What We Learned From 18 Months on the Bench Rig
Before any of the field deployments, the platform lived on a bench rig that staged thousands of thermal events. Most of what we know came from there.
The bench rig is undramatic — a steel frame, a thermal source we can move and modulate, and twelve sensor cells in fixed positions. Over 18 months it produced about 3,200 catalogued events spanning early-stage thermal anomalies, frank fires, and a long catalogue of nuisances.
The most useful thing about it
Reproducibility. Every algorithm change is regressed against the same event catalogue. We can quote false-positive rate against a fixed dataset rather than against the most recent voyage, which is how detection products usually drift into being unmeasurable.
Three things the bench rig got wrong
- Solar gain — we had no solar source on the bench. The first sea trial caught us.
- Vehicle motion — engine bays cool unevenly when the ship rolls.
- Cargo density — close-packed vehicles share thermal mass in ways we did not model.
All three of these are now in the rig in some form. None of them were obvious before sea trials.