Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

DDD-GenDT: Dynamic Data-driven Generative Digital Twin Framework

Created by
  • Haebom

Author

Yu-Zheng Lin, Qinxuan Shi, Zhanglong Yang, Banafsheh Saber Latibari, Shalaka Satam, Sicong Shao, Soheil Salehi, Pratik Satam

Outline

DDD-GenDT is a dynamic data-driven generative digital twin framework based on the Dynamic Data-Driven Application Systems (DDDAS) paradigm. To address the high data requirements, data constraints, and inability to adapt to environmental changes of existing digital twins, it consists of a Physical Twin Observation Graph (PTOG), which represents the operating state of a physical twin; Observation Window Extraction, which captures temporal sequences; a Data Preprocessing Pipeline for sensor structuring and filtering; and an LLM ensemble for zero-shot predictive inference. Leveraging generative AI, it enables digital twin construction even in data-poor environments while maintaining industrial data privacy. Through the DDDAS feedback mechanism, it autonomously adapts predictions of physical twin wear and tear to support digital twin aging (DT-aging), ensuring gradual synchronization with the evolution of the physical twin. Validated using the NASA CNC milling dataset, the GPT-4-based digital twin achieved an average RMSE of 0.479 A (4.79% of 10 A spindle current) at zero-shot settings, accurately modeling nonlinear process dynamics and physical twin aging.

Takeaways, Limitations

•
Takeaways:
◦
Building digital twins even with limited data
◦
Ensuring industrial data privacy
◦
Data-efficient digital twin modeling using generative AI
◦
Improved adaptability to wear and deterioration of physical twins (DT-aging)
◦
Achieving high prediction accuracy through zero-shot learning
◦
Providing a generalizable and adaptable digital twin modeling approach.
•
Limitations:
◦
Further research is needed to determine generalizability, with validation using only the NASA CNC milling dataset.
◦
Lack of detailed description of the specific configuration and optimization strategy of the LLM ensemble.
◦
Applicability verification is required for various industrial environments and data types.
◦
Further research is needed on potential problems and solutions that may arise when applying this technology to actual industrial environments.
👍
Made with Slashpage