Machine learning-driven prediction of accompanying droplet structures based on primary droplet shape

New applications of drop-on-demand inkjet printing, including the printing of electronic devices and biological structures, have accelerated the need to improve the efficiency and precision of microfluidic dispensers. Therefore, it is essential to identify the optimal operating regimes of inkjet printers to control droplet characteristics, particularly to reduce remnants, which is crucial when trying to optimize ink waste. In the era of big data and the pursuit of systems that enable accurate real-time decision-making, data-driven models have become an essential tool. In this context, this work focuses on implementing neural networks to predict morphological characteristics of droplets from the operating parameters of a piezoelectric dispenser. The developed model allows us to infer the presence or absence of accompanying structures, by identifying the morphology of the primary droplets. To carry out this study, a set of droplet images was used together with a database that contains information on the parameters of the dispenser's voltage pulses used in their production, along with geometric data that characterizes each droplet present in the images.