Collaborative Research: Self-Learning Digital Twins for Robotic Remanufacturing under Uncertainty

Digital twins have emerged as an important technology for predicting system behavior and supporting decision-making in manufacturing systems. Remanufacturing presents one of the most challenging environments for such technologies as products arrive with varying designs, levels of wear, and uncertain histories, making automated decision-making difficult. Current digital twin technologies often require extensive computational resources and struggle to adjust when operating conditions change. These limitations restrict their broader deployment in advanced manufacturing environments where flexibility and rapid decision-making are essential. This project seeks to establish new scientific foundations for self-learning digital twins that can model, predict, and adjust to uncertain manufacturing conditions in real time. The research will strengthen the competitiveness and sustainability of United States manufacturing by promoting more practical recovery and reuse of valuable products and materials, reducing waste generation, and supporting circular economy. The project will also provide interdisciplinary education and workforce development opportunities in artificial intelligence, robotics, manufacturing, and data science through student research experiences and outreach activities. The specific goal of the research is to develop a new computational framework for self-learning digital twins that unifies predictive modeling and autonomous decision-making under uncertainty. The research investigates three interconnected objectives: (1) development of a hybrid modeling framework that combines physics-based models and data-driven surrogate models through Bayesian multi-fidelity fusion to balance computational cost and predictive accuracy; (2) establishment of a hierarchical reinforcement learning framework that links strategic planning and operational decision-making while incorporating self-supervised and semi-supervised learning mechanisms; and (3) creation of a robotic disassembly testbed for consumer electronic products to evaluate prediction, planning, and decision-making under varying product conditions. The overarching focus is on developing digital twins that continuously improve their predictive and decision-making capabilities as new information becomes available while maintaining computational tractability in complex manufacturing environments. The resulting advances will contribute new knowledge in digital twin modeling, machine learning under uncertainty, and autonomous manufacturing systems through a collaborative effort that brings together complementary expertise from the University of Florida, the University of California, Riverside, and Florida International University to establish the scientific foundations of next-generation self-learning digital twins. 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.