Virtual Surface Morphology Generation of Ti-6Al-4V Directed Energy Deposition via Conditional Generative Adversarial Network

Abstract
The core challenge in directed energy deposition is to obtain high surface quality through process optimisation, which directly affects the mechanical properties of fabricated parts. However, for expensive materials like Ti-6Al-4V, the cost and time required to optimise process parameters can be excessive in inducing good surