2. Crash test data sources and instrumentation

Restraint performance analysis lives or dies on the quality of the measurement chain. A frontal sled test or full-vehicle crash produces a few hundred milliseconds of violent motion, and every conclusion you draw about belt loading, airbag deployment timing, or occupant kinematics depends on sensors that were selected, mounted, and sampled correctly before the event. This section walks through the four primary data sources used in restraint work — accelerometers, load cells, high-speed video, and the data acquisition (DAQ) system that ties them together — and shows how to configure them so the resulting channels are actually usable for injury analytics.

Accelerometers: capturing the crash pulse and occupant response

Accelerometers fall into two functional groups in a restraint test. Vehicle-side accelerometers record the crash pulse that the restraint system must respond to: a tunnel-mounted or sill-mounted sensor measures longitudinal deceleration, and this channel typically drives the airbag ECU firing decision in a full-vehicle test. Occupant-side accelerometers — head, chest, and pelvis — record what the occupant actually experienced, and these are the channels that feed injury criteria such as HIC and chest deflection-derived metrics.

Selection matters more than most people expect. A 2000 g full-scale sensor with a ±5% linearity band is appropriate for a barrier pulse; a 50 g sensor is appropriate for a chest. Mounting a low-range sensor on the tunnel will clip the signal, and mounting a high-range sensor on the chest wastes resolution on a signal that never exceeds 60 g. For occupant instrumentation, use piezoresistive or capacitive accelerometers with a range of roughly 2–3× the expected peak, and mount them rigidly to the underlying structure — a chest accelerometer on a padded surface will show a filtered, phase-shifted version of the true signal.

Orientation conventions must be fixed before the test and carried through every channel name and plot. A common convention is SAE J211: positive X forward, positive Y to the occupant's left, positive Z up, with rotational signs following the right-hand rule. If your head accelerometer is mounted with its sensitive axis reversed, the resulting HIC will be computed from a sign-flipped pulse and the injury number will be wrong without any obvious symptom. Label channels with the axis and sign explicitly, for example HEAD_AX_POS_FWD.

Load cells: measuring what the restraint actually does

Accelerometers tell you how the occupant moved; load cells tell you what the belt and airbag did to make that happen. The workhorses are belt load cells (shoulder and lap), buckle load cells, and seat pan / cushion load cells. A shoulder belt load cell is typically a strain-gauge link inserted in the belt path with a range of 10–20 kN and a sampling rate matched to the DAQ. Its output is the tension history, and the peak and the time-to-peak are the numbers you compare against pretensioner and load-limiter design targets.

Two practical points dominate load cell quality. First, calibration and zeroing: zero the channel with the belt routed and at its installed pre-tension, not with the cell lying on a bench. A 50 N offset at zero becomes a systematic bias in every peak reading. Second, series stiffness: a load cell inserted into the belt path adds compliance, which slightly changes the restraint response. Use a cell whose stiffness is high relative to the belt and document the insertion point so the data can be compared across builds.

For airbag work, pressure transducers inside the inflator or module and force sensors on the module cover are common additions. These are not load cells in the classical sense but belong to the same measurement family and are sampled on the same DAQ.

High-speed video: the ground truth for kinematics

Video does not produce a number directly, but it is the only source that lets you verify that the sensors measured what you think they measured. A typical restraint test uses 1000 fps at the occupant compartment, with a resolution sufficient to resolve belt geometry and head position — 1280×1024 is a common working point. Higher frame rates (2000–4000 fps) are used for airbag deployment and pretensioner actuation, where the event of interest lasts only tens of milliseconds.

Lighting is the usual failure mode. At 1000 fps the exposure per frame is short, and under-lit footage produces motion blur that makes digitization unreliable. Use synchronized high-intensity LED arrays, and place a synchronization flash or LED pulse in frame at t=0 so video and DAQ share a common time origin. Without that marker, aligning a belt load peak to a video frame is guesswork. For quantitative kinematics, add targets to the head, chest, and pelvis — retro-reflective markers tracked by photogrammetry software — so head excursion and belt-to-body relative motion can be extracted as time histories rather than eyeballed.

The DAQ chain: sampling, filtering, and channel integrity

The DAQ system is where all of this becomes a dataset. Three settings determine whether the data survives analysis. Sample rate: SAE J211 recommends a minimum of 10× the channel's frequency content; in practice, 10 kHz for accelerometers and load cells is standard, with 20 kHz for airbag pressure. Anti-alias filtering: apply a hardware low-pass at or below half the sample rate before digitization, then apply the J211 digital filter (CFC 60 for occupant accelerometers, CFC 180 or 600 for belt loads depending on the analysis) in post-processing. Skipping the hardware anti-alias filter lets high-frequency content fold back into the passband and corrupt peaks. Channel integrity: verify every channel with a shunt calibration or known input before the test, and record a pre-trigger window (typically 100–500 ms) so the zero level and any drift are visible.

The figure below shows how the four sources connect in a typical frontal sled test and where the synchronization and filtering decisions sit in the chain.

Putting the chain together: a reproducible pre-test checklist

The following checklist is the minimum set of decisions to lock down before a restraint test. It is written as a working document, not a summary — each line corresponds to a setting or verification you can actually perform.

When a test produces an unexpected result — a belt load peak far below the design target, or a head acceleration that looks filtered — the first diagnostic step is to check the chain in reverse: confirm the channel was not clipped, confirm the filter setting matches the analysis convention, and confirm the video time origin aligns with the DAQ trigger. In practice, most "anomalous" restraint data traces back to one of these three instrumentation decisions rather than to the restraint hardware itself. Getting the measurement chain right is therefore not a preliminary formality; it is the first act of restraint performance analysis.