Product / Simulate
A stress number you can defend.
Studies belong to the part. Fixtures, loads, material, and mesh settings are saved in the document and follow it. And a static result arrives with a statement about its own convergence rather than a fineness slider position.
Simulate workspace
The result, the mesh it came from, and the numbers behind it.
A finished static study reports von Mises, Tresca shear, both principal stresses, peak displacement, and the minimum factor of safety — alongside the node, element, and degree-of-freedom counts that produced them. The mesh is part of the result, not a setting you configured once and stopped looking at.
Mesh independence
Is that peak stress a property of the part, or of the mesh?
Every stress result depends on how finely the part was divided. A mesh that is too coarse under-reports peak stress, which is the dangerous direction — it reads as a passing factor of safety. The usual answer to that is a fineness slider and a hope.
Chudware ships two independent mesh generators. The default fills the exact solid from its own surfaces with curvature-driven refinement; the second lays a graded lattice through the volume and projects the boundary. They agree on nothing except the answer. Switch the generator on a study, re-run it, and compare: when two unrelated discretizations land on the same number, that number belongs to the part.
Measured, not asserted
A cantilever with a closed-form answer. Analytic tip deflection: 0.2000 mm. Solved on a mesh from the default generator: 0.19270 mm. Solved on a mesh from the second generator: 0.19068 mm.
3.7% and 4.7% under theory, and 1.0% of the analytic value apart from each other. Two generators that share no algorithm, converging on the same beam.
Adaptive refinement
Refine where the error is, not everywhere.
Refining uniformly is the usual workaround for a mesh you do not trust, and it is expensive: element count grows with the cube of the size step, and almost all of those elements land where the stress field is already smooth.
An adaptive study closes the loop instead. Mesh, solve, score the discretization error element by element, tighten the mesh only around the elements carrying the worst of it, and repeat until the error target is met or the pass budget runs out. You set both: one to six passes, and any positive error target.
The score is a Zienkiewicz–Zhu recovery estimator — the solved stress field compared against a smoothed reconstruction of itself — and the worst elements are marked before the next pass, so refinement is spent where it changes the answer.
Every pass reuses the ordinary static study verbatim: the same fixture resolution, the same bonded ties, the same material model, the same result fields. An adaptive result is not a different kind of answer, it is the same answer with its error stated.
Why it needs the second generator
Refining selectively means handing the mesher a spatial size field — a map of how fine to be, place by place. Only one of the two generators accepts one, so adaptive studies run on that generator, and the study says so in its notes rather than switching silently.
That is the honest reason a second mesher exists in the first place. It was not added as a robustness fallback, and it is not described as one.
Worked example
A plate with a hole, and an error indicator you can watch.
A 120 × 60 × 8 mm plate with a 16 mm hole, in tension. Three adaptive passes took the error indicator from 0.157 to 0.100 and the mesh from 41.5k to 187.5k elements. The peak stress settled at 170.0 MPa against a 56.8 MPa net-section nominal — a stress concentration factor of 2.99, computed from the model, against the classical value of 3 for a circular hole.
Note what the run does not claim. 0.100 is a real improvement, and it is not the 0.05 that the error target defaults to. The run does not paper over the difference: the results panel reports the achieved indicator, whether the target was met, how many passes ran, and the reason it stopped. A number that agrees with a textbook is pleasant. A number that shows you its own convergence history is usable.
| Passes run | 3 |
|---|---|
| Error indicator | 0.157 → 0.100 |
| Elements | 41.5k → 187.5k |
| Peak stress | 170.0 MPa |
| Net-section nominal | 56.8 MPa |
| Concentration factor | 2.99, computed |
| Classical value | 3, for a circular hole |
Defining a study
Five things, all of them explicit.
Nothing in a study is inferred from the geometry on your behalf. Fixtures and loads are faces you picked, and you can see them highlighted before the solve starts.
Create the study on the part
Studies live in the document, listed as a tree beside the model. They save inside the native
.step+file, follow the tab when you switch documents, and are still there when the part is reopened months later.Pick the fixtures
Fixture faces are picked in the 3D view and held at zero displacement in all three directions. Picked faces are previewed in the viewport before you run, so you check what you actually selected rather than what you meant to select.
Apply the loads
A load is either a force vector in newtons or a pressure in megapascals, applied to a set of picked faces. For an imported assembly whose parts carry design clearances, bonded ties across the gap keep the load path continuous instead of letting the system go singular.
