2. Crash test data landscape

Before any CAE model can be judged against a physical crash test, the analyst must understand exactly what the test produced. A full-scale frontal or side impact test is not a single measurement — it is a synchronized, multi-domain data acquisition event involving dozens to hundreds of channels, each with its own physical principle, mounting location, sampling rate, and calibration. The discrepancy analysis in later chapters depends entirely on how well this raw landscape is understood, because a large share of apparent "CAE error" actually originates from test-side instrumentation effects: filtering, mounting resonance, channel offsets, or mislabeled metadata.

Physical Test Setups and Their Instrumentation Zones

A typical frontal offset deformable barrier (ODB) test at 64 km/h places the vehicle on a fixed rigid barrier with a deformable honeycomb face. The vehicle is towed by a propulsion system, released at a defined velocity, and impacts the barrier. Instrumentation is distributed across several zones, each with distinct measurement objectives:

• Vehicle body / occupant cell: accelerometers on the B-pillar, rocker, floor tunnel, and rear seat cross-member to capture structural deceleration and intrusion.
• Restraint system: seat belt load cells, buckle load cells, retractor pretensioner signals, and airbag pressure or firing-time channels.
• Dummies (ATDs): head, chest, pelvis, and femur accelerometers; chest deflection potentiometers; neck load cells; and, in some setups, angular rate sensors.
• Barrier and environment: barrier face load cells, high-speed cameras, and sometimes a moving deformable barrier (MDB) for side impact.

Each zone has a defined coordinate convention. ISO 6487 and SAE J211 define the vehicle coordinate system: X forward, Y to the right, Z upward, with the origin typically at the vehicle centerline at ground level. Accelerometer polarity must be recorded explicitly — a positive X reading means forward acceleration of the sensor housing, which corresponds to a rearward inertial force on the occupant. Confusing this sign convention is one of the most common sources of sign-flipped comparisons between test and CAE.

Sensor Channels: Accelerometers, Load Cells, and Beyond

Accelerometers dominate crash test instrumentation. Piezoresistive and capacitive MEMS accelerometers are common, with full-scale ranges from ±50 g (body) to ±2000 g (head impact). Key channel attributes include:

Load cells measure force directly and are used for belt forces, buckle forces, barrier face loads, and seat anchorage. They are typically strain-gauge based with ranges from 5 kN to 100 kN. Unlike accelerometers, load cells are sensitive to off-axis loading and mounting preload, so their installation torque and orientation must be documented. A belt load cell reading 8 kN in the test but 11 kN in CAE often indicates a belt routing or slip difference rather than a material model error.

High-speed imaging provides the spatial context that point sensors cannot. Cameras at 1000–2000 fps capture dummy kinematics, airbag deployment, belt slip, and structural deformation. Camera calibration, frame timing, and the synchronization signal (typically a common trigger or time-zero pulse) are essential metadata. Without a shared time-zero, comparing a camera-derived displacement to a CAE node position is meaningless.

Data Acquisition Metadata and Channel Mapping

The metadata layer is what makes the raw signals interpretable. A well-structured test data package includes a channel list with sensor ID, location, axis, range, filter class, sample rate, and calibration factors. It also includes the test configuration: impact velocity, barrier type, vehicle mass, dummy type and position, seat position, and ambient conditions. In practice, this metadata is often stored in a TDMS, ISO 13499, or manufacturer-specific format, and the analyst must map each channel to the corresponding CAE output request.

Time-zero alignment is a recurring practical issue. The test time-zero is usually defined by the trigger signal at barrier contact or by a dedicated zero-time channel. CAE time-zero is the start of the explicit integration. If the test time-zero is set at first barrier contact but the CAE model includes a 5 ms pre-impact approach, every peak will appear shifted. The analyst must document and apply a consistent time offset before any correlation metric is computed.

Practical Implications for Discrepancy Analysis

When a CAE result diverges from a test signal, the first diagnostic step is not to adjust the material model. It is to verify that the comparison itself is valid. Three checks resolve a large fraction of apparent discrepancies:

1. Filter equivalence: Apply the same CFC filter to the CAE output as was applied to the test channel. A CFC 60 filter can reduce a 120 g peak to 95 g, which alone explains many "CAE over-predicts peak" observations.
2. Time-zero and sign alignment: Overlay the first 5 ms of both signals. If the test shows a sharp rise at t=0 while CAE rises at t=5 ms, the offset is a setup issue, not a physics issue.
3. Channel definition match: Confirm that the CAE output request corresponds to the same physical location, axis, and coordinate system as the sensor. A B-pillar sensor mounted on the outer panel and a CAE node on the inner panel can differ by 10–20 g during structural bending.

Understanding the test data landscape is therefore not a preliminary formality — it is the foundation of credible discrepancy analysis. The physical setup defines what was measured, the sensor layer defines how it was measured, the acquisition layer defines how it was conditioned, and the metadata layer defines how it should be interpreted. Only when all four layers are aligned can the analyst begin to ask whether the remaining differences are due to modeling assumptions, material parameters, or genuine physical effects not captured in the CAE model.