Large-area OLED substrate printing path planning method based on multi-head GAT imitation learning to solve partitioned integer programming

In the realm of organic light-emitting diode (OLED) substrate manufacturing, inkjet printing emerges as a highly promising technology. In comparison to the evaporation process, inkjet printing boasts advantages such as simplicity in processing, high material utilization rates, and suitability for various display manufacturing processes. However, in the inkjet manufacturing process, the resolution of the printhead (nozzle per inch, NPI) typically does not align with the pixel resolution of the substrate (pixel per inch, PPI). It is essential to strategize the movement path of the printhead module to minimize the number of printing cycles needed for pattern completion. This presents a challenging multi-objective optimization process. The difficulty intensifies when producing large-area OLEDs, where angular alignment errors in the substrate are magnified, resulting in an exponential increase in the number of pixels involved in the planning process, the number of all pixel pits reaches hundreds of millions of levels. Additionally, the time complexity of the planning problem grows exponentially O(m ⁿ ), rendering it an NP-hard problem. This issue significantly impacts the production efficiency of the manufacturing process. This paper establishes an inkjet integer programming model based on parallel modelling of substrate partitioning (PMIP). PMIP provides partitioning rules and parallel modeling methods specifically for substrate misalignment angles, and proposes imitation learning of the SCIP solver based on a multi-head graph attention network, which is applied to solving the printing planning problem. The model was implemented on a G4.5-size substrate with a resolution of 394 PPI, achieving pixel CF layer printing.