1. Occupant restraint systems and crash response fundamentals
Every restraint injury analysis begins with a deceptively simple question: what did the occupant actually experience during the crash pulse, and what did the restraint system do about it? Answering that question requires connecting three domains that are often taught separately — the mechanical behavior of restraint hardware, the biomechanics of injury, and the data pipeline that turns sensor signals into interpretable metrics. This chapter establishes the shared vocabulary and the analytical scope that the remaining chapters build on.
Restraint System Taxonomy
Restraint systems are conventionally divided into primary (energy-absorbing) and supplemental categories, but for analytics purposes a more useful split is by load path and deployment logic.
- Three-point seat belts — the baseline restraint. The lap portion loads the pelvis through the iliac crests; the shoulder portion loads the clavicle and sternum via the diagonal webbing. Load limiters (typically 3–6 kN force-limiting torsion bars or stitch patterns) and pretensioners (pyrotechnic or motorized, retracting 40–80 mm within ~10–20 ms of the crash pulse) define the belt's force–time signature.
- Front airbags — driver (steering-wheel-mounted) and passenger (instrument-panel-mounted) units. Deployment is triggered by a crash sensor algorithm evaluating longitudinal deceleration; typical threshold logic uses a velocity change (Δv) integral over a 20–40 ms window rather than raw acceleration alone, to reject rough-road and hammer-blow events.
- Side airbags and curtains — thorax/pelvis side bags and roof-rail curtains, triggered by lateral acceleration and pressure sensors in the door cavity. Curtains typically remain inflated for 3–5 s to provide rollover ejection mitigation.
- Knee bolsters, anti-submarining pans, and active seat-belt features — secondary load paths that shape occupant kinematics rather than absorb primary energy.
Each device contributes a distinct force–time curve to the occupant's loading history. The analytics problem is that these curves overlap in time and are only partially observable from vehicle-mounted sensors.
Injury Mechanisms and Their Measurement
Injury criteria exist because raw acceleration is not a good injury predictor on its own — duration and direction matter enormously. The most common criteria encountered in restraint analytics are:
| Criterion | Body region | Formulation (simplified) | Typical reference threshold |
|---|---|---|---|
| HIC15 / HIC36 | Head | max over window of [ (1/(t₂−t₁)) ∫ a dt ]^2.5 · (t₂−t₁) | HIC15 ≈ 700 |
| 3 ms clip | Chest / head | max acceleration sustained ≥ 3 ms | chest ≈ 60 g |
| Chest deflection | Thorax | peak sternal displacement | ≈ 50–75 mm (age-dependent) |
| Nij (neck injury) | Cervical spine | F_z/F_crit + M_y/M_crit | Nij ≤ 1.0 |
| Viscous criterion VC | Thorax | v(t) · C(t) (rate × compression) | ≈ 1.0 m/s |
These criteria are computed from occupant kinematics, which in a real vehicle are rarely measured directly. That gap is the central challenge of the analytics domain: we infer occupant loading from belt force sensors, seat-track load cells, accelerometers, and event data recorder (EDR) outputs, then map those inferences onto injury criteria.
From Crash Pulse to Dashboard: The Analytics Scope
A crash response dashboard is not a single chart — it is a layered analytical product. The figure below shows the signal chain from physical event to decision-support output, with the observable quantities at each stage.
The chain is deliberately drawn with a feedback path because the dashboard is not a terminal artifact. Field-return data — real crashes with known outcomes — is used to recalibrate sensor thresholds and to validate the injury-criterion mapping. A dashboard that never feeds back into the sensing or reconstruction layer is a reporting tool, not an analytics system.
What the Dashboard Must Answer
Scope discipline is essential. A crash response dashboard for restraint analytics should answer four bounded questions, and should be explicit about what it cannot answer:
- Did the restraint system perform as designed? Deployment timing versus the crash pulse, belt force within the load-limiter band, airbag inflation onset relative to occupant forward excursion.
- What loading did the occupant receive? Reconstructed belt and airbag forces, estimated chest deflection, head acceleration history.
- How close to injury thresholds was the occupant? Criterion values as a fraction of reference thresholds, with the time window at which the peak occurred.
- Which events are outliers? Statistical deviation from the fleet baseline, flagged for engineering review.
What it should not claim is a diagnosis. A dashboard can report that HIC15 reached 620 in a 56 km/h frontal barrier event with a 40 ms pulse; it cannot assert that a specific occupant sustained a concussion. That boundary between measurement and medical inference must be visible in the dashboard's own labeling.
A Worked Example: Reading a Single Event
Consider a frontal impact event with the following EDR-derived parameters (values are illustrative, constructed for teaching): Δv = 48 km/h, pulse duration 95 ms, driver belted, pretensioner fired at 12 ms, load limiter peak 4.2 kN, driver airbag deployment commanded at 18 ms. The reconstruction produces a head resultant acceleration peaking at 78 g with a 22 ms duration above 50 g.
Computing HIC15 over the worst 15 ms window: if the mean resultant acceleration over that window is approximately 62 g (≈ 608 m/s²), then HIC15 ≈ (608)^2.5 × 0.015 ≈ 700 × 0.015-adjusted value in the low 600s — comfortably below the 700 reference but not with large margin. The dashboard should present this as a ratio (≈ 0.88 of threshold) alongside the time of peak, because a value near the threshold at an unusual time (e.g., late in the pulse, after airbag bottom-out) tells a different engineering story than the same value at 30 ms.
Analytical Boundaries and Data Quality
Three limitations shape everything that follows. First, sampling: EDR data is often recorded at 100–500 Hz, which is adequate for pulse-level metrics but marginal for HIC, where the 15 ms window contains only a few samples at the low end. Second, observability: belt force sensors measure webbing tension, not occupant chest force; the mapping between them depends on geometry and clothing. Third, population variance: thresholds derived from mid-sized male surrogates do not transfer directly to small occupants or to out-of-position scenarios.
These are not reasons to avoid dashboards — they are the reasons dashboards must display uncertainty explicitly. A criterion value shown without its window, sample rate, and reconstruction assumptions is a number without a meaning. The chapters that follow build the reconstruction methods, the criterion computations, and the dashboard design patterns that keep those assumptions visible.