Mesh quality calculator
Type the corner coordinates of a quad or tria and get its aspect ratio, warpage, skew, taper, Jacobian, interior angles, minimum length and Ansys equiangle skewness, each worked out the way Altair's HyperMesh help defines it, drawn on the element and ticked off against your quality criteria.
Presets hold the criteria of Skill-Lync's hood, door, sheet metal, plastic and instrument panel meshing challenges. The second tab checks a whole shell mesh pasted from a Nastran, Abaqus or LS-DYNA file; the third shows which criterion fails first as an element distorts. Free, no sign-up, nothing leaves your browser.
Altair HyperMesh help · Ansys Meshing · 4,230 Skill-Lync meshing reports · Updated 20 Nov 2026
Edit the criteria
Leave N4 empty for a tria. Nodes go round the element in order (either direction). The "Start from" shapes are scaled to the edge size.
The file is read by your browser and never uploaded. Up to 200,000 elements. 8-node quads, 6-node trias and solids are skipped. The demo is a 16 × 10 panel of 5 mm elements on a 400 mm radius with a few problems planted in it.
| Check | Value | Limit | |
|---|---|---|---|
| Aspect ratio | 2.06 | ≤ 3 | ✓ |
| Warpage (°) | 0.0 | ≤ 15 | ✓ |
| Skew (°) | 0.0 | ≤ 45 | ✓ |
| Taper | 0.333 | – | |
| Jacobian | 0.677 | ≥ 0.7 | ✗ |
| Min angle (°) | 76.0 | ≥ 45 | ✓ |
| Max angle (°) | 104.0 | ≤ 135 | ✓ |
| Min length | 2.5 | ≥ 2 | ✓ |
| Max length | 5.2 | ≤ 7 | ✓ |
| Equiangle skew (Ansys) | 0.156 | – |
- Fails: Jacobian 0.677 (limit 0.7).
- Warpage on the two diagonal splits: N1-N3 0.00°, N2-N4 0.00°. Jacobian determinants at the Gauss points: 5.59, 5.59, 3.785, 3.785.
- Ansys equiangle skewness 0.156: excellent. Min length by shortest edge 2.500, by normalised height 2.425.
| Check | Failed | Worst | Element |
|---|---|---|---|
| Load the demo mesh or paste yours, then Check mesh. | |||
What do the mesh quality checks measure?
Each one measures how far an element is from its ideal shape, a square for a quad and an equilateral triangle for a tria, in one particular way. A linear shell element gives its best answers when it is close to that shape. Stretch, shear, taper or bend it and its stiffness and stresses drift, and a folded element cannot be solved at all. Meshing tools report a handful of numbers per element and highlight the ones outside your limits:
| Check | What it looks at | Ideal |
|---|---|---|
| Aspect ratio | Longest edge over the shortest edge, or over the minimal normalised height (a setting) | 1 |
| Warpage | A quad split along a diagonal into two trias: the angle between their normals (quads only) | 0° |
| Skew | 90° minus the angle between lines through the element's mid-sides | 0° |
| Taper | The four corner triangles of a quad against half its area | 0 |
| Jacobian | Smallest over largest determinant of the element's mapping, at the Gauss points or corners | 1 |
| Interior angles | Smallest and largest corner angle, quads and trias set separately | 90° / 60° |
| Min / max length | Shortest edge (or minimal normalised height) and longest edge | near the target size |
The formulas below are the ones in Altair's HyperMesh help. ANSA, Abaqus, Ansys and Nastran have their own versions of several of these checks (HyperMesh can switch to each), so the same element can get slightly different values in different tools. Where HyperMesh's help leaves a detail open, the page says what it assumes.
How is skew calculated?
Quads: join the midpoints of opposite edges with two lines; skew is 90° minus the angle between them. Trias: at each node, take the angle between the line to the opposite mid-side and the line joining the two neighbouring mid-sides; skew is 90° minus the smallest of the three.
For a parallelogram the mid-side lines run parallel to its edges, so its skew is just 90° minus its smaller corner angle: a 45° rhombus has skew 45.0°. That is why, for this kind of distortion, the challenge limits skew 45, quad min angle 45 and quad max angle 135 all fail at exactly the same shear. A right-angled tria made by splitting a square has a skew of 26.6° (tan⁻¹ ½) and corner angles of 45° and 90°.
