2. Data acquisition in physical crash tests

Physical crash testing is the ground truth against which every simulation model is eventually judged. Before any correlation study can begin, the test itself must produce trustworthy measurements. That means the data acquisition chain — sensors, signal conditioning, cabling, digitizers, and logging software — has to be designed as carefully as the vehicle structure being tested. A sled test or full-scale barrier impact lasts only 80–150 ms, yet during that window dozens to hundreds of channels must capture peak accelerations exceeding 500 g, structural deformations of several hundred millimetres, and occupant kinematics at millisecond resolution. If any link in the chain saturates, aliases, or drops samples, the resulting dataset is not merely noisy; it is misleading, and any correlation conclusion drawn from it is invalid.

Accelerometers: The Primary Crash Signal

Piezoresistive accelerometers dominate crash testing because they respond to DC (steady-state) acceleration, which is essential for integrating velocity and displacement. A typical uniaxial accelerometer used on a vehicle body has a full-scale range of ±500 g to ±2000 g, a resonant frequency above 50 kHz, and a sensitivity around 0.1–0.5 mV/g when excited by a 5–10 VDC bridge supply. Piezoelectric types offer higher sensitivity and wider bandwidth but cannot measure DC, so they are generally reserved for high-frequency events such as component-level shock rather than vehicle pulse integration.

Mounting is where most accelerometer errors originate. A stud-mounted sensor on a machined flat surface achieves the intended frequency response; a sensor attached with wax, tape, or a magnetic base introduces a mechanical low-pass filter that can attenuate peaks by 10–30% and shift phase. For a B-pillar or tunnel-mounted sensor, the mounting surface must be rigid enough that its own resonance stays well above the measurement band. The rule of thumb is that the mounted resonance should be at least five times the highest frequency of interest — for a 1000 Hz channel, that means 5 kHz minimum.

Channel naming and polarity conventions must be fixed before the test. A common convention defines the vehicle X-axis as forward, Y as lateral (right positive), and Z as vertical (up positive). Every accelerometer record should carry its axis, location, and sign convention in the channel metadata, because a polarity error discovered after the test cannot be corrected by post-processing — it will flip the sign of integrated velocity and corrupt every downstream comparison.

Load Cells and Force Measurement

Load cells quantify the forces transmitted through structural interfaces: barrier face loads, seat belt anchor forces, steering column axial force, and pedal forces. Strain-gauge load cells with a full Wheatstone bridge output 1–3 mV/V at rated capacity. A barrier load cell wall, for example, may contain 30–60 individual load cells, each rated 50–200 kN, sampled simultaneously so that the total barrier force can be reconstructed by summation.

Two practical issues dominate load cell data quality. First, the bridge excitation must be stable; a 1% drift in excitation voltage produces a 1% error in force that is indistinguishable from a real structural effect. Second, cable routing matters — long runs near high-current ignition or pyrotechnic firing lines can couple noise into the millivolt-level signal. Twisted-pair shielded cable with the shield grounded at the amplifier end is standard practice. Where a load cell cannot be placed directly, a strain gauge rosette on a structural member can be calibrated to force using a known static load before the test.

High-Speed Cameras and Optical Measurement

High-speed cameras provide the spatial context that point sensors cannot. A typical setup uses 1000–2000 frames per second at 1280×720 or higher for full-vehicle views, and up to 10,000 fps for localized events such as airbag deployment or dummy head contact. At 1000 fps, the inter-frame interval is 1 ms, which is adequate for tracking gross occupant motion but marginal for capturing the sharp acceleration onset in the first 10–20 ms of impact.

Camera synchronization is critical. All cameras and the data logger must share a common trigger — usually a contact closure from the barrier or a pre-trigger signal from the test controller. Without a common time base, correlating a camera-derived displacement with an accelerometer-derived displacement introduces an unknown time offset that can easily exceed 5 ms, which at 50 g is a velocity error of 2.5 m/s. Modern systems use IRIG-B time code or a dedicated sync pulse distributed to every device.

For quantitative work, cameras are calibrated using a target grid or known reference markers in the field of view. Photogrammetry software then converts pixel coordinates to millimetres. Tracking a target on the dummy chest at 1000 fps typically yields displacement accuracy of ±2–5 mm, which is sufficient for occupant kinematics but not for structural strain measurement.

Data Logging Systems and Signal Conditioning

The data logger is the central acquisition hub. In a full-scale crash test, a typical configuration might include 64–128 analog channels, 16–32 digital channels, and 4–8 camera sync outputs, all sampled at 10–20 kHz per channel with 16-bit resolution. Anti-alias filters are mandatory: a low-pass filter set at 1/2.5 of the sample rate (e.g., 4 kHz for a 10 kHz sample rate) prevents high-frequency content from folding back into the measurement band. Without it, a 15 kHz structural ringing mode can appear as a false 5 kHz signal.

Signal conditioning includes bridge completion, excitation, amplification, and filtering. For piezoresistive accelerometers, a completion module supplies the bridge and a differential amplifier boosts the millivolt signal to the ±5 V or ±10 V input range of the digitizer. Gain should be set so that the expected peak occupies 70–90% of the full-scale range — high enough for resolution, low enough to avoid clipping. A channel that clips at 500 g when the true peak is 520 g will report a flat-topped pulse, and the integrated velocity will be underestimated by the area lost above the clip line.

Pre-trigger buffering is standard. The logger continuously samples into a circular buffer and, on trigger, retains a configurable pre-trigger window — typically 50–100 ms — so that the zero-acceleration baseline and any pre-impact motion are captured. This baseline is essential for offset correction: accelerometer zero offsets of even 0.5 g, integrated over 100 ms, produce a 0.5 m/s velocity error.

From Raw Signals to Usable Data

After the test, raw voltages are converted to engineering units using each channel's calibration factor. Accelerometer data is typically filtered with a CFC 60 or CFC 180 filter (SAE J211) depending on the application — CFC 60 for occupant response, CFC 180 for structural response. The filtered acceleration is then integrated to velocity and double-integrated to displacement, with offset correction applied to the pre-trigger baseline. Load cell data is summed across the barrier wall to produce total force. Camera data is digitized and synchronized to the common time base.

The output is a time-aligned dataset where every channel shares a common trigger and a known time zero. This dataset becomes the reference for simulation correlation. If the acquisition chain introduced errors — clipped channels, aliased frequencies, unsynchronized cameras, or uncorrected offsets — those errors propagate directly into the correlation metrics and can lead to incorrect conclusions about model fidelity. The discipline of crash test data acquisition is therefore not a preliminary formality; it is the foundation on which the entire correlation exercise rests.