Welcome to APVIA!
中文
APVIA

Improving roll-to-roll slot-die coating via machine learning

Source：pv magazine

Researchers at the University of Sheffield in the United Kingdom have developed a machine learning framework to improve coating properties in a roll-to-roll slot die coating process.

“To our knowledge, this work is the first machine learning optimisation of roll-to-roll slot die coating, looking at changing fundamental coating parameters to improve coating performance,” Christopher Passmore, corresponding author of the research, told pv magazine.

Slot die coating is used for the precise deposition of slurry liquids on a variety of substrate materials to make thin films. The slurries used in the experiment, according to the researchers, resembled those used to make thin films for solar PV devices, such as perovskite solar cells, as well as thin films for lithium-ion batteries and polymer electrolyte membrane fuel cells.

The team selected Radial Basis Function Neural Network (RBFNN) as the surrogate model,  combined with a Reference Vector Guided Evolutionary Algorithm (RVEA) to assist in the optimization.

Surrogate models are particularly effective when analytical models are unavailable, explained the scientists, further specifying that surrogate optimization is “well-suited” for slot die coating, due to the complexity of the process, and the numerous interacting inputs and outputs.

The model was trained on a small experimental dataset to be able to predict new optimal parameter sets. The slot die coating parameters included in the model were as follows: substrate velocity, pump rate, coating gap, shim thickness and composition of coating solution.

“When these new sets were experimentally tested, they gave substantial improvements in coating properties. We also utilized machine learning predicted trends to highlight the impact of individual parameters on coating properties; something that has historically been challenging for complex coating formulations, such as those used in solar PV, often requiring extensive experience to decipher,” explained Passmore.

The team found that the models predict coating thickness and uniformity with mean absolute errors below 11.5 %.

It noted that shim thickness and substrate velocity were found to have the greatest impact on coating uniformity, while coating gap played a lesser role.  They said that the evolutionary algorithm “identified new operating parameters, leading to improved coating properties.”

Compared to traditional trial and error optimization approaches, the surrogate model-assisted optimization offers increased understanding of slot die coating behavior alongside providing rapid and large improvements in coating properties, stated the researchers.

The group sees the work as an “initial step” toward the “broader integration” of machine learning guidance in a metrology and data-driven approach to optimizing slot-die coating.

It proposed further a list of model enhancements, such as using larger and more diverse training data sets, using other sampling technology to manage multivariate parameters and incorporating the ability to repeat or replicate measurements at selected parameter settings to enable “more accurate assessment” of process and measurement variability, and to improve model reliability.

Noting the potential to include other coating properties, provided that reliable metrology is available, the researchers said, “Moreover, integrating additional parameters, such as substrate pre-treatment, surfactant concentration, and slot die manifold design into the algorithm and employing physics-informed machine learning based on fundamental slot die coating principles may offer deeper insights into complex coating behaviors and improve model generalizability.”

Both industrial and R&D slot-die coating equipment users can benefit from the new methods to optimize coatings and a reliable way to quantify how good the coating is, according to Passmore.

The study is described in “Surrogate-assisted optimization of roll-to-roll slot die coating,” in nature Scientific Reports.

Looking at upcoming research work, Passmore said that the team is investigating the production of large-area colloidal crystals and mesoscale structured thin films, with potential applications in photovoltaics and other optoelectronic devices.