Federated Learning for Privacy-Preserving Collaboration in Smart Manufacturing

Abstract

Manufacturers today are increasingly connected as part of a smart and connected community. This transformation offers great potential to deepen their collaborations through resource and knowledge sharing. While the benefits of artificial intelligence (AI) have been increasingly demonstrated for data-driven modeling, data privacy has remained a major concern. Consequently, information embedded in data collected by individual manufacturers is typically siloed within the bounds of the data owners and thus under-utilized. This paper describes an approach to tackling this challenge by federated learning, where each data owner contributes to the creation of a global data model by computing a local update of relevant model parameters based on its own data. The local updates are then aggregated by a central server to train a global model. Since only the model parameters instead of the data are shared across the various data owners, data-privacy is preserved. Evaluation using sensor data for machine condition monitoring has shown that the global model produced by federated learning is more accurate and robust than the local models established by each of the single data owners. The result demonstrates the benefit of secure information sharing for individual manufacturers, especially Small and Mid-Sized Manufacturers (SMMs), for improved sustainable operation.