The CarbonSense Flare Monitor watches every flare with LWIR thermal imaging, RGB vision and an acoustic proxy. It infers combustion state on the device and quantifies vented CH₄ from process data for confirmed venting events.
Sense is not a molecular methane sensor. It is combustion-state inference, with quantification that is defensible by design.
CH₄ mass quantified only during confirmed VENTING, using SCADA flow — never derived from smoke or opacity. Every event carries duration, thermal evidence and an auditable trail.
States, events, vented mass and device health — from onshore clusters to FPSOs.
Embedded INT8 models run on the device itself — no data link required to infer the state of the flame.
160×120 @ 8.7 Hz. Thermal signature of the flame — the first gate of the flameout detector, with sun-glint and hot-surface rejection via FFT.
Three-class smoke classification (Ringelmann proxy 0/1/2), paired frame by frame with the thermal channel.
Flow state from the sound signature — additional evidence to tell venting apart from no flow.
Confirmed events flow into CarbonOS: vented mass enters the GHG inventory with full lineage in the Proof Engine, including duration, SCADA flow, thermal evidence and acknowledgement trail.
Short answers on how Sense infers flare combustion state and records venting events. To learn about the pilot, book a demo.
Carbonova Sense is an edge AI system for flares, currently at pilot stage. On the device itself, it infers each flare’s combustion state (lit and normal, smoking, degraded, flameout, venting or no flow) from radiometric LWIR thermal imaging, RGB vision and an acoustic proxy. Vented CH₄ is quantified only for confirmed venting, from SCADA flow data. Sense does not measure methane in the air.
Inventories usually assume flares stay lit and burn well, but airborne measurements across three US basins found that flares effectively destroyed only 91.1% of methane, against the 98% assumed (Plant et al., Science, 2022). Sense is for operators of flares in production, processing and refining: operations, HSE and emissions inventory teams that need to know when a flame went out and how much gas was vented in that window.
Three signals reach the device: LWIR thermal imaging is the first gate of the flameout detector, the RGB camera classifies smoke into three classes, and the acoustic proxy helps tell venting apart from no flow. Embedded INT8 models give one verdict per frame without needing a data link. During confirmed venting, CH₄ mass comes from SCADA flow (m = Q · h · x · 0.717), never from smoke or opacity.
Operators on the UK Continental Shelf need North Sea Transition Authority consent to flare or vent during production, and the NSTA expects zero routine flaring and venting by 2030, with new developments planned on that basis (NSTA). Sense infers each flare’s combustion state at the edge and logs flameouts and confirmed venting, with vented methane calculated from SCADA flow, giving operators an auditable record to set against their consents and emissions reports.
Every confirmed event flows into CarbonOS with its duration, SCADA flow, thermal evidence and acknowledgement trail. The CH₄ mass enters the Carbon Intelligence Scope 1 inventory with full lineage in Proof Engine, and alerts go through triage, acknowledgement and resolution. The edge-to-cloud link runs through Build’s AIoT Hub, which also pushes over-the-air model updates to the whole fleet.
No. Sense is not a methane sensor or a leak detector, and it does not measure methane in the air. It infers the flare’s combustion state and quantifies CH₄ only when it confirms venting, using the flow measured by SCADA and the methane fraction of the gas. Fugitive leaks from valves and flanges are out of scope, and an unlit flare with no flow is not counted as an emission event. That boundary is what makes each number defensible.
We are selecting operators for the CarbonSense Flare Monitor pilot — starting with sites that have critical flares and available process data. Developed within the Qualcomm AI Program for Innovators 2026.
.avif)
