Droplet Characterization through Neural Network Geometric Analysis and Classification for Drop-on-Demand Inkjet Systems
Please login to view abstract download link
The rising importance of ink-jet printing in hi-tech manufacturing highlights the need to understand how operational parameters and ink properties relate to the characteristics of produced droplets [1]. Thus, creating models that closely match real conditions is key to fine-tuning operational settings, and hence, this study aims to delve into these relationships [2]. In the present work, using an experimental set-up Microdrop MD-E-5000 involving a micro-dispenser head, droplets were generated using a range of operation parameters. The droplet generation process involved three pulses characterized by voltage, pulse width, and inter-pulse time, providing images that were posteriorly processed using the OpenCV library. Image recognition facilitated the droplet count as well as the geometric characteristics of the droplets, subsequently compiled into an HDF5 database. Such database is a basis for machine learning training with the aim of obtaining a neural network capable of quickly predicting the outcome of the printing [3,4]. In our work, such neural network was constructed. We show that our model allows to estimate the optimal operation parameters for producing the desired droplet ejection regimes. This initial analysis forms the foundation for an expanded project aiming to improve the model using more robust data, incorporating various physical and operational attributes as well as CFD simulation results based on our in-house computational high-fidelity model [5,6].
