EDGE-BASED FAULT DETECTION WITH LIGHTWEIGHT CNNS FOR IIOT GATEWAYS

The Edge computing has proved to be vital for implementing intelligent systems in real-time usage, especiallywhere data privacy, latency, and bandwidth are an issue. Compact convolutional neural networks (CNNs) aredesigned and tested to suit deployment on ARM-based edge devices in this paper. To implement with efficientlow-power consumption and minimal computational capability, these models are maximally suited for IoTecosystems. Our system is optimized to stream preprocessed sensor data from the edge to the cloud only upondetection of anomalies, thus reducing data transmission and cloud dependency. We combine lightweight CNNarchitectures with anomaly detection mechanisms at the edge layer, utilizing sensor data from diverse industrialand healthcare sources. The framework ensures real-time monitoring with adaptive thresholds and dynamicinference techniques. Large-scale experimentation on synthetic and public datasets confirms the system'saccuracy, responsiveness, and efficiency of use. The networks are evaluated using metrics such as inferencelatency, power consumption, and detection accuracy. The outcomes report up to 70% bandwidth savings and 40%reduction in cloud-processing overhead relative to conventional full-stream systems. In addition, a real-time ehealthcare case study illustrates the application of the solution in practice in patient monitoring settings. TheARM-powered edge processors accurately detect abnormalities like abnormal heart rates or fever peaks, triggeringalerts to the cloud and medical staff without constant data streaming. The modular and scalable architecturefacilitates deployment in multiple domains ranging from smart industry to personalized medicine with quickresponse times and better data security. The architecture also facilitates simple retraining and redeployment withlow engineering overhead, which makes it ideal for large-scale edge AI rollouts.