Checked against known answers
Every case below runs through the program the way you would run it: STEP file in, faces clicked, adaptive passes. The misses are published with the hits.
| Case | What is compared | Standard (3 passes) | Accurate (4) |
|---|---|---|---|
| Cantilever, RHS 100×50×4 | Tip deflection / bending stress | −0.10% / −0.10% | same |
| Cantilever, round tube 88.9×4 | Tip deflection / bending stress | −0.19% / −0.01% | same |
| Round tube in torsion | Twist / shear stress | 0.00% / +0.03% | same |
| RHS under its own weight, 4 m | Tip deflection / bending stress | −0.12% / −0.26% | same |
| Plate with a hole | Peak stress at the hole | +2.39% | −0.32% |
| Thick cylinder (Lamé) | Hoop stress at the bore | +0.56% | −0.15% |
| Column (Euler) | Buckling load factor | −0.08% | −0.08% |
| Printed PLA bar, layers stacked along it | Tip deflection / factor of safety | −0.42% / +0.09% | same |
| Printed PLA plate with a hole | Peak stress at the hole | +2.62% | −0.36% |
| Plywood plate with a hole (experimental) | Peak stress at the hole | −6.2% | −0.06% |
| Woven carbon fibre plate with a hole (experimental) | Peak stress at the hole | −5.3% | −0.21% |
| NAFEMS benchmark | What is compared | Result at the last pass |
|---|---|---|
| LE1 elliptic membrane | Stress at point D | −0.04% |
| LE5 Z-section in torsion | Stress at point A | +0.58% (three passes) |
| LE6 skew plate | Principal stress at the centre | −0.05% (−3.0% at three passes) |
| LE10 thick plate | Stress at point D | +0.09% (four passes) |
| FV32 tapered membrane | Six natural frequencies | within 0.05% |
| FV52 thick plate | Seven natural frequencies | in-plane 0.00%; bending −1.1% to −3.6% |
Stresses at holes and bores close in as passes are added. A hole in plywood or carbon fibre sheet needs the fourth pass. LE5 stops at its third pass, with nothing left to refine. FV52's bending modes settle below a thick-plate theory value, as a conventional meshed model of the same solid does. Read the full verification document (PDF).
Real parts: jet engine brackets
The cases above have exact answers and simple shapes. These do not: four brackets from SimJEB, a public set of 381 designs for a jet engine bracket, each under four loads. Each was run at Standard from its STEP file and compared with a separate solution of the same part by a conventional method: quadratic tetrahedra, meshed with gmsh and solved with scikit-fem. Figures are the median of the four loads.
| Bracket | Largest deflection | Stress on the surface, mean | High stress (top 0.1%) | Stress field, RMS difference |
|---|---|---|---|---|
| 40 | +0.2% | +1.4% | +1.9% | 9% |
| 148 | −2.9% | +0.7% | 0.0% | 9% |
| 570 | +0.1% | +1.1% | +0.7% | 14% |
| 488, a lattice of struts 1.6 mm wide | −4.0% | +5.2% | +24.9% | 30% |
Bracket 488 is the miss. Its struts are narrower than the cells Standard gives it on a 16 GB computer, and the program says so: it marks the result as low confidence and names the thin material. The reference has about one element across the same struts, so its stresses there are not settled either. Eleven brackets were run in all; against the dataset's own results, which come from first-order tetrahedra and read about 7% stiff and 12% low in surface stress beside the quadratic ones, Gusset's largest deflection is a median 5% higher. Read the full comparison (PDF).
What the settings mean
Standard and Accurate are the number of passes the solver makes, each finer than the last where it matters. Solving and accuracy explains them, and the verdict the program gives on every result.
On your own model
These cases have answers to compare with. Yours does not, so Gusset says over the results whether they have settled, and flags the peaks that never will. Reading results shows how to judge them.
See the answer on your own frame
Download it, open a STEP file, and solve. It costs nothing to find out.