An Edge-Intelligent Vision Transformer and TinyML-Based Predictive Framework for Autonomous Fault Diagnosis in Smart Industrial IoT Systems

Abstract

The rapid adoption of Industry 4.0 technologies has transformed conventional manufacturing environments into intelligent Industrial Internet of Things (IIoT) ecosystems characterized by interconnected sensors, edge devices, autonomous controllers, and cloud-based analytics. Although these smart industrial systems significantly improve operational efficiency and production flexibility, unexpected equipment failures continue to cause production downtime, financial losses, reduced product quality, and increased maintenance costs. Traditional fault diagnosis methods largely depend on periodic inspections or cloud-centric data processing, introducing communication latency, bandwidth limitations, and privacy concerns that reduce the effectiveness of real-time predictive maintenance. To address these challenges, this research proposes an Edge-Intelligent Vision Transformer and TinyML-Based Predictive Framework for autonomous fault diagnosis in smart Industrial IoT systems. The proposed framework integrates industrial sensor networks, thermal and visual image acquisition, embedded edge computing, Vision Transformer (ViT)-based feature extraction, TinyML-enabled lightweight inference, multimodal sensor fusion, anomaly detection, fault classification, predictive maintenance analytics, and autonomous maintenance decision support. The framework performs intelligent diagnosis directly on low-power edge devices, minimizing cloud dependency while enabling rapid fault detection and prediction. Experimental evaluation demonstrates superior fault classification accuracy, reduced inference latency, lower energy consumption, improved predictive reliability, and enhanced maintenance efficiency compared with conventional cloud-based diagnostic systems. The proposed framework provides a scalable, intelligent, and computationally efficient solution suitable for smart manufacturing, predictive maintenance, industrial automation, cyber-physical production systems, digital factories, intelligent robotics, process industries, energy systems, oil and gas facilities, smart grids, and future autonomous Industrial IoT ecosystems.