Secure Federated Learning for Collaborative Fault Detection in Distributed Industrial Systems
Keywords:
Secure Federated Learning; Collaborative Fault Detection; Distributed Industrial Systems; Differential Privacy; Predictive Maintenance.Abstract
This study proposes a secure federated learning framework for collaborative fault detection in distributed industrial systems, focusing on improving diagnostic accuracy, data privacy, and operational reliability. The proposed methodology enables multiple industrial facilities to collaboratively train machine learning models without exchanging sensitive operational data. Local models are developed using equipment sensor measurements, vibration signals, temperature variations, and historical fault records, while encrypted model updates are aggregated through a centralized coordination mechanism. Secure aggregation and differential privacy techniques are incorporated to reduce information leakage and protect proprietary manufacturing data. An adaptive federated optimization strategy addresses heterogeneous equipment characteristics, imbalanced fault distributions, and communication constraints across participating facilities. The framework evaluates fault classification accuracy, precision, recall, convergence stability, communication overhead, and resilience against model poisoning attacks. The proposed approach establishes a scalable foundation for privacy-aware industrial fault diagnosis, collaborative predictive maintenance, and secure intelligent monitoring across geographically distributed manufacturing environments.