EDGE-BASED VISION FOR INDUSTRIAL IOT: REAL-TIME QUALITY INSPECTION AND PREDICTIVE MAINTENANCE

Abstract

Industrial Internet of Things (IIoT) implementations increasingly move computer-vision analysis from the cloud to local hardware to satisfy strict latency, secrecy, and dependability prerequisites on manufacturing floors. Current progress indicates that small CNN/Transformer models and abnormality-detection workflows can attain instant visual examination on integrated platforms while maintaining production-line capacities. Practical examples and comparisons—notably concerning self-supervised abnormality detection (UAD) on collections like MVTec AD and the more recent MVTec AD-2—show dependable flaw pinpointing across varied materials and items, advancing the discipline toward workable in-line quality assurance. Local execution lessens backhaul and reaction durations, and when coupled with predictive-maintenance (PdM) schemes, permits sooner fault identification and enhanced Overall Equipment Effectiveness (OEE).