1. THUMS Total Human Model for Safety: Model Variant Selection for a Target Injury Scenario

Why Variant Selection Is the First Engineering Decision

THUMS (Total Human Model for Safety) is distributed as a family of finite element models rather than a single mesh. Each variant encodes a specific posture, a specific anatomical region set, and a specific set of boundary conditions that the developers validated. Choosing the wrong variant does not merely reduce accuracy — it can invalidate the entire load path. A pedestrian model loaded as a seated occupant will produce belt interaction artifacts that never occur in reality; a standing model used for a frontal sled test will lack the seat pan contact geometry that governs pelvis kinematics.

The selection task therefore has three coupled inputs: the impact scenario (what is hitting what, from which direction, at what velocity), the occupant posture implied by that scenario, and the anatomical coverage required by the injury metrics you intend to extract. The variant must satisfy all three simultaneously. This case walks through a documented selection for one specified scenario and produces a rationale record that can be attached to the study.

Defining the Target Scenario Precisely

Before any model is opened, the scenario must be reduced to a set of selection-driving attributes. Vague descriptions such as "frontal crash" are insufficient because they do not constrain posture or coverage. The following attribute set is the minimum needed to discriminate between THUMS variants:

• Loading direction and contact partner — e.g., frontal barrier, steering-wheel airbag plus three-point belt, or vehicle front-end contact with a standing pedestrian.
• Occupant posture and restraint state — seated upright with belt and airbag, seated with belt only, standing, or walking gait.
• Required anatomical coverage — which regions must be meshed in enough detail to output the injury criteria of interest (head/brain, cervical spine, thorax with ribs and internal organs, abdomen, pelvis, lower extremities).
• Output metrics — the specific criteria the study will report, since a variant that lacks a detailed brain mesh cannot support a brain-tissue metric regardless of how well it matches posture.

For this case the specified scenario is: a belted driver in a nominal upright seated posture subjected to a 40 km/h frontal full-width impact, with a steering-wheel airbag deploying, and injury metrics required for head, cervical spine, thorax, and pelvis. This is a concrete, reproducible specification that can be checked against each candidate variant.

Comparing Candidate Variants Against the Specification

THUMS variants differ along the three axes identified above. The comparison below is organized by the attributes that actually discriminate between candidates for this scenario, not by a generic feature list. Values are stated as selection-relevant characteristics; exact mesh counts and version strings must be confirmed against the specific THUMS release documentation in use, since they change between releases.

The seated occupant variant is the only candidate that satisfies posture, restraint state, and coverage simultaneously. The pedestrian and standing variants fail on posture and contact partner; the rear-seat variant fails on restraint state and steering-wheel interaction. This is a clean single-candidate outcome, which is the common case — but the rationale must still be recorded, because a reviewer will ask why the alternatives were excluded.

Selection Logic and Exclusion Path

The figure below encodes the decision as a discrimination structure rather than a linear procedure: each candidate is tested against the three selection axes, and the first failing axis excludes it. This makes the exclusion reasoning auditable and shows at a glance which axis did the excluding.

Documenting the Rationale

The selection is only defensible if the reasoning is written down in a form a reviewer can audit. The rationale record should contain, at minimum: the scenario specification as reduced attributes, the selected variant with the release identifier, the exclusion of each rejected candidate with the specific failing axis, and any open items that must be resolved before the simulation is trusted. The open item in this case is coverage depth: the seated occupant variant includes head and cervical spine detail, but the specific mesh resolution must be checked against the brain and spine metrics the study intends to report. If the required metric needs a finer brain mesh than the variant provides, the selection is not complete and a different release or a refined variant must be considered.

A practical way to keep this honest is to record the version string and file path at the moment of model import, not from memory afterward. THUMS releases change mesh detail and posture between versions, and a rationale that cites "the seated model" without a version is not reproducible. The study log entry should read something like the variant name plus the release identifier plus the import path, so that a later reviewer can load exactly the same model.

Verification Before Proceeding

Before the selected variant is used for any loading, three checks confirm the selection is sound. First, confirm the posture in the model matches the specified posture — inspect the pelvis and torso orientation rather than assuming from the variant name. Second, confirm the restraint geometry exists: belt segments and airbag contact surfaces must be present and positioned for the seated configuration. Third, confirm the anatomical regions required by the output metrics are meshed at the needed detail. If any check fails, return to the discrimination step; the selection is not final until all three pass. Once they do, the rationale record is complete and the model is ready for the coordinate and unit consistency checks that precede comparative simulation.