Laser and electron beam welding simulation: what FEA predicts and what it cannot
A laser seam takes seconds. The flatness it costs, the joint gap it tolerates and the pores it traps decide whether a fixture, a joint design or a line survives. An equivalent heat source predicts the first two. Melt-pool CFD and a test plan address the third.
Laser welding simulation predicts the weld pool, the narrow heat-affected zone and the distortion that a laser or electron beam seam leaves in a thin part, and how sensitive that result is to clamping and joint gap. Porosity is the exception, which no finite element model predicts, so we build both the workflow that answers the first questions before tooling is cut and the test plan that covers the rest, on your existing welding solver.
The seam takes two seconds. What follows takes weeks. A thin battery lid leaves the fixture out of flat, the next seam starts with a gap the drawing never allowed, and a section through a weld that passed every visual check shows pores. The decisions waiting are specific: clamp pattern and release order, weld path, how much gap the joint design must tolerate. Each answer taken from the line costs a fixture revision.
What does laser welding simulation have to predict?
Per seam, three things: the fusion zone and heat-affected zone, the distortion and residual stress after cooling, and how both move when fit-up and clamping change. The first depends on the mode. TWI puts conduction-limited welding at power densities typically below 105 W/cm2, where the beam is absorbed at the surface and the weld is wide for its depth. Keyhole welding starts typically above 106 to 107 W/cm2: the metal vaporises, the beam drills a vapour-filled cavity, and the weld is deep and narrow.
Electron beam welding belongs to the keyhole family. TWI notes that it runs in a high vacuum because gas scatters the beam, and credits its small heat-affected zone with minimal distortion. A conduction weld can be loaded as a surface flux. A keyhole weld puts its heat in at depth, and a model that deposits it at the surface gets the pool and the shrinkage wrong.
Why is a keyhole weld hard to simulate?
Heat input per unit length, E = ηP/v, sets the average and not the shape. Inside the pool, SaediArdahaei and Pham (Materials, 2024) describe recoil pressure from vaporisation opening the keyhole while surface tension closes it, and tie keyhole instability to porosity. For a cavity of radius R the closing pressure is of order γ/R, so a narrow keyhole closes harder. Surface tension also varies with temperature, and the shear stress τ = (∂γ/∂T) ∇sT drives the Marangoni flow that sets pool width and depth. Add scale: the pool is a millimetre-scale feature that forms and freezes in a fraction of a second, while the part is far larger and keeps cooling for much longer. No single model resolves both.
- Conduction mode. The beam is absorbed at the surface and heat spreads inwards by conduction, so the pool is wide for its depth. A heat flux on the surface is a fair way to load the model.
- Keyhole. At keyhole power density the metal vaporises and the beam drills a vapour-filled cavity. The beam is absorbed down the walls of the cavity, so the heat arrives at depth and the weld is deep and narrow. A model must put the heat at depth too.
- Recoil pressure against surface tension. Vapour leaving the surface pushes the cavity wall outwards, and surface tension pushes the liquid back in. When the balance tips, the keyhole can pinch off.
- Marangoni flow. Surface tension changes with temperature, so a temperature gradient along the free surface drives a flow in the pool. Its direction depends on the alloy and its surface-active elements, and it shapes the width and depth of the pool.
- Pore from a collapsing keyhole. If the keyhole pinches off, the gas in it can be trapped in the solidifying metal. A model with an equivalent heat source has no keyhole, so it cannot predict this.
- Heat-affected zone. Metal that is heated but not melted. It stays narrow in both modes because the heat is concentrated, and in keyhole mode it is long and thin along the cavity.
- From pool to part. Melt-pool CFD resolves the pool, the keyhole and the flow over a short length of seam. It hands an equivalent heat source, a fitted cone of heat, to the part-scale model that predicts distortion.
Equivalent heat source or melt-pool CFD?
