3. THUMS and GHBMC Model Alignment: Coordinate System and Unit Consistency Verification

Why alignment must be verified before any comparative run

THUMS and GHBMC are developed by different organizations with different conventions. THUMS (Total Human Model for Safety) is distributed by Toyota Central R&D Labs and typically uses millimetre–tonne–second (mm–t–s) consistent units with a gravity vector of −9.81 m/s² expressed as −9810 mm/s². GHBMC (Global Human Body Models Consortium) models are commonly distributed in kg–mm–ms units, where the consistent force unit is the millinewton and gravity is −0.00981 mm/ms². If you drop a THUMS occupant and a GHBMC occupant into the same deck without reconciling these conventions, the two models will experience accelerations that differ by a factor of 1000, and every injury metric you extract will be meaningless.

This case walks through a concrete verification and reconciliation procedure. The deliverable is a signed-off alignment record that states, for each model, the length unit, mass unit, time unit, derived force unit, gravity magnitude and direction, global origin location, and the transformation applied to bring both models into a common working frame.

Step 1 — Read the unit system from each model's control cards

Do not trust the filename or the README alone. Read the actual control cards. In LS-DYNA, the unit system is implicit in *CONTROL_UNITS (if present) and in the magnitude of physical constants. A reliable cross-check is the density of cortical bone: approximately 2.0e-9 t/mm³ in mm–t–s, 2.0e-6 kg/mm³ in kg–mm–ms, and 2000 kg/m³ in m–kg–s. Query the density of a known part and compare.

*KEYWORD
*CONTROL_UNITS
$     LENGTH      MASS      TIME
         1.0       1.0       1.0
$ (mm, tonne, s) — THUMS convention
*END

For GHBMC, the corresponding block typically reads:

*CONTROL_UNITS
$     LENGTH      MASS      TIME
         1.0       1.0       1.0
$ (mm, kg, ms) — GHBMC convention
*END

Both blocks look identical because *CONTROL_UNITS only scales relative to the deck's native units; the actual convention is encoded in the numeric values of material densities, gravity, and initial velocities. Extract a representative density from each model and record it in the alignment table.

Step 2 — Extract and compare gravity and initial velocity

Gravity is the single most common source of silent mismatch. In THUMS decks you will typically find *LOAD_BODY_Z with a curve whose ordinate is −9810 (mm/s²). In GHBMC decks the same load is often −0.00981 (mm/ms²). If you merge the two decks and keep both load curves, one model will be pulled down 1000× harder than the other.

*LOAD_BODY_Z
$#    LCID        SF        LCIDDR       XC          YC          ZC
         1       1.0         0         0.0         0.0         0.0
*DEFINE_CURVE
$#    LCID        SIDR       SFA        SFO        OFFA        OFFO
         1          0       1.0        1.0        0.0        0.0
$#                 A1                   O1
             0.000000            -9810.000000
             1.000000            -9810.000000

Record the gravity magnitude, direction, and the load-curve ID for each model. If the two models use different time units, the gravity ordinates will differ by the square of the time-unit ratio (1000² = 10⁶ between s and ms, but the mass unit also changes, so the net force scaling is 10³). Do not attempt to reason this out from first principles under time pressure — use the density cross-check from Step 1 to confirm which convention each deck actually uses.

Step 3 — Reconcile coordinate systems and global origin

THUMS and GHBMC both nominally use a right-handed SAE-style vehicle coordinate system: X forward, Y to the occupant's left, Z up. However, the global origin placement differs. THUMS occupant models are frequently positioned with the H-point near the origin, while GHBMC M50 models are often positioned with the origin at the vehicle reference point or at the seat cushion. A naive *INCLUDE of both models into one deck will place them at different heights and longitudinal positions.

Determine the offset by querying the coordinates of a common anatomical landmark — the H-point, the T1 vertebral body centre, or the top of the head — in each model. The difference vector is the rigid translation you must apply to one model to bring it into the other's frame. Apply it via *INITIAL_TRANSLATION on a part set, or by transforming the include file with a pre-processor, and record the applied vector.

Step 4 — Verify with a static settle run

Before running the comparative impact simulation, run each model through a short gravity-only settle (e.g. 50 ms of simulated time with no impactor and no belt) and check that the model does not drift, sink through the seat, or oscillate. A model with a unit mismatch will show an implausible acceleration magnitude — for example, a head acceleration of 9.81e6 mm/s² instead of 9.81e3 mm/s² — or will collapse through the seat within the first few milliseconds. A model with a coordinate mismatch will show a non-zero initial velocity or a sudden jump at t = 0.

Step 5 — Record the alignment in a signed-off table

The alignment record is the artifact that prevents the next engineer from repeating the investigation. Fill in every field; leave nothing as "assumed".

Common failure symptoms and their causes

• Model collapses through the seat in the first millisecond. Gravity ordinate is 1000× too large for the deck's time unit — check whether the GHBMC gravity curve was left in mm/ms² while the deck was converted to seconds.
• Head acceleration reads ~9.81e6 mm/s² in a gravity-only settle. Mass unit mismatch: the model is being accelerated as if it were 1000× heavier or lighter than intended.
• Occupant floats above the seat or intersects it at t = 0. Coordinate origin mismatch — the translation vector was not applied, or was applied to the wrong part set.
• Contact forces are orders of magnitude off but kinematics look correct. Force unit mismatch — the contact stiffness or penalty scale was not converted alongside the mass and time units.

Once the static settle passes and the alignment table is complete, the two models are in a common frame and a common unit system, and the comparative impact simulation can proceed with confidence that any difference in injury response reflects the models' biomechanical behaviour rather than a bookkeeping error.