Multi-Parameter Prediction of Inkjet Droplet Characteristics Based on Dual-Pulse Waveform and Ensemble Learning

Inkjet printing technology has emerged as an efficient solution for large-scale manufacturing of electronic devices, sensors, and light-emitting diodes (LEDs) due to its ability to directly deposit functional materials onto flexible substrates according to predetermined patterns, while offering advantages of simple processing, low cost, and high adaptability. In the field of inkjet printing, piezoelectric waveform parameters directly influence droplet characteristics and quality, which in turn affect the final print quality. Therefore, accurate prediction of droplet parameters is crucial for optimizing printing quality. This research addresses the challenge of droplet parameter prediction in piezoelectric inkjet printing by proposing a multi-parameter collaborative prediction method for jetting performance based on dual-pulse waveforms. By constructing a six-dimensional parameter space including primary/secondary pulse voltages, dwell times, and interval times, and employing Sobol sequences to generate initial samples alongside an adaptive sampling strategy based on global sensitivity analysis, we innovatively designed an ensemble learning model based on stacking strategy to predict droplet parameters. Experimental results demonstrate that the adaptive sampling strategy based on global sensitivity analysis significantly enhances data acquisition efficiency (1408 valid datasets) and greatly increases data density in sensitive regions. The heterogeneous ensemble model achieves precise mapping between waveform parameters and droplet velocity (R²=0.91), volume (R²=0.94), and quality score (R²=0.82), enabling accurate prediction of droplet velocity, volume, and quality score.