Abstract
Real-time quality control in Industrial Internet of Things (IIoT) environments demands low-latency data processing and intelligent decision-making at the network edge. This article investigates the integration of edge computing and artificial intelligence (Edge-AI) to enable real-time defect detection and process monitoring in manufacturing. We propose a hierarchical Edge-AI architecture that deploys lightweight convolutional neural networks on edge devices for local inference, while cloud servers handle model retraining and aggregation. A prototype was implemented using Raspberry Pi nodes equipped with camera modules and accelerometers, connected via MQTT to a local edge server. Experimental evaluation on a simulated assembly line achieved a mean detection latency of 18.3 ms and an accuracy of 94.7% for surface defects, outperforming cloud-only approaches by reducing latency by 62%. Resource utilization analysis shows that model quantization and pruning reduced memory footprint by 45% without significant accuracy loss. The findings demonstrate that Edge-AI provides a viable solution for real-time quality control in bandwidth-constrained industrial settings, offering a balance between accuracy, latency, and computational cost.