3. Crash pulse data acquisition

A frontal crash pulse is only as trustworthy as the measurement chain that produced it. The pulse used to calibrate airbag deployment thresholds, pretensioner timing, and load-limiter behavior is a derived quantity: it comes from accelerometers mounted on the vehicle body, digitized by a data acquisition system, filtered to remove non-physical content, and validated before it is accepted into a restraint model. Errors introduced anywhere in that chain propagate directly into deployment decisions. This section walks through the four decisions that define acquisition quality — where the accelerometers go, how the signal is filtered, how fast it is sampled, and how the resulting data is proven fit for purpose.

Accelerometer placement and the rigid-body assumption

The crash pulse is intended to represent the deceleration of the vehicle's rigid body, not the local vibration of a panel or a bracket. Placement therefore follows one rule above all others: mount the sensor on a structurally stiff, massive part of the body-in-white that stays attached to the main load path throughout the event. The longitudinal tunnel, the rocker section, the base of the B-pillar, and the rear cross-member are typical locations. Thin sheet-metal panels, closures, and components that tear away or buckle early are unsuitable because their local acceleration is dominated by plate modes rather than whole-vehicle deceleration.

Orientation and polarity must be fixed before the test and documented. The standard convention is a right-handed vehicle coordinate system with X forward, Y to the left, and Z upward; a frontal deceleration therefore appears as a negative X acceleration. Every downstream plot, filter, and calibration table must use the same sign convention, because a polarity inversion silently flips the pulse and produces a restraint system that fires late or not at all. Sensors should be bonded or bolted with a stiff, high-frequency-capable mount; a soft adhesive layer acts as a mechanical low-pass filter and can attenuate the sharp onset that matters most for deployment timing.

Redundancy is standard practice. Two accelerometers on opposite sides of the vehicle, or a body sensor paired with a tunnel-mounted reference, allow cross-checking and reveal mounting or cabling faults. The difference between redundant channels should remain small during the rigid-body phase; a growing divergence indicates that one sensor is seeing local structural response.

Filtering: separating the crash pulse from structural noise

Raw accelerometer output contains the crash pulse plus high-frequency structural ringing, sensor resonance, and electrical noise. The industry convention for occupant-safety crash pulses is a low-pass filter with a corner frequency in the range of roughly 100–300 Hz, commonly specified as a Butterworth response applied in both forward and reverse directions (zero-phase, or "filtfilt") so that the filter introduces no time shift. A causal single-pass filter delays the pulse and distorts onset timing, which is unacceptable when the pulse drives deployment logic.

The choice of corner frequency is a trade-off. A lower corner (for example 100 Hz) produces a smooth, repeatable pulse suitable for restraint calibration and for comparing tests, but it rounds off the initial crash onset. A higher corner (for example 300 Hz or above) preserves sharp features but admits structural ringing that can create spurious peaks. The filter specification must be stated with the data: type, order, corner frequency, and whether zero-phase processing was used. Two engineers filtering the same raw file with different settings will produce different pulses and different calibration conclusions.

Always retain and archive the unfiltered raw channel alongside the processed pulse. Filter settings can be revisited; raw data cannot be recovered once discarded.

Sampling rate and anti-alias protection

Crash pulses contain meaningful energy well above the filter corner, and the onset transient can be very sharp. A sampling rate of at least 10 kHz per channel is common for vehicle crash testing, with 20 kHz used when precise onset timing or high-frequency structural content is of interest. The governing constraint is the Nyquist criterion: the sample rate must exceed twice the highest frequency present in the analog signal, and an analog anti-alias low-pass filter must be applied before digitization. Without it, high-frequency content folds back into the passband and appears as false low-frequency oscillation that no digital filter can remove.

Resolution and range also matter. A sensor range of ±500 g to ±2000 g covers frontal crash severity, and 16-bit digitization gives adequate dynamic range for the onset region where accelerations are still small. Time synchronization across channels — accelerometers, seat-belt load cells, and the event trigger — should be verified to within a fraction of a millisecond, since deployment timing decisions are made on that scale.

