This paper presents an Industrial Internet of Things (IIoT)-enabled AC–DC power supply with a multi-model embedded machine learning framework for secure, real-time fault detection in oil and gas applications. Five classifiers (DT, GNB, SVM, ELM, QNN) execute on an STM32H7 MCU within an adaptive scheduler that applies early-exit logic and confidence-aware logging. Telemetry uses HTTPS/TLS and Modbus RTU for continuous SCADA integration under EMI-prone, intermittently connected conditions, with an edge-to-cloud feedback loop enabling targeted offline retraining. A custom IP68-rated prototype with shielded sensing, galvanic isolation, and statistical feature extraction achieved up to 97.4 accuracy (F1 = 0.96) across seven fault categories, with all models meeting a sub-1 ms inference budget. The framework demonstrated deterministic execution, stable memory usage, and compliance with industrial standards, enabling robust, cloud-independent deployment in harsh environments.
