Simulation-to-view handoff

In engineering simulation, two ideas often get lumped together: where the work runs, and what still means something after data leaves the solver.

Scenaven is built on both.

  • Simulation stays on the server — VTK holds the file you opened, every time step, and the grids filters and Python use.
  • The browser is the view — camera, colormap, probes, and the scene tree stay fast because they do not reload or re-solve the full model.
  • OpenUSD is the handoff — the server writes a scene the viewer reads. That scene is more than a bag of triangles: it carries structure, field names, time, and enough lineage to know what you are looking at.
VTK on the server VTK on the server
OpenUSD handoff OpenUSD handoff
Three.js view Three.js view

The handoff between server and browser should preserve meaning, not only mesh quality. In many industrial workflows, *the contract matters as much as frame rate.*

The concern: not mesh-only delivery

When a web viewer only receives raw vertices and indices, you can still get a pretty picture at high frame rate — and still lose what matters for engineering review:

Idea Why it matters
Units Wrong length or pressure scale invalidates comparisons
Field semantics “Temperature” and “Temp (K)” must map to the array you color by
Variants Different configurations, parts, or layers of the same product
Provenance Which file, block, or run step produced this surface

A serious pipeline keeps those ideas explicit in the handoff between compute and view — not only in someone’s head or in a side spreadsheet.

OpenUSD is often cited here because it can describe scene structure and attributes, not just geometry. Scenaven uses it that way on the path from server to browser. It does not replace VTK for simulation math on the server.

What Scenaven preserves today

One simulation for the UI, filters, and Python

Everything that changes the underlying result — clip, warp, extract, time preparation, Python cells — runs against the same server session as the 3D view. A cut from Scene → Selected prim and a cut from the Python session are not two parallel worlds.

That is the architectural split industrial workflows want: compute next to the data, view in the tab.

Scene structure you can navigate

After conversion (and again after compile when you add filters), the Scene card lists the server’s OpenUSD prim layout—geometry paths, sources, filters, parts, and multi-asset rows:

  • A source row for each dataset you loaded (labeled by name, not internal asset ids)
  • Geometry and nested prims under the visual stage
  • Clip, Warp, and Extract when those operations produced new surfaces
  • Multi-asset sessions with more than one source in the same tree
  • Assembly datasets where parts appear as separate rows, tied to stable paths in the catalog

Visibility, active object, and which layer you color follow that tree. See Scene tree.

Field semantics on the surface

VTK point arrays from the simulation become primvars on the USD mesh the viewer loads. Names are carried through in a stable, legal form (digits, spaces, and punctuation in VTK names are normalized so the viewer can list them reliably).

In the product that means:

  • Scalar names appear in Scene → Selected prim and the legend
  • Ranges drive colormaps and probes
  • Vector fields with three components are supported where the export provides them

The browser does not invent a second naming scheme in a client-side VTK stack.

Time

When your dataset has more than one time step, the server can bake time samples into the scene — positions and field values per step. The viewer scrubs or plays from that description (and may ask the server to prepare animation compile for smooth playback on live sessions). See Time animation.

Prepared Examples on cloud preview can scrub time locally because the example was already converted and stored that way.

Provenance (session scope)

Scenaven records enough lineage to orient you in the session, not a full enterprise audit trail:

What you get Where it shows up
Original filename Scene labeling and server metadata on the mesh
Which asset in a multi-file scene Catalog and prim paths such as /Root/Asset_1/Geometry
Assembly part ↔ simulation block Catalog paths linked to VTKHDF assembly layout
Which geometry branch you fetched Server response metadata the viewer uses when loading

The canonical simulation file (VTK ingest) stays on the server. OpenUSD is the visual and session scene the viewer loads, not a replacement for your solver archive.

Surfaces, not full volumes in the browser

Scenaven visualizes the extracted boundary surface of volumetric data (the same surface path used for standard and assembly VTKHDF uploads). The full grid remains on the server for filters and Python.

That is inspection-oriented: what you can shade, clip, and probe interactively — while volume-scale analytics stay on the server with VTK.

What travels on the wire (conceptual)

Your VTK file (upload or example seed)
    → server reads and caches the simulation (VTK session)
    → server writes OpenUSD (USDC) for geometry and compile results
    → browser fetches bytes, reads the scene, builds the Three.js mesh
    → you orbit, color by field, filter, script, scrub time

You upload VTK (.vtk, .vtkhdf, .vtp, .vtu). You do not upload .usd / .usdc as the source of truth. The server produces the scene after conversion and after each compile. See Load simulation data.

What Scenaven does not include

Use this table when you are comparing Scenaven to a broader OpenUSD or PLM pipeline.

Industrial expectation Scenaven today
Explicit unit contract in USD (e.g. stage metersPerUnit, upAxis as a shared reference across teams) Coordinates and field values pass through as in your VTK file. Scenaven does not add a separate unit layer on top of the solver output.
OpenUSD VariantSets (product options, LOD variants, configuration switches) Not used. Alternatives show up as different prims (sources, filter outputs, assembly parts), not variant authoring.
Rich provenance (run ID, solver settings, full lineage graph) Light — filename, paths, catalog, and session structure. Not a PLM or job-scheduler audit model.
Field metadata schema (SI unit per array, zone type, boundary condition tags in the handoff) Mostly VTK array names → primvars. No separate semantic schema in the scene file.
Full simulation model in the browser By design, no. No WebAssembly VTK port; no shipping the volume to the tab for clip/warp. See Why Three.js?.
OpenUSD as user upload format No. VTK in; USD out for the viewer.

None of that diminishes the core bet: do not reduce industrial review to dumb mesh in the tab — use a shared scene description for structure, fields, and time while VTK remains the compute source of truth.

It does mean Scenaven is strongest for interactive surface inspection, server-side post-processing, and a single session contract between view and scripts — and weaker as a full OpenUSD digital-twin or variant-management hub until those features are explicitly added.

Where the full tool set runs

The handoff architecture is the same in cloud and on-prem. Which actions you can perform depends on connection mode:

Cloud preview On-prem (full tool set today)
Learn the viewer on Examples Yes Yes
Open your own VTK files No (preview) Yes
Clip / warp / extract, dataset info On your data: on-prem Yes
Python session on your simulation On-prem Yes

See Cloud and On-prem for where each action is available today.