Inkjet printing offers significant benefits for additive manufacturing (AM) and printed electronics, such as cost-effectiveness, scalability, non-contact printing, and the flexibility for ad-hoc customization. However, some challenges such as stable jetting states and defined printing zone still needs attention. Data driven modeling such as machine/deep learning (ML/DL) as a predictive methodology has proven to reduce the experimental cost and workload in AM for low-viscosity inks. However, there is an oversight in ML extension to high-viscosity inks due to some inherit challenges such as irregular shape formation, and adhesiveness. Therefore, this study is focused on the prediction of jetting states and defining the printing zone for three-dimensional (3D) inkjet printing of high-viscosity ink. The experimental data is comprised of equipment parameter settings, material properties, and camera-captured features. The jetting behavior is recorded with a high-speed camera and carefully categorized into five classes: no jetting, orifice adhesion, droplet jetting, orifice tail, and beads hanging. A robust and efficient high-viscosity 3D printing U-Net (HV3DP-UNet) model is proposed, that achieved the jetting state and printing zone prediction accuracy of 97.98% and 100% respectively. For the fair comparison, three traditional ML and two more DL models are tested and analyzed in detail. The robustness and efficacy of the proposed model is supplemented with four performance metrics, i.e., accuracy, precision, recall and f1-score. The models’ efficacy has been proved by achieving improved results on the public dataset, the proposed model has achieved overall prediction accuracy of 92.95%. The presented data-driven approach serves as a systematic framework for enhancing quality of inkjet-based 3D printing utilizing high-viscosity ink.
Read full abstract