I-DEAS Universal — UNV (.unv)

The I-DEAS Universal File format is an ASCII interchange format used by SDRC I-DEAS, Salome, Code-Aster and many FE tools. A file is a sequence of datasets, each delimited by a line containing only -1, followed by a dataset-id line and the dataset's records.

Format name	unv
Extensions	.unv
Read / Write	✓ / ✓
Extra dependencies	—

Reading & writing

import meshioplusplus

mesh = meshioplusplus.read("mesh.unv")          # or meshioplusplus.unv.read("mesh.unv")
meshioplusplus.unv.write("out.unv", mesh)

write accepts two optional keyword arguments:

• code_aster=False — when True, field data is written as the legacy datasets 55 (node data) and 57 (element data) instead of dataset 2414, matching the Code-Aster convention.
• node_dataset=2411 — the node dataset id to emit; 781 is also accepted.

read takes no keyword arguments.

File structure

    -1
  2411
<node records>
    -1
    -1
  2412
<element records>
    -1
    -1
  2467
<group records>
    -1

Datasets are split by scanning for lines that strip to exactly "-1"; the line right after such a delimiter is the numeric dataset id, and everything up to the next -1 is the dataset body.

2411 / 781 — nodes

Two-line records: label CS1 CS2 color (only label, field 0, is used), followed by a coordinate line. Coordinates may use Fortran exponent notation (D/d instead of E/e), which is normalized before float() parsing. Node labels are arbitrary integers (not necessarily 1..N or contiguous); a label → 0-based index map is built while reading so later datasets (elements, groups) can resolve them.

2412 — elements

Record 1 is 6 integers: label fedesc pid ... ... num_nodes (fields 0, 1, 2, 5 are used; fields 3-4 are ignored). If fedesc is one of the four beam descriptors (11, 21, 22, 24), an extra 3-integer orientation record follows immediately and is discarded on read, always rewritten as 0 0 0 on write — beam orientation data does not round-trip through meshioplusplus. Node-label records follow, gathered (possibly across several lines) until num_nodes labels have been collected.

2467 / 2477 / 2452 / 2435 / 2432 / 2430 — permanent groups

All six of these dataset ids share the same record layout and are accepted on read (FEconv parity); dataset 2467 is used on write. Record 1 has ≥8 integers; field 7 is the entity count n. The next line is the group name. Then 4*n integers follow, laid out as (entity_type, tag, 0, 0) quadruples (typically 2 quadruples — 8 integers — per line). Entity type 8 = node → the group becomes a point_sets entry; entity type 7 = element → cell_sets. Other entity types are collected internally but never emitted.

2414 / 55 / 56 / 57 — field (results) data

Result fields map to point_data (data at nodes) and cell_data (data on elements). The 2414 dataset (the default on write) carries a dataset name (→ the data key, de-duplicated on collision), a location code (1 = nodes, 2 = elements, 3 = nodes-on-elements), and a data characteristic / value count that gives the number of components: 1 (scalar), 3 (vector), 6 (symmetric tensor) or 9 (general tensor), stored as the array's inner dimension. The legacy datasets 55 (nodes) and 57 (elements) carry the same information in a shorter header and are what write(..., code_aster=True) emits; 56 (nodes-on-elements) is read but, like 2414 location 3, is skipped with a warning (barycenter averaging is not implemented). Complex data is likewise skipped. The reserved key unv:pid is never written as a field (it is the element property id carried by dataset 2412).

Cell types & node ordering

The FE descriptor id (record-1 field 1 of a 2412 dataset) selects the element type:

descriptor(s)	meshio++ type	descriptor(s)	meshio++ type
11, 21	line	112	wedge
22, 24	line3	113	wedge15
41, 81, 91	triangle	115	hexahedron
42, 82, 92	triangle6	116	hexahedron20
44, 84, 94, 122	quad	111	tetra
45, 85, 95	quad8	118	tetra10

On write, one canonical descriptor is chosen per meshio++ type: line→21 (beam), line3→24 (beam), triangle→91, triangle6→92, quad→94, quad8→95, tetra→111, tetra10→118, wedge→112, wedge15→113, hexahedron→115, hexahedron20→116.

Cell types with no standard UNV descriptor (e.g. pyramid, quad9, 0-D vertex) are skipped with a warning on write rather than emitted with an invented, non-interoperable descriptor; an unrecognized descriptor on read is likewise skipped with a warning rather than being a fatal error.

Parabolic (second-order) elements use the Salome/Code-Aster mid-node "sandwich" ordering — corner, mid-node, corner, mid-node, … — which differs from meshio++'s "all corners, then all edge nodes" convention. The full permutation table (meshio_position[i] = unv_position[table[i]] on read; inverse applied on write) is:

type	permutation
line3	[0, 2, 1]
triangle6	[0, 3, 1, 4, 2, 5]
quad8	[0, 4, 1, 5, 2, 6, 3, 7]
tetra10	[0, 4, 1, 5, 2, 6, 7, 8, 9, 3]
wedge15	[0, 6, 1, 7, 2, 8, 9, 10, 11, 3, 12, 4, 13, 5, 14]
hexahedron20	[0, 8, 1, 9, 2, 10, 3, 11, 12, 13, 14, 15, 4, 16, 5, 17, 6, 18, 7, 19]

Data mapping

• cell_data["unv:pid"] — the element's physical/material property id (record-1 field 2 of the 2412 dataset); defaults to 1 on write if absent.
• point_sets / cell_sets — from permanent-group datasets, keyed by group name (node groups → point_sets, element groups → cell_sets).
• point_data / cell_data (other than unv:pid) — from / to field datasets 2414 (or 55/57), keyed by the field name.

Quirks & limitations

• Beam-type orientation data (the extra record after beam elements) is read and discarded; the writer always emits a placeholder 0 0 0 record — this is a genuinely lossy round-trip for beam-oriented meshes.
• Node/element labels need not be contiguous or 1-based; meshio++ maintains explicit label→index maps throughout so groups referencing raw labels resolve correctly regardless of numbering gaps.
• The C++ core implements nodes (2411/781), elements (2412), permanent groups (2467 and its aliases, via the UnvInfo side-channel) and field datasets (2414 + legacy 55/57); the shim only falls back to the Python reference for buffer targets. Groups (point_sets/cell_sets) travel out of band because they are not part of the numpy conversion layer — see the UnvInfo side-channel, analogous to AnsysInfo.
• Nodes-on-elements field data (2414 location 3, dataset 56) and complex field data are read but skipped with a warning (barycenter averaging / complex values are not implemented).

Notes

• Implemented against the FEconv format documentation and its FE-descriptor table.
• No reference fixture exists under tests/meshes/unv/; tests round-trip synthetic meshes covering every supported linear and parabolic type, plus an explicit test_groups exercising point_sets/cell_sets.