4. CAE model architecture for occupant safety

Occupant safety CAE is not a single model but a layered architecture in which each layer answers a different question. The dummy model answers "how does the occupant respond?" The sled or full-vehicle model answers "what environment does the occupant experience?" The restraint submodels answer "how does the system manage that energy?" Correlation quality is bounded by the weakest layer: a perfectly tuned airbag model cannot compensate for a dummy that is positioned incorrectly or a seat structure that is too stiff. This chapter defines the architecture, the fidelity decisions at each layer, and the interfaces that must be controlled before any correlation run begins.

Layer 1: Dummy Models

The dummy model is the measurement instrument of the entire program. Two families dominate occupant safety work: the Hybrid III family (50th percentile male, 5th percentile female, and the 95th percentile variant used mainly for packaging checks) and the THOR family (THOR-50M and THOR-AV), with Q-series dummies for rear-seat child occupant work. Each dummy model is delivered as a validated LS-DYNA or PAM-CRASH deck with a defined certification basis.

Three properties of the dummy model must be verified before it is used in correlation:

  • Certification status. The model must reproduce the physical dummy's calibration corridors — head drop, neck pendulum, thorax pendulum, knee impact, and pelvis impact. If the delivered model does not match the corridors of the physical dummy actually used in the test, the correlation has a built-in bias that no restraint tuning can remove.
  • Joint stiffness and friction. Neck, lumbar, and hip joint properties control the dummy's kinematics. These are often the first parameters adjusted when head excursion or chest deflection does not match, and they must be documented so that adjustments are not silently repeated across programs.
  • Instrumentation output definitions. The model must output the same channels as the physical dummy at the same locations: head CG accelerations (three axes), upper and lower neck loads and moments, chest deflection (IR-TRACC or chest band), pelvis acceleration, and femur loads. Channel naming and filtering (SAE J211 CFC 600 for accelerations, CFC 180 for deflection) must match the test data processing chain.
A common correlation failure is comparing an unfiltered simulation channel against a CFC-filtered test channel. Always confirm that the same filter class and cutoff frequency are applied to both sides before evaluating any curve match.

Layer 2: Sled and Full-Vehicle Setups

The sled model is the workhorse for restraint tuning because it isolates the occupant-restraint system from the vehicle crush. The full-vehicle model is used for final confirmation because it adds the structural deceleration pulse, steering column and knee bolster intrusion, and seat belt anchor motion.

The critical fidelity decision in a sled model is the pulse definition. The pulse is applied as a prescribed acceleration or velocity boundary condition on the sled buck. It must be taken from the physical sled test's measured sled acceleration, not from a generic corridor. The pulse should be applied at the same location where the physical accelerometer was mounted, and the integration of the pulse must produce a velocity change (delta-v) that matches the test within a few percent. A pulse that matches peak acceleration but not delta-v will produce a systematically wrong occupant response.

In the full-vehicle model, the additional fidelity requirements are:

  • Structural deceleration pulse at the seat mounting points. This is extracted from the vehicle model at the seat track attachment nodes, not at the B-pillar or tunnel.
  • Intrusion and steering column motion. Steering column collapse, knee bolster deformation, and pedal box intrusion must be represented because they directly load the occupant.
  • Seat structure compliance. Seat cushion and back frame stiffness affect occupant kinematics and must be correlated separately, typically through a seat pull or seat push test.

Layer 3: Restraint Submodel Fidelity

Restraint submodels are where most correlation effort is spent. Each submodel has a fidelity ladder, and the correct level depends on what the correlation is trying to resolve.

Submodel Low fidelity High fidelity When high fidelity is required
Seat belt 1D seatbelt elements with constant stiffness 2D shell belt with slip ring, retractor, pretensioner, load limiter When belt load, spool-out, or pretensioner timing is a correlation target
Airbag Uniform pressure (UP) volume Control volume (CV) or corpuscular particle method (CPM) with fabric permeability and venting When bag pressure, vent flow, or out-of-position loading is a target
Retractor / load limiter Rigid belt anchor Torsion bar with force-displacement curve and spool-out When chest deflection or belt force is a target
Pretensioner Not modeled Pyrotechnic or motorized with time-dependent pull-in When belt slack removal timing affects occupant kinematics

The airbag model deserves particular attention. A uniform-pressure bag is adequate for a centered, in-position occupant where the bag is fully deployed before contact. It fails when the occupant interacts with the bag during deployment, when venting is asymmetric, or when the bag folds and contacts the occupant unevenly. In those cases, a control-volume or CPM approach is required, and the gas inflow curve, fabric permeability, and vent hole area become calibration parameters.

Interface Control Between Layers

The layers connect through defined interfaces, and each interface must be frozen before correlation begins. The dummy model connects to the seat through contact definitions and initial positioning. The seat connects to the vehicle through the seat track attachment. The restraint system connects to the dummy through belt-to-dummy contact and bag-to-dummy contact.

Initial positioning is the most common source of interface error. The dummy H-point, torso angle, and limb positions must match the physical test setup within the tolerances defined by the test protocol (typically ±10 mm for H-point and ±2° for torso angle). A dummy positioned 20 mm forward of the test position will show earlier belt engagement and different chest deflection, and this error will be misattributed to the restraint model.

Occupant Safety CAE Model Architecture Layer 1: Dummy Model Hybrid III 50th / THOR-50M Certification corridors Joint stiffness / friction Head CG, neck, chest, pelvis SAE J211 CFC 600 / 180 Output: occupant response Layer 2: Environment Sled buck / full vehicle Measured pulse at mount Delta-v match Intrusion / column collapse Seat structure compliance Output: occupant loading Layer 3: Restraint Belt: 1D / 2D shell Airbag: UP / CV / CPM Retractor + load limiter Pretensioner timing Vent / permeability Output: energy management interface interface Interface Control (freeze before correlation) H-point ±10 mm | Torso angle ±2° | Belt routing | Contact definitions | Channel filtering Correlation Targets and Typical Failure Symptoms Head excursion mismatch → check pulse delta-v, belt slip, pretensioner timing Chest deflection mismatch → check load limiter curve, belt stiffness, seat back compliance Neck load mismatch → check dummy joint stiffness, head contact timing, bag pressure

Practical Correlation Workflow

A reproducible correlation run follows a defined sequence. First, freeze the dummy model and verify its certification status against the physical dummy used in the test. Second, extract the measured sled or vehicle pulse and confirm delta-v. Third, position the dummy to the test setup and document H-point and torso angle. Fourth, run the baseline simulation with the as-delivered restraint models. Fifth, compare channels one at a time, starting with the earliest event (belt engagement) and progressing to the latest (bag interaction).

When a channel does not match, the diagnostic order is: timing first, then magnitude, then shape. A timing error (peak occurs too early or too late) usually indicates a pulse, pretensioner, or contact issue. A magnitude error with correct timing usually indicates a stiffness or force-level issue. A shape error (the curve has the right peak but wrong rise) usually indicates a rate-dependent effect such as venting or fabric permeability.

Document every parameter change with its justification and the channel it was intended to improve. Undocumented tuning is the primary reason correlation results cannot be reproduced by a different engineer or in a later program.

The architecture described here is the foundation for all subsequent correlation work. Dummy fidelity, environment fidelity, and restraint fidelity are independent axes, and each must be validated on its own before the combined model can be trusted. The next step in the program is to apply this architecture to a specific correlation case and evaluate the result against the defined targets.

← Previous 📖 Contents Next →