Multinozzle Droplet Volume Distribution Control in Inkjet Printing Based on Multiagent Soft Actor–Critic Network
Xiao Yue, Jiankui Chen, Hua Yang, Xin Li, Jiacong Xiong, Zhouping Yin
Abstract
Inkjet printing has become an essential technology for manufacturing large-area, flexible organic light-emitting diode (OLED) displays because of its low cost, high efficiency, and flexible manufacturing characteristics. Printing large-area OLED displays requires thousands of nozzles to jet droplets, and the volume of droplets varies from nozzle to nozzle. However, the volume distribution of multinozzle droplets affects the accuracy and consistency of pixel film thickness. Therefore, controlling multinozzle droplet volume distribution has become a crucial challenge for printing large-area displays. This article proposes a multiagent deep reinforcement learning algorithm to extract the strategy from industrial data for the distribution control of multinozzle droplet volumes. First, we construct an offline policy experience dataset using historical state data and update it using online regulation data and new detection state data. Then, based on this dataset, we train the multiagent soft actor–critic networks offline to obtain the control policy. Finally, based on visual feedback information, the extraction strategy is used to achieve online regulation. Experiments were conducted in an industrial environment, and the mean volume of multinozzle droplets was controlled within <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\pm 4\%$</tex-math></inline-formula> of the target volume, with the standard deviation reduced to 0.07.