Surrogate-assisted optimization of roll-to-roll slot die coating.

Roll-to-roll slot die coating is a key wet processing technique, where achieving a specific thickness with minimal variability is crucial. However, the numerous input parameters make optimization complex. Despite its advanced applications, computer-aided optimization remains underutilized, leaving potential performance improvements untapped. Due to the lack of accurate first-principle models, machine learning offers a promising approach. This study employs Radial Basis Function Neural Networks as surrogate models trained on experimental data to optimize roll-to-roll slot die coating. These models predict coating thickness and uniformity with mean absolute errors below 11.5 %. Key process parameters are identified, with shim thickness and substrate velocity having the greatest impact on coating uniformity, while coating gap played a lesser role. An evolutionary optimization approach identified new operating parameters, leading to improved coating properties. Experimentally, these optimized conditions achieved the five lowest recorded uniformity values and increased the hyper-volume fraction from 0.68 to 0.84. Some prediction inconsistencies were observed, likely due to the high sensitivity of lab-scale equipment, which is expected to improve at an industrial scale. This work paves the way for wider adoption of machine learning and accurate metrology in slot die coating.