Self-Supervised Data Mining Techniques for Industrial Internet of Things (IIoT) Anomaly Detection: A Comprehensive Review

Abstract

The rapid expansion of the Industrial Internet of Things (IIoT) across Industry 4.0 manufacturing plants, power grids, and automated supply chains has led to an exponential explosion of high-dimensional, temporal, and heterogeneous sensor telemetry data. Detecting operational anomalies, physical equipment faults, and cyber-physical attacks in real time is paramount to preventing catastrophic industrial failures and economic losses. However, traditional supervised machine learning and data mining techniques suffer from severe limitations due to the overwhelming scarcity of labeled anomaly data, heavy class imbalance, sensor noise, and dynamic operational drift. Self-Supervised Learning (SSL)—an innovative data mining paradigm that constructs pretext learning tasks directly from unlabeled raw data—has emerged as a transformative solution for learning robust representation spaces without manual annotation. This paper provides a state-of-the-art review on self-supervised data mining techniques for IIoT anomaly detection. We systematically categorize and evaluate mainstream SSL paradigms, including contrastive learning, predictive temporal pretext tasks, masked autoencoding, and generative adversarial frameworks. Furthermore, we analyze key data mining strategies for feature extraction, temporal windowing, multi-sensor correlation, and noise reduction in industrial environments. A thorough quantitative comparison across standardized IIoT benchmarks (such as SWaT, WADI, and SMD) highlights the superiority of contrastive and masked SSL models in ultra-low label regimes. Finally, critical open challenges including edge deployment resource constraints, model interpretability, adversarial robustness, and concept drift are critically evaluated, accompanied by promising future research directions. KEYWORDS: Industrial Internet of Things (IIoT); Anomaly Detection; Self-Supervised Learning; Data Mining; Contrastive Representation; Time-SeriesMining; Cyber-Physical Systems.