A machine learning framework for process optimization in aerosol jet 3D printing

Abstract

As a burgeoning three-dimensional (3D) printing technology, aerosol jet printing (AJP) technique has the characteristics of direct writing and customizing microelectronic components with flexible substrates. Therefore, it has been widely applied to manufacture different electronic devices. Although AJP has unique advantages over traditional methods, the electrical performance of printed electronic devices is significantly reduced because of the inferior printing qualities, such as high overspray, low level of line thickness and high level of edge roughness. Therefore, producing lines with high-controllability and high-aspect ratio is urgent for AJP technology. In this research, a machine learning scheme is developed for process optimization in AJP. In the proposed scheme, a support vector machine is combined with Latin hyper sampling to determine an optimal operating window of AJP, producing conductive lines with better edge definition and reduced overspray. Then, based on the identified 3D operating window, the conflicting relationship between the deposited line width and thickness was revealed based on the developed Gaussian process regression models. Following that, via a non-dominated sorting genetic algorithm, the conflicting printed line morphology was further optimized under dual conflicting targets for maximizing line thickness and customizing line width, which helps to produce high-controllability and high-aspect ratio lines for AJP. The optimization results demonstrated the validity of the proposed approach, which is beneficial to the systemic optimization of the entire printing process.