How is warpage calculated, and how much out-of-plane is 15°?
Warpage only applies to quads: three points always lie in a plane. HyperMesh splits the quad along a diagonal into two trias and reports the angle between their normals:
The help does not say which diagonal. The two splits give different angles, so this calculator works out both and reports the larger (the drawing shows which). The one exception is a concave quad: one of its diagonals runs outside the element, that split folds over to about 180° whatever the shape, so the calculator uses the other. Lift one corner of a square of edge a by h and the larger split gives
so the challenges' 15° is a corner 0.188 × a out of plane, 0.94 mm on a 5 mm element. Altair's "generally acceptable" 5° is 0.062 × a, 0.31 mm. A 0.3 × a lift gives 23.4° on one split and 23.0° on the other. Warpage is what you see when quads go over a curved surface with an element size too coarse for the curvature. Split the worst quads into trias or refine.
What is taper?
At each corner of a quad, take the triangle made by that corner and its two neighbours, and compare its area with half the quad's area. Taper is 1 minus the smallest ratio, so a rectangle or any parallelogram has 0, and trias are given 0:
For a trapezoid whose top edge is shorter than the bottom by a fraction t, that works out to
so the hood challenge's limit of 0.5 lets the top edge shrink to one third of the bottom (t = 2/3). A 5 mm trapezoid with a 2.5 mm top edge has taper 0.333.
What is the Jacobian, and why above 0.7?
A 4-node element is drawn in the solver as a perfect square in its own coordinates (ξ, η from −1 to 1) and mapped onto the real shape. The determinant of that mapping says how much area each bit of the square is stretched to. HyperMesh works it out at the 2 × 2 Gauss points (or at the corners) and reports the smallest divided by the largest:
A parallelogram is a uniform stretch, so its Jacobian is exactly 1 however sheared it is: skew catches shear, the Jacobian does not. Taper and bent-in corners make the stretch uneven. The 5 mm trapezoid above has 0.677 at the Gauss points. At or below 0 the element is folded (a concave quad) and the solver cannot integrate it. Altair calls 0.7 and above generally acceptable at the Gauss points. The corner-node option gives lower numbers for the same element (the corners are further apart in ξ, η), so the same 0.7 is a stricter limit there. A linear tria has a constant determinant, so its ratio is always 1.
On a trapezoid, the sheet metal set's Jacobian 0.7 fails once the top edge is 53% of the bottom (t = 0.468), well before a taper limit of 0.5 would (t = 0.667). So the sheet metal briefs, which have no taper limit, still catch tapered quads.
Aspect ratio: shortest edge or minimal normalised height?
Aspect ratio is the longest edge divided by the minimum length, and HyperMesh offers two ways to get the minimum length (a Check Element setting that changes both checks):
- Shortest edge. Simple, and it misses a flat tria with three fair-length edges.
- Minimal normalised height. For a tria, the shortest distance from a corner to the opposite edge, times 2/√3 so an equilateral tria scores its edge length. For a quad, the smallest of the 8 corner-to-opposite-edge distances and the 4 edges.
The split-square tria shows the difference: on a 5 mm element its aspect ratio is 1.41 by edges but 1.73 by normalised height, and its minimum length drops from 5 to 4.08 mm. Its long edge is 7.07 mm, too, more than the 6 mm maximum length in the door and instrument panel briefs. Splitting 5 mm quads into trias to fix warpage can trip the length check instead.
What quality criteria do Skill-Lync's meshing challenges use?
