Lightweight Edge based Anomaly Detection Framework for Real Time Industrial IoT Monitoring

Real-time anomaly detection in Industrial Internet of Things (IIoT) environments is critical for minimizing equipment failure, improving operational efficiency, and ensuring predictive maintenance. Traditional models often require heavy computational resources and centralized architectures, making them unsuitable for lightweight edge deployment. The paper presents LEAD-XNet, a Lightweight Edge-based Anomaly Detection system, which can directly execute the system on resource-restrained edge devices. The model proposed incorporates such rare and new things like the Temporal Heterogeneous Information Network (T-HIN) embeddings, spiking graph neural layers, and a hybrid Tiny Temporal Conformer module to be able to capture the temporal-topological patterns in the most effective way possible. A large number of experiments was carried out based on real-world IIoT datasets, which included multivariate time-series data of machinery work. LEAD-XNet obtained the highest accuracy of 96.7% compared to the base models such as GRU-AE (91.3%), LSTM-VAE (93.5%), and Edge-AttentionNet (94.1%). It also showed granted minimal inference latency (22.5 ms) and decreased model size (6.9 MB), which validated its suitability as an edge. The model also has a drift-aware retraining scheme, which is founded on Wasserstein distance, which is resilient to environmental and operational shifts. These findings confirm the usefulness of LEAD-XNet in industrial anomaly detection in real-time with scalability. Its hybrid architecture guarantees high detection accuracy and cost-efficiency, which makes it a solid option to use on the modern smart factories and Industry 4.0 ecosystems.