Governance-as-Code for Deterministic, Reliable and Reproducible AIoT Systems in Smart Buildings

Reliable and reproducible operation remains difficult in Artificial Intelligence of Things(AIoT) systems for smart buildings, where rapid software updates must coexist with safety, privacy,resilience and energy-efficiency requirements. This article presents a shortened journal version ofa governance-as-code framework that embeds executable policies into both CI/CD pipelines andruntime control. The proposed monolithic-modular architecture centralizes admission control,intent routing, capability resolution, scheduling, health management and rollback in a deterministicorchestrator, while allowing independently versioned AI service modules through standardizedcontracts. A normalized device layer abstracts heterogeneous building protocols such as BACnet/IP,Modbus/TCP, OPC-UA and MQTT, preventing direct module-device interaction and supportingcoordinated recovery. The deployment problem is formulated as a constrained multi-objectivetrade-off over tail latency, mean time to repair (MTTR), availability, energy-normalized comfortand governance overhead. Evaluation follows a stage-consistent pipeline spanning simulation,hardware-in-the-loop experiments and a production-like prototype. Across these stages, theproposed architecture reduces P90 action latency by approximately 40%, reduces MTTR by about60%, approaches 99.9% availability under fault injection, improves energy-normalized comfort byapproximately 15% and keeps governance overhead within 3-5%. These findings suggest thatgovernance-as-code can transform compliance from static documentation into a continuouslyenforced operational mechanism for trustworthy and sustainable AIoT systems.

Hali tarjima qilinmagan