1. Crash Pulse Data Acquisition and Channel Verification

Why Channel Verification Comes Before Any Kinematic Diagnosis

A crash pulse is only useful for occupant kinematics diagnosis if every number in it can be defended. Before onset detection, baseline correction, or double integration, the acquisition chain must be proven: the accelerometer is mounted with a known axis orientation, the channel is wired with a known polarity, the scaling converts raw counts to physical units correctly, and the sample rate is high enough and stable enough that the pulse's frequency content is preserved. A single inverted channel or a mis-scaled gain will produce a plausible-looking pulse that integrates into a completely wrong velocity and displacement — and every downstream conclusion about occupant motion will inherit that error.

This case establishes a trustworthy pulse dataset from one accelerometer channel. The deliverable is not a corrected pulse (that is the next case) but a verified one: a raw record with documented polarity, scale factor, sample rate, and a signal-chain record that another engineer can audit.

The Signal Chain and What Each Stage Can Corrupt

A typical crash-pulse acquisition chain runs: accelerometer → charge amplifier or IEPE conditioning → anti-alias filter → ADC → digital record. Each stage has a characteristic failure mode that shows up in the data:

• Accelerometer orientation — a reversed mounting flips the sign of the entire pulse. The crash deceleration appears as acceleration.
• Polarity convention — SAE and many OEM conventions define positive as the direction of vehicle travel (forward), so a frontal crash produces a negative longitudinal pulse. If the channel is wired or configured with the opposite convention, the sign is wrong even with correct mounting.
• Scaling — sensitivity in mV/g (or pC/g) times amplifier gain times ADC volts-per-count gives the physical scale. A factor-of-10 error is common when g vs. m/s² is confused.
• Sample rate — crash pulses carry energy up to roughly 300–1000 Hz depending on structure; 10 kHz is a common minimum, and the anti-alias filter must sit below Nyquist.
• Baseline/offset — a DC offset before trigger corrupts integration; it is corrected in the next case but must be recorded here.

Acquiring and Verifying the Channel

The verification is done with two records: a static calibration record and the crash record. The static record uses a known 1 g reference (accelerometer face-up under gravity, or a calibrated shaker) to confirm scale and polarity. The crash record is then checked against the same scale factor and against the pre-trigger baseline.

Assume the channel is configured as: sensitivity 10 mV/g, amplifier gain 100, ADC 16-bit bipolar ±5 V, sample rate 20 kHz. The nominal scale factor is derived as follows:

ADC LSB size      = (2 × 5 V) / 2^16 = 152.59 µV/count
Signal at ADC     = 10 mV/g × 100 = 1.0 V/g
Scale factor      = 1.0 V/g ÷ 152.59 µV/count = 6554 counts/g
Physical scale    = 1 g / 6554 counts = 152.6 µg/count

With a 1 g static reference the channel should read approximately 6554 counts (plus any DC offset). If it reads 655 counts, the gain is off by 10×; if it reads −6554, the polarity is inverted. Both are caught here, before the pulse is ever integrated.

The crash record is then checked for three things: (1) the pre-trigger window (100 ms before trigger) has a mean near zero and low noise — this is the baseline that will be removed in the next case; (2) the pulse peak, converted with the verified scale, falls within the accelerometer's ±500 g range without clipping; (3) the pulse's dominant frequency content sits well below the 4 kHz anti-alias corner, confirming the 20 kHz sample rate is adequate.

Signal-Chain Record (Deliverable)

The following record should accompany the raw pulse file. It is the audit trail that makes the dataset trustworthy for downstream diagnosis.

Common Failure Symptoms and What They Point To

• Pulse looks positive in a frontal crash — polarity inverted at the channel or the axis is mounted backward. Re-run the tap test; do not "fix" it by flipping the sign in post-processing without documenting it.
• Peak reads ~10× too small or too large — gain or g-vs-m/s² confusion. Re-derive from the 1 g static record.
• Pulse looks smooth and rounded, missing sharp structure — anti-alias filter set too low, or sample rate too low. Check the FFT of a known impulse.
• Pre-trigger mean far from zero — DC offset or sensor bias. Record it; it is corrected in the next case, not here.
• Clipped flat-top at peak — sensor range exceeded. The pulse is unusable for peak-sensitive diagnosis; flag it.

Once the signal-chain record is complete and every pass criterion is met, the raw pulse is a verified dataset. Only then does it make sense to move on to baseline correction and onset alignment — the steps that turn a verified pulse into a kinematic reconstruction.