Federated Causal Learning for Distributed Manufacturing Networks

Nagi Gebraeel
Georgia Power Early Career Professor
H. Milton Stewart School of Industrial and Systems Engineering

Tuesday, August 26
12 – 1 p.m.
Location: Callaway/GTMI bldg.,
Room 114

Lunch provided for in-person attendees on a first come first serve basis.

If you can’t join us in-person, just us virtually via Zoom.

Whether you are joining virtually or in person, please register here prior to attending.

Abstract: Industry 4.0 has transformed geographically distributed manufacturing enterprises into digitally interconnected networks. This shift has introduced significant challenges in maintaining operational efficiency and managing process disruptions across networked sites. Traditional analytics struggle in this environment because collecting and centralizing high-frequency sensor data from multiple facilities places heavy demands on bandwidth and latency, while differences in local control systems add further complexity and heterogeneity. Decentralized machine-learning frameworks, such as Federated Learning (FL), help mitigate these challenges by enabling sites to collaboratively train shared models without exchanging raw data. However, conventional FL approaches still fall short of capturing how disturbances at one site reverberate through the broader network.  In this talk, I will introduce a new methodology that addresses this limitation by uncovering interdependencies across heterogeneous operations at different sites. Using a federated form of Granger causality, our approach enables each local controller not only to diagnose its own equipment but also to reason about upstream and downstream effects across sites. By capturing these cross-site relationships, the framework supports more effective process optimization, accelerates root-cause analysis, and enhances cybersecurity resilience.

Speaker Bio: Professor Nagi Gebraeel is the Georgia Power Early Career Professor in the Stewart School of Industrial and Systems Engineering at Georgia Tech.  Dr. Gebraeel’s research sits at the nexus of industrial analytics, machine learning, and decision-making under uncertainty. He develops advanced statistical and causal-learning algorithms for machine diagnostics and prognostics, as well as optimization models that guide subsequent operational and logistical decisions in capital-intensive sectors, such as manufacturing, power generation, and aerospace. Dr. Gebraeel leads the Predictive Analytics and Intelligent Systems (PAIS) research group in ISyE. He is a Fellow of the Institute of Industrial and Systems Engineers and was the former president of its Quality and Reliability Engineering Division.