Real-Time Predictive Maintenance in Smart Factories Using Industrial IoT Sensor Fusion and Edge-AI: An Empirical Assessment Across Manufacturing Sectors

Abstract

The convergence of Industrial Internet of Things (IIoT) sensor networks, edge computing architectures, and artificial intelligence- driven analytics is fundamentally reshaping predictive maintenance (PdM) capabilities in smart manufacturing  environments. This paper presents an original empirical investigation of a real- time IIoT- enabled predictive maintenance framework — the Sensor Fusion Edge- AI Predictive Maintenance (SF- EAPM) system deployed and evaluated across six manufacturing plants spanning three industry sectors: aerospace component machining, semiconductor fabrication, and precision injection moulding. The SF- EAPM system integrates multi- modal sensor fusion (vibration, acoustic emission, \ thermal imaging, and current signature analysis) with an on- device edge- AI inference engine based on a lightweight Temporal Convolutional Network (TCN) architecture, enabling sub- second fault detection latency without cloud dependency. A 24- month deployment study across 214 critical production assets reveals that the SF- EAPM system achieves a mean fault detection accuracy of 94.7%, a false positive rate of 3.2%, and an average remaining useful life (RUL) prediction error of 6.8 hours. Compared to traditional time- based maintenance schedules, SF- EAPM deployment reduces unplanned downtime by 67.4%, maintenance cost per asset by 41.3%, and overall equipment effectiveness (OEE) improvement of 12.8 percentage points. A novel Industry 4.0 Maintenance Readiness Index (I4MRI) is introduced to benchmark organisational and technical preparedness for IIoT- enabled predictive maintenance adoption. Sector- specific implementation findings, edge- AI model performance benchmarks, and a validated deployment roadmap are presented.