Edge-AI Predictive Monitoring for Vaccine Refrigeration in Rural Ghanaian Health Clinics

A Combined Edge AI, Embedded Systems, and IoT Approach to Cold-Chain Integrity and Predictive Fault Detection

Abstract

Vaccine cold-chain failures remain a major and preventable cause of vaccine wastage worldwide, and the problem is especially acute in rural, resource-constrained health facilities where unreliable electricity, limited technical support, and unreliable internet connectivity compound the risk. This work presents the design, validation, and prototype implementation of an edge-AI-informed monitoring system for vaccine refrigeration units, motivated specifically by the documented cold-chain challenges faced by rural health clinics in Ghana. Unlike existing systems, which are largely reactive (alerting only after a temperature threshold has already been breached) and cloud-dependent, the proposed system combines three complementary detection layers running locally on a low-cost ESP32-S3 microcontroller: (1) predictive compressor-fault detection derived from a machine-learning analysis of cycling behaviour, (2) gas-based detection of likely non-vaccine items placed in the unit, and (3) cumulative vaccine-exposure-risk tracking using a real-time clock. The predictive detection approach was first validated against a genuine, publicly available industrial dataset of a failing air compressor, achieving 95.3% recall in identifying documented failure periods, before being applied to a purpose-built simulation of vaccine-fridge degradation. The resulting detection logic was distilled into lightweight, real-time embedded logic and implemented and verified in a simulated hardware environment (Wokwi), including a live IoT reporting layer built on MQTT and a web-based dashboard. This document describes the related work that motivated the system, the validation methodology, the complete system design, results, and an honest discussion of the system's current limitations and next steps.