4. Human body model architecture

A human body model (HBM) is not a single mesh but a layered architecture in which geometric fidelity, constitutive behaviour, and kinematic actuation are chosen together to match a specific loading regime. When you assess whether an HBM is applicable to a new scenario — a far-side impact, a low-speed pedestrian strike, a blast load, or a sports injury — the first question is not "how many elements does it have" but "which anatomical structures are discretized, with what material laws, and is the musculature passive or actively recruited?" This chapter unpacks that architecture so you can read a model's documentation critically and predict where it will diverge from reality.

Mesh architecture: from rigid-body shells to voxelized continuum

HBMs occupy a spectrum of geometric resolution. At the coarse end, multibody models represent the skeleton as rigid segments connected by joints and wrap soft tissue as contact surfaces; they are fast and stable but cannot predict local strain. At the fine end, detailed finite element (FE) models discretize bone as cortical shell plus trabecular core, soft organs as hexahedral or tetrahedral continua, and skin/fat as layered shells. Between them sit hybrid models that use deformable FE regions only where injury is expected (head, thorax, pelvis) and rigid bodies elsewhere.

For a detailed FE HBM, mesh density is deliberately non-uniform. Cortical bone is typically meshed with shell or solid elements of 1–3 mm edge length to capture bending and fracture initiation, while the trabecular core uses coarser solids. Solid organs such as the liver and spleen use hexahedral meshes of roughly 5–10 mm to keep hourglass control manageable while resolving gross deformation. Membranes — dura, pleura, pericardium — are modelled as thin shells because their bending stiffness is negligible but their tensile role in containing tissue is not. A practical check when reviewing a model: confirm that element size in the region of interest is small enough that a strain gradient across one element is physically plausible; if a 10 mm solid element spans the full thickness of a rib, fracture prediction is not credible regardless of the material law.

Mesh quality also governs stability. Negative-volume errors, excessive aspect ratios in thin soft-tissue layers, and unconstrained hourglass modes are the usual causes of early termination. Before trusting any output, run a short rigid-wall indentation at low speed and confirm that energy balance closes (hourglass energy below roughly 5% of internal energy) and that no elements invert.

Material models for soft tissue

Soft tissue is nearly incompressible, nonlinear, and rate-dependent. The standard choice is a hyperelastic formulation — often a reduced polynomial or Ogden form — paired with a viscoelastic overlay. A representative set for a liver-like solid organ in an explicit solver might use a first-order Ogden model with shear modulus on the order of tens of kilopascals and a bulk modulus two to three orders of magnitude higher to enforce near-incompressibility, plus a Prony-series relaxation to capture rate sensitivity. Adipose tissue is softer and more compliant; muscle in its passive state sits between the two.

Two failure modes deserve attention. First, fully incompressible formulations in explicit codes cause volumetric locking and prohibitively small stable time steps; use a slightly compressible bulk modulus instead. Second, strain-rate sensitivity is often omitted, which makes a model acceptable in quasi-static compression but over-stiff in blast or high-speed impact. If your scenario involves strain rates above roughly 10 s⁻¹, verify that the model documentation reports rate-dependent parameters rather than a single quasi-static curve.

Injury in soft tissue is usually predicted by strain or stress thresholds rather than by element erosion. For example, a common thoracic criterion flags lung contusion when first principal strain exceeds a calibrated limit, and aortic rupture when circumferential strain in the vessel wall passes its threshold. These thresholds are model-specific; never transplant a threshold from one HBM to another without revalidation.

Material models for bone

Bone is anisotropic, asymmetric in tension and compression, and strain-rate sensitive. Detailed HBMs typically use a piecewise-linear plasticity model with different yield stresses in tension and compression, an elastic modulus around 15–20 GPa for cortical bone, and a failure strain that triggers element deletion. The trabecular core is modelled with a much lower modulus (hundreds of megapascals) and a crushable foam or plasticity formulation that captures compaction under compression.

Element deletion is convenient but dangerous: deleted elements remove load path, so a rib cage can appear to "absorb" an impact it would physically resist. Use deletion only with a calibrated failure strain and check that the post-failure response remains physically sensible. Where fracture pattern matters — rib fracture location for thoracic injury, pelvis ring disruption — verify that the model reproduces published cadaveric force–deflection corridors, not merely that it produces "some" fracture.

Active musculature options

Passive muscle is just another soft tissue. Active muscle adds a contractile element whose force depends on activation, fibre length, and shortening velocity. Three implementation levels are common:

• Passive-only: muscle modelled as hyperelastic/viscoelastic tissue with no activation. Adequate for relaxed-occupant crash scenarios.
• Hill-type lumped actuators: discrete one-dimensional elements spanning joints, with activation input and force–length–velocity curves. Cheap and stable; good for bracing and posture studies.
• Distributed active fibre models: activation applied along the fibre direction of continuum muscle elements, capturing three-dimensional force transmission and deformation. Expensive but necessary when muscle deformation itself is the quantity of interest.

Activation is a control input, not a material property. A bracing study must specify onset time, rise time, and activation level; a model run with 100% activation from t = 0 is not a realistic occupant response. When assessing applicability, ask whether the study's activation profile was calibrated against volunteer or PMHS data for the same posture and warning time.

Reading the architecture as an applicability decision

The three columns of the diagram are not independent choices; they must be matched to the injury target. A model with cortical rib shells but a passive-only muscle layer is well suited to a relaxed-occupant frontal impact, where muscle activation is secondary. The same model is poorly suited to a braced-occupant study, where pre-impact muscle tension changes chest deflection by a meaningful margin. Conversely, a model with distributed active fibres but coarse organ meshes will predict muscle force well and organ contusion poorly.

A defensible applicability statement therefore names four things explicitly: the finest resolved anatomical structure, the constitutive family used for each tissue class, whether musculature is passive or active (and with what activation profile), and the validation corridors the model reproduces. If any of these is missing from the documentation, treat the model as unvalidated for that scenario until you can supply the evidence yourself.

With this architecture in mind, the next assessment step is to compare the model's resolved structures and constitutive assumptions against the specific loading and injury mechanism of your scenario — which is exactly what the applicability-domain workflow formalizes.