Quality assessment and fast geometry prediction in paperboard forming

Paperboard forming is a productive process that meets the increasing demands for recyclability in packaging. However, it remains underutilized in mass production due to open challenges, for instance volatile material behavior and control challenges. Moreover, wrinkles are inevitable during paperboard forming. To produce numerous small wrinkles that do not compromise part quality and can improve its stability, the process must be set up appropriately. Achieving this efficiently requires minimal testing, which is complicated by rapid variations in material properties due to external factors. One solution to improve process design and stabilize a running process is integrating machine learning (ML) models, which represent the forming process, into a stroke-to-stroke control. Therefore, this study presents an approach consisting of Finite Element (FE) simulations to generate data, an ML model trained on them, and quality assessment for the predicted parts with an optimization strategy to determine process parameters. FE simulations were built to represent the forming process and ML models were trained on their results to establish a fast correlation between material properties, press settings, and the resulting part geometry, including wrinkles. An automated quality assessment using Fast Fourier Transform quantifies the quality of the part in terms of wrinkles. Results show that the ML model is integrated into an optimization strategy to propose enhanced process parameters if quality deviations occur. The ML model accurately represents FE simulation results and collaborates with quality assessment and optimization fast enough to ensure high part quality in a forming process with cycles of up to 60 strokes per minute.