A knowledge transfer framework to support rapid process modeling in aerosol jet printing

Abstract

Aerosol jet printing (AJP) technology recently gained considerable attention in an electronic manufacturing industry due to its ability to fabricate parts with fine resolution and high flexibility. However, morphology control has been identified as the main limitation of AJP process, which drastically affects the electrical performance of printed components. Even though previous researches have made significant efforts in process modeling to improve the controllability of the the printed line morphology, the modeling process is still inefficient under modified operating conditions due to the repeated experiments. In this paper, a knowledge transfer framework is proposed for efficient modeling of the AJP process under varied operating conditions. The proposed framework consists of three critical steps for rapid process modeling of AJP. First, a sufficient source domain dataset at a certain operating condition is collected to develop a source model based on Gaussian process regression. Then, the representative experimental points are selected from the source domain to construct a target dataset under different operating conditions. Finally, classical knowledge transfer approaches are adopted to extract the built-in knowledge from the source model; thus, a new process model can be developed efficiently by the transferred knowledge and the representative dataset from the target domain. The validity of the proposed framework for the rapid process modeling of AJP is investigated by case study, and the limitations of the classical knowledge transfer approaches adopted in AJP are also analyzed systematically. The proposed framework is developed based on the principles of knowledge discovery, which is different from traditional process modeling approaches in AJP. Therefore, the modeling process is more systematic and cost-efficient, which will be helpful to improve the controllability of the line morphology. Additionally, due to its data-driven based characteristics, the proposed framework can be applied to other additive manufacturing technologies for process modeling researches.