Measurement-Informed Deep Learning for Real-Time Monitoring of Hydrodynamic Drying Dynamics in Inkjet-Printed Display Pixels via Low-Resolution Imaging

Real-time monitoring of drying dynamics is critical for ensuring high-quality thin-film fabrication in inkjet printing. Here, we present a classification framework for the drying stages of ZnO nanoparticle inks on 200-PPI patterned substrates using low-resolution optical images. By correlating 3D surface profiles with visual textures, we show how the volatility mismatch and viscosity contrast of binary solvents govern hydrodynamic evolution, generating distinct image textures for each stage. Based on this, we define three drying stages for binary mixed-solvent inks and two for single-solvent systems. A lightweight convolutional neural network, trained on grayscale images downsampled to 115 × 40 pixels, achieved a 99.4% test accuracy. Explainable AI analyses, utilizing input gradient saliency and gradient-weighted class activation mapping (Grad-CAM), revealed that the model autonomously learns to focus on the central pixel region-where physical variations are most pronounced-without any spatial supervision. Furthermore, the lightweight model requires only 0.64 ms per image for inference on a GPU, representing a 16.4-fold reduction in theoretical computational cost compared to the baseline. This image-only approach enables reliable, computationally efficient process monitoring, establishing a practical pathway for real-time quality control specifically for the tested ZnO nanoparticle inkjet printing configuration.