Privacy-Preserving Federated Learning for Equipment Failure Detection in Smart Manufacturing

Smart manufacturing, a cornerstone of Industry 4.0, relies on interconnected devices and real-time data analysis to ensure efficient operations and prevent costly equipment failures. Traditional approaches to equipment failure detection often involve centralized data collection, raising critical concerns around data privacy and security. This paper presents a novel federated learning (FL) framework for privacy-preserving equipment failure detection, enabling collaborative anomaly detection without exposing sensitive data. The proposed method is based on autoencoders for anomaly detection, and local models are trained on each machine and then the updates are aggregated to form a global model. Centralized models and standalone models were outperformed by the proposed FL model with an F1 score of 96.1%, while reducing communication overhead by 85%. In addition, to improve security, differential privacy (DP) techniques and secure aggregation were also incorporated. On synthetic datasets, with very strict privacy constraints (ϵ = 0.1), PrivFL shows that accuracy remains high (92.5%). These results show the framework’s scalability, efficiency, and privacy preservation, and thus it can be a practical solution for industrial applications.