4. Finite element occupant model setup

In a multi-objective HIC/Nij/femur study, the occupant model is not a passive visualization asset — it is the measurement instrument. Every injury metric you later optimize is computed from nodal accelerations, contact forces, and element-level deflections produced by this model. If the dummy is the wrong variant, seated at the wrong H-point, or constrained by a belt that slips through a gap in the contact definition, the optimizer will faithfully converge on numbers that mean nothing. This chapter treats model setup as a measurement-chain problem: choose the right dummy, place it deterministically, define contacts that transmit load the way the physical hardware does, and apply boundary conditions that isolate the response you intend to measure.

Selecting the dummy model and its calibration state

For frontal chest and femur work, the two workhorses are the Hybrid III 50th percentile male (HIII-50M) and the THOR-50M. The choice is not cosmetic. Hybrid III chest deflection is measured by a single sternal potentiometer (or a chestband in modified builds), and its ribcage is comparatively stiff; Nij computed from Hybrid III upper-neck load cells is well characterized but known to be conservative in some loading regimes. THOR-50M provides four-quadrant chest deflection, a more biofidelic clavicle/rib structure, and additional instrumentation (e.g., upper and lower neck load cells, chest IR-TRACC units), which changes both the magnitude and the spatial resolution of the metrics you can extract. If the study's objective function includes chest deflection distribution or rib-level criteria, THOR is the defensible choice; if the study must be comparable to a legacy regulation or a prior Hybrid III dataset, stay with Hybrid III and state that explicitly in the model documentation.

Equally important is the calibration state of the model. A production LS-DYNA or Radioss dummy deck ships with a documented validation matrix (component-level: head drop, neck pendulum, chest impact, knee slider; full-body: sled). Record the deck version, the units system (mm–kg–ms–kN is common in LSTC decks; verify before mixing in any submodel), and whether the model is rigid-body or deformable in the regions of interest. Mixing a rigid femur with a deformable tibia will produce a femur force time history that is numerically clean but physically misleading at the knee joint.

Positioning: from H-point to a repeatable posture

Positioning is where most reproducibility is lost. The dummy must be placed relative to the seat and restraint geometry using a defined procedure, not eyeballed. A practical, deterministic sequence for a frontal sled or crash-pulse simulation is:

  1. Establish the seat reference frame. Define the seat H-point and the seat-back angle from the CAD or the seat FE model, and record them. All dummy placement is expressed relative to this frame.
  2. Set the pelvis first. Translate and rotate the pelvis so its H-point coincides with the target H-point within a stated tolerance (e.g., ±5 mm, ±1°). Lock pelvis translation before adjusting the spine.
  3. Adjust the spine and neck. Apply the torso angle by rotating the lumbar/thoracic joints, then set head angle so the Frankfort plane is approximately horizontal (or per the test standard). Do not use the head to compensate for a torso error.
  4. Set the limbs. Position upper and lower extremities to the target joint angles (hip, knee, ankle) and confirm that the feet contact the toe pan or floor without interpenetration.
  5. Apply the belt. Route the belt through the defined belt-element path, tension it to the specified pre-tension (or zero, if the study isolates the dummy), and verify that the belt lies on the dummy surface, not inside it.

Record every joint angle and the final H-point offset in a positioning log. When the optimizer perturbs seat parameters, the dummy must be re-positioned by the same scripted procedure, not re-placed by hand — otherwise posture variation contaminates the objective function.

Contact definitions: the load path is the measurement

Contacts determine how force enters the dummy, and therefore what the load cells and accelerometers report. Three contact families matter most here:

  • Dummy–seat contact: define surface-to-surface contact between the dummy's pelvis/back surfaces and the seat cushion/back. Use a penalty formulation with a stiffness scaled to the dummy's mass; a too-soft penalty lets the pelvis sink into the cushion and shifts the H-point during the pulse.
  • Dummy–belt contact: the belt must be modeled as a shell/seatbelt element with a defined friction coefficient (typically 0.1–0.3 for webbing–clothing). Verify that the belt does not pass through the shoulder or lap region; a single missed contact segment produces a local force spike that corrupts chest deflection and Nij.
  • Dummy self-contact and limb–interior contact: arms against torso, thighs against the seat pan edge. These are often the source of spurious high-frequency content in femur force; enable them and check the contact energy.

