01 — Problem

Manual FEA is Brittle

Manual FEA workflows are brittle: geometry cleanup, mesh controls, boundary condition mapping, and result extraction must be repeated for every design iteration. A typical accelerometer simulation requires 40+ clicks and 15 minutes of setup before the solver even starts.

500 Simulations automated end-to-end
45 min → 2 min Setup time per run (geometry → mesh → solve → extract)
0 Manual interventions after pipeline deployment
02 — Pipeline

Automation Stack Architecture

The pipeline spans the full ANSYS Workbench stack: parametric geometry in SpaceClaim, automated meshing with adaptive controls, boundary condition mapping via named selections, distributed solving, and programmatic extraction via PyDPF.

Geometry SpaceClaim Parametric SCDM API 6 Params Mesh Mechanical Auto-sweep Sizing ctrl Setup Mechanical BC Mapping Named Sels Solve Solvers DDP/Shared Memory opt Post PyDPF Auto-extract CSV/JSON 6 Outputs FIG 1.0 — End-to-End ANSYS Automation Pipeline
03 — Capabilities

Key Automation Capabilities

Each stage of the pipeline addresses a specific pain point in manual FEA workflows, enabling the 500-simulation DoE that feeds the surrogate model.

Parametric Geometry

SCDM Python API drives SpaceClaim — all 6 geometric parameters linked to driving dimensions. Geometry regenerates cleanly at any DoE point.

Adaptive Meshing

Mesh sizing scales with beam thickness; sweep controls on thin sections; automatic defeaturing of small fillets below mesh resolution.

Boundary Condition Mapping

Named selections persist across geometry updates. Fixed support on base, acceleration load on proof mass — mapped via body/scoping IDs.

DesignXplorer Integration

Direct coupling to DesignXplorer for DoE (Latin Hypercube, CCD), response surfaces, and sensitivity analysis — all driven from Workbench journal.

PyDPF Post-Processing

Results extracted programmatically: modal frequencies, mode shapes, stress/deformation fields. Output as structured CSV for surrogate training.

04 — Journal Script

Workbench Journal Excerpt

The core automation loop: parameterize → mesh → solve → extract. Driven entirely from Python journal files with zero GUI interaction. 

# Workbench Journal - parametric automation
from ansys.workbench import Journal
import ansys.mechanical.core as mech

# Parameter setup
params = {
    'L': [8, 16], 'W': [2, 6], 'H': [0.5, 2.0],
    'Lm': [4, 10], 'Wm': [4, 10], 'R': [0.2, 1.0]
}

# Generate DoE - 500 points Latin Hypercube
doe = latin_hypercube(params, n=500, seed=42)

for i, point in enumerate(doe):
    # Update SpaceClaim parameters
    scdm = SpaceClaim.Api.V19.Geometry
    for name, val in point.items():
        scdm.Parameters[name].Value = val
    scdm.Regenerate()

    # Mechanical: mesh, solve, extract
    mech_app = mech.App(version=232)
    mech_app.open(project)
    mech_app.mesh.generate()
    mech_app.solve()

    # PyDPF extraction
    results = mech_app.results
    freqs = results.modal_frequencies()
    stress = results.max_principal_stress()
    deform = results.max_deformation()

    save_csv(f'doe_results_{i:03d}.csv', point, freqs, stress, deform)

print(f'Completed {len(doe)} simulations')
05 — Impact

Reproducible Simulation at Scale

  • Throughput: 500 full FEA runs completed in 48 hours vs 6+ weeks manual.
  • Reproducibility: 100% identical results for identical inputs — eliminates human variance.
  • Data Quality: Structured CSV output enables direct ML pipeline ingestion without cleaning.
  • Scalability: Pipeline runs on DDP (Distributed Domain Parallel) for linear scaling across cores.
Technical Repository