Abstract
Advanced sensors and IoT devices have improved the monitoring and control ofcomplex industrial enterprises. They have also created an interdependent fabricof geographically distributed process operations (clients) across theseenterprises. Granger causality is an effective approach to detect and quantifyinterdependencies by examining how one client's state affects others over time.Understanding these interdependencies captures how localized events, such asfaults and disruptions, can propagate throughout the system, possibly causingwidespread operational impacts. However, the large volume and complexity ofindustrial data pose challenges in modeling these interdependencies. This paperdevelops a federated approach to learning Granger causality. We utilize alinear state space system framework that leverages low-dimensional stateestimates to analyze interdependencies. This addresses bandwidth limitationsand the computational burden commonly associated with centralized dataprocessing. We propose augmenting the client models with the Granger causalityinformation learned by the server through a Machine Learning (ML) function. Weexamine the co-dependence between the augmented client and server models andreformulate the framework as a standalone ML algorithm providing conditions forits sublinear and linear convergence rates. We also study the convergence ofthe framework to a centralized oracle model. Moreover, we include adifferential privacy analysis to ensure data security while preserving causalinsights. Using synthetic data, we conduct comprehensive experiments todemonstrate the robustness of our approach to perturbations in causality, thescalability to the size of communication, number of clients, and the dimensionsof raw data. We also evaluate the performance on two real-world industrialcontrol system datasets by reporting the volume of data saved bydecentralization.