Comparison and Identification of Optimal Machine Learning Model for Rapid Process Modeling of Aerosol Jet Printing Technology

Aerosol Jet Printing (AJP) offers high resolution and flexible working distances, making it a promising technology for the customization of complex electronic devices. However, devices fabricated through the AJP process often suffer from reduced electrical performance due to limited control over printed line width, which constrains its applicability in advanced electronic manufacturing. Consequently, achieving high precision in line width control is of paramount importance for optimizing AJP technology. In this research, a machine learning framework is proposed to enable rapid modeling of printed line width. The framework considers sheath gas flow rate, carrier gas flow rate, and print speed as input variables, with line width as the target output. Three representative machine learning algorithms – support vector regression, artificial neural networks, and XGBoost – were employed to develop predictive models. The modeling performance of these algorithms was systematically compared using four conventional evaluation metrics. Ultimately, the optimal machine learning model identified through this process was selected for the rapid modeling of printed line width in the AJP process.