3. Time-zero alignment and event synchronization across accelerometers, load cells, and high-speed video.

In a crash test, every sensor and camera records on its own clock. The accelerometers on the vehicle body, the load cells in the barrier or sled, and the high-speed cameras all begin sampling at slightly different moments, and each has its own latency, trigger behavior, and timestamp convention. If these streams are not aligned to a common time zero, the resulting analysis is quietly wrong: peak acceleration appears to precede the barrier force that caused it, occupant kinematics in video no longer match the measured pulse, and injury criteria computed from misaligned channels drift by several percent. Time-zero alignment is therefore not a cosmetic preprocessing step — it is a prerequisite for any physically meaningful correlation between channels.

What "time zero" actually means

Time zero (t₀) is the instant the event begins, defined by a single reference. In most full-scale barrier tests, t₀ is the moment of first contact between the vehicle and the barrier; in sled tests, it is the start of the sled pulse. Every other stream is then expressed relative to that reference, so that a sample at t = +10 ms means "10 ms after first contact" regardless of which device recorded it.

Two conventions must be fixed before any alignment work begins, and kept consistent throughout the analysis:

• Sign convention for acceleration. Vehicle-body accelerometers are typically mounted so that the crash direction (vehicle forward, toward the barrier) reads positive, following the SAE J211 sign convention. Confirm this per channel rather than assuming it; a reversed channel will produce a mirrored pulse that looks plausible but integrates to the wrong velocity change.
• Timestamp basis. Decide whether each stream is stored with an absolute trigger timestamp, a sample index from the start of acquisition, or a device-internal counter. Mixing these bases is the most common source of a constant offset that survives every later correction.

Sources of misalignment

Misalignment between streams comes from several distinct mechanisms, and each needs a different remedy:

• Trigger latency. A data acquisition system may begin recording some milliseconds after the hardware trigger edge, and different subsystems (accelerometer DAQ, load-cell amplifier, camera) can have different latencies. This produces a constant offset per stream.
• Camera frame timing. High-speed cameras expose frames at fixed intervals, but the first frame after trigger is rarely coincident with t₀. The camera's own trigger delay and exposure timing shift the video timeline relative to the sensor timeline.
• Clock drift. Over a test of 100–200 ms, drift is usually negligible, but across long pre-trigger recordings or when devices free-run on independent oscillators, drift accumulates and cannot be corrected by a single shift.
• Filtering and group delay. Anti-alias and channel filters introduce phase delay. If one channel is filtered at a different corner frequency than another, their apparent timing differs even when their raw timestamps agree.

Establishing t₀ from physical evidence

The most reliable alignment uses a physical event visible in more than one stream. Common choices:

• Contact switch or tape switch on the barrier face, which produces a clean electrical edge recorded by the DAQ and simultaneously visible as the instant the vehicle bumper meets the barrier in video.
• First non-zero load-cell response at the barrier, which marks the onset of contact force.
• Accelerometer onset on the vehicle structure closest to the impact point, which rises sharply at contact.

In practice, the tape-switch edge is the cleanest electrical marker, and the video frame in which the bumper first touches the barrier is the cleanest optical marker. Aligning these two gives the sensor-to-video offset directly. Where no switch is available, the load-cell onset and the nearest accelerometer onset can be cross-correlated to estimate the offset, but this is less precise because both signals have finite rise times.

Worked alignment procedure

The following example uses representative values to show the mechanics; the numbers are illustrative, not measured results.

1. Extract raw timestamps. Suppose the accelerometer DAQ reports its first sample at trigger + 0.8 ms, the load-cell amplifier at trigger + 1.5 ms, and the camera's first frame at trigger + 2.3 ms. These are the raw offsets to be corrected.
2. Locate the tape-switch edge in the DAQ record. Say it appears at sample index corresponding to 3.2 ms after DAQ start. That sample is the electrical t₀ candidate.
3. Locate first contact in video. Say the bumper first touches the barrier in frame 14, and the camera runs at 10,000 fps (0.1 ms per frame). Frame 14 occurs at 14 × 0.1 = 1.4 ms after the camera's first frame, i.e., at trigger + 2.3 + 1.4 = 3.7 ms.
4. Compute the sensor-to-video offset. The electrical t₀ is at 3.2 ms, the optical t₀ at 3.7 ms, so the video timeline leads the sensor timeline by 0.5 ms. Shift the video timestamps by −0.5 ms (or equivalently shift sensor timestamps by +0.5 ms) so both agree.
5. Verify with a second event. Check a later feature visible in both streams — for example, the moment the vehicle's rear structure begins to move, or a marker on the hood passing a fixed reference. If the residual offset is within one video frame (0.1 ms here), the alignment is acceptable; if it grows with time, suspect clock drift rather than a constant offset.

# Illustrative alignment check (values are example inputs, not measured data)
daq_t0_ms      = 3.2      # tape-switch edge in DAQ record
cam_first_ms   = 2.3      # camera first frame after trigger
cam_fps        = 10000    # 0.1 ms per frame
contact_frame  = 14
cam_t0_ms      = cam_first_ms + contact_frame * (1000.0 / cam_fps)
offset_ms      = cam_t0_ms - daq_t0_ms   # 0.5 ms: video leads sensors
print(f"video-sensor offset = {offset_ms:.2f} ms")

Verifying synchronization across all channels

After applying the offset, confirm that every stream is consistent. A useful check is to compare the integrated velocity change from the vehicle accelerometers against the momentum change implied by the barrier load cells and the known masses. If the two agree within a few percent, the channels are aligned and the sign conventions are correct. A persistent discrepancy that grows with time points to drift; a constant discrepancy points to a residual offset or a scale/sign error.

For video, overlay a synchronized time readout on each frame and step through the impact sequence frame by frame alongside the acceleration trace. The frame at which the acceleration first rises should coincide with the frame of first contact. If the acceleration rise appears one or two frames early, the video is lagging and needs a small additional shift.

Documenting the alignment

Record the final offsets, the reference event used, the verification result, and any residual uncertainty in the test report. A short alignment table makes the analysis reproducible and lets a reviewer confirm that the same t₀ was used for every channel.

The table above uses illustrative values. In a real test, populate it from the actual trigger logs and frame counts, and state the camera frame rate and the frame index of first contact explicitly so the offset can be recomputed independently.

The diagram summarizes the alignment logic: each stream has its own raw offset, a single physical reference defines t₀, and each stream is shifted by a measured correction. The verification step at the bottom is what separates a defensible alignment from a cosmetic one — it confirms that the corrected streams are not only visually consistent but physically consistent, so that acceleration, force, and video can be analyzed together with confidence.