3. CAE role in DFMEA

In a DFMEA for a restraint system, CAE is not a decorative appendix. It is an evidence source that either supports a severity/occurrence rating or exposes a gap that must be closed before the design can be considered robust. The central question for the DFMEA team is not "did we run a simulation?" but "does this simulation represent the failure mode with enough fidelity that its result can be trusted as detection or prevention evidence?" Answering that requires three linked judgments: which simulation type addresses the failure mode, what model fidelity is appropriate, and how well the model correlates with physical test.

Mapping failure modes to simulation types

Restraint DFMEAs typically contain failure modes across several physical domains, and each domain has a natural simulation family. Structural failure modes — bracket cracking, anchor pull-out, retractor housing deformation — are addressed with explicit or implicit nonlinear structural FEA under crash-level loading. Occupant performance failure modes — excessive head excursion, late airbag deployment, submarining, out-of-position injury — are addressed with occupant simulation, either multibody (e.g., MADYMO-class) or finite-element occupant models coupled to a sled or full-vehicle model. Deployment and gas-flow failure modes — insufficient bag fill, asymmetric deployment, module rupture — require fluid-structure interaction or at minimum a validated pressure-volume characterization. Sensing and timing failure modes — late or missed crash detection — are addressed with sensor-to-deployment timing simulation using vehicle pulse data.

The DFMEA discipline is to state, for each failure mode, the simulation type and the specific output metric that would reveal it. "Bracket cracking" is not a metric; "peak plastic strain at the weld toe exceeding the material's elongation limit under the 50 km/h frontal pulse" is. Without a quantified metric, the simulation cannot be used as detection evidence, and the detection rating must reflect that weakness.

Model fidelity: matching the model to the decision

Fidelity is not "higher is better." It is the minimum modeling resolution that makes the predicted metric insensitive to further refinement. A useful way to think about it is in three tiers.

  • Concept-level models — lumped masses, rigid bodies, simplified joints. Fast, useful for packaging and kinematic feasibility, but not credible for strain or injury metrics.
  • Detailed component/subsystem models — shell/solid meshes with material cards, contact, and welds. Appropriate for bracket strength, anchor integrity, and retractor housing deformation.
  • System-level correlated models — full occupant or full-vehicle models with validated restraint components, belt routing, and airbag characterization. Required for occupant injury metrics and deployment timing.

Fidelity decisions must be documented in the DFMEA because they directly affect the credibility of the detection rating. A common failure in DFMEA reviews is citing a concept-level model as evidence for a detailed structural failure mode. The model may run, but it cannot resolve the failure mechanism, so the detection rating is overstated.

A practical fidelity check: refine the mesh or add a physics feature (e.g., strain-rate sensitivity, weld failure) and re-run. If the metric moves by more than the decision threshold, the original model was under-resolved for that decision.

Correlation to physical tests

Correlation is the bridge between simulation and DFMEA confidence. It should be treated as a documented activity with defined metrics, not a qualitative "looks good" review. For restraint performance, correlation typically proceeds through a hierarchy: component-level (belt, buckle, retractor, airbag module), sled-level (occupant kinematics and injury metrics), and vehicle-level (full crash pulse and deployment timing).

A workable correlation protocol uses objective metrics alongside visual comparison. For occupant kinematics, common metrics include head resultant acceleration peak and timing, chest deflection peak, HIC15, and belt force time histories. For structural components, force-displacement curves and peak strain locations are compared. The correlation target should be stated in advance — for example, peak value within ±10% and timing within ±5 ms for the primary injury metric — and the DFMEA should record whether the target was met.

Correlation record (example structure)
Failure mode:      Late airbag deployment — driver out-of-position
Simulation type:   Occupant FE, sled model, 50 km/h frontal pulse
Model fidelity:    System-level, validated module PV curve, belt routing per CAD
Output metric:     Head-to-airbag contact time; HIC15
Test correlation:  Sled test, 3 repeats
Target:            Contact time within ±5 ms; HIC15 within ±10%
Result:            Contact time +7 ms (outside target) — model updated
DFMEA action:      Detection rating revised; design change to sensor calibration

When correlation fails, the DFMEA must reflect it. An uncorrelated model is weak detection evidence, and the corresponding occurrence or detection rating should be adjusted until correlation is re-established. This is where CAE and DFMEA genuinely interact: the simulation result is only as strong as its correlation record.

Reading the simulation-to-DFMEA chain

The figure below shows how a failure mode flows through simulation type, fidelity, correlation status, and finally into the DFMEA rating decision. The key branch is the correlation gate: a model that has not met its correlation target cannot be used to justify a low detection rating, regardless of how detailed it is.

Failure Mode Simulation Type Fidelity Tier Correlation Gate Bracket cracking / anchor pull-out Nonlinear structural FEA (explicit) Detailed component shell/solid + welds Force–displacement vs. component test Excessive head excursion / HIC Occupant simulation (MB or FE) System-level, validated components Kinematics + injury metrics vs. sled test Late / missed deployment Sensor-to-deploy timing simulation Vehicle pulse + module PV curve Contact time vs. sled / vehicle test Correlation met? ±10% peak / ±5 ms YES Low D NO — raise D rating or add design action Failure mode Simulation type Fidelity tier Correlation evidence

Practical implications for the DFMEA team

Three habits make CAE evidence usable in a restraint DFMEA. First, every simulation cited in the DFMEA should carry a one-line evidence tag: simulation type, fidelity tier, output metric, and correlation status. Second, correlation targets should be agreed before the simulation is run, so the result cannot be retrofitted to a convenient threshold. Third, when correlation is not met, the DFMEA should record the gap explicitly and adjust the detection rating rather than leaving the original rating in place.

A useful review question for each failure mode is: "If this simulation were removed, would the detection rating still hold?" If the answer is no, the simulation is load-bearing evidence and must be traceable, correlated, and current with the released design. If the answer is yes, the simulation is supporting context, and the DFMEA should say so. This distinction keeps the DFMEA honest about what CAE actually contributes to restraint system confidence.

Do not treat a converged simulation as a validated one. Convergence confirms numerical stability; correlation confirms physical credibility. DFMEA detection ratings depend on the latter.
← Previous 📖 Contents Next →