Structural FEA on 3D-scanned geometry, 2020
Idealized, about 2. As welded, almost 10.
Two 10 mm steel plates were welded, then 3D-scanned as built. Symmetry models of the joint put the stress concentration at the seam at about 2. Modelling the full, distorted assembly raised it to between 5 and almost 10, mainly through bending and torsion caused by post-weld misalignment.
Stress concentration at the weld seam
Range across the three evaluated cross-sections, per model
The plates were no longer flat
On a drawing, this is a simple butt-welded joint in tension. In reality, the two plates rotated relative to each other during welding. Under load, that misalignment adds bending and torsion to the nominal tensile stress, concentrated exactly where the weld toe already raises it.
The question for the analysis: how much of the local stress comes from the weld itself, and how much from the shape of the welded assembly?
To answer it, the model had to start from the measured part, not the nominal geometry.
From scan to analysis-ready mesh
A raw scan is a surface with holes, spatter and noise. Four steps turned it into a volume mesh that resolves the weld toe.
-
1Inspect the raw scan
Holes, weld droplets and scanning noise identified on the measured surface.
-
2Repair the surface
Holes closed and droplets simplified, while the seam features that drive local stress are kept.
-
3Resolve the weld toe
Surface elements of 0.01–0.03 mm at the weld-to-plate transition on the finest mesh.
-
4Refine the volume
Refinement zones along both sides of the seam, at three mesh levels up to 6.4 million cells.
Four models of the same joint
Two symmetry models isolate the weld on one plate and suppress the misalignment. Two full models keep the real geometry: one loaded on the free edge, one loaded through holes, as a test grip would load it. All four carry the same 13 kN static load.
Peak von Mises is the highest value on the X = 25, 75 and 120 mm sections at the finest mesh. Stress-concentration ranges span all three sections. Values are read from the report's plots, so treat them as approximate.
Where the difference appears
Each point is one reading from the report's stress-concentration plot, at three sections along the seam. The symmetry models sit in a tight band near 2 everywhere. The full models are three to four times higher at every section. Select a model to isolate it; hover or focus a point for its value.
Filled points: left plate. Open points: right plate. Read from figure 17 of the 2020 report; some overlapping markers in the original could not be separated.
Why the full models are higher
The symmetry models cut the joint at the seam and hold one plate in its own plane, so the weld toe sees mostly tension. The full models keep the angle between the plates. Pulled straight, a kinked joint tries to straighten itself, which adds a bending moment, and some torsion from the uneven distortion, right at the weld toe. Loading through holes, as a test machine would grip the part, changes both the peak values and how they are distributed along the seam.
How much the peaks depend on the mesh
Peak von Mises stress on each section at three mesh levels, plotted on the same scale. Between the medium and finest mesh, the symmetry models change by at most 6%. The full models change by up to 8% edge-loaded and up to 20% hole-loaded, so their absolute peaks are still mesh-dependent, which is typical at a sharp weld toe in a linear-elastic model. The gap between the idealized and full models is far larger than this sensitivity.
The right-plate symmetry plot labels its middle section 70 mm; the report text says 75 mm.
What the results support
This was a linear-elastic static analysis with S235-like properties (E = 210 GPa, ν = 0.30). Several local peaks exceed the yield strength of S235, so the absolute values describe elastic stress concentration, not the stress the real steel would carry.
The study shows
- Idealized symmetry models substantially underestimate local stress at this joint.
- Post-weld misalignment is the main driver of the extra stress, through bending and torsion.
- How the load is introduced (free edge or holes) changes the peak and where it occurs.
- Where along the seam the stress is concentrated, for targeting further checks or tests.
A design or fatigue decision would add
- Elastic-plastic material behaviour, to see how the peaks redistribute after yield.
- A fatigue assessment with a recognised method, such as hot-spot or effective notch stress.
- Weld residual stress, where it matters for the decision.
- Comparison with strain-gauge or test measurements on the real part.
How the study was set up
- Software
- Salome-Meca / Code_Aster
- Geometry
- 3D scan of two 10 mm welded plates, repaired and retriangulated; incomplete seam-edge regions excluded
- Material
- Linear elastic, isotropic, S235-like: E = 210 000 N/mm², ν = 0.30
- Load
- 13 kN static, four boundary-condition configurations
- Mesh
- Three levels per model, 0.68–6.40 million cells; seam elements 0.03–0.09 mm (coarse) to 0.01–0.03 mm (fine)
- Evaluated
- Peak von Mises stress and stress concentration on sections X = 25, 75 and 120 mm
Have a welded or scanned part that doesn't behave like its drawing?
Send a short description of the part, the loads and what you need to decide. I can scan it, model it as built and tell you where the stress really goes.