📚 Course ContentsChapter 3 Finite element modeling of airbag inflation

3. Finite element modeling of airbag inflation

An airbag is a thin, highly flexible membrane that folds into a compact package, unfolds explosively, and then presses against an occupant or a rigid barrier. Capturing this behavior numerically requires three coupled decisions: how to discretize the fabric, how to couple the internal gas to the membrane, and how to integrate the equations of motion in time. Each decision constrains the others, and a poor choice in any one of them typically manifests as either a spurious wrinkle pattern, an unstable inflation, or an unacceptably small stable time increment. This chapter works through the three decisions in the order an analyst actually faces them, using a driver-side airbag as a running example.

Shell formulations for the fabric membrane

Airbag fabric is a woven nylon (typically 420–470 dtex, 0.3–0.4 mm thick) with negligible bending stiffness relative to its in-plane stiffness. The natural finite element idealization is therefore a shell, not a solid, and the analyst must choose among three families of shell formulation:

Thin shell (Kirchhoff–Love). Assumes zero transverse shear. Accurate for the fabric's in-plane response but prone to locking and to poor behavior in the highly folded, wrinkled configurations that dominate the pre-deployment state.
Thick shell (Mindlin–Reissner). Retains transverse shear and is more robust under large rotations, but requires shear correction and can exhibit transverse shear locking when the element is very thin.
Membrane-only (no bending). The most common choice for airbag fabric. Bending stiffness is set to a small regularization value purely to suppress hourglassing and to stabilize the folded geometry; it carries no physical meaning.

In practice, most explicit codes use a four-node quadrilateral membrane or a membrane-enhanced shell with reduced integration and hourglass control. A representative setup for the driver airbag example is a 4-node membrane element with a uniform mesh of roughly 8–10 mm edge length, giving about 12,000–18,000 elements for a 60 L bag. The fabric is modeled as an orthotropic elastic–plastic material: warp and weft moduli around 1.0–1.5 GPa, Poisson's ratio near 0.3, and a yield stress of 60–80 MPa with modest hardening to represent the weave's crimp interchange. The thickness is entered as a section property (0.35 mm) and is not meshed through the thickness.

Because membrane elements have no bending resistance, the initial folded state must be pre-stressed or geometrically stabilized. A common approach is to run a short quasi-static "packing" step with artificial bending stiffness, then release it before the inflation step. Skipping this step usually produces an immediate hourglass mode or an element inversion at the fold creases.

Fluid–structure interaction: the gas side

The inflator delivers gas at a mass flow rate that rises to a peak within a few milliseconds and then decays. Two FSI strategies are used in production:

Uniform pressure (UP) method. The bag cavity is treated as a single control volume. At each time step the solver computes the cavity volume V from the deformed membrane, applies the ideal-gas relation p = mRT/V (with mass m updated from the inflator mass flow and temperature T from an energy balance), and applies the resulting pressure uniformly to all membrane facets. This is cheap, robust, and adequate when the bag inflates fast enough that internal pressure equalizes quickly — which is the case for most driver and passenger bags during the first 30–50 ms.

Arbitrary Lagrangian–Eulerian (ALE) or particle-based CFD coupling. The cavity is meshed with a fluid domain, and gas flows through an ALE or SPH/particle formulation that exchanges traction with the membrane at every step. This captures jet effects from the inflator, non-uniform pressure during the early fill, and venting through the bag's exhaust holes. It is substantially more expensive and is reserved for cases where the pressure field is genuinely non-uniform — for example, side curtains with long, narrow cavities, or out-of-position occupant studies where the inflator jet impinges directly on the membrane.

The coupling itself is a partitioned scheme: the structural solver advances the membrane, passes the deformed surface to the fluid solver, the fluid solver updates pressure, and the pressure is returned as a traction load. In explicit codes this exchange happens every time step or every few steps, and the fluid domain must be re-meshed or re-particleized as the cavity grows. A practical rule is to keep the coupling interval at or below the structural stable time increment; loosening it introduces a lag that shows up as pressure oscillations in the first few milliseconds.

Explicit dynamics: time integration and stability

Airbag inflation is a wave-propagation problem with contact and large deformation, so an explicit central-difference integrator is the standard choice. The solver advances nodal accelerations from the assembled internal and external forces, then updates velocities and displacements. There is no global stiffness matrix to invert, which makes each increment cheap and the method robust for contact, but the scheme is only conditionally stable: the time increment must satisfy the Courant condition

Δt ≤ Δt_crit = L_e / c_d

where  L_e = characteristic element length
       c_d = sqrt(E / ρ)  (dilatational wave speed)

For the fabric example, with E ≈ 1.2 GPa and ρ ≈ 1140 kg/m³, the wave speed is about 1025 m/s. An 8 mm membrane element gives Δt_crit ≈ 7.8 µs. A 80 ms event therefore needs on the order of 10,000 increments — which is why mass scaling is so tempting and so dangerous. Scaling the density to raise Δt_crit also raises the inertial resistance of the fabric, and an over-scaled bag inflates sluggishly and under-predicts the peak pressure. A defensible practice is to limit mass scaling to a factor that keeps the added kinetic energy below about 1–2 % of the internal energy, and to verify that the scaled and unscaled runs agree on peak cavity pressure within a few percent.

Two further solver settings matter for airbag models. First, contact between the folded fabric layers must be defined before inflation; a general contact with a penalty stiffness scaled to the membrane stiffness and a friction coefficient of 0.1–0.2 is typical. Second, the inflator mass flow and gas temperature are entered as tabulated time histories, and the cavity energy balance must include the work done by the expanding membrane — omitting it causes the predicted pressure to decay too slowly after peak.

Verification before prediction

Before using the model for occupant or out-of-position studies, run three checks. (1) Free inflation: inflate the bag against no obstacle and confirm that the cavity pressure rises to a plausible peak (for a 60 L driver bag, roughly 150–250 kPa gauge within 20–30 ms) and then decays through the vents. (2) Energy balance: confirm that the ratio of added kinetic energy to internal energy stays small and that hourglass energy is below about 5 % of internal energy. (3) Mesh sensitivity: refine the membrane from 10 mm to 6 mm and confirm that peak pressure and bag stand-off change by less than about 5 %. If any check fails, the usual culprits are an under-stabilized folded state, an over-aggressive mass scale, or a coupling interval that is too coarse relative to the structural time step.

Do not tune the inflator mass flow to compensate for a mesh or stability problem. A model that only matches the pressure curve after inflator tuning will mispredict the deployment kinematics, and the error will reappear as soon as the occupant or barrier geometry changes.

With the shell formulation, the FSI strategy, and the explicit time step settled and verified, the model is ready to serve as the forward solver for the pressure and deformation field predictions that the rest of this course builds on.