An equivalent heat source replaces the pool physics with a fitted volume of heat. For keyhole welds a conical Gaussian source is common: q(r, z) = 9ηP e3 / [π(e3 − 1) H (rt2 + rtrb + rb2)] · exp(−3r2/r02), where r0 narrows linearly from rt at the surface to rb at depth H and the constant makes the absorbed power ηP. We fit η, rt, rb and H so the fusion boundary matches your cross-section. An outline does not fix efficiency and shape separately, so we also match thermal cycles or measured distortion.
Melt-pool CFD solves the flow, the free surface, the recoil pressure and the Marangoni stress, and shows when the keyhole turns unstable, which no equivalent source can. It covers a short length of seam, not a housing. The two belong together: CFD calibrates the pool for a new alloy, spot size or speed, and the equivalent source carries it into the part-scale model where distortion is decided.
How do you predict distortion in thin sheets and thin-walled parts?
With a sequentially coupled model: the thermal solution first, then a thermo-elastic-plastic solution that reads the temperature history. The heated zone is held back by cold material and yields; on cooling it shrinks against that material, and the mismatch εth = αΔT is left as plastic strain that bends the part. A narrow heat-affected zone leaves less plastic strain, but a thin lid has little stiffness, so what remains can still break a flatness tolerance. Clamps decide what you see: the part stays flat while stress is locked in, and release is a springback step. The model needs clamp contacts, forces, seam order and release order, or it predicts a part nobody builds.
Joint gap is the second sensitivity. TWI notes the relatively low tolerance of laser welding to joint position, misalignment and gaps, with heads aligned to within a few tenths of a millimetre or less, typically, to avoid porosity or lack of sidewall fusion. The gap you measured is not the gap at the beam: the seam behind it shrinks, and the plates ahead rotate against tacks and clamps. We model that movement, then sweep gap, offset and clamp pattern, so your drawing carries a tolerance map instead of a guess.
- Beam and heat input. The beam travels along the seam. Heat input per unit length, E = ηP/v, sets how much metal is heated for each millimetre of seam, and the equivalent heat source carries it into the model.
- Shrinkage behind the beam. The seam cools and contracts across its width and pulls the plates towards it. The plastic strain this leaves behind is what distorts the part.
- Gap ahead of the beam. The dashed lines are the joint as fitted. Shrinkage behind the beam turns the plates, so the gap at the beam narrows and the far end opens. Tacks and clamps decide how far.
- Clamps. Clamps hold the part flat during welding, so the stress stays locked in. The model needs their positions, their forces and the order in which they open.
- Angular distortion after release. When the clamps open, the part springs into the shape the locked-in strain dictates, with its edges lifting towards the weld side. The dashed line is the flat shape it was held in.
- The model chain. A thermal step with the equivalent heat source, a mechanical step with the clamps on, then a release step. Each step reads the one before, and each can be checked against a measurement.
Battery parts show both. The open-access review of tab-to-busbar interconnections in Micromachines (2025) finds laser welding of tabs to busbars fast and precise, but in need of careful control of spatter, porosity and brittle intermetallics in dissimilar joints, with no process universally superior. A thermal-mechanical model shows how much heat reaches thin foils and what stress the tab keeps; intermetallic growth and joint strength come from your tests. For the weld as a connection in a pack model, see our EV battery pack crash article.
What can finite element analysis not predict in a laser weld?
Porosity. TWI on defects in laser welding traces it to gas trapped in the solidifying metal, from surface contamination, dissolved gas or an unstable keyhole that collapses, and finds partial-penetration welds more prone. A finite element model with an equivalent source has no keyhole, so no pores. Melt-pool CFD resolves collapse and shows what makes it likelier, but contamination, gas content and shielding are not in the model. The same page lists solidification cracking and brittle microstructures from the high cooling rate, which depend on composition and test results too.
