Parameter optimization of electrohydrodynamic inkjet printing based on numerical simulation and machine learning

Abstract Electrohydrodynamic (EHD) inkjet printing has gained widespread attention in electronics, biomedicine, and materials science for its exceptional resolution and printing versatility. However, the droplet formation process is governed by complex interactions between driving waveform parameters and fluid properties, making traditional trial-and-error optimization inefficient. To address this, a hybrid approach combining numerical simulation, machine learning regression, and genetic algorithm optimization is proposed to achieve precise control of droplet diameter. A multiphysics numerical model is established in COMSOL Multiphysics to simulate the complete cycle of Taylor cone formation, jetting, and droplet deposition under pulsed electric fields. Parametric studies are conducted to investigate the influence of waveform characteristics and fluid properties on droplet size and jetting stability. Based on these simulations, a dataset of 912 samples is constructed for machine learning analysis. Among seven regression models evaluated, the artificial neural network (ANN) shows the best predictive performance and is further integrated with a genetic algorithm to optimize the driving parameters for different target droplet diameters. Experimental validation is performed using a Super Inkjet (SIJ) printing system. The results confirm the effectiveness of the proposed method: the average droplet diameter error ranges from 1.00 μm to 1.89 μm, and 84.21% of the droplets fall within ±5% of the target diameter. This study demonstrates a practical and data-driven framework for enhancing precision and process control in EHD printing.