Always-on, ultra-low-power neuromorphic gas-leak detection — and the operational dashboard + 3D facility twin that visualises it.
ANFAS ("breaths" in Arabic) puts gas-detection intelligence on event-driven neuromorphic compute (a Spiking Neural Network, prototyped on Intel Loihi 2 / BrainChip Akida) so a node can idle at microwatts. That makes per-valve sensor density economically feasible — closing the blind spots that power-hungry, $1,000–$10,000-installed conventional fixed detectors leave uncovered. Each node is dual-use: it protects a worker from an H₂S knockdown and flags a methane leak for climate reporting.
This repository is the front-end: a Palantir-style control room for a refinery's sensor fleet, centred on an interactive 3D facility twin where every node is a live gas sensor.
Concept · sim data. Telemetry is simulated for demonstration (Intel AI Global Impact Festival 2026). The numbers below are sourced market and product context; see
ANFAS Market.mdfor the full competitive analysis, sources and caveats.
The problem ANFAS attacks
- Conventional fixed point detectors cost ≈$1,000–$3,000 each (up to $10,000+ installed); portables $100–$200.
- ~3.5M gas detectors are deployed globally — only ~¼ are IoT-connected.
- The barrier to per-valve density isn't unit price — it's power, wiring, recurring calibration, and explosion-proof install cost.
The neuromorphic edge
- Compute idles at microwatts (µW) — event-driven, only "wakes" to transmit on a spike.
- Intel–Cornell Loihi olfaction (Nature Machine Intelligence, 2020) matched a deep-learning system needing ~3,000× more training samples, learning each odour from a single sample.
- Hardware-portable: prototypes on BrainChip Akida (buyable today, $249–$1,495), with Loihi 2 as the research-optimisation target.
The market (ranges, not point estimates)
- Global gas-detection equipment: $5.60B (2024) → $10.66B (2030), fixed = 65% (Grand View Research).
- MEA gas detection → ≈$551M by 2030 (CAGR 11.8%).
- Methane monitoring: $3.5–4.8B (2024–25) → $8.9–11.6B (2034).
- Realistic 3–5-yr obtainable market (SOM): $1–15M — conservative $1–2M, base $5–8M, optimistic $15M+ (certification + hardware path permitting).
Why MENA / GCC
- ADNOC's Ruwais complex processes up to 922,000 bpd — the largest single-site refinery in the Middle East, 4th-largest worldwide.
- Global Methane Pledge: 159 countries + EU to cut methane 30% below 2020 levels by 2030.
- COP28 Oil & Gas Decarbonisation Charter: 50 producers (incl. 29 NOCs — ADNOC, Saudi Aramco, PDO, Bapco) targeting methane down to 0.2% of production by 2030 and net-zero operations by 2050.
Honest scope (we don't over-claim)
- The sensor front-end is conventional (electrochemical / MOS / NDIR) — the innovation is in compute, deployment economics and dual-use system design, not the transducer.
- ANFAS enters as supplementary dense monitoring alongside certified detectors; SIL 2 / ATEX-IECEx certification is staged, not claimed.
- The SNN is paired with deterministic raw-concentration thresholds, so the life-safety trip stays auditable — the network never silently owns the safety decision.
A shared shell (top bar + left rail) over several views:
- Overview — the main control room: fleet KPIs, 2D facility map, multi-gas sparklines, live alert feed, emissions band, and a node inspector. Centred on a preset critical alarm at valve V‑237 (H₂S, Zone 4 · Sulfur Recovery).
- Live Map — the interactive 3D facility twin (raw three.js): 41 sensor nodes across 5 zones, pulsing alarms, orbit/zoom, hover + click to inspect. Click a silo/vessel to step inside — the camera fades into a fully enclosed, properly-scaled interior (walls, catwalk, liquid at fill level, machinery, wall-mounted sensor units you can click).
- Nodes · Alerts · Emissions · Network · Settings — fleet table, event log, climate detail, mesh/RF health, and thresholds/alarm-routing/neuromorphic config.
- ⌘K / Ctrl+K command palette — search nodes, zones and pages from anywhere.
- Next.js 16.2.9 (App Router, Turbopack) · React 19.2
- Tailwind CSS v4 (
@themedesign tokens) - three.js r185 — raw, not react-three-fiber (UnrealBloom + RoomEnvironment + OrbitControls)
- lucide-react icons · fonts: Inter, JetBrains Mono, Noto Sans Arabic
npm install
npm run devOpen http://localhost:3000.
npm run build # production build
npm start # serve the production buildANFAS is a student-origin concept for the Intel AI Global Impact Festival 2026. This app uses simulated data for demonstration. Market figures are third-party estimates that disagree by up to ~2× and are presented as ranges; vendor µW/mW power figures are best-case/idle. This is competitive/market intelligence and a UI prototype — not safety-certification or investment advice.