Researchers at PUSAN National University reveal new calibration frameworks for digital twins

summary

Busan, South Korea – July 22, 2025 – Industry uses an automated material handling system (AMHSS) to handle the increasing complexity of semiconductors and the expansion of display manufacturing. Digital twins offer better visibility and control, but discrepancies with actual conditions can affect production and cause delays.

AMHSS’s digital twins face two major issues: parameter uncertainty and inconsistency. Parameter uncertainty arises from actual parameters that are difficult to measure accurately, but are essential for accurate modeling. On the other hand, discrepancies stem from differences in operational logic between the actual system and the digital twins. Over time, these problems reduce the accuracy of predictions. However, most calibration methods focus solely on parameter uncertainty, require extensive field data, and often ignore discrepancies.

To address this gap, a research team led by Professor Sungdo Hong of the Faculty of Industrial Engineering at Busan National University in Korea has developed a new Bayesian calibration framework. “Our framework allows us to simultaneously optimize calibration parameters and compensate for inconsistencies,” explained Professor Hong. “It is designed to expand across large smart factory environments, providing reliable calibration performance with much less field data than traditional methods.” Their research was made available online on May 8, 2025 and published in Volume 80 of the Journal of Manufacturing Systems on June 1, 2025.

Researchers applied modular Bayesian calibration to various operating scenarios. This approach can estimate uncertain parameters and explain discrepancies using sparse actual data. Combined with digital twin simulations through probabilistic models to generate a posterior distribution of calibration results, with field observations and prior knowledge, particularly through Gaussian processes. They evaluated three models.

Representation of fields only that directly predict actual behavior from observed data. Baseline digital twin model using only calibrated parameters. Calibrated twin model considering both parameter uncertainty and inconsistency.

The calibrated digital twin model significantly outperformed field-only surrogates, showing measurable improvements in prediction accuracy in the baseline digital model. “Our approach allows for effective calibration even with small real-world observations, taking into account the inherent model inconsistencies,” says Professor Hong. “The important thing is that it provides practical, reusable calibration procedures validated through experimental experiments and can be customized to the characteristics of each facility.”

The developed framework is a practical, reusable solution for precisely tuning and optimizing digital twins, otherwise hampered by scale, inconsistency, complexity, or the need for industry-wide flexibility. It accurately predicts field system responses for large-scale systems with limited observations, allowing for rapid calibration of future production schedules on real systems. Calibration systems are also suitable for contradictory digital models that behave differently from their actual counterparts due to simplified logic or code. When manual optimization is difficult, high multiple production and material processing environments can also benefit from this calibration framework. This allows for the development of sustainable, reusable digital twin frameworks that can be relocated across a variety of industries. The framework is currently being applied and expanded to Samsung Displays. There, researchers work with their operations teams to adapt the system to actual complexity.

This new framework has the potential to translate the applicability and effectiveness of AMHSS. Professor Hong concluded, “Our research offers a path to self-adaptable digital twins and could become a core enabler of smart manufacturing in the future.”

The original paper was titled “Digital Twin Calibration of Semiconductor Fab’s Automatic Material Processing System.” Featured in the Journal of Manufacturing Systems.