Skip to main content

From USD Geometry to a Reusable Airflow Field

Use this workflow to turn an approved USD scene into a versioned airflow field that can be reused in repeated equipment or moving-object simulations.

Workflow

Prerequisites

RequirementWhy it matters
Approved scene versionGeometry changes invalidate the relationship between the field and the scene.
Engineering ownerOwns source assumptions, domain choices, acceptance criteria, and calibration.
Runtime ownerConfirms object identity, mass properties, collision representation, and runtime settings.
Reference evidenceMeasurements, public benchmarks, equipment information, or prior approved runs support review.

1. Select the scene and geometry

Start from a fixed scene and model-asset version. Extract the surfaces that materially affect the selected airflow region. Preserve equipment, covers, walls, guides, and openings that shape the flow. Retain visual detail when it affects the engineering question; omit decorative detail that increases grid or review cost.

Record:

  • scene and asset version;
  • coordinate system, origin, and units;
  • included and excluded geometry;
  • geometry simplifications and their expected effect;
  • source ownership and permitted use.

2. Define the simulation domain

Set domain bounds around the region needed for the decision. The domain should include enough upstream, downstream, and lateral space for the selected boundary treatment. Choose a base resolution that represents important openings and obstacles, then document any local refinement used near an air source or wall.

Convert the selected surfaces into a solid representation for the flow solver. Review the resulting mask or voxel view before starting the solve. Small gaps, thin covers, inverted surfaces, or incorrect scale can create a plausible-looking but invalid flow path.

3. Define air sources and boundaries

For each source, record its position, direction, effective opening, operating state, and input provenance. Define walls, openings, inlet treatment, and outlet treatment consistently with the engineering question.

Base tuning changes on traceable engineering evidence. When an effective source value is calibrated, retain both the original evidence and the calibrated value with an explanation.

4. Run and review the field

Use the analytical path for early source-placement and directional studies. Use the GPU LBM path when the decision depends on source interaction, recirculation, walls, or enclosure geometry.

Before accepting the field, review:

  • stability and residual behavior over the selected run window;
  • conservation checks appropriate to the domain;
  • grid sensitivity near important sources and surfaces;
  • expected symmetry or asymmetry;
  • velocity profiles at agreed review planes;
  • sensitivity to changed boundary assumptions.

5. Bake and version the field

Store the reviewed three-dimensional field with enough metadata to reproduce and interpret it:

MetadataRequired record
IdentityField ID, version, scenario ID, and owner.
GeometryScene version, asset versions, bounds, and solid representation version.
ModelModel type, solver version, resolution, time window, and boundary settings.
SourcesSource configuration version and input provenance.
ScalingUnit conversion, density assumption, and any calibration factor.
EvidenceValidation checks, comparison plots, limitations, and reviewer decision.

The runtime should load an immutable field version. A changed field must receive a new version rather than replacing evidence from an earlier decision.

6. Couple the field to moving objects

At runtime, sample airflow at reviewed surface points on each moving object. Combine local airflow with object velocity and rotation, integrate surface forces and torques, and pass the resulting wrench to the selected physics runtime.

Keep the airflow model independent from the runtime adapter. PhysX and Newton may use different contact, collision, and stepping behavior, so compare results statistically and rerun runtime-specific validation instead of assuming frame-by-frame equivalence.

Expected output

The completed package contains the scene version, geometry review, domain definition, source configuration, baked field, runtime settings, force and torque records, scenario results, and reviewer decision.

Failure handling

SymptomResponse
Flow crosses a wall or coverReview scale, surface extraction, thin geometry, mask resolution, and boundary assignment.
Results change sharply with resolutionKeep the study exploratory and run a grid-sensitivity review.
Object behavior looks unrealisticReview field scale, mass properties, center of mass, collision representation, force limits, and runtime stepping.
PhysX and Newton differReview each runtime's contact and collision settings; compare decision metrics rather than individual frames.
A source value has no traceable evidenceHold the field in review until provenance or calibration evidence is available.