Autonomous Closed-Loop Plastic Forming Process with Real-Time Optimization

The rapid changes in the social environment and the accelerating data society have necessitated a shift in research paradigm in various fields. In this context, there is a need to reconsider how to utilize data generated from materials and processes more effectively. In this area, we have made efforts in different areas, such as exploring multi-faceted analysis methods to extract new insight [1], utilizing multimodal deep learning to predict and optimize material properties by integrating multiple data streams [2], and developing closed-loop autonomous processes to remove human intervention [3].
The forming of plastics into diverse shapes is industrially needed to provide function to a relatively inexpensive material. However, the plastic forming is a complex process which requires the control of viscoelastic fluids melted at high temperatures to produce the desired products. Fluidity, a critical factor in the forming process, is dependent on numerous factors, such as molecular structure, molecular weight distribution, additives, recycled plastic blends, storage and thermal history, and environmental conditions (e.g. atmospheric temperature). Thus, identifying the optimal forming conditions for a given materials is difficult particularly for new materials, which lack significant prior data. To address these issues, we developed a prototype autonomous closed-loop plastic forming process based on an active learning decision-making algorithm. In the forming component of our system, polycarbonate plastic pellets were fed, melted, transported, and extruded through a die (ie. slit) to form a film, then cooled and collected onto a roll. Equipped with real-time in-situ evaluation, consisting of cameras and laser displacement sensors, the film dimensions (width and thickness) could be accurately captured and used as inputs to the autonomous processor as the measurement-adjustment iteration to realize autonomous control of process conditions governing material input, draw rate, and applied heat. The use of an active learning algorithm afforded our system to proceed optimization in the absence of training data, which is unavailable due to the complex relationships between the control factors (material supply rate, applied force, material viscosity) within the plastic forming process. Application of this system towards nine distinct film width-thickness conditions demonstrated the ability of the system to quickly determine the appropriate and stable process conditions (average 11 characterization-adjustment iterations, 19 minutes) and the ability to avoid traps, such as repetitive over-correction. Furthermore, comparison of the achieved film dimensions to the target values showed a high accuracy (R2 = 0.87, 0.90) for film width and thickness, respectively. The proposed method in this study is general and can be extended to most any types of research and manufacturing processes involving continuous forming of materials from synthesis to slurries, pastes, and melts. In this presentation, we will discuss the details of this autonomous process and our vision for the future transformation of R&D.

References:
[1] Muroga et al., Appl. Spectrosc. (2024). DOI:10.1177/00037028241228865. arXiv:2311.15703.
[2] Muroga et al., Adv. Sci., 10, 2302508 (2023). arXiv:2303.16412.
[3] Muroga et al., arXiv:2311.15703 (2023).

Acknowledgements: The part of the work was supported by a project (JPNP16010) commissioned by the New Energy and Industrial Technology Development Organization (NEDO).