Stochastic Modeling for NDIR Gas Sensor Drift: Definition, Decomposition, Dynamics and Calibration

Drift undermines low-cost sensor measurements over time but is often treated only as calibration, despite unclear definition, causes and timing. The thesis focuses on non-dispersive infrared CO2 sensors, defining and addressing drift. Errors are stochastic processes; drift means their statistics change. An additive model separates reversible environmental effects from irreversible instrument changes; field data support this.

After environmental correction, most sensors show mean-reverting instrumental errors, not a common deterministic trend. An Ornstein–Uhlenbeck process captures each sensor's baseline error, reversion speed and random fluctuations, guiding calibration.

Calibration without a co-located reference uses pseudo-references from same-type sensors, reference stations and local networks via environmental transfer, fusion and rendezvous detection. Field and indoor data test them.