A Fail-Closed Edge-AI Reference Architecture for Occupational RF/EMF Exposure Monitoring and Safety Governance

Abstract Personnel working in the vicinity of high-power radio-frequency (RF) and electromagnetic field (EMF) sources may face elevated long-term health risks if exposure is insufficiently monitored, interpreted, and governed at the operational level. Existing approaches often rely on static zoning, offline assessments, or centralized monitoring, which are insufficient for dynamic, heterogeneous, and safety-critical environments. This repository presents RF-Safety Edge, a fully specified, fail-closed Edge-AI reference architecture for occupational RF/EMF exposure safety. The system is designed to operate locally at the edge, integrating sensor fusion, uncertainty-aware inference, drift detection, and policy-based decision logic to support human-centric operational guidance such as zoning, dwell-time limits, and personnel rotation — without automating hazardous actions. The architecture deliberately avoids any form of tactical optimization, targeting, or system placement. Its sole purpose is exposure risk reduction, governance, and auditability under uncertainty. What This Work Provides This release contains: A complete Edge-AI specification set (YAML/Markdown) covering: signal processing pipelines, anomaly and drift detection, uncertainty handling and fail-closed policies, update/rollback mechanisms, security and provenance constraints, operational runbooks and verification invariants. A self-contained executable notebook that: generates synthetic RF exposure streams, executes edge-style preprocessing and detection, applies fail-closed zoning and policy logic, performs Monte-Carlo risk simulations, enables reproducible safety testing without external data or uploads. All components are designed to be vendor-neutral, audit-ready, and deployable on constrained edge hardware. Design Principles The system follows five core principles: Fail-Closed SafetyIn the presence of uncertainty, missing data, calibration ambiguity, or drift, the system degrades conservatively to alarm-only or human-in-the-loop modes. Human-Centric GovernanceThe architecture supports decision-making and leadership (e.g., dwell-time limits, rotation guidance) rather than automated enforcement or control. Edge SovereigntySafety decisions do not depend on cloud connectivity and remain operational under intermittent or offline conditions. Reproducibility and ProvenanceEvery artifact is versioned, typed, and bound to explicit compatibility and provenance declarations. Non-Tactical ScopeThe work explicitly excludes system placement optimization, coverage analysis, targeting, or any operational effectiveness enhancement. Intended Use RF-Safety Edge is intended for: occupational RF/EMF safety in industrial, infrastructure, and critical environments, regulatory and compliance testing, safety engineering research, edge-AI governance and assurance studies. It is not intended for weapons design, targeting, or tactical system optimization. Citation and DOI This repository is published to enable transparent citation, reuse, and peer discussion of edge-based safety governance architectures under real-world uncertainty.