3. Crash sensing and event detection

An airbag ECU must answer one question within a few milliseconds: is this a crash that requires deployment? The answer depends on three tightly coupled elements — where the accelerometers are mounted, what the crash pulse actually looks like at those locations, and how the deployment algorithm converts that pulse into a fire/no-fire decision. This section walks through each element and shows how they interact in a real calibration workflow.

3.1 Accelerometer Placement and Sensing Architecture

Modern restraint systems use a distributed sensing architecture rather than a single central sensor. The central ECU is typically mounted on the vehicle tunnel or floor pan near the center of gravity, where it measures longitudinal and lateral acceleration of the vehicle body. Satellite sensors (front-end sensors, pressure or acceleration based) are mounted on the radiator crossmember, crash box, or B-pillar to detect deformation earlier than the central sensor can.

Placement is driven by three constraints:

• Early signal onset. A front satellite on the crash box sees deceleration 5–15 ms before the tunnel sensor, because the crumple zone deforms before the body decelerates. This time advantage is what allows a 20–30 ms deployment decision.
• Discrimination capability. Sensors must distinguish a 50 km/h full frontal barrier impact from a 15 km/h parking bump, a curb strike, or a hammer blow during repair. Mounting position determines which frequency content and amplitude range reaches the sensor.
• Survivability and diagnostics. The sensor must survive the crash it is measuring long enough to report data, and must support plausibility checking against other sensors.

A typical frontal configuration uses two front satellites plus the central ECU, with the algorithm requiring agreement between at least one satellite and the central sensor before firing. This redundancy suppresses single-sensor false positives from rough roads or door slams.

3.2 Reading the Crash Pulse

The raw accelerometer output is a time-domain signal, but calibration engineers work with the crash pulse — the deceleration history expressed in g, with a defined sign convention. In this course we adopt the SAE convention: positive acceleration points forward (vehicle direction of travel), so a frontal impact produces a negative longitudinal pulse. All thresholds and figures below follow this convention.

Three quantities characterize a pulse for deployment purposes:

The integral relationship is the physical foundation of most algorithms: Δv(t) = ∫ a(t) dt. A 50 km/h (13.9 m/s) barrier impact must produce a total Δv of roughly 13.9 m/s, but the rate at which that Δv accumulates separates a real crash from a misuse event. A pothole can generate 20 g for 2 ms — large amplitude, negligible Δv. A barrier impact generates 40 g sustained for 60 ms — large Δv. This is why amplitude-only thresholds are insufficient and why energy- or velocity-based criteria dominate.

3.3 From Pulse to Decision: Algorithm Structure

A production deployment algorithm is not a single threshold. It is a layered structure:

1. Signal conditioning. Anti-alias filtering, offset compensation, and a low-pass filter (commonly 300–1000 Hz) remove sensor noise and structural ringing that carries no crash information.
2. Feature extraction. The filtered signal feeds parallel computations: short-window velocity integration, energy accumulation, and frequency-domain measures.
3. Arming and discrimination. An arming condition (e.g. |a| > 3 g for > 1 ms) wakes the algorithm; a discrimination function then compares the extracted features against calibrated thresholds.
4. Fire decision and timing. If the discrimination criterion is satisfied before a time limit (typically 20–40 ms), the deployment command is issued. A separate safing sensor — an electromechanical or solid-state switch — must close in series, providing an independent physical confirmation.

The most common discrimination form is a velocity threshold with a time window: fire if Δv(t) exceeds V_th within T_window. A representative calibration for a frontal airbag might use V_th = 4 m/s evaluated at t = 15 ms, with a second, higher threshold V_th2 = 8 m/s at t = 30 ms for late-but-severe events. The two-threshold structure prevents both late firing on moderate crashes and missed firing on soft, long-duration impacts.

3.4 A Worked Calibration Example

Suppose a front satellite records the following simplified pulse during a 50 km/h barrier test (values are illustrative, sampled every 5 ms):

t (ms) :   0    5   10   15   20   25   30   35   40
a (g)  :   0  -12  -38  -52  -48  -40  -30  -18   -6

Integrating with the trapezoidal rule and converting g to m/s² (1 g = 9.81 m/s²):

Δv(10 ms) ≈ 0.5*(0+12)*5e-3*9.81 + 0.5*(12+38)*5e-3*9.81
          ≈ 0.294 + 1.226 ≈ 1.52 m/s
Δv(15 ms) ≈ 1.52 + 0.5*(38+52)*5e-3*9.81 ≈ 1.52 + 2.21 ≈ 3.73 m/s
Δv(20 ms) ≈ 3.73 + 0.5*(52+48)*5e-3*9.81 ≈ 3.73 + 2.45 ≈ 6.18 m/s

With V_th = 4 m/s at t = 15 ms, this event fires at approximately 16 ms — well inside the target window. Now compare a curb strike: 25 g for 4 ms yields Δv ≈ 0.5 × 25 × 4e-3 × 9.81 ≈ 0.49 m/s, far below threshold, so no deployment. The same amplitude (25 g) appears in both events, but the Δv criterion separates them cleanly.

3.5 Verification and Failure Symptoms

Calibration is validated against a test matrix that includes barrier impacts at multiple speeds and offsets, pole impacts, underride, and the misuse cases above. The pass criteria are: fire for all required deployment events within the timing window, and no-fire for all misuse events. A common failure mode is late firing on offset deformable barrier tests, where the pulse rises slowly; the fix is usually a second, lower threshold active in a later time window rather than a global reduction of V_th. Another is false fire on rough roads, which typically points to insufficient low-pass filtering or a satellite mounted at a structural resonance node.

The figure summarizes the full chain: two sensor locations feed conditioned signals into parallel feature extractors, whose outputs are compared against a velocity threshold inside a bounded time window, gated by an independent safing sensor. The lower plot shows why the Δv criterion — not raw amplitude — is the discriminating feature: the barrier pulse crosses V_th at roughly 16 ms, while the curb strike never approaches it despite comparable peak g.

3.6 Practical Calibration Checklist

The essential engineering insight is that accelerometer placement determines what the algorithm can see, the crash pulse determines when information arrives, and the threshold logic determines whether that information is sufficient to fire. All three must be calibrated together; optimizing any one in isolation produces a system that either misses real crashes or deploys on potholes.