So the model plans the tests. It finds the fit-up, speed and penetration conditions that deserve coupons and sections; sections and your monitoring confirm them; acceptance levels come from your application standard. ISO 15614-11:2025 specifies requirements for qualification testing of welding procedure specifications for electron or laser beam welding of metallic materials, new parts and repair included. Simulation prepares that test matrix and does not replace it.
A prediction is not a qualification
| Question | Heat source | Model | Decision it supports |
|---|---|---|---|
| Thin lid or wall: flatness after welding | Equivalent source fitted to your cross-sections | Thermal then elastic-plastic, with clamps and release | Clamp pattern, release order, weld path, pre-set |
| How much joint gap or beam offset is tolerable | Same source, gap and offset as inputs | Part-scale model with plate movement | Tolerance on incoming parts and fixture accuracy |
| Keyhole stability and porosity risk | Melt-pool CFD with recoil pressure and Marangoni stress | Pool-scale CFD, then coupons and sections | Which parameter window to test and monitor |
| Tab, foil stack or busbar joint | Source fitted to the thin stack | Thermal-mechanical on the stack, tests for metallurgy | Heat reaching the cell side and the stress in the tab |
| Many seams, many fit-up cases | Planned runs of the calibrated model | Surrogate trained on them, with a gate | Which cases get a full run |
Where does AI help laser welding simulation, and where does it not?
- Heat-source calibration. An optimiser fits efficiency and source shape to your sections and thermal cycles and reports the spread of solutions that fit equally well.
- Fit-up screening. Planned runs over gap, offset, speed, power, clamps and seam order train a surrogate that returns distortion in seconds, so full runs go to cases near your limit. See surrogate models for CFD and FEA.
- A signal gate. Whatever your line records per weld is classified against sectioned parts, so flagged welds reach inspection first.
It cannot find pores in a model that has none, cannot extrapolate beyond its training runs, and never decides whether a distortion or defect level is acceptable. At the conduction-to-keyhole boundary the request goes to the full model. See AI surrogate models you can sign off.
From one weld study to a fit-up and fixture capability you keep
The return comes when the next lid, housing or busbar joint runs through the same chain, without a rebuilt model or a fixture trial to learn what the model could have said.
- Physics model. CFD and FEA set up once, as templates: geometry clean-up, mesh, solver settings and post-processing that every later run reuses.
- Validation. Mesh and time-step studies, and comparison with tests or full-scale data, so the model's uncertainty is known before it is trusted.
- Automated runs. Scripted runs over the parameters and operating conditions that matter, on your hardware or in the cloud.
- AI surrogate. A machine-learning model trained on those runs answers in seconds and says how sure it is; low-confidence questions go back to the solver.
- Decision tool. Optimisation, trade-off charts or operating guidance, in a form your engineers and operators use every day.
- Field data. Test, trial and monitoring data flow back to recalibrate the model, so it improves with every project instead of being rebuilt.
- A calibration set: your cross-sections, thermal cycles and measured distortion, stored with the heat-source parameters fitted to them.
- A weld template in your existing welding solver: seam list, order, clamp pattern and release step, so a new part is an input file.
- A fit-up sweep over gap, beam offset, speed, power and clamps that writes the tolerance map your drawings use.
- A validation record against your last measured parts, with tolerances set before the run.
- A surrogate with a stated range and a gate at the conduction-to-keyhole boundary.
- Handover and training, so your engineers extend it themselves. See our training and software development services.
Questions engineers ask about laser welding simulation
What is the difference between conduction and keyhole mode?
Can FEA predict porosity in a laser weld?
How do you simulate laser welding distortion?
Is electron beam welding simulated differently from laser welding?
Can simulation replace welding procedure qualification?
If a fixture, a joint design or a launch is waiting on a distortion or fit-up answer, send us two weld cross-sections with their process parameters, the drawing with seam order and clamp layout, and a measured flatness or gap record from your last build. We reply with how we would calibrate the heat source, which fit-up cases we would sweep first, and what the validation record should hold. Start with a conversation. See also the welding simulation hub and the laser and electron beam welding page.
