잉크젯 프린팅에서 기계학습을 활용한 프린팅 조건 최적화

Abstract

Elimination satellite drops in Drop-on-demand inkjet is a challengeable problem because it has a complex relationship with viscosity, surface tension, and condition of a waveform. We propose a method to optimize a waveform about unknown fluid using machine learning. We observe drop formation and drop velocity in random waveform each fluid and design the architectures using Jettability Number and information of waveform as inputs and drop formation and drop velocity as outputs. Prediction results are used to find an optimal condition of a waveform. Neural Network(NN) has more significantly improved performance than other machine learning model. We have found optimal hyper-parameters in NN to obtain superior performance. NN achieves approximately 88.0% accuracy of prediction drop formation and 5.62 Root Mean Square Error(RMSE) of prediction drop velocity. This method will decrease consumption of the inks and efforts to find the condition of a waveform makes a single drop.