GOALI: Federated Interdependency Learning for Securing Distributed Manufacturing Systems

This Grant Opportunities for Academic Liaison with Industry (GOALI) award supports research focused on developing a scientific methodology for strengthening the cybersecurity and operational resilience of distributed manufacturing enterprises. Manufacturing enterprises comprise digitally interconnected plants and operations that need to synchronize production schedules, ensure consistent quality and reliability, and efficiently manage inventory. The same interconnections create complex interactions between processes across different sites. Thus, operational disruptions caused by cyber intrusions at a single location can rapidly propagate and adversely impact operations throughout the entire enterprise. Current threat-detection tools are primarily designed for information-technology environments or, when tailored to operational technology, monitor only a single machine or process in isolation. Sophisticated attacks, however, exploit this blind spot by introducing subtle changes that mimic normal behavior at each individual site while jointly perturbing the coordinated operation of multiple plants. Such attacks result in cascading failures and significant financial losses. This research utilizes machine learning and artificial intelligence to look to develop decentralized data analytics models that learn normal patterns of interactions among distributed control systems to detect these stealthy cyberthreats. A partnership with a major aluminum manufacturing enterprise seeks to ensure the practicality and relevance of this approach in real-world, multi-site manufacturing environments. The research is complemented by an educational plan that includes curriculum development and workforce development, featuring professional training and micro-credentials focused on the cybersecurity of operational technology. The primary goal of this research is to advance both the methodological foundations and practical implementation strategies necessary for detecting cyber-physical threats that exploit latent interdependencies among geographically distributed manufacturing assets. This project looks to address the limitations of conventional methods that rely on centralized data aggregation, which is often impractical in settings with large volumes of high-resolution data and strict data privacy constraints. The research pursues four primary objectives: (i) develop the algorithmic foundations of a decentralized interdependency learning methodology that utilizes process-level interdependencies to distinguish between localized anomalies and system-wide, cyber-induced disruptions, (ii) rigorously establish the theoretical properties that characterize convergence, stability, and estimation accuracy of the methodology, (iii) generalize the learning capabilities to accommodate complex process dynamics using advanced artificial intelligence methods, and (iv) engineer a deployable solution that integrates seamlessly with existing industrial control environments to support real-time cyberthreat monitoring and detection. The project looks to enable a new class of interdependency-aware cybersecurity solutions, offering manufacturing enterprises the ability to detect, localize, and respond to complex, coordinated cyberthreats before they compromise safety, reliability, and productivity. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.