2. Crash test measurement systems

Correlation between a physical crash test and its CAE counterpart is only as trustworthy as the measurement chain that produced the test data. A CAE model can be refined indefinitely, but if the accelerometer channel was over-ranged, the load cell was mounted with a non-representative preload, or the acquisition filter was set differently from the CAE output filter, the resulting CORA score will describe the instrumentation rather than the structure. This chapter therefore treats the measurement system as a signal chain with defined transfer characteristics, and examines each element — accelerometers, load cells, high-speed cameras, and the data acquisition (DAQ) system — from the perspective of what a correlation engineer needs to know before comparing curves.

Accelerometers: Range, Mounting, and Frequency Content

Piezoresistive and piezoelectric accelerometers dominate vehicle crash instrumentation. Piezoresistive (MEMS or strain-gauge) sensors are DC-capable, which matters for occupant and restraint measurements where the signal must be integrated to velocity and displacement; piezoelectric sensors offer wider bandwidth and higher range but cannot hold a static offset. For correlation work, the decisive parameters are the measurement range, the resonant frequency of the mounted sensor, and the mounting method.

A practical rule is to select a range at least 1.5× the expected peak, because clipped channels destroy the peak and the subsequent integration. Consider a B-pillar sill accelerometer in a 64 km/h frontal ODB test: expected peak around 400–600 g, so a 1000 g sensor is a reasonable choice, while a 500 g sensor risks clipping. The mounted resonance should sit well above the highest frequency of interest — typically the channel is filtered to CFC 60 (cut-off ≈ 100 Hz) or CFC 180 for structural work, and a mounted resonance above 20 kHz keeps the sensor safely out of band.

Mounting is where correlation is most often lost. A stud-mounted sensor on a flat, machined pad behaves very differently from a sensor glued to a curved, painted surface or fixed with a plastic bracket. The bracket introduces its own resonance and can rattle, producing high-frequency content that the CAE model does not contain. The correlation engineer should record, for every channel, the sensor type, serial number, range, mounting method, and location coordinates in the vehicle coordinate system (X forward, Y left, Z up, origin at the vehicle datum), because the CAE output must be extracted at the same point and in the same axis sense.

Load Cells and Force Measurement

Load cells appear in crash testing in several distinct roles: barrier load cells (for example, the 36-element load cell wall used in frontal impact), seat-belt force transducers, steering-column force measurement, and pedal force sensors. Each has a different correlation implication. Barrier load cells measure the spatial distribution of impact force and are compared against the sum of contact forces in the CAE model; belt load cells measure the restraint force path and are compared against belt element forces.

Two practical issues dominate. First, preload and zeroing: a belt load cell must be zeroed after installation with the belt routed and tensioned as in the test configuration, otherwise the recorded force carries a constant offset that shifts the entire curve. Second, cross-talk and off-axis sensitivity: a load cell loaded predominantly in one direction still responds to moments and lateral forces, and the CAE model must extract the force in the same direction and at the same physical location. For the load cell wall, the correlation is usually performed on the total force–time curve and on individual cell curves, with the total curve being far more robust to small positioning differences.

High-Speed Imaging

High-speed cameras provide the deformation and kinematics that no accelerometer can: intrusion profiles, dummy kinematics, airbag deployment, and belt routing. For correlation, the camera system contributes in two ways. First, it supplies qualitative and semi-quantitative validation of the CAE animation — the timing of airbag contact, the direction of occupant excursion, the sequence of structural folding. Second, when the cameras are calibrated (target-based photogrammetry), it supplies displacement and velocity traces at tracked points that can be compared directly with CAE nodal output.

Typical settings for a frontal test are 1000 fps at 1280×800 for the vehicle exterior and 1000–2000 fps for the occupant view, with exposure times short enough to freeze motion (often 1/2000 s or shorter) and synchronized lighting. The correlation engineer should note the frame rate and the synchronization offset between camera trigger and the DAQ time zero, because a 1 ms offset at 1000 fps is a full frame and can shift the apparent contact timing by an amount comparable to the CORA time-shift tolerance.

Data Acquisition and the Shared Signal Path

The DAQ system defines the sampling rate, anti-alias filtering, and time base for all channels. Crash DAQ systems typically sample at 10–20 kHz per channel, with hardware anti-alias filters and a common time zero derived from the trigger. The critical correlation requirement is that the test data and the CAE output pass through the same post-processing: the same CFC filter class, the same resampling, and the same time window. If the test channel is filtered at CFC 60 and the CAE output is unfiltered, the comparison will show differences that are purely filter artefacts.

The following signal-chain diagram summarizes how the four measurement subsystems feed a common, synchronized data set that is then conditioned identically for test and simulation before CORA evaluation.

Building a Channel Map That Supports Correlation

The single most useful deliverable from the measurement side is a channel map that the CAE engineer can use directly. For each channel it should carry: channel name, physical quantity and unit, sensor type and serial number, measurement range, mounting method, location in vehicle coordinates, axis direction and sign convention, CFC filter class, and the corresponding CAE node or element set. A compact example for three channels is shown below; the values are illustrative and should be replaced with the actual test documentation.

With this map in hand, the correlation workflow becomes reproducible: extract the CAE signal at the listed node and axis, apply the listed CFC filter to both signals, align time zero, and only then compute the CORA metrics. When a discrepancy appears, the channel map lets the engineer distinguish a genuine model error from a measurement artefact — for example, a high-frequency spike present only in the test channel usually points to bracket resonance or a loose connector, not to a modelling deficiency.

In practice, the measurement system is not a passive recorder but an active participant in the correlation result. Treating sensor selection, mounting, zeroing, filtering, and time-base alignment as explicit, documented decisions — rather than defaults inherited from the test house — is what allows a CORA score to be interpreted as a statement about the CAE model rather than about the instrumentation.