afflu ignite continuously monitors industrial pilot flames using fused thermal and optical AI — detecting failure the moment it starts, not after the next inspection.
Trusted by teams backed by
Built for the demands
of critical infrastructure
From flare tips to compliance reports — ignite handles the full detection pipeline without requiring manual intervention or cloud connectivity.
Multi-Modal Detection
Thermal and RGB cameras fused into a single 0–1000 confidence score per frame. No single sensor point of failure.
Edge-First Architecture
All inference runs on-prem. No raw video leaves the facility. Designed for air-gapped and secure environments.
Sub-2s Alert Delivery
Flame-out events trigger alerts in under two seconds. Integrate with existing DCS, SCADA, or notification systems.
Compliance Automation
Continuous timestamped detection logs with confidence scores. Audit-ready evidence without manual data collection.
Three layers.
One score. Zero guesswork.
ignite processes every frame through three independent analysis layers before fusing them into a single, auditable confidence score. Each layer is designed to degrade gracefully — if one modality fails, the others still produce a meaningful signal.
Thermal ROI Analysis
Extracts three measurements from every frame of the thermal camera.
- Intensity vs Background — Ambient-invariant ratio comparing pilot ROI to a cool reference patch. Robust to AGC and temperature drift.
- Variance Over Rolling Window — Standard deviation of mean ROI intensity over 1–2s. Live flames flicker; static heat sources don't.
- Spatial Gradient — Edge strength at fixture boundary. A real pilot creates a sharp, localized gradient — diffuse radiation doesn't.
RGB / Optical Analysis
Three optical features mapped to the thermal coordinate space via calibration homography.
- Motion + Flicker — Frame differencing + bandpass filter extracts dominant flicker frequency. Natural gas pilots flicker at 5–15Hz — a highly distinctive temporal signature.
- Edge Variance — Flame boundaries shift every frame. Fixed objects — glare, metal, reflections — have stable or zero edge movement.
- Color Gate (Veto) — Loose HSV range check. Does not contribute a positive score — only vetoes an otherwise-confident detection when the ROI is definitively non-flame colored.
Fusion Intelligence
Weighted combination of all six feature scores into a single 0–1000 integer per frame.
- Weighted Scoring — Thermal intensity 30%, thermal variance 30%, RGB motion 25%, RGB edge 15%. Weights reflect robustness per environment.
- Color Gate Cap — If the color veto fires, confidence is capped at 200 regardless of other scores — preventing confident false positives.
- Graceful Degradation — Losing one modality entirely still yields a meaningful, lower-confidence score with wider uncertainty bands — never a silent failure.
Fusion score output
Each camera frame produces a single integer from 0–1000. Thresholds are configured at commissioning per site. Above the threshold: pilot confirmed. Below: flame-out alert triggered. The score is logged with every frame, providing a continuous evidence trail for compliance and incident review.
What each layer handles —
and what the fusion solves.
No single feature handles every operating condition in isolation. The power of ignite lies in the fusion — each layer compensates for the blind spots of the others. Hover any cell for a detailed explanation.
| Thermal Layer | RGB / Optical Layer | Fusion Score | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Situation | Thermal Intensity30% | Thermal Variance30% | Spatial Gradient | RGB Motion + Flicker25% | RGB Edge Variance15% | Color Gate (Veto)veto | Outcome | ||||||
Pilot flame — normal burn | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | HIGH | ||||||
Pilot absent — cold stack | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | LOW | ||||||
Residual heat — post flame-out | ⚠ | ✓ | ⚠ | ✗ | ✗ | ⚠ | MED→LOW | ||||||
Solar heating of stack tip | ✗ | ✓ | ⚠ | ✓ | ✓ | ⚠ | LOW | ||||||
Main flame radiation spill | ✗ | ✗ | ✓ | ⚠ | ⚠ | ✓ | MED | ||||||
Static glare / reflections | ✗ | ✓ | ⚠ | ✓ | ✓ | ⚠ | LOW | ||||||
Wind-blown debris in ROI | — | ⚠ | — | ✗ | ⚠ | ✓ | MED–LOW | ||||||
Nighttime / low light | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | HIGH | ||||||
Hazy or foggy conditions | ✓ | ✓ | ⚠ | ✗ | ✗ | ✗ | MED | ||||||
Very stable flame — low flow | ✓ | ⚠ | ✓ | ⚠ | ⚠ | ✓ | MED–HIGH | ||||||
Built for where
failure has consequences
ignite is designed for environments where a missed flame-out means regulatory violations, safety hazards, and uncontrolled emissions.
Oil & Gas
- Upstream flare and pilot monitoring
- Refinery and processing unit compliance
- Continuous regulatory evidence collection
- No scheduled inspection dependencies
Industrial Utilities
- Combustion equipment pilots
- Boiler and furnace flame monitoring
- Remote and unmanned facility sites
- Integration with existing SCADA systems
Remote & Critical Sites
- Operates without cloud connectivity
- Edge inference in air-gapped environments
- Mining and resource extraction sites
- Harsh weather and extreme temperature range
From install to monitoring
in weeks, not months
Connect Your Cameras
Mount thermal and RGB cameras at the pilot fixture. Afflu engineers handle the calibration homography and ROI commissioning. Compatible with FLIR Boson and most industrial camera platforms.
Run Fused AI Analysis
ignite processes every frame through three detection layers — thermal intensity, thermal variance, spatial gradient, optical motion, edge variance, and color gate — fusing them into a per-frame score.
Operate With Confidence
Real-time alerts via your existing DCS or notification system. Every detection event logged with timestamp and confidence score. Audit-ready without additional work from your team.
What your team
actually gets
Faster incident response
Sub-2s detection means your team knows about a flame-out before the next scheduled walkround. Respond in real time, not retrospectively.
Audit-ready compliance
Every frame produces a timestamped confidence score. Continuous evidence collection — no manual data entry, no gaps in the record.
Fewer manual checks
Replace scheduled physical inspections with continuous automated monitoring. Redeploy technician time to higher-value work.
Transparent confidence scoring
Unlike black-box models, ignite's per-feature scoring is explainable. Operators see exactly which signals contributed to each detection.
No infrastructure overhaul
Runs on edge hardware at the site. No cloud dependency, no network connectivity requirement, no changes to existing plant systems.
Graceful under adversity
Haze, night, solar heating, main flame spill — each condition that degrades one layer is compensated by the others. Never a silent failure.
Common questions
Ready to see ignite
at your facility?
Book a 30-minute demo. We'll walk through the detection architecture, discuss your site conditions, and define what a pilot would look like.
No hard sell. 1-day response. Pilot in 2–4 weeks.