2. LS-DYNA model setup for DOE

A Design of Experiments study is only as trustworthy as the baseline model it is built on. Before a single run is submitted, the occupant safety model must be reduced to a stable, repeatable reference configuration in which every design variable is exposed as a controllable input and every response is measured at a fixed, well-defined location. This chapter covers the practical work of converting a working LS-DYNA occupant safety deck into a DOE-ready baseline: selecting the airbag and seatbelt variables worth studying, parameterizing the relevant keywords, and verifying that the baseline runs cleanly and produces consistent responses.

From a Working Deck to a DOE-Ready Baseline

Most production occupant safety decks are built for a single confirmation run, not for a matrix of hundreds of runs. They often contain hard-coded values, include files with absolute paths, restarts, and output requests tuned for one specific load case. The first task is to freeze a baseline: a single, self-contained input deck that represents the nominal design, runs to termination without numerical issues, and produces the occupant responses of interest at a stable, converged state. Everything the DOE varies later is measured as a delta from this baseline.

A practical baseline checklist includes: a single master input file with all includes resolved to relative paths; a consistent unit system (typically mm, ms, kg, kN, GPa for automotive crash work); a defined gravity and initial velocity; a settled occupant position (pre-simulated or prescribed); and a stable termination time that captures the full occupant response. Run the baseline twice with identical inputs and confirm that the response values are bit-for-bit or numerically identical — any run-to-run variation indicates a non-deterministic element (random seeds, parallel decomposition effects, or uninitialized state) that will corrupt the DOE.

Identifying Airbag and Seatbelt Design Variables

Design variables should be chosen for engineering relevance, not convenience. A useful filter is: does this parameter plausibly move an occupant injury metric (HIC, chest deflection, neck force, femur load), and can it be changed within a realistic design space? For airbags, the most commonly studied variables are:

• Venting area / vent hole diameter — controls bag pressure decay and occupant loading; typically 20–60 mm equivalent diameter.
• Inflator mass flow rate and gas temperature — set by the inflator model (e.g., *AIRBAG_HYBRID or *AIRBAG_PARTICLE); the mass flow curve is often scaled by a multiplier.
• Bag fabric permeability — expressed through the *MAT_FABRIC leakage coefficient or *AIRBAG_REFERENCE_GEOMETRY porosity.
• Bag volume / geometry scale — a shape or scale factor applied to the folded mesh.
• Tethered length or internal strap — where a tether is modeled, its slack length controls deployment shape.

For seatbelts, the high-leverage variables are:

• Seatbelt webbing stiffness — the load-extension curve slope in *MAT_SEATBELT or the *ELEMENT_SEATBELT force-deflection table.
• Pretensioner force and stroke — the retractor pretensioner pull-in force and maximum stroke.
• Load limiter force level — the force plateau in the retractor or buckle load limiter.
• Anchor point locations — D-ring, buckle, and outboard anchor coordinates, which change belt routing and chest loading.
• Belt slip / friction coefficient — the *CONTACT friction between belt and occupant or belt and D-ring.

A sound DOE typically selects 4–8 variables total across both systems, each with a defined lower and upper bound. Too many variables inflate the run count and confound the analysis; too few miss the interaction effects that matter most in occupant safety (for example, the interaction between load limiter force and pretensioner stroke).

Parameterizing Keywords

LS-DYNA offers several mechanisms for parameterization. The most portable and DOE-friendly is the *PARAMETER keyword combined with & references in the input deck. Define each design variable once, then reference it wherever it appears:

*PARAMETER
$#   name       value
     VENT_DIA    40.0
     LL_FORCE     3.5
     PRET_STROKE  120.0
     BELT_STIFF   1.0

*AIRBAG_HYBRID
$#     sid     sidtyp     rbid      lcid     lcidd    ...
         1         0         0         0         0
$#   ventdia   ...
     &VENT_DIA

*ELEMENT_SEATBELT_RETRACTOR
$#    sid       sidtyp    ...
         1           0
$#   lc1       lc2       ...
     &LL_FORCE  &PRET_STROKE

For DOE drivers that generate decks programmatically (Python, LS-OPT, modeFRONTIER, or a custom script), the cleanest approach is to keep a template deck with placeholder tokens and substitute values per run. A minimal Python substitution pattern looks like:

template = open("baseline_template.k").read()
for run_id, params in design_matrix.iterrows():
    deck = template
    for name, value in params.items():
        deck = deck.replace(f"&{name}", f"{value:.4f}")
    with open(f"run_{run_id:04d}.k", "w") as f:
        f.write(deck)

Two practical rules apply. First, never parameterize a value that also appears in a contact or control card in a way that changes the model topology — only numeric magnitudes should vary. Second, keep the parameter names identical to the DOE matrix column headers so that traceability from run ID to deck is trivial. After substitution, always run a syntax check (ls-dyna i=run_0001.k -s or the LS-PrePost keyword checker) before submitting the batch.

Preparing the Baseline Occupant Safety Model

The baseline must be validated against a known reference before it is trusted as the DOE origin. A typical validation sequence is: (1) run the baseline and confirm normal termination with no negative volume or mass-scaling warnings; (2) extract the occupant responses at the standard measurement locations (head CG acceleration, chest deflection at the sternum, neck axial force, femur load); (3) compare against the physical sled or vehicle test corridor; (4) confirm that the response curves are smooth and free of numerical spikes that would make DOE sensitivity analysis meaningless.

The SVG below summarizes the baseline preparation workflow and the parameter-to-keyword mapping that the DOE will exercise.

Verifying the Baseline Before the DOE

Once the baseline is parameterized, run a small screening set — typically a two-level, one-at-a-time or fractional-factorial sweep of the chosen variables — to confirm that each parameter actually influences the response and that the model remains stable across the design space. A variable that produces no measurable change in any occupant metric is either mis-referenced in the deck or physically irrelevant, and should be removed before the full DOE. A variable that causes negative volume, contact instability, or premature termination at its bounds indicates that the design space is too wide or that the parameterization touches model topology; narrow the bounds or fix the underlying issue.

With the baseline locked, the parameter names fixed, and the screening set confirming that each variable is live and stable, the model is ready for the design matrix. The next step is to choose the DOE strategy — full factorial, fractional factorial, Latin hypercube, or an optimization-driven design — and to define the response extraction that will turn each run's d3plot output into the occupant metrics the study will analyze.