Skill-Lync read all 4,230 public 2D-meshing reports on 20 Nov 2026 (hood, door inner panel, backdoor, floor panel, sheet metal, plastic components, instrument panel, pressure valve; 1,566 name ANSA, 490 HyperMesh). 408 print the criteria table as text, and they agree on two sets. Below, the count is how many of the 408 give that value:
| Criterion | Hood, door, backdoor, floor panel | Sheet metal, plastic, instrument panel | Reports |
|---|---|---|---|
| Aspect ratio | 5 | 3 | 139 give 5, 261 give 3 |
| Skew | 45° | 45° | 402 |
| Warpage | 15° | 15° | 395 (30° in 6) |
| Taper | 0.5 | – | 139 |
| Jacobian | – (19 of 85 hood reports list 0.7) | 0.7 | 256 |
| Quad angles | 45° – 135° | 45° – 135° | 399 |
| Tria angles | 20° – 120° | 30° – 120° | max 120 in 373 (110 in 22); min 20 in 118, 30 in 70 |
| Trias | ≤ 10% | ≤ 15% | 15% in 277, 10% in 116 |
| Min / max length | hood 2 – 8 mm, door 2 – 6 mm | sheet metal 2 – 7, plastic 0.5 – 3, instrument panel 2 – 6 mm | most common per challenge |
The calculator's criteria list has each of these as a preset. Two things stand out. The challenges allow three times the warpage Altair calls generally acceptable (15° against 5°). And for sheared elements the skew limit duplicates the angle limits, as shown above, so the angle checks and taper/Jacobian do most of the work.
Which check fails first?
The third tab of the calculator distorts a square one way and plots every check divided by its limit, so you can see which line crosses 1 first. At the default settings (shortest edge, Gauss points), the distortion t at which each check fails:
| Distortion | Hood criteria: fails first | Sheet metal criteria: fails first |
|---|---|---|
| Shear (parallelogram), top edge slides by t | Skew at 1, Min angle at 1, Max angle at 1 | Skew at 1, Min angle at 1, Max angle at 1 |
| Taper (trapezoid), top edge shrinks by t | Taper at 0.667, Aspect ratio at 0.785 | Jacobian at 0.468, Aspect ratio at 0.65 |
| Stretch (rectangle), length t | Aspect ratio at 5 | Aspect ratio at 3 |
| Warp, one corner lifted by t | Warpage at 0.188 | Warpage at 0.188 |
| Corner pulled in along the diagonal by t | Max angle at 0.293, Taper at 0.333, Min angle at 0.5 | Jacobian at 0.234, Max angle at 0.293, Min angle at 0.5 |
Stretch fails on aspect ratio and nothing else, shear fails skew and both angles together, and warping fails warpage only. Taper and a pulled-in corner are where the sets differ. On a trapezoid the hood set fails taper at t = 0.667; the sheet metal set has no taper limit, and its Jacobian fails first, at t = 0.468. Pull one corner in along the diagonal and the sheet metal Jacobian fails at t = 0.234, while the hood set waits for the 135° max angle at 0.293. Skew 45 only fails there at 0.586.
How does Ansys skewness compare?
Ansys Meshing and Fluent use a skewness from 0 to 1 instead of an angle. The equiangle version works from the largest and smallest corner angles, with θₑ = 60° for a tria and 90° for a quad:
| Skewness | Ansys calls it |
|---|---|
| 0 | equilateral |
| 0 – 0.25 | excellent |
| 0.25 – 0.5 | good |
| 0.5 – 0.75 | fair |
| 0.75 – 0.9 | poor |
| 0.9 – 1 | bad (sliver) |
| 1 | degenerate |
Ansys says a quality mesh averages about 0.1 in 2D and 0.4 in 3D. A 45° parallelogram scores 0.500 ("good"), and the split-square tria 0.250 (just "good"). Ansys's default for triangles is a second version, based on how much smaller the tria is than an equilateral one in the same circumscribed circle. HyperMesh calls the same quantity "area skew"; for the right-angled tria it is 0.230.
What this calculator does not cover
- Chordal deviation: it needs the CAD surface, not just the nodes.
- ANSA's own definitions. ANSA can compute several checks to match other solvers, and its manual is not public, so values from an ANSA quality report can differ a little from these for badly shaped elements.
- Warped Jacobians. For a non-flat quad the page projects onto the element's mean plane; HyperMesh does not publish its method, so expect the 2nd or 3rd decimal to differ.
- Solids and second-order elements (CHEXA, CTETRA, CQUAD8, CTRIA6) are skipped, as are HyperMesh's Quality Index and the HyperMesh-Alt methods.
Meshing a hood, a door inner panel or a plastic part to a set of quality criteria is the first real job in a CAE team, and every one of these challenges is graded on it. They are part of Skill-Lync's HyperMesh and CAE programs.
