Optimizing Thermal Pressing of Airlaids with Machine Learning

Airlaying is a promising alternative to conventional papermaking that does not require extensive drying and can potentially decrease the energy consumption of the forest industry. The key limitation of airlaids is weak fiber bonding, which results in low strength. Strength can be improved with thermal pressing, which involves multiple process parameters whose relationships with strength are not known. Here, we addressed this problem by combining deterministic linear models and probabilistic machine learning to improve airlaid properties by optimizing the conditions in thermal pressing. Our approach starts with a fractional factorial design as the initial sampling strategy to quantify the independent and interpretable variable effects and their interactions. We show how these resource-efficient designs can be easily complemented with a few additional experiments to identify more complicated behavior using a formal statistical test. We then identified three main challenges in optimizing the pressing conditions for our airlaids and tackled them with Bayesian optimization. Bayesian optimization improved the mechanical and physical properties of our airlaids, which showed tensile performance comparable to or 10% higher than traditional wet-laid paper, although bulk was still 30% lower than wet-laid paper. Our results are important, as they suggest that the middle layer of cardboard could potentially be replaced with a thermally pressed airlaid.