Rapid Process Modeling of the Aerosol Jet Printing Based on Gaussian Process Regression with Latin Hypercube Sampling

• Published in: International journal of precision engineering and manufacturing Vol. 21; no. 1; pp. 127 - 136
• Main Authors: Zhang, Haining, Moon, Seung Ki, Ngo, Teck Hui, Tou, Junjie, Bin Mohamed Yusoff, Mohamed Ashrof
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society for Precision Engineering 2020; Springer Nature B.V
• Subjects: Carrier gases; Engineering; Flow velocity; Gas flow; Gaussian process; Hypercubes; Industrial and Production Engineering; Jet printing; Latin hypercube sampling; Machine learning; Materials Science; Model accuracy; Modelling; Morphology; Process parameters; Regression models; Regular Paper; Sheaths; Three dimensional printing; Gaussian process regression; Process modeling; Aerosol jet process; 3D printing; Machine learning
• ISSN: 2234-7593; 2005-4602
• DOI: 10.1007/s12541-019-00237-3

Aerosol jet printing (AJP) technology is a relatively new 3D printing technology for producing customized microelectronic components due to its high design flexibility and fine feature deposition. However, complex interactions between machine, process parameters and materials will influence line morphology and remain a challenge on modeling effectively. And the system drift which induced by many changing and uncertain factors will affect the printing process significantly. Hence, it is necessary to develop a small data set based machine learning approach to model relationship between the process parameters and the line morphology. In this paper, we propose a rapid process modeling method for AJP process and consider sheath gas flow rate, carrier gas flow rate, stage speed as AJP process parameters, and line width and line roughness as the line morphology. Latin hypercube sampling is adopted to generate experimental points. And, Gaussian process regression (GPR) is used for modeling the AJP process because GPR has the capability of providing the prediction uncertainty in terms of variance. The experimental result shows that the proposed GPR model has competitive modeling accuracy comparing to the other regression models.