A Safety-Critical Edge Intelligence Architecture for Real-Time Fault Diagnosis and Predictive Maintenance in Electronic Systems

The safety-critical electronic systems installed in power grids, (industrial) automation, transport, and medical technologies require not only intelligent but also deterministic, explainable, and safety-conformant maintenance solutions. The traditional reactive, preventive, and cloud-centric predictive maintenance methods remain limited by reliance on networks, communication latency, and insufficiently guaranteed runtime safety. The objective of this study is thus to design and test a safety-related edge-wise smart framework that provides real-time fault diagnosis and predictive preventive maintenance at the system edge. A safety-focused design-implementation-evaluation approach was adopted, combining hierarchical edge computing with lightweight deep-learning-based fault diagnosis, hierarchical-logic probabilistic reasoning, and explicit safety techniques such as redundancy, confidence validation, and fail-safe logic control. Several exceptional results were obtained by means of experimental testing in simplicity of fault injection under controlled conditions and high degradation limits: the proposed architecture demonstrated a high fault diagnosis rate of nearly 98%, deterministic end-to-end delays of around 2,308 milliseconds, a low false alarm (2.1) and missed detection (1.3) rates, and high availability (99.94) in operation under fault and unpredictabilities. These findings indicate that edge-deployed intelligence can meet the analytical performance, real-time performance, and functional safety needs simultaneously- something that has not been realised by many centralised approaches. The assertion is that safety-critical edge intelligence is a reflective, scalable paradigm for reliable electronic systems, and it is suggested that future systems adopt edge-centric, confidence-conscious intelligence and that functional safety criteria evolve to explicitly account for the practice of deploying artificial intelligence at the edge.