Frequently asked questions
How is skew calculated in HyperMesh?
For a quad, HyperMesh joins the midpoints of opposite edges with two lines and reports 90° minus the angle between them. For a tria, at each node it takes the angle between the line to the opposite mid-side and the line joining the two neighbouring mid-sides, and reports 90° minus the smallest. A square or equilateral tria has 0°.
What is a good warpage value for a shell mesh?
Altair's HyperMesh help calls up to 5° generally acceptable. The Skill-Lync hood, door, sheet metal, plastic and instrument panel challenges allow 15° (395 of the 408 reports that print the criteria). On a 5 mm square element, 15° is one corner 0.94 mm out of plane and 5° is 0.31 mm.
Why should the Jacobian be above 0.7?
The Jacobian ratio compares the smallest and largest determinant of the element's mapping at its integration points. 1.0 is a parallelogram; a tapered or bent-in quad drops below it, and a value at or below 0 means the element is folded and the solver cannot integrate it. Altair calls 0.7 and above generally acceptable when measured at the Gauss points.
Is a skew of 45° the same as a minimum quad angle of 45°?
For a parallelogram, yes: its mid-side lines run parallel to its edges, so skew = 90° − the smaller corner angle, and skew 45, min angle 45 and max angle 135 all fail at exactly the same shear. For other shapes they part ways: a trapezoid hardly moves either (taper and the Jacobian catch it), and when a corner is pulled in, the 135° max angle fails well before skew 45.
What does taper mean in mesh quality?
Taper compares the four corner triangles of a quad with half its area: 1 − the smallest ratio. A rectangle or parallelogram has 0. A trapezoid whose top edge is shorter by a fraction t of the bottom has taper t/(2 − t), so 0.5, the limit in the hood challenge, is a top edge one third the length of the bottom.
What is a good skewness in Ansys Meshing?
Ansys uses a different skewness, 0 (equilateral) to 1 (degenerate). Its table: below 0.25 excellent, 0.25-0.5 good, 0.5-0.75 fair, 0.75-0.9 poor, 0.9-1 bad (sliver). It says quality meshes average about 0.1 in 2D and 0.4 in 3D. A right-angled isosceles tria scores 0.25 on the equiangle version.
Related on Skill-Lync
- HyperMesh courses compared and ANSA courses compared
- HyperMesh and ANSA project ideas
- Student project: 2D meshing for sheet metal
- Why meshing matters in FEA and CFD
- Stress concentration factor calculator (how mesh density changes the FEA peak stress)
- Section modulus calculator (the hood section modulus challenge)
- FEA courses compared
- All free engineering tools
Sources
- Altair, HyperMesh help, "Element Quality Calculations: HyperMesh" (2024-2025 help): aspect ratio, minimal normalised height (tria min h × 2/√3; quad, 8 heights and 4 edges), tria skew, taper, warpage ("the quad is divided into two trias along its diagonal"), Jacobian at Gauss points or corner nodes, smallest/largest, 0.7 acceptable; warpage up to 5° acceptable.
- Altair, HyperMesh 2022.1 help, Check Elems panel, 2-d subpanel: "Skew in quads is calculated by finding the minimum angle between two lines joining opposite mid-sides of the element. Ninety degrees minus the minimum angle found is reported"; aspect ratio below 5:1 in most cases; area skew of trias.
- Altair, HyperMesh help, "Element Quality Calculations: Nastran" (for comparison: Nastran's warping coefficient 2H / (D1 + D2)).
- Ansys Meshing User's Guide 2025 R2, "Skewness": equilateral-volume and normalised equiangular skewness, the 0-1 quality table, 0.1 in 2D and 0.4 in 3D. Ansys ICEM CFD help 2024 R2, "Equiangle Skewness".
- Skill-Lync, all 4,230 public 2D-meshing reports, read 20 Nov 2026: 408 criteria tables, the software named, the most common value per challenge.
- Skill-Lync, scripts/verify/mesh-quality.py: re-implements every check, checks them against closed-form answers (parallelogram skew, trapezoid taper and Jacobian, lifted-corner warpage, equilateral and right-angled trias), writes 300 random elements and a demo mesh in three file formats, and the page checks itself against them at build time.