Always inspect the contact energy and the number of penetrations in the initial state (before the pulse). A non-zero initial penetration means the model is pre-loaded, which biases every subsequent metric.

Boundary conditions and metric extraction

Boundary conditions should isolate the response you intend to measure. For a sled-style frontal study, the seat and floor are prescribed a crash pulse (acceleration time history) and the dummy is free to respond; the belt anchorages move with the seat. For a component-level chest or femur study, the dummy may be fixed at the pelvis and the impactor prescribed a velocity. In both cases, define the gravity direction and magnitude explicitly, and confirm that the initial state is in equilibrium (no drift before the pulse).

Metric extraction is a post-processing contract. HIC is computed from the resultant head CG acceleration over the 36 ms window that maximizes the integral; Nij combines upper-neck axial force and bending moment normalized by critical values; femur force is taken from the femur load cell (or a defined cross-section) and filtered per the relevant standard (e.g., CFC 600). Fix the filter class, the sampling rate, and the window definitions before running the optimization, and apply them identically to every design point.

A useful sanity check: run a single baseline design, extract HIC, Nij, and femur force, and compare against the dummy deck's published validation corridor. If the baseline is outside the corridor, fix the model setup before optimizing — an optimizer cannot repair a mis-positioned dummy.
Do not let the optimizer vary dummy positioning, contact friction, or filter settings. These are measurement-chain parameters, not design variables. Freezing them is what makes the multi-objective trade-off between HIC, Nij, and femur force interpretable.
Occupant Model Setup as a Measurement Chain 1. Dummy Model HIII-50M / THOR-50M Deck version + units Rigid vs deformable Instrumentation set 2. Positioning Seat H-point frame Pelvis → spine → head Joint angles logged Belt routed + tensioned 3. Contacts Dummy–seat (penalty) Dummy–belt (μ 0.1–0.3) Self / limb–interior Zero initial penetration 4. Boundary Cond. Crash pulse / impactor Gravity direction Initial equilibrium Anchorage motion Metric Extraction Contract (frozen before optimization) HIC — head CG resultant, 36 ms window Nij — upper-neck Fz, My, normalized Femur force — load cell, CFC 600 Filter class · sampling rate · window definitions applied identically to every design point Baseline Validation Gate Compare HIC / Nij / femur to deck corridor Outside corridor → fix setup, do not optimize Frozen (not design variables): dummy variant · positioning procedure · contact friction · filter settings Design variables belong to seat, restraint, and structure — not to the measurement chain

The figure above summarizes the chain: each stage feeds the next, and the metric-extraction contract is fixed before any optimization begins. The baseline validation gate is the single most valuable checkpoint — it separates a study that can be trusted from one that merely converges.

A minimal positioning and extraction log

To make the setup reproducible across design iterations, keep a short machine-readable log alongside each model. A practical schema (values below are illustrative, not measured):

model: HIII-50M_LSTC_v1.2
units: mm-kg-ms-kN
seat_Hpoint_offset: [0.0, 0.0, 0.0]   # mm, relative to seat frame
torso_angle_deg: 25.0
head_frankfort_deg: 0.0
hip_angle_deg: 95.0
knee_angle_deg: 100.0
belt_friction_mu: 0.20
belt_pretension_kN: 0.0
gravity: [0, -9.81, 0] m/s^2
pulse: 50g_peak_120ms
filters: {head: CFC1000, neck: CFC1000, femur: CFC600}
windows: {HIC: 36ms, Nij: 0-150ms}
baseline_check: PASS | FAIL (corridor ref: deck_validation.pdf)

When the optimizer proposes a new seat-back angle, the positioning script re-runs from this log, the dummy is re-placed deterministically, and the same filters and windows are applied. The resulting HIC, Nij, and femur force are then comparable across the Pareto front. Without this discipline, the multi-objective trade-off is an artifact of setup drift rather than a property of the design.

Finally, treat the dummy deck as a validated instrument with a known domain. If the study pushes into loading regimes outside the deck's validation envelope — very high chest deflection, unusual belt routing, or a femur load path that bypasses the load cell — document the extrapolation explicitly. The optimizer will still return a Pareto front; the engineer's job is to state honestly what that front is measuring.

← Previous 📖 Contents Next →