| dc.description.abstract | Abstract. Additive manufacturing (AM) techniques, such as inkjet printing (IJP) and aerosol jet printing (AJP), enable the fabrication of high-resolution, customizable printed electronic (PE) devices. However, postprinting sintering is required to establish and enhance their electrical conductivity. Intense pulsed light (IPL) sintering, a photonic-based process, efficiently produces uniform, dense conductive AM-printed traces. Optimizing IPL parameters (e.g., number of pulses, irradiation duration, and pulse interval) is crucial for achieving high conductivity but remains challenging due to material costs and fabrication time, resulting in limited tabular experimental data. This study enhances the regression attention-generative adversarial network (RA-GAN) by integrating a Gaussian process (GP) model to capture input–output relationships during data augmentation in limited tabular data scenarios. The proposed approach leverages GP predictions, including mean response estimation and uncertainty quantification, to train RA-GAN and generate synthetic samples that maintain the relationships between independent (IPL process parameters) and dependent (PE conductivity) variables. Real and synthetic data are combined for modeling and evaluation using three well-established regression models—lasso, GP, and deep GP (DGP). A case study on aerosol jet-printed sensor pads demonstrates that the proposed GP-enhanced RA-GAN outperforms alternative tabular data augmentation methods. The general nature of our approach suggests that it can be applied to tuning other manufacturing processes with limited experimental data. | |
