Privacy-Preserving Analytics for Smart Manufacturing

The industrial internet of things (IIoT) has ushered in unprecedented opportunities for the evolution of distributed AI embedded in, for example, smart manufacturing machines, automated guided vehicles, and sensors. These AI models serve not only as predictive tools but autonomously make pivotal decisions based on real-time information within the system. However, the predominant focus is on improving effectiveness and efficiency of AI models, with critical consideration of cybersecurity risks often neglected. AI models that lack privacy-preserving algorithms are susceptible to exploitation, including the compromise of sensitive information and disturbances to the decision-making system.

These attacks, meanwhile, pose a substantial threat, risking unexpected disruptions to manufacturing operations and potential business losses. Therefore, addressing these vulnerabilities is imperative for sustaining the integrity and efficiency of advanced manufacturing processes and of such technological advances as digital twins. Digital twins have brought innovation to manufacturing by enhancing system and operational monitoring, minimizing disruptions, and optimizing production planning and scheduling.

Building on the mosaic neuron perturbation-based algorithm that was developed during the related Year 1 project, the team led by Pennsylvania State University and Siemens will expand its conceptual framework to devise these new privacy-preserving algorithms. These will be specifically crafted to protect sensitive information in distributed AI systems from privacy-leaking attacks and reduce vulnerability to adversarial attacks. A comprehensive set of experiments will benchmark the efficacy of the proposed techniques in terms of model accuracy and attack-prevention robustness, with a subsequent technical demonstration providing comparative analysis of the algorithm's productivity against state-of-the-art approaches.

The privacy-preserving AI algorithms derived from this research will serve as a pivotal tool to help alleviate risks associated with privacy leaks and system disturbances within autonomous manufacturing systems. The project additionally aims to help small- and medium-sized manufacturers harness the benefits of intelligent and secure digital twin technology. Notably, these advancements are poised to enhance the efficacy of decision-making processes in manufacturing operations and management.