Intelligent Waveform Control for Optimizing the Jetting Behavior by Reinforcement Learning

Jetting behavior plays a critical role in the quality of inkjet printing patterning. The ideal jetting behavior is achieved by optimally designing the waveform to match the ink characteristics. Even after the optimal waveform has been found, the jetting behavior must be optimally maintained during printing. This study proposes the strategy of not only optimal waveform design, but also adaptive waveform control to improve the jetting reliability through the machine learning technique. We built the jetting prediction model of the complex ink, which has the complicated viscoelastic properties, based on supervised learning. The jetting image data at different waveforms was collected from the drop watching system. The learning model predicts the drop velocity and jetting morphology from the rheological parameters and the waveform. The prediction model shows reasonable accuracy of the characteristics of the jetting behavior. Reinforcement learning was used to build the algorithm to design the optimal waveform that generates a single drop at 3 ms^-1 using the prediction model. The trained agent successfully recommends the optimal waveform from a random initial waveform within 30 steps. We transfer the trained agent to the drop watching system for adaptive waveform control. The trained agent automatically manipulated the waveform to achieve the optimal jetting behavior, even though the jetting temperature is increased.