Process mapping of additively-manufactured metallic wicks through surrogate modeling

No Thumbnail Available
Borumand, Mohammad
Borujeni, Sima E.
Nannapaneni, Saideep
Ausherman, Moriah
Madiraddy, Guru
Sealy, Michael
Hwang, Gisuk
Issue Date
Conference paper
Additive manufacturing , Bayesian optimization , Laser powder bed fusion , Data classification , 3D printed wick , Porous materials , Gaussian , Support vector machine , Random forest
Research Projects
Organizational Units
Journal Issue
Borumand, M, Esfandiarpour Borujeni, S, Nannapaneni, S, Ausherman, M, Madiraddy, G, Sealy, M, & Hwang, G. "Process Mapping of Additively-Manufactured Metallic Wicks Through Surrogate Modeling." Proceedings of the ASME 2021 International Mechanical Engineering Congress and Exposition. Volume 2A: Advanced Manufacturing. Virtual, Online. November 1–5, 2021. V02AT02A013. ASME.

Tailored wick structures are essential to develop efficient two-phase thermal management systems in various engineering applications, however, manufacturing a geometrically-complex wick is challenging using conventional manufacturing processes due to limited manufacturability and poor cost effectiveness. Additive manufacturing is an ideal alternative, however, the state-of-the-art metal three-dimensional printers have poor manufacturability when depositing pre-designed porous wicks with pore sizes below 100 μm. In this paper, a powder bed fusion 3D printer (Matsuura Lumex Avance-25) was employed to fabricate metallic wicks through partial sintering for pore sizes below 100 μm with data-driven control of process parameters. Hatch spacing and scan speed were selected as the two main AM process parameters to adjust. Due to the unavailability of process maps between the process parameters and properties of printed metallic wick structures, different surrogate-based models were employed to identify the combinations of the two process parameters that result in improved manufacturability of wick structures. Since the generation of training points for surrogate model training through experimentation is expensive and time-consuming, Bayesian optimization was used for sequential and intelligent selection of training points that provide maximum information gain regarding the relationships between the process parameters and the manufacturability of a 3D printed wick structure. The relationship between the required number of training points and model prediction accuracy was investigated. The AM parameters’ ranges were discretized using six values of hatch spacing and seven values of scan speed, which resulted in a total of 42 combinations across the two parameters. Preliminary results conclude that 80% prediction accuracy is achievable with approximately forty training points (only 10% of total combinations). This study provides insights into the selection of optimal process parameters for the desired additively-manufactured wick structure performance.

Table of Contents
Click on the DOI link to view this conference paper (may not be free).
Book Title
ASME 2021 International Mechanical Engineering Congress and Exposition;2021
PubMed ID