Digital twin of a roll-to-roll DTF line with web deviation forecasting. Поліграфія і видавнича справа. Інститут поліграфії та медійних технологій. НУ «Львівська політехніка»

Author(s)	Collection number	Pages	Download abstract	Download full text
Шепіта П. І.	№ 2 (90)	196-209

Summary
References

The article presents the results of developing and modeling a digital twin of a roll-to-roll DTF (Direct-to-Film) printing line with an embedded web deviation forecasting system based on a Physics-Informed LSTM neural network. The aim of the study is to enhance the stability of web movement during printing by integrating physical laws into the machine learning process. The proposed approach combines a mathematical model of web dynamics—taking into account tension forces, damping, and aerodynamic pressure—with a recurrent neural network capable of short-horizon edge deviation prediction. The web-based implementation of the digital twin provides tools for adjusting environmental parameters (airflow speed and direction, web tension, turbulence, temperature, and feed speed) and features an interactive visualization interface with real-time adaptive LSTM retraining capability.

Experimental results demonstrate that when airflow velocity increases to 0.5 m/s, the web edge deviation reaches –12 mm; however, increasing the web tension to 100 N reduces the deviation by half and shortens the stabilization time from 5 s to 2 s. The mean absolute error (MAE) of prediction decreased to 0.3 mm, while the coefficient of determination R2 reached 0.98, confirming the high precision and robustness of the model under dynamic external disturbances. These results highlight the advantages of using the digital twin not only as a production tool for process stabilization but also as an educational and diagnostic module for printing enterprises.

The proposed system aligns with the Industry 4.0 concept, ensuring real-time interaction between the physical process and its virtual replica through machine learning–based predictive control. The developed digital twin serves as a universal platform for optimizing roll-to-roll printing processes, can be scaled to other types of printing lines (offset, screen, packaging), and contributes to material waste reduction, energy efficiency, and automation of print quality control.

Keywords: digital twin, DTF printing, roll-to-roll line, web deviation forecasting, Physics-Informed LSTM, machine learning, web tension, aerodynamic effects, web-based simulation, Industry 4.0.

doi: 10.32403/0554-4866-2025-2-90-196-209

1. The future of sustainable printing: How AI revolutionised the printing and packaging industry. Printing Review. 2024. March–April, с. 4–7.
2. Physics Informed LSTM Network for Flexibility Identification in Evaporative Cooling System / M. Lahariya et al. IEEE Transactions on Industrial Informatics. 2022. P. 1. URL: https://doi.org/10.1109/tii.2022.3173897.
3. Baty H. Modelling Lane–Emden type equations using physics-informed neural networks. Astronomy and Computing. 2023. P. 100734. URL: https://doi.org/10.1016/j.ascom.2023. 100734.
4. Yavari M., Ghahramani N. A., Zardashti R., Karimi J. (2023) Incremental Predictive Control with Input Constraints: Formulation and Cost Function. Journal of Aerospace Science and Technology, Vol. 16, №2.
5. Direct Printing of Ultrathin Block Copolymer Film with Nano-in-Micro Pattern Structures / T. W. Park et al. Advanced Science. 2023. URL: https://doi.org/10.1002/advs.202303412.
6. An overview of first Doppler Weather Radar inducted in the cyclone detection network of India Meteorological Department / P. R. RAO et al. MAUSAM. 2022. Vol. 55, no. 1. P. 155–176. URL: https://doi.org/10.54302/mausam.v55i1.963.
7. Emerging market trends: the cultural designs printed with digital printing technology: an overview of Ajrak design / S. P. Simair et al. Digital Textile Printing. 2023. P. 225–240. URL: https://doi.org/10.1016/b978-0-443-15414-0.00008-x.
8. Yang S. Flight Trajectory Prediction Based on LSTM. Transactions on Computer Science and Intelligent Systems Research. 2024. Vol. 8. P. 81–92. URL: https://doi.org/10.62051/f3hwwa06.
9. Zhang R., Liu Y., Sun H. Physics-informed multi-LSTM networks for metamodeling of nonlinear structures. Computer Methods in Applied Mechanics and Engineering. 2020. Vol. 369. P. 113226. URL: https://doi.org/10.1016/j.cma.2020.113226.
10. Kidie F. M., Ayaliew T. G., Mekonone S. T. Optimizing 3D printing process parameters to improve surface quality and investigate the microstructural characteristics of PLA material. The International Journal of Advanced Manufacturing Technology. 2025. URL: https://doi.org/10.1007/s00170-025-15543-6.
11. Rotor Dynamic Response Prediction Using Physics-informed Multi-LSTM Networks / D. Jiang et al. Aerospace Science and Technology. 2024. P. 109648. URL: https://doi.org/10.1016/ j.ast.2024.109648.
12. Research on Digital Design Platform of Gravure Printing Press / P. Dou et al. Innovative Technologies for Printing and Packaging. Singapore, 2023. P. 306–314. URL: https://doi.org/10.1007/978-981-19-9024-3_40.
13. Simulation analysis and research of printing cylinder of gravure printing press based on Adams / K. Chang et al. International Conference on Mechanical Design and Simulation (MDS 2022), Wuhan, China, 18–20 March 2022 / ed. by D. Shi, G. Wu. 2022. URL: https://doi.org/10.1117/12.2638865.