Physics-informed machine learning for optimizing the coating conditions of blade coating

Abstract

Large-scale liquid coating has been used in various industrial fields, such as the fabrication of functional panels or surfaces. Blade coating with continuous liquid supply is a cost-effective method for large-scale coatings. To coat a liquid without defects, it is essential to maintain the coating bead trapped under the blade stably. However, numerous experiments are required to obtain the optimal conditions. We developed a novel strategy for acquiring coating conditions using physics-informed neural networks (PINNs) to avoid this laborious effort. Although standard neural networks (NNs) predict the coating performance directly from the operating parameters, PINNs predict parameters related to the state of coating bead to enhance its predictive performance. Our results revealed that the PINNs performed better than the standard NNs. Furthermore, we derived a damped harmonic oscillator model that provides physical insight into the correlation between the coating performance and coating conditions. Finally, a parametric study was performed using the PINN-based model to determine the optimal coating conditional zones, and we experimentally demonstrated the stable coating with the optimized operating parameters.