Building Per-Vehicle Thermal Baselines

A per-vehicle thermal baseline is a rolling estimate of what "normal" looks like for one vehicle, so the system can trip on a departure from that vehicle's own history rather than a deck-wide constant. This is the reference layer beneath the trip logic — separate from the magnitude/dwell thresholds that act on it and from the spatial coherence check that suppresses false positives. Get the baseline wrong and every threshold above it is measuring from the wrong zero.

Static thresholds are simple and almost useless on cargo decks. Engine bays cool at different rates, hot-ambient sailings push background temperatures up, and a fixed 70 °C trip line either fires constantly or never fires until it is too late. The baseline exists to remove that ambient floor so the residual signal is the vehicle, not the voyage.

How the baseline is built

Each sensor cell maintains an exponentially weighted moving average (EWMA) of the surface temperature for the vehicle in its field of view. The EWMA half-life sets the core trade-off: too short and the baseline chases transients into itself, masking a slow real rise; too long and it lags a genuine post-loading cooldown and flags cooling as anomaly. The half-life is a per-deck parameter, swept against replayed calibration voyages rather than shipped as a universal constant.

Why per-vehicle, not per-deck

A deck-wide baseline averages thousands of vehicles with different thermal mass, fuel type, and recent running state, so the early single-cell rise that matters is buried in the mean. A per-vehicle reference keeps each car's zero local, which is what lets a sustained delta of roughly 6 °C stand out instead of being lost against deck-scale variation. The magnitude and dwell that act on that delta are covered separately in our thresholds write-up; here the point is only that the delta is meaningful because the reference is local.

Holding the baseline through loading and voyage

The hardest period is the first hours after loading, when engine-bay heat, solar gain, and ventilation changes all move the floor at once. A vehicle needs a warm-up window after parking before its delta is trusted, so residual drivetrain heat is absorbed as baseline rather than alarmed. Re-baselining is then paused while a candidate delta is open, so a slowly developing fault cannot quietly redefine "normal" underneath itself. Major structural or ventilation changes to a deck warrant re-tuning the warm-up and half-life and re-running the calibration replay.

False-positive control

The dominant nuisance source is solar heating through deck openings on coastal sailings. A cross-cell coherence check suppresses it: if the delta is shared by neighbouring cells, it is environmental, not vehicular. The coherence-window mechanics are detailed in the companion article; from the baseline's perspective, coherence is what stops a shared ambient swing from being read as many simultaneous anomalies.

Sources

• 1. RoRoSafe bench testing — per-vehicle EWMA baseline vs deck-zone smoke detection [VERIFY: ~6 °C delta, 18–25 min lead].
• 2. Companion method detail — "Setting Thermal Anomaly Detection Thresholds: ΔT and Dwell" (magnitude/dwell on top of this baseline).
• 3. Companion method detail — "Tuning Coherence Windows to Kill False Positives" (spatial suppression layer).