Take the material from the part
The study uses the active part's own material rather than a throwaway override. Change the material on the model and the study changes with it — there is no second value to keep in sync, and no way for the study to quietly analyze a different alloy than the drawing calls out.
Set the mesh, or let it adapt
Element order, fineness, curvature refinement, size overrides, and the generator are explicit controls. Second-order elements are the default because linear tetrahedra shear-lock in bending, and their mid-side nodes are projected onto the real surface so stress recovery on a fillet or a hole is not reading off a flat facet. Turn on adaptive refinement and the fineness guess is replaced by an error target.
What comes out
Fields, quality metrics, and what the solver actually did.
Results are a full metric table, not one headline. Every stress field is recovered from the same averaged tensor, so switching the plotted field — von Mises to maximum principal to factor of safety — recolors the model without a re-solve.
Mesh quality is reported rather than assumed. Aspect ratio comes back as a minimum, an average, a maximum, and a ten-bucket histogram; the worst scaled Jacobian in the mesh is a single number; the percentage of poorly shaped elements is stated outright. Moving the fineness control has a visible consequence in those numbers, which is the only way to know it did anything.
Solver notes are echoed verbatim: how a pressure area was resolved, which generator produced the mesh, what fell back to what and why. If a mesh option did not apply to the generator you chose, that is a note, not a silent no-op.
| Displacement | Per-node vector and magnitude, in millimeters |
|---|---|
| Stress | von Mises, maximum and minimum principal, maximum shear, and the three normal components |
| Factor of safety | Against yield, with von Mises, Tresca, or maximum-normal-stress as the failure theory |
| Mesh quality | Aspect-ratio min, average, max and histogram; worst scaled Jacobian; percentage of poor elements |
| System size | Nodes, elements, and degrees of freedom of the mesh that was actually solved |
| Convergence | Per adaptive pass: element size, elements, degrees of freedom, error indicator, peak stress, peak displacement |
| Solver notes | Echoed verbatim, including which generator ran and any option that did not apply |
The rest of the study family
Static stress is the one with the convergence story.
The other study kinds are real and in the product, and they are listed here at the resolution they deserve rather than dressed up to match. Where a result is qualitative, the product says so permanently, not behind a beta badge.
| Modal | Natural frequencies and mode shapes for a fixtured part |
|---|---|
| Frequency response | Harmonic base-excitation response by modal superposition; the damping ratio you supply is echoed back as the assumption it is |
| Thermal conduction | Steady-state or transient temperature field. Fixture faces hold a temperature; loads are heat flux or convection with a film coefficient and an ambient. Transient flux loads accept a duty cycle or a piecewise schedule |
| Fatigue | Evaluated on every mean-stress theory — Goodman, Gerber, Soderberg, ASME elliptic, Morrow — with endurance corrections for surface finish, section size, load case, temperature, and design reliability |
| External flow | Drag, lift, side force and their coefficients, plus pressure-coefficient and velocity fields, at an angle of attack and sideslip |
| Motion | Drives a single dimensional mate in an assembly through a range, re-solving the mate graph at every step, optionally checking interference at each sample |
Labeled, permanently
Every flow result carries a banner that does not go away, and it says what the product means rather than what marketing would prefer: the result is qualitative, and a full CFD solver should be used for engineering decisions. The banner reports the Reynolds number it actually ran at, so you can see for yourself which regime you were in.
Where the line is
Linear, elastic, and small.
The static solver is linear and elastic on an isotropic material. Loads are applied at once rather than stepped, displacements are assumed small, and there is no plasticity and no geometric non-linearity. The only joint model is a bonded tie across an assembly clearance — there is no sliding or frictional contact.
A part that would yield in reality does not yield in the model. It reports a stress above the yield strength, which is exactly what a factor of safety is for, and exactly why a factor of safety is not a proof.
What simulation does not do
A study is evidence for an engineering decision, not a certification of a design. Chudware reports the fields, the mesh quality, and how far the error indicator got. It does not sign off on a part.
A result whose mesh you did not inspect, on a load case you did not question, is a color picture. The value of the convergence table is that it makes that distinction checkable instead of a matter of taste.
Next
Run the study on the part, not on an export.
Open a browser workspace, or read how the second mesher and the error estimator are built in the open.