers heat along a different path, so each one is limited by a different thing.
| Method | How heat gets to the joint | What limits it | Where it fits |
|---|---|---|---|
| Ultrasonic | High-frequency vibration applied at the joint line through a sonotrode | Energy delivery into thick sections; tooling access on both faces | Small stiff parts; ribs spot-welded into a shell |
| Resistance | Current passed through a heating element laid at the interface | The element stays in the joint as a foreign layer | Large areas and long seams |
| Induction | Eddy currents in a susceptor or in the conductive fibre network itself | Uniformity over a large area; one coil per geometry | Closed structures with limited joint access |
| Laser-assisted tape placement | Laser heating during continuous lay-down, consolidated in the nip | Narrow process window that the machine must hold along the whole pass | Long profiles and winding |
Four joining routes compared on the axis that decides a method choice: what limits it, not what it is called.
Ultrasonic welding puts high-frequency energy at the joint line through a sonotrode. Cycle times are short, and the equipment suits small stiff parts and spot-welding ribs into a shell. Its limits are energy delivery into thick sections and the need for tooling access on both sides.
Resistance welding places a heating element at the interface and runs current through it. It handles large areas and long seams. The element stays in the joint afterwards, so it becomes a foreign layer that has to be accounted for in the mechanical design rather than ignored.
Induction welding heats a susceptor or the conductive fibre network itself by eddy currents. No contact is needed at the joint, which helps with closed structures. Uniformity across a large area is the constraint, and each geometry wants its own coil.
Laser-assisted tape placement consolidates continuously and needs no autoclave, which is what makes it attractive for long profiles and for winding. Its weakness is that thermal history is a function of speed. The process window is narrow, and the machine has to hold it for the full length of the pass. The winding study cited above is, in effect, a measurement of what is lost when it does not.
Specifying a Weld You Can Actually Inspect
A drawing that says “welded” is not a specification. What a buyer can reasonably require from a supplier is narrower, and more useful, than a strength number.
- Process parameters recorded per part or per batch: heating method, interface temperature, pressure, hold time and cooling rate.
- Coupons cut from production parts or from a sacrificial tab, not from a separate plaque made for the test.
- Single-lap shear strength together with a peel-type test, because the two fail differently and a joint can pass one while failing the other.
- Void content and crystallinity measured on a polished section, sampled at the weld line rather than in the parent laminate.
- An ultrasonic scan of the weld line, with the acceptance criteria written down before the first part is made.
Design allowables for a weld are normally taken as a fraction of the laminate allowable, and that conservatism is not laziness. Non-destructive testing can find voids and disbonds in a weld line. It cannot find an interface that is bonded but under-diffused, because there is nothing there to reflect an echo. A joint of that kind passes inspection and fails in service.
This is our assessment, not a statement of fact: for structural parts, the process card and the coupon record matter more than the nominal strength printed on a data sheet, because the printed figure describes a coupon consolidated under ideal conditions that the production cell will not reproduce exactly.
Matching the Resin to the Weld
Resin selection sets the equipment, the drying routine and the failure mode.
Polypropylene melts lowest and is the most forgiving to weld, which is why large housings and ducts are usually PP. PA6 gives higher strength and toughness, but it is hygroscopic, and moisture at the interface turns to steam during welding and leaves voids behind. Drying before welding is part of the process, not a housekeeping detail. PPS melts near 280 °C, keeps its properties at temperature and resists most service fluids, which is why the winding data cited above was generated on PPS tape. PAEK and PEEK need interface temperatures above 340 °C plus tooling that survives them, and they are justified when the service temperature leaves no alternative. Amorphous matrices such as PC and PES tolerate a wider temperature band but are more sensitive to creep and to solvents.
For a welded assembly, the prepreg properties that matter most are not the headline mechanical figures. They are the amount of resin available at the moulded surface to bridge the joint, and the fibre architecture that decides how that resin flows under pressure. Areal weight and tow size set both. We supply thermoplastic prepregs across CF/PP, CF/PA6, CF/PPS and CF/PAEK, with carbon fibre from 3K to 24K tow and fabric areal weight from 50 to 600 g/m², slit to a maximum width of 1270 mm. If you send us the joint geometry, the service temperature and the consolidation method, we will tell you which of those three is the real constraint before you cut metal. Specifications are confirmed per grade, and the thermoplastic prepreg range lists the resin systems we hold.