3. Dummy model architecture

Anthropomorphic test device (ATD) models used in occupant-safety simulation are not monolithic. Every production dummy model is a hybrid assembly in which the choice of representation for each body region, the way joints are constrained, and the contact definitions that govern how the dummy interacts with restraints and interior surfaces jointly determine whether the model can be trusted for a given applicability domain. This section examines those three architectural decisions and how they propagate into simulation fidelity.

Multibody and finite element representations

A multibody (MB) representation treats each rigid segment as a body with six degrees of freedom, connected by kinematic joints and characterized by mass, centre of gravity, and inertia tensor. Contact is handled through geometric primitives—ellipsoids, cylinders, planes—whose penetration is converted into force via a penalty or impulse law. Because the equations of motion are formulated in generalized coordinates, an MB dummy is computationally cheap: a full Hybrid III 50th percentile MB model can run a 150 ms frontal pulse in seconds on a single core, making it well suited to design-of-experiments sweeps, optimization loops, and real-time hardware-in-the-loop rigs.

The cost of that efficiency is geometric and material abstraction. An MB dummy cannot represent local deformation of the ribcage under a shoulder-belt load path, nor can it capture the strain-rate-dependent stiffness of foam in a knee bolster. Where the applicability domain requires injury metrics derived from local response—chest deflection time histories, tibia bending moments, acetabular force distribution—a finite element (FE) representation is mandatory. FE dummies discretize each segment into shells and solids with explicit material models (elastic-plastic foams, viscoelastic rubbers, rate-dependent rib cortical bone surrogates) and resolve contact through penalty or Lagrange-multiplier formulations on the actual mesh surface.

In practice, most validated models are hybrid: an FE thorax and pelvis combined with MB limbs, or an FE head-neck complex riding on an MB torso. The architectural question is therefore not "MB or FE" but "which regions need resolved geometry for the metrics this study will report." A useful rule of thumb: if the injury criterion is computed from a rigid-body acceleration at a segment CG (HIC from head acceleration, for example), MB is usually adequate; if it is computed from a deformation or a local force (chest compression, femur force through a load cell), the region must be FE.

Joint definitions and their consequences

Joints are where MB and FE regions meet, and they are the most common source of applicability-domain failure. A dummy joint is not a simple hinge; it is a rotational constraint with a defined torque–angle characteristic, damping, and hard stops. In an MB formulation the joint is a generalized-coordinate constraint with a user-supplied stiffness curve; in an FE formulation it is typically realized as a connector element (bushing or kinematic) between two reference nodes, or as a set of rigid links with a rotational spring.

Consider the lumbar joint of a Hybrid III 50th percentile model. A representative definition uses a stiffness of approximately 1.2 N·m per degree in flexion, damping near 0.8 N·m·s per degree, and rotational limits of about ±35° flexion and ±25° extension. These values are not arbitrary—they are calibrated so that the model reproduces the dummy's physical response in a sled test. If a study replaces the lumbar joint with a free hinge to simplify the model, the torso will pitch forward under belt loading far more than the physical dummy, chest deflection will be over-predicted, and any chest-injury criterion derived from that run is outside the validated domain.

Two practical rules follow. First, joint stiffness and damping must be carried over verbatim from the validated model definition; they are part of the model's identity, not tunable parameters. Second, when a joint connects an MB segment to an FE segment, the reference node placement and the local coordinate system orientation must match the validated configuration. A 5 mm offset in the neck reference node, or a flipped local x-axis at the hip, will produce a systematic bias in head excursion or femur force that is easy to mistake for a restraint-design effect.

Contact formulations

Contact is the third architectural pillar and the one most sensitive to the applicability domain. MB dummies use primitive-based contact: an ellipsoid representing the chest interacts with a belt represented as a series of beam or shell segments, and the contact force is a function of penetration depth and a penalty stiffness. FE dummies use surface-to-surface contact with a penalty or Lagrange formulation, friction coefficient, and—critically—a mesh density fine enough to resolve the contact patch.

The choice matters because contact stiffness and friction directly shape the load path into the dummy. A shoulder belt with a friction coefficient of 0.3 sliding over an FE thorax shell will generate a different chest deflection history than the same belt over an MB ellipsoid, even if the gross kinematics look similar. For restraint-optimization studies where the metric of interest is chest compression, the contact formulation must be the FE one; for studies where the metric is head excursion or belt payout, the MB formulation is often sufficient and far cheaper.

Putting the architecture to work

A defensible workflow starts by listing the injury metrics the study will report, then mapping each metric to the body region and response quantity it depends on. Metrics that depend on local deformation or local force require an FE region with adequate mesh and a validated material model; metrics that depend on rigid-body acceleration can use MB. Joint definitions are then inherited unchanged from the validated model, and contact formulations are selected to match the interacting surface—belt, airbag, or interior trim. The resulting hybrid architecture should be documented explicitly, because the applicability domain of the assembled model is the intersection of the domains of its parts, and a single substituted joint or contact law can silently shrink that intersection.

The architectural decisions described here are not independent: an FE thorax demands an FE-compatible belt contact, which in turn demands a joint definition that transmits the resulting moment without introducing artificial compliance. Treating the dummy as a system—rather than a collection of interchangeable parts—is the precondition for any applicability-domain claim built on top of it.