New AI Model Enhances Droplet Evolution Prediction for Inkjet Printing.

Predicting the behavior of droplets in material jetting, such as inkjet printing, is critical for maintaining high print quality. However, forecasting these evolutions over long periods is challenging due to the accumulation of errors and the complex interplay of various process variables. A new framework, DiffARFNO (Diffusion-corrected Auto-Regressive Fourier Neural Operator), has been developed to address these difficulties. DiffARFNO operates in two stages. First, an autoregressive Fourier-MIONet acts as a coarse predictor, deployed for long-horizon forecasting. This initial stage provides a general trajectory of droplet evolution. In the second stage, a conditional Denoising Diffusion Implicit Model (DDIM) corrector refines these coarse predictions. This corrector works within a sliding window, iteratively denoising and restoring fine details that might be lost in the initial coarse prediction. By combining the broad predictive power of Fourier-MIONet with the detail-enhancing capabilities of the DDIM corrector, DiffARFNO achieves high-fidelity predictions for extended timeframes, demonstrating superior performance on droplet datasets compared to current models.