2. Airbag CAE correlation fundamentals

Airbag CAE correlation is the discipline of proving that a numerical occupant-restraint model reproduces the physical crash event with sufficient fidelity to be trusted for design decisions. The model is not validated against "a crash" in the abstract; it is validated against specific regulatory and consumer-test load cases, each with its own barrier, impact speed, dummy, seating position, and injury criteria. Understanding those load cases and the role of the measured test signals is therefore the prerequisite for any meaningful correlation work.

Occupant Safety Requirements and the Criteria That Matter

Occupant safety requirements define what the restraint system must achieve. In the US, FMVSS 208 establishes the federal performance floor; in Europe, UN Regulation No. 94/95 and the Euro NCAP consumer protocol set both legal and rating targets. The criteria are computed from dummy instrumentation, and each one maps to a physical injury mechanism:

• HIC15 / HIC36 — head injury criterion, derived from the resultant translational acceleration at the head CG. HIC15 is the FMVSS 208 basis; Euro NCAP also uses HIC15 with a 700 limit for a 5th-percentile female in the frontal ODB test.
• Head resultant acceleration (3 ms clip) — a peak-hold measure that limits short-duration spikes.
• Neck injury criteria — Nij (combined tension/compression and flexion/extension moments) and, in Euro NCAP, neck tension and shear limits.
• Chest deflection and Viscous Criterion (VC) — chest compression in mm and the rate-dependent VC = 1.3 × (deflection rate × deflection).
• Femur axial force — typically 10 kN per leg in FMVSS 208, with knee slider displacement limits.
• Knee, tibia, and foot criteria — tibia index (TI) and foot/ankle moments in Euro NCAP.

These criteria are not independent. A stiffer airbag that reduces chest deflection may increase neck tension; a later trigger may reduce HIC but raise chest VC. Correlation work must therefore reproduce the whole time history, not just the peak values, because the trade-offs live in the timing and shape of the signals.

FMVSS 208 and Euro NCAP Load Cases

The two regimes differ in barrier type, speed, dummy, and belt configuration, and these differences directly change what the model must capture.

The ODB case is the harder correlation problem. The deformable barrier crushes progressively, so the vehicle deceleration pulse is longer and lower in amplitude than the rigid-barrier pulse, and the 40% overlap induces yaw and lateral occupant motion. The airbag must therefore be modeled with asymmetric deployment and the seat belt with realistic slip through the D-ring. A model tuned only to the rigid-barrier pulse will typically under-predict chest deflection in ODB because the longer pulse allows more occupant excursion before the bag is fully pressurized.

The Role of Test Signals in Model Validation

Test signals are the ground truth against which the model is judged. They fall into three groups, and each group validates a different part of the model:

1. Vehicle-level signals — accelerometers on the B-pillar, sill, and tunnel, plus the vehicle velocity obtained by integration. These validate the structural model and the crash pulse that drives the occupant simulation.
2. Restraint-level signals — belt load cell (shoulder and lap), belt payout, pretensioner firing current, and airbag pressure or tank pressure. These validate the restraint model: belt stiffness, load-limiter force, and inflator mass flow.
3. Occupant-level signals — head CG accelerations, chest accelerations and deflection, neck loads, femur loads. These are the final correlation targets and the basis for the injury criteria.

A common mistake is to correlate only the occupant-level signals. If the B-pillar pulse is wrong, the occupant response will be wrong for the wrong reason, and the model will not extrapolate to a new speed or barrier. The correct sequence is to correlate the vehicle pulse first, then the restraint signals, then the occupant response.

Practical Correlation Workflow and Tolerances

A reproducible correlation exercise starts with a defined set of inputs and a defined acceptance criterion. The following example uses a generic frontal ODB model and is illustrative, not a measured result.

Inputs:
  - Vehicle pulse: B-pillar accel, 0–150 ms, 10 kHz
  - Belt: shoulder load cell, lap load cell, payout
  - Airbag: tank pressure, inflator mass flow curve
  - Dummy: Hybrid III 50th, HIC15, chest deflection, femur load

Steps:
  1. Align all signals to T0 (barrier contact) using the vehicle pulse.
  2. Compare B-pillar accel: peak within ±10%, pulse duration within ±5 ms.
  3. Compare belt shoulder load: peak within ±15%, time-to-peak within ±5 ms.
  4. Compare airbag tank pressure: peak within ±10%, rise time within ±3 ms.
  5. Compare occupant: HIC15 within ±10%, chest deflection within ±5 mm.
  6. If occupant error exceeds tolerance, trace back to the first
     upstream signal that is out of tolerance and correct that model
     parameter before re-running.

Tolerances are not arbitrary. A ±10% peak tolerance on the B-pillar pulse is typical because the pulse is the integration of many structural deformations; a tighter tolerance is rarely achievable without over-fitting. For occupant criteria, ±10% on HIC15 and ±5 mm on chest deflection are common internal targets, but the final acceptance depends on the program. The key is to define tolerances before correlation and to apply them consistently across all load cases.

Typical Failure Symptoms and Troubleshooting

When correlation fails, the symptom usually points to a specific model deficiency. The table below maps common symptoms to likely causes and corrective actions.

The troubleshooting logic is always upstream-first: if the vehicle pulse is wrong, no amount of restraint tuning will produce a physically meaningful occupant response. Once the pulse and restraint signals are within tolerance, the occupant response becomes a genuine validation of the restraint model rather than a curve-fitting exercise.

The diagram above summarizes the signal chain and the upstream-first validation logic. The vehicle pulse is the input that drives the restraint and occupant models; the restraint signals are the intermediate validation; the occupant response is the final target. The two load cases differ in pulse shape and occupant loading, which is why a model correlated only to FMVSS 208 must be re-validated against the Euro NCAP ODB case before it can be used for European programs.

In practice, correlation is an iterative process. Each iteration should change one model parameter at a time, and the effect should be checked against the signal that parameter most directly influences. Documenting the parameter change, the signal response, and the resulting tolerance status creates an audit trail that makes the correlation defensible and reproducible. This discipline is what separates a validated model from a curve-fitted one.