Edge Intelligence Model for Autonomous Decision-Making in IoT-Enabled Smart Manufacturing Enterprises
Annotatsiya
Smart manufacturing environments generate IoT sensor data at rates that cloud AI architectures cannot serve within the sub-100ms decision latency that real-time process control requires: network round-trips, bandwidth limits, and cloud inference queuing collectively produce latencies an order of magnitude above the defect-propagation threshold. Edge intelligence-deploying optimized AI models directly on edge compute nodes co-located with the manufacturing process-resolves this latency constraint while reducing bandwidth consumption and eliminating cloud dependency during network partitions. This paper proposes a hierarchical edge-intelligence architecture (cloud-to-regional-edge-to-local-edge-to-IoT-device), deployed across 12 manufacturing plants (48,000 IoT sensors, 24,000,000 annotated data points, 12-month evaluation; 80/20 chronological split: 19,200,000 train / 4,800,000 test), achieving decision accuracy 0.884 at 48ms mean latency, outperforming cloud-only AI (0.784, 840ms; +12.8% accuracy, −94.3% latency [(840-48)/840]), edge-only ML (0.724, 84ms), and rule-based PLC (0.684, 2,400ms; +29.2% [(0.884-0.684)/0.684]). Deployed outcomes: −48.4% unplanned downtime, −34.2% defect rate, +28.4% energy efficiency, −78.4% bandwidth vs. cloud-only AI. Sub-domain F1: Predictive Maintenance (0.924), Defect Detection (0.884), Quality Control (0.844), Energy Optimisation (0.784). Sensor drift degrades accuracy to 0.784 (−11.3%) and network partitioning to 0.724 (−18.1%), each with a deployed autonomous mitigation.
Hali tarjima qilinmagan