BROWSE
Detail
Journal Article
ACL-DT: adaptive continuous learning digital twin framework for resilient manufacturing
Cited 0 time in 
Share 
Share
- Authors
- Yoo Ho Son, Eun Seo Lee, Ji Yeon Son
- Issue Date
- 2026-05
- Citation
- International Journal of Advanced Manufacturing Technology, v.144, no.5, pp.3597-3610
- ISSN
- 0268-3768
- Publisher
- Springer Nature
- Language
- English
- Type
- Journal Article
- Abstract
- Accurate and timely productivity prediction is critical in manufacturing, where uncertainties from operator variability, machine failures, and urgent orders frequently disrupt production schedules. Conventional approaches, such as discrete event simulation (DES) and digital twin (DT)-based methods, provide accurate insights but are often limited by high computational cost and static surrogate models that fail to adapt to evolving factory conditions. To address these challenges, this study proposes the Adaptive Continuous Learning Digital Twin (ACL-DT) framework. Rather than explicitly modeling individual sources of uncertainty, ACL-DT captures and compensates for their effects through a unified learning and adaptation framework. The proposed framework integrates simulation-driven surrogate learning, reality-based correction, and an adaptation mechanism aligned with factory operation cycles. This design enables fast predictions while continuously incorporating real production records to maintain robustness against production drift. The framework was validated through an industrial case study on a vibration welding process involving human–machine collaboration. Experimental results show that ACL-DT consistently outperformed a wide range of static machine learning baselines under real-time inference constraints. It achieved a mean prediction error of about 7% while sustaining sub-millisecond inference. Compared with DES, ACL-DT delivered comparable or better accuracy while operating more than 80 times faster. Accurate cycle-level predictions directly support delivery reliability by reducing the risk of accumulated delays. As a framework-level contribution, ACL-DT is not tied to a specific learning algorithm and can accommodate different surrogate models that satisfy deployment constraints, advancing DT-based productivity prediction toward adaptive decision support in dynamic manufacturing environments.
- Keyword
- Continuous learning, Digital twin, Productivity prediction, Smart manufacturing, Surrogate modeling
- KSP Keywords
- Adaptive Decision, Continuous learning, Digital Twin, Discrete Event(DE), Learning algorithms, Learning and adaptation, Machine Collaboration, Prediction error, Real-time inference, Smart Manufacturing, Surrogate modeling