The design of an inkjet drive waveform using machine learning

Abstract

A drive waveform, which needs to be optimized with ink's fluid properties, is critical to reliable inkjet printing. A generally adopted rule of thumb for its design is mostly dependent on time-consuming and repetitive manual manipulation of its parameters. This work presents a closed-loop machine learning approach to designing an optimal drive waveform for satellite-free inkjet printing at a target velocity. Each of the representative 11 model inks with different fluid properties was ink-jetted with 1100 distinct waveform designs. The high-speed images of their jetting behaviors were acquired and the big datasets of the resulting drop formation and velocity were extracted from the jetting images. Five machine learning models were examined and compared to predict the characteristics of jetting behavior. Among a variety of machine learning models, Multi-layer Perceptron affords the highest prediction accuracy. A closed-loop prediction algorithm that determined the optimal set of waveform parameters for satellite-free drop formation at a target velocity and employed the most superior learning model was established. The proposed method was confirmed through the printing of an unknown model ink with a recommended waveform.