MAYI-EDGE: An Offline-First Edge-AI and Digital-Twin Architecture for Intermittent Water Systems

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Intermittent water supply creates nonstationary pressure, uncertain availability, leakage risk, and operational blind spots, while cloud-only monitoring is fragile where connectivity and power are unreliable. This article presents MAYIEDGE as an architecture-led reference study with a reproducible algorithmic demonstration, an executable offline transaction contract, and a prespecified confirmatory protocol. The proposed offline-first cyber-physical architecture couples flow, pressure, and tank-level sensing with ESP32-S3-class edge processing, explicitly frozen quality control and inference, transactional local buffering, MQTT-TLS communication, server-side services, and a versioned EPANET/WNTR hydraulic-twin design. A deterministic synthetic fixture demonstrates the frozen detector logic under a declared surrogatereference substitution, alert persistence within the prespecified 60-s window, backlog behavior, checksum-confirmed acknowledgement, and server de-duplication. It does not claim that an EPANET twin, complete physical prototype, MQTT deployment, or field system has already been executed. The defensible novelty is the joint treatment of intermittent supply, offline-first edge inference, transactional store-and-forward, a versioned hydraulic-twin design, resource-constrained deployment, and confirmatory evaluation prespecified before testing. Four statistical hypotheses (H1–H4) and a separate engineering validation gate (H5) define the future campaign. Parameters, temporal splits, baseline selection, ablations, calibration, missingness, network stress, energy profiling, human evaluation, multiplicity, and fixed sample sizes are specified in advance.