We are excited to share that our latest research, “Machine Learning-Driven Prediction of Accompanying Droplet Structures Based on Primary Droplet Shape,” has just been published in Physics of Fluids . In this paper, we implement neural networks to infer the presence or absence of secondary droplet structures from the morphology of primary droplets, offering real-time predictive capabilities crucial for optimizing ink-waste and precision in drop-on-demand inkjet printing.
Study Highlights
- We trained our neural network model on a comprehensive dataset combining high-resolution droplet images with a database of piezoelectric dispenser voltage-pulse parameters and geometric descriptors for each droplet.
- The model accurately predicts whether secondary (accompanying) droplets form, based solely on the shape features of the primary droplets.
- By enabling real-time inference of droplet morphology, this approach dramatically reduces ink remnants and optimizes operating regimes in microfluidic dispensers.
- Such predictive tools are vital for applications in printed electronics, bioprinting, and any domain where microfluidic precision and efficiency are paramount.
- This work represents a significant step toward fully automated, data-driven control of droplet formation in inkjet-based microfluidic systems.
Data Availability
- The data that support the findings of this study are openly available in Zenodo at http://doi.org/10.5281/zenodo.13862494.
- Our image‐processing code is available under an open-source license at https://github.com/ARHashemi/MorphoContour.
- The pretrained model files and accompanying usage scripts can be found at https://github.com/DropletDynamics/EllipseAI_PredictDroplets.
Read the full article for a detailed description of our neural network architecture, training methodology, and experimental validation:
🔗 Machine Learning-Driven Prediction of Accompanying Droplet Structures Based on Primary Droplet Shape, Phys. Fluids 37(4), 042019 (2025)