Enhancing the performance of Industrial IoT LoRaWAN-enabled federated learning frameworks: A case study

Abstract

The ongoing development of Industrial Internet of Things (IIoT) smart systems is transforming industrial maintenance by improving operational efficiency. In this context, anomaly detection within IIoT architectures is crucial for early issue identification in industrial machinery. However, many systems generate vast sensor data while operating in environments with poor accessibility and network coverage, making centralized training impractical. Federated learning (FL) offers a solution by enabling distributed training on local devices, reducing bandwidth usage by transmitting models instead of raw data, and enhancing privacy. Despite these advantages, applying FL in IIoT resource-constrained devices — characterized by limited storage, processing capacity, and high-frequency heterogeneous data — remains challenging. This study showcases FL framework performance enhancement in LoRaWAN-enabled IIoT environments through optimized local machine data management. The improvements explore three key approaches: (1) techniques to manage high-variability, high frequency data from multiple sources via LoRaWAN-enabled prototypes, (2) an adaptive optimization approach addressing industrial machinery’s sensory diversity, and (3) strategies to reduce false alarms by refining the management system to categorize risk levels based on proximity to anomaly detection thresholds. The enhanced framework achieves an F1-score of 97%, TPR of 96%, and TNR of 80%, with the positive class representing normal conditions and the negative class indicating anomalies. Moreover, the false alarm reduction strategy decreases false positives by at least 72%, preventing values near the threshold from being misclassified as high risk anomalies.