A Robust Anomaly Detection Framework in Industrial Internet of Things

The rapid growth of the Industrial Internet of Things (IIoT) has highlighted the crucial need for robust anomaly detection systems to prevent operational disruptions and maintain the stability and reliability of industrial systems. This study presents an innovative methodology that combines Wasserstein Generative Adversarial Networks (WGAN) with isolation forest algorithms to address the challenge of anomaly detection in IIoT amid the frequent issue of missing data. The proposed Isolation Forest with Wasserstein Generative Adversarial Networks (iForest-WGANs) framework exploits the robust data imputation capabilities to reconstruct incomplete sensor data, thereby enriching the feature space for the subsequent anomaly detection executed by the isolation forest model. We thoroughly evaluate the performance of the integrated model against a comprehensive dataset of sensor fault detection, demonstrating a significant enhancement in anomaly detection compared to conventional benchmarks. Our integrated approach achieved 97.2% detection accuracy and 93.9% recall on a comprehensive sensor fault dataset, significantly outperforming conventional benchmarks by 5-12% and reducing false positives. We thoroughly evaluate the model against baseline methods and recent state-of-the-art techniques, demonstrating substantial improvements in precision and F1-score. These results confirm the effectiveness of the proposed iForest-WGAN framework, particularly in scenarios with sparse data, and underline its potential to enhance the resilience of IIoT systems.