AI-Driven Digital Twin by Scientific Reports Optimizes Roll-to-Roll Manufacturing
AI-Driven Digital Twin by Scientific Reports Optimizes Roll-to-Roll Manufacturing
A recent study published in Scientific Reports (6 July 2025) introduces a groundbreaking AI-driven digital twin framework for autonomous web tension control in roll-to-roll (R2R) manufacturing. Developed by Anton Nailevich Gafurov, Sooyoung Lee, Uzair Ali, Muhammad Irfan, Inyoung Kim, and Taik-Min Lee, the research addresses the complex challenge of maintaining precise web tension in high-throughput manufacturing environments. The team demonstrates that leveraging AI and digital twin integration can significantly reduce tension variation and enhance the stability of R2R systems.
Roll-to-Roll Manufacturing: Challenges and Requirements, Scientific Reports
R2R manufacturing is a continuous, high-speed process crucial for industries producing flexible electronics, displays, semiconductors, batteries, and photovoltaic devices. Its continuous nature brings substantial advantages in productivity and cost, but also presents major technical challenges. Key among these is the need for microscale accuracy and consistent web tension, as interactions between web material properties and roller mechanics can cause non-uniform stress, deformation, and manufacturing defects. Common issues include surface roughness, linewidth inaccuracies, and defects such as wrinkles or telescoping, often exacerbated by heat exposure during processing.
Traditional Control Approaches and Their Limitations, Scientific Reports
Conventional web tension control methods are divided into feedback-based and model-based categories. Feedback-based systems use sensors to measure tension, but face issues with response delays and measurement errors—limiting their effectiveness at high speeds. Model-based strategies such as model predictive control (MPC) are more accurate, but depend on precise modeling and come with high computational demands. As manufacturing becomes faster and more complex, both approaches struggle to deliver the adaptability and real-time optimization required.
The Rise of AI and Machine Learning in Manufacturing, Scientific Reports
To overcome these constraints, the integration of machine learning (ML) and digital twins is gaining momentum in advanced manufacturing. ML approaches, especially Bayesian optimization, have shown remarkable efficiency in optimizing control parameters with fewer experiments compared to traditional methods. Bayesian optimization is sample-efficient and excels at quickly finding optimal solutions, making it highly suitable for R2R systems where computational cost and adaptation speed are critical.
Digital twins, meanwhile, provide a virtual representation of the physical manufacturing system, enabling continuous data exchange, monitoring, and model updates. This allows for real-time optimization and system adaptation to changing production conditions, setting the stage for fully autonomous manufacturing processes.
The Scientific Reports AI-Driven Digital Twin Solution
The proposed solution combines Bayesian optimization with Gaussian process modeling inside a real-time digital twin (DT) platform. The system architecture features real-time data communication between the physical twin (PT)—the actual R2R hardware—and the digital environment, managed through an OPC UA server and secure file transfer. This setup enables the DT to issue control commands, receive sensor data, and refine its models iteratively.
Experimental Setup
The research team validated the approach using a real-world R2R system designed for flexible and printed electronics. The system consists of three main zones (unwinder, main operation, rewinder) and includes multiple tension control mechanisms. Each zone’s tension is managed using a combination of fixed torque and closed-loop PI (proportional-integral) control, with the main operation zone using phase shifting and advanced averaging to minimize errors.
Optimization Workflow
The digital twin initiates optimization by sampling a range of controller parameters (Kp and Ki), running experiments, and recording the system’s step response—analyzing features like time constant, overshoot, and settling time. A quality score is then calculated as a weighted sum of these features, with the process repeating until optimal control parameters are found.
Bayesian optimization enables rapid convergence by using Gaussian process models to predict the most promising parameter sets. The researchers introduce safety constraints to avoid instability and penalize unsuccessful experiments, ensuring safe exploration during optimization.
Results and Key Findings
The team conducted 100 experiments at a web speed of 50 mm/s, with the digital twin autonomously refining PI gains for optimal performance. The best results were obtained at the 81st experiment, achieving a quality score of 0.193, with Kp = 7.5 and reciprocal Ki = 730. The corresponding control response metrics were:
- Time constant: 0.207 s
- Overshoot: 0.105 kgf
- Settling time: 0.396 s
These results represented improvements in speed and stability compared to initial settings, confirming the efficiency and effectiveness of the AI-driven digital twin method. The DT platform’s real-time operation was supported by low communication latency (about 100 ms per cycle) and scalable computation using advanced Gaussian process techniques.
Scalability and Adaptability
Generalization tests demonstrated the method’s adaptability across different web speeds (25, 50, 75 mm/s) and tension transitions (3 to 5, 3 to 7, and 3 to 9 kgf), consistently converging to effective control parameters. This confirms the robustness and scalability of the approach, positioning it as a promising solution for diverse R2R manufacturing scenarios.
Conclusion
The Scientific Reports study showcases a novel, AI-driven digital twin framework for autonomous web tension control in roll-to-roll manufacturing. Through integration of Bayesian optimization, Gaussian process modeling, and real-time digital-physical communication, the research demonstrates substantial gains in precision, adaptability, and efficiency. As manufacturing industries continue to pursue greater automation and intelligent process control, this work provides a clear path forward for scalable, self-optimizing systems.
More info here – Have a Story? Submit it here
About Scientific Reports
Scientific Reports is a peer-reviewed, open-access journal published by Nature Portfolio, part of Springer Nature. Since its launch in 2011, the journal has become one of the world’s largest and most influential multidisciplinary scientific journals. Scientific Reports publishes high-quality research from across the natural and clinical sciences, covering subjects such as biology, chemistry, physics, earth sciences, medicine, and engineering. The journal’s open-access policy ensures that published research is freely available to the global scientific community and the public.
Manuscripts are assessed for scientific rigor, validity, and ethical standards, rather than perceived impact, enabling a broad spectrum of research to be disseminated. Scientific Reports upholds transparency, rapid publication, and reproducibility as its core values. With a wide international readership and high visibility, the journal serves as a trusted platform for advancing scientific knowledge and fostering collaboration across disciplines.
Featured image source: Youtube
Disclaimer
The information provided in this article is for general informational purposes only and from publicly available sources. While we strive for accuracy, we do not make any representations or warranties, express or implied, regarding the completeness, reliability, or validity of the content. This article does not make any direct claims about specific companies, individuals, or organizations. Any references to reports or external sources are for context and do not imply endorsement or verification of any specific allegations. Readers are encouraged to conduct their own research and seek professional advice before making business decisions. We disclaim any liability for any losses or damages incurred as a result of reliance on the information provided.