#South Korea #Pusan National University #Busan #Digital Twins #Calibration Framework

Pusan National University Introduces New Calibration Framework for Digital Twins

In the fast-evolving landscape of semiconductor and display manufacturing, the ability to adapt to increasing complexity is crucial. Industries are progressively adopting automated material handling systems (AMHS), which have revolutionized the way production is executed. However, the integration of digital twins—virtual representations of physical systems—has presented challenges concerning prediction accuracy. Recent research from Pusan National University has unveiled a groundbreaking Bayesian calibration framework that promises to tackle these challenges head-on.

Addressing the Dual Challenges of Parameter Uncertainty and Discrepancy

Digital twins in the context of AMHS face two significant issues: parameter uncertainty and discrepancy. Parameter uncertainty arises when real-world parameters, critical for precise modeling, cannot be accurately measured. Discrepancy, on the other hand, originates from the differing operational logic between the real plant and its digital counterpart. Over time, these problems can lead to a decline in prediction accuracy, hampering production efficiency and potentially resulting in delays. Unfortunately, many existing calibration methods focus exclusively on parameter uncertainty, often requiring extensive field data without adequately addressing discrepancies.

Recognizing these gaps, a research team led by Professor Soondo Hong from the Department of Industrial Engineering developed the innovative Bayesian calibration framework. According to Prof. Hong, “Our framework allows for the simultaneous optimization of calibration parameters and compensates for discrepancies, offering a notable improvement in decision-making and production performance.” The study detailing their work was published in the Journal of Manufacturing Systems in June 2025.

The Mechanics of the Bayesian Calibration Framework

The modular Bayesian calibration technique utilized in this framework is adept at managing various operational scenarios. Unlike traditional methods, this approach not only estimates uncertain parameters but also accounts for discrepancies, leveraging minimal real-world data efficiently. By integrating field observations with prior knowledge through probabilistic models—specifically Gaussian processes—the framework generates a posterior distribution of calibrated outcomes.

The researchers conducted evaluations using three distinct models: 1) a field-only surrogate predicted behaviors based solely on observed data; 2) a baseline digital twin model reliant exclusively on calibrated parameters; and 3) the newly calibrated twin model that addresses both parameters and discrepancies. Results indicated that the calibrated digital twin significantly outperformed the field-only surrogate, with measurable gains in prediction accuracy over the baseline model. As Prof. Hong highlights, “Our approach effectively calibrates even with limited real-world observations and accounts for model discrepancies.”

Practical Applications and Future Implications

The implications of this framework extend far beyond immediate applications, positioning it as a sustainable and reusable solution for calibrating digital twins. The system works effectively in large-scale environments with limited observations, enhancing responsiveness in future production schedules. The framework is particularly beneficial for environments where high complexity and manual optimization present obstacles. Moreover, it is adaptable across diverse industries, linking different sectors through its versatility.

Currently, this framework is being implemented at Samsung Display, where researchers collaborate with operational teams to tailor the calibration system to real-world complexities. As Prof. Hong notes, this innovative work could pave the way for self-adaptive digital twins, becoming a core feature of smart manufacturing in the upcoming years.

Conclusion

The groundbreaking calibration framework developed by Pusan National University signifies a substantial leap toward optimized digital twin applications in semiconductor and display manufacturing. As the industry continues to evolve, this research not only curtails operational inefficiencies but also fosters scalability and adaptability, vital for future advancements in smart manufacturing technologies. For more information on the original study, refer to the paper titled “A digital twin calibration for an automated material handling system in a semiconductor fab,” published in the Journal of Manufacturing Systems.

Reference

• - Title of original paper: A digital twin calibration for an automated material handling system in a semiconductor fab
• - Journal: Journal of Manufacturing Systems
• - DOI: 10.1016/j.jmsy.2025.04.015