Secure and Efficient Decentralized Analytics on Digital Twins Using Federated Learning

Digital Twin, the Internet of Things, and Machine Learning have the potential to redefine our imagination of innovation and impact every sector in the future. A digital twin is a virtual replica of a physical object that facilitates real-time data collection from IoT sensors and uses machine learning/artificial intelligence to analyze and predict data. In this paper, we investigate the use case of federated machine learning for analyzing data on digital twins. Since digital twins capture data about their physical counterparts, they may contain private and sensitive information. This poses challenges in performing machine learning tasks on digital twins due to potential data privacy leakage threats. The motivation behind this research is to explore data ownership and privacy issues in the context of the vast amounts of data available in digital twins and to apply federated learning as a potential solution to address these challenges. We investigated how federated learning can enhance the predictive capability of digital twins by promoting knowledge sharing while maintaining privacy. Our findings indicate that federated learning is an effective approach for facilitating learning between twins and can significantly reduce communication overhead in comparison to training on physical edge devices. Additionally, we show that the convergence rate of federated learning on digital twins improves due to the greater amount of training data available on digital twins compared to the baseline federated learning approach.