A secure multi-layer embedded framework for real-time fault detection in Industrial IoT power supply systems

Abstract

Industrial power supplies are traditionally designed as passive components with limited diagnostic visibility and secure remote supervision. This thesis presents the design and experimental validation of a secure, multi-layer Industrial Internet of Things (IIoT)-enabled power-supply platform that integrates deterministic protection, embedded diagnostics, and authenticated remote monitoring within a unified architecture. The platform combines real-time hard-fault protection with quantized neural network (QNN)-based soft-fault diagnostics executed on a microcontroller-class device, enabling early detection of incipient degradation with bounded latency. Industrial interoperability is maintained through Modbus RTU communication, while secure access is provided via TLS-encrypted HTTPS and a web-based dashboard. Experimental results demonstrate stable electrical operation under sustained thermal stress, consistent communication timing with an average latency of 5.76 ms, and diagnostic performance exceeding 96%, with sub-millisecond inference latency. These results demonstrate that intelligence, connectivity, and cybersecurity can be co-designed while preserving bounded real-time behavior and operational reliability.