4. Data acquisition in crash tests

Crash test data acquisition is the front end of the entire CAE correlation workflow. If the acquisition chain distorts, aliases, clips, or misaligns the physical signals, no amount of downstream filtering or model tuning can recover the truth. This chapter examines the four acquisition parameters that most directly determine whether measured acceleration, force, and displacement signals are trustworthy enough to correlate against simulation: sampling rate, anti-aliasing filtering, dynamic range, and synchronization across channels and video.

Sampling Rate and the Nyquist Constraint

The sampling rate fs determines the highest frequency that can be represented without ambiguity. The Nyquist–Shannon theorem states that a signal component at frequency f is uniquely representable only if fs > 2f. In crash testing, the useful signal content is not limited to the rigid-body deceleration pulse (typically below 100 Hz). Structural ringing of the vehicle body, sensor resonance, and contact-induced high-frequency transients can extend well into the kilohertz range. ISO 6487 and SAE J211 recommend a minimum sampling rate of ten times the channel class frequency (CFC), which is why 10 kHz to 20 kHz per channel is common for accelerometers, and 20 kHz to 50 kHz for load cells and high-bandwidth strain gauges.

Undersampling does not merely lose high-frequency detail; it folds that energy back into the low-frequency band as aliasing. A 3 kHz structural ring sampled at 5 kHz appears as a spurious 2 kHz oscillation that has no physical origin. Because aliasing is irreversible after digitization, the sampling rate must be chosen before the test, not corrected afterward.

Anti-Aliasing Filters

An anti-aliasing filter is a low-pass filter placed before the ADC to attenuate frequencies above the Nyquist limit. Its critical parameters are the cutoff frequency fc, the roll-off order, and the stopband attenuation at fs/2. A practical rule is to set fc at or below one-fifth of fs, so that the filter has at least one octave of transition band before Nyquist. For a 20 kHz sampling rate, a 4 kHz cutoff with a fourth-order Butterworth response provides roughly 24 dB/octave roll-off, giving about 48 dB attenuation at 10 kHz — usually sufficient for accelerometer channels.

Two filter types dominate crash instrumentation. Analog anti-aliasing filters in the data acquisition system (DAS) are mandatory and must be matched to the sampling rate. Digital filters applied after acquisition (for example, the CFC 60 or CFC 180 filters defined in SAE J211) are used for channel-class compliance and for comparison with simulation, but they cannot undo aliasing that already occurred. A common mistake is to rely on the post-processing CFC filter as if it were an anti-aliasing filter; it is not, because it operates on already-sampled data.

Dynamic Range and Quantization

Dynamic range is the ratio between the largest and smallest signals a channel can resolve. For a 16-bit ADC with a ±500 g full-scale range, the least significant bit (LSB) corresponds to approximately 1000 g / 65536 ≈ 0.015 g. This is the quantization floor. If the crash pulse of interest has a peak of 40 g, the effective resolution is about 0.04% of full scale — acceptable. But if the same channel is used to capture a 2 g low-level vibration, the signal occupies only about 130 LSBs, and quantization noise becomes significant.

The practical guidance is to set the full-scale range so that the expected peak occupies 50–80% of the ADC range. This preserves headroom for unexpected spikes while keeping quantization noise low. For accelerometers, a ±500 g range is typical for vehicle body channels, while ±2000 g may be needed for near-crush-zone or wheel-hub locations. Load cells often use ±50 kN or ±100 kN ranges. Always record the actual full-scale setting and the resulting LSB value in the test log, because these determine the noise floor of every subsequent correlation plot.

Synchronization Across Channels and Video

Crash tests combine many channel types — accelerometers, load cells, displacement sensors, seat belt force transducers, pressure sensors, and high-speed video. If channels are sampled by different DAS units or if video is recorded on a separate system, time alignment errors of even a few milliseconds can shift the apparent peak force or the airbag deployment timing, corrupting the CAE correlation.

Three synchronization mechanisms are used in practice. First, a common hardware clock or trigger bus distributes a single time base to all DAS units, ensuring sample-level alignment. Second, an event signal — typically a contact-closure or voltage step at the moment of impact — is recorded on every system, including the video system, so that all records share a common t = 0. Third, for video, a timecode or LED event marker visible in the frame provides a visual reference that can be aligned to the analog event signal during post-processing.

The verification step is straightforward: inject a known simultaneous pulse (for example, a square wave from a signal generator) into all channels and confirm that the recorded edges align within one sample period. For video, flash an LED at a known time and confirm that the frame containing the flash corresponds to the same timestamp as the analog event channel. A residual offset greater than 1 ms should be corrected by shifting the video or the affected channel before correlation.

Practical Configuration Example

Consider a frontal 56 km/h rigid-barrier test with 24 accelerometer channels, 8 load cells, and two high-speed cameras. A representative acquisition setup might be:

Channel group A (body accelerometers):
  fs = 20 kHz, analog AA filter fc = 4 kHz, 4th-order Butterworth
  ADC full scale = ±500 g, 16-bit → LSB ≈ 0.015 g
  CFC post-filter = 60 Hz (SAE J211)

Channel group B (load cells):
  fs = 50 kHz, analog AA filter fc = 10 kHz, 4th-order Butterworth
  ADC full scale = ±50 kN, 16-bit → LSB ≈ 1.5 N
  CFC post-filter = 600 Hz

Synchronization:
  Common clock bus, 1 MHz reference
  Event signal: 5 V step at t = 0, recorded on all DAS and video
  Video: 1000 fps, LED marker visible in frame 0

Verification:
  Inject 1 kHz square wave into one channel per group
  Confirm edge alignment within 1 sample (50 µs at 20 kHz)
  Confirm video LED frame aligns with analog event within 1 ms

After the test, the first check is not the correlation plot but the acquisition integrity: verify that no channel clipped (peak below 80% of full scale), that the anti-aliasing filter was active, and that the event signal appears at the same sample index across all records. Only then should the data be filtered to CFC 60 or CFC 180 and compared with the CAE model. If a channel shows a suspicious high-frequency oscillation that does not appear in the simulation, check the sampling rate and filter settings before adjusting the model — the discrepancy may be an acquisition artifact, not a modeling error.

Summary

Sampling rate, anti-aliasing filtering, dynamic range, and synchronization form a chain in which each link constrains the others. The sampling rate sets the Nyquist limit; the analog anti-aliasing filter must attenuate above that limit before digitization; the ADC full scale determines quantization noise and clipping risk; and the synchronization scheme ensures that all channels and video share a common time base. In crash test data acquisition for airbag CAE correlation, these parameters are not optional settings — they are the boundary conditions that determine whether the measured data can legitimately be used to validate a simulation model.