Insider Temporary
Researchers have developed a brand new Bayesian calibration framework that considerably improves the accuracy of digital twin fashions for automated materials dealing with programs (AMHSs) by addressing each parameter uncertainty and system discrepancy.
The framework makes use of sparse discipline knowledge and probabilistic modeling to calibrate digital twins, outperforming typical fashions and enabling sooner, extra dependable predictions in complicated manufacturing environments.
The strategy has been validated via empirical testing, utilized at Samsung Show, and is designed to scale throughout varied industries looking for correct, self-adaptive digital twin options.
PRESS RELEASE — Digital twins for automated materials dealing with programs (AMHSs) of semiconductor and show fabrication industries undergo from parameter uncertainty and discrepancy. This results in inaccurate predictions, in the end affecting efficiency. To deal with this, researchers have developed a brand new Bayesian calibration framework that concurrently accounts for each parameter uncertainty and discrepancy, bettering the prediction accuracy of digital twin fashions. This progressive framework holds nice potential for enhancing digital twin applicability throughout various industries.
To handle more and more complicated manufacturing programs, involving materials flows throughout quite a few transporters, machines, and storage areas, the semiconductors and show fabrication industries have carried out automated materials dealing with programs (AMHSs). AMHSs usually contain complicated manufacturing steps and management logic, and digital twin fashions have emerged as a promising answer to boost the visibility, predictability, and responsiveness of manufacturing and materials dealing with operation programs. Nonetheless, digital twins don’t all the time totally replicate actuality, doubtlessly affecting manufacturing efficiency and should end in delays.
Digital twins of AMHSs face two main points: parameter uncertainty and discrepancy. Parameter uncertainty arises from real-world parameters which might be tough to measure exactly however are important for correct modeling. For instance, the acceleration of an automatic car in AMHSs can range barely within the discipline however is fastened within the digital twin. Discrepancy, however, originates from the distinction in operational logic between the real-world system and the digital twin. That is particularly necessary since digital twins usually simplify or resemble the true processes, and discrepancies collected over time result in inaccurate predictions. Regardless of its significance, most performance-level calibration frameworks overlook discrepancy and focus solely on parameter uncertainty. Furthermore, they usually require a considerable amount of discipline knowledge.
To deal with this hole, a analysis group led by Professor Soondo Hong from the Division of Industrial Engineering at Pusan Nationwide College, South Korea, developed a brand new Bayesian calibration framework. “Our framework permits us to concurrently optimize calibration parameters and compensate for discrepancy,” explains Prof. Hong. “It’s designed to scale throughout massive good manufacturing facility environments, delivering dependable calibration efficiency with considerably much less discipline knowledge than typical strategies.” Their research was made obtainable on-line on Might 08, 2025, and revealed in Quantity 80 of the Journal of Manufacturing Methods on June 01, 2025.
The researchers utilized modular Bayesian calibration for varied working situations. Bayesian calibration can use sparse real-world knowledge to estimate unsure parameters whereas additionally accounting for discrepancy. It really works by combining discipline observations and obtainable prior information with digital twin simulation outcomes via probabilistic fashions, particularly Gaussian processes, to acquire a posterior distribution of calibrated digital twin outcomes over varied working situations. They in contrast the efficiency of three fashions: a field-only surrogate that predicts real-world habits straight from noticed knowledge; a baseline digital twin mannequin utilizing solely calibrated parameters; and the calibrated digital twin mannequin accounting for each parameter uncertainty and discrepancy.
The calibrated digital twin mannequin considerably outperformed the field-only surrogate and confirmed concrete enhancements in prediction accuracy over the baseline digital fashions. “Our method permits efficient calibration even with scant real-world observations, whereas additionally accounting for inherent mannequin discrepancy.” notes Prof. Hong, “Importantly, it provides a sensible and reusable calibration process validated via empirical experiments, and may be personalized for every facility’s traits.”
The developed framework is a sensible and reusable method that can be utilized to precisely calibrate and optimize digital twins, in any other case hindered by scale, discrepancy, complexity, or the should be versatile for widespread cross-industry utility. This method precisely predicted discipline system responses for large-scale programs with scarce discipline observations and supported fast calibration of future manufacturing schedules in real-world programs. The calibration system can also be apt for discrepancy-prone digital fashions that behave otherwise than their real-world counterparts attributable to simplified logic or code. Excessive-complexity manufacturing and materials dealing with environments, the place handbook optimization is difficult, may also profit from this calibration framework. It additionally permits the event of reusable and sustainable digital twin frameworks that may be utilized to totally different industries. Moreover, this method is being utilized and scaled at Samsung Show, the place the researchers have carefully collaborated with operation groups to customise the framework for the real-world complexities.
General, this novel framework has the potential to vary the applicability and effectivity of AMHSs. Trying forward, Prof. Hong concludes, “Our analysis provides a pathway towards self-adaptive digital twins, and sooner or later, has sturdy potential to turn into a core enabler of good manufacturing.”