Data validation before the pulse enters calibration

Validation is a structured check, not a glance at a plot. The following sequence catches the majority of acquisition faults:

  1. Pre-event baseline: confirm the signal sits near zero with low noise before the trigger. A nonzero offset or excessive noise indicates a mounting, grounding, or cabling problem.
  2. Onset consistency: compare the time of first significant acceleration across redundant channels. A large spread suggests a loose mount or a channel that is not on the rigid body.
  3. Pulse shape sanity: a frontal pulse should show a rapid rise, a peak, and a decay consistent with the vehicle's crush behavior. Double peaks, ringing, or a plateau that does not match the structural response warrant investigation.
  4. Integration check: integrate the filtered acceleration to obtain velocity change (delta-v) and compare against the independently measured impact velocity. A mismatch beyond a few percent points to a scaling, polarity, or filter error.
  5. Redundancy comparison: overlay redundant channels and confirm agreement through the rigid-body phase.

The integration check is the most powerful single validation. If the measured impact speed is 50 km/h (13.9 m/s) and the integrated delta-v from the pulse is 12.1 m/s, the discrepancy is too large to accept and the acquisition chain must be reviewed before the pulse is used.

Crash Pulse Acquisition Chain and Validation Gates 1. Placement Rigid body-in-white Tunnel / rocker / B-pillar base X fwd, Y left, Z up Redundant channels 2. Anti-Alias Analog low-pass before digitization Prevents fold-back into passband 3. Sampling ≥ 10 kHz per channel 16-bit, ±500–2000 g Channel sync < 0.1 ms Nyquist respected 4. Digital Filter Butterworth 100–300 Hz Zero-phase (filtfilt) No time shift Settings documented Validation Gates — all must pass before the pulse enters calibration Baseline Pre-event ≈ 0 Low noise floor Onset Redundant channels agree in time Shape Rise–peak–decay matches crush Integration Δv vs impact speed within few % Redundancy Overlay agreement Accepted pulse → restraint calibration Raw + filtered + settings archived

Worked example: a validation failure and its diagnosis

Consider a frontal sled or full-vehicle test at a nominal 50 km/h. The tunnel-mounted accelerometer and a redundant rocker sensor are sampled at 20 kHz, anti-alias filtered in hardware, and digitally filtered with a 4th-order Butterworth at 165 Hz applied zero-phase. The filtered pulse peaks at roughly 38 g and the integrated delta-v comes out at 12.4 m/s against a measured impact speed of 13.9 m/s — an 11% shortfall. The shape looks plausible, so the fault is not obvious from the plot alone.

Working through the chain: a polarity error would flip the sign, not shrink the magnitude; a filter corner set too low would round the onset but rarely lose 11% of the integral; a scaling error in the sensor sensitivity would scale the whole pulse uniformly. The most likely cause is a sensor that is not fully on the rigid body — for example, a rocker sensor mounted on a section that begins to buckle early, so it stops tracking whole-vehicle deceleration partway through the event. Comparing the two redundant channels confirms it: they agree for the first 20 ms and then diverge, with the rocker channel falling away. The fix is to relocate the sensor to a stiffer section and repeat the test; the tunnel channel alone integrates to within 2% of impact speed and is accepted.

This example illustrates why validation must be quantitative. A pulse that "looks like a crash pulse" can still be wrong by enough to shift a deployment threshold by several milliseconds. The integration check against independently measured impact velocity, combined with redundancy comparison, is what converts a plausible-looking trace into a defensible calibration input.

Never calibrate restraint timing against a pulse whose delta-v has not been reconciled with the measured impact velocity. A few percent of integral error can move airbag trigger timing enough to change occupant injury metrics.

Documentation as part of acquisition

Finally, the acquisition record must travel with the data. At minimum it should capture sensor type and serial number, mounting location and orientation, coordinate convention and polarity, sample rate, anti-alias filter specification, digital filter type/order/corner/phase behavior, trigger source, and the validation results including the delta-v reconciliation. This record is what allows a calibration engineer — or a reviewer months later — to reproduce the pulse exactly and to defend the restraint settings derived from it. Acquisition is not finished when the file is written; it is finished when the data has been validated and its provenance documented.

← Previous 📖 Contents Next →