Intelligent path planning algorithm system for printed display manufacturing using graph convolutional neural network and reinforcement learning

Abstract

Inkjet printing technology is considered one of the core components of next-generation display technologies for manufacturing organic light-emitting diode (OLED). However, the patterning process for novel display inkjet printing involves diverse characteristics across different dimensions, such as varying printing scales and resolutions. Existing patterning modules using a single planning algorithm for all inkjet printing scenarios often result in long planning times and unstable planning quality. Therefore, a more comprehensive algorithm system is needed to evaluate inkjet planning problems and select the most suitable planning algorithm. This paper proposes a multi-algorithm integrated online patterning intelligence planning system, which includes three patterning algorithms specific to the inkjet display field and an algorithm selection network based on Proximal Policy Optimization (PPO). We first identify the core metrics of the inkjet planning problem as planning time and solution quality, analyzing how different characteristics of the planning problem affect these metrics. We then propose three algorithms suited to different performance needs: an integer programming method based on graph convolutional neural networks, a binary greedy algorithm, and a maximum contiguous interval search algorithm, each corresponding to high overall performance, high solution quality, and short solution time, respectively, to address complex inkjet planning scenarios. Additionally, the PPO-based algorithm selection network refines the features of the inkjet planning problem to achieve intelligent algorithm selection. Finally, we validate the multi-algorithm integrated online patterning intelligence planning system using the self-developed NEJ-PRG4.5 inkjet equipment.