4. Mapping dummy measurements to HBM responses

Interpreting injury metrics from an anthropomorphic test device (ATD) is only meaningful when every measurement can be traced back to a defined physical location, a defined axis convention, and a defined acquisition chain. A head injury criterion value or a chest deflection reading is not a property of the dummy alone — it is the output of a specific sensor, mounted at a specific point, expressed in a specific coordinate frame, sampled at a specific rate, and filtered by a specific channel class. This section establishes the mapping from dummy-mounted transducers to the human body model (HBM) response quantities used in tissue risk assessment, and it covers the practical mechanics of getting that mapping right: coordinate system definitions, instrumentation selection and placement, and data acquisition configuration.

Coordinate system conventions and the sign problem

Every ATD family publishes a local axis system, and these systems are not interchangeable. The two conventions that cause the most mapping errors in practice are the SAE J211 / ISO 6487 vehicle axis system and the dummy-local anatomical axes defined in the ATD user manual. In the vehicle system, +X points forward, +Y points to the driver's left, and +Z points upward, with rotations defined right-handed about those axes. In the dummy-local system, the head and chest transducers are typically referenced to anatomical directions: +X anterior, +Y left lateral, +Z superior, with the origin at the head center of gravity or the thoracic spine reference point.

The mapping is a rigid-body transform. If R is the 3×3 rotation matrix from dummy-local to vehicle coordinates and t is the translation from the vehicle origin to the dummy origin, then a measured acceleration vector a_local maps to a_vehicle = R · a_local, and a measured position p_local maps to p_vehicle = R · p_local + t. The critical discipline is to fix the convention once, document it in the test log, and apply it consistently to accelerometers, angular rate sensors, and displacement measurements. A sign flip on the lateral axis of a chest accelerometer will invert the apparent loading direction and can turn a plausible HBM strain pattern into a physically impossible one.

Instrumentation classes and what each one feeds

The mapping from dummy channel to HBM response is not one-to-one. A single accelerometer triplet at the head CG feeds translational kinematics, from which rotational acceleration is derived by differentiation of angular rate; the resulting six-degree-of-freedom rigid-body motion is then applied as a boundary condition to a finite element head model. Chest deflection measured by a potentiometer or an IR-TRACC (infrared telescoping rod for assessment of chest compression) feeds a different path: it constrains the anterior-posterior compression of the rib cage, which the HBM uses to calibrate rib strain and the associated fracture risk.

The table is a starting point, not a specification. The channel class (CFC 60, 180, 600, or 1000) must match the injury metric being computed, because the filter changes the peak value. A chest deflection filtered at CFC 1000 will report a higher peak than the same signal at CFC 180, and the HBM calibration must use the same filter class as the metric definition. Mixing filter classes between the dummy data and the HBM boundary condition is a silent source of bias.

Data acquisition configuration in practice

A defensible acquisition setup for a sled test with HBM follow-on analysis typically uses a 20 kHz sample rate on all dynamic channels, with an anti-alias filter set at 40% of the sample rate (8 kHz) and the metric-specific CFC filter applied digitally after acquisition. The pre-test sequence matters as much as the sample rate. Zero the channels with the dummy in its final seated posture, record the zero offsets, and run a shunt calibration on every bridge channel so that a gain error can be detected after the test. If a chest deflection channel is zeroed before the dummy is fully settled into the seat, the entire deflection trace will carry a constant offset that propagates directly into the rib strain boundary condition.

Time alignment is the second practical trap. All channels should share a common trigger, and the residual skew between the accelerometer and deflection channels should be verified to be under 1 ms. A 2 ms skew between chest acceleration and chest deflection will produce a viscous criterion that is systematically wrong, because the viscous criterion is the product of velocity and compression at the same instant. The verification method is straightforward: apply a tap or a known step input to a spare channel and confirm that the recorded onset times differ by less than one sample period.

Worked mapping example

Consider a frontal sled test at 30 g peak with a Hybrid III 50th percentile male. The head CG accelerometer reports a peak resultant of 82 g at CFC 1000, and the head angular rate sensor reports a peak of 28 rad/s about the Y-axis. The dummy-local to vehicle rotation is a pure yaw of 0° in this test, so R is the identity and the transform reduces to a sign check. The HBM head model is then driven with the six-DOF rigid-body motion reconstructed from these channels. If the reconstructed angular acceleration, obtained by differentiating the rate signal and filtering at CFC 180, peaks at approximately 4,200 rad/s², that value is the boundary condition the model receives — not the raw differentiated signal, which would be dominated by high-frequency noise.

The chest channel illustrates the deflection path. An IR-TRACC reading of 42 mm at CFC 180, combined with a chest acceleration peak of 55 g, gives a viscous criterion of roughly 0.9 m/s when computed as the maximum of the instantaneous product of the velocity of deformation and the normalized compression. That value is then compared against the HBM-predicted rib strain field to check whether the model reproduces the same injury severity. A mismatch of more than about 15% between the dummy-derived metric and the HBM-predicted response usually indicates a coordinate, filter, or alignment error rather than a genuine model discrepancy, and the acquisition chain should be re-examined before the HBM result is trusted.

The mapping from dummy measurement to HBM response is therefore a chain of explicit, documented decisions rather than a single conversion. Coordinate conventions fix the direction and sign of every vector; instrumentation selection fixes what physical quantity is actually observed; acquisition configuration fixes the frequency content and timing that the HBM receives. When all three are consistent and recorded, the dummy data become a valid boundary condition for tissue risk assessment. When any one of them is left implicit, the resulting strain or risk prediction may look plausible while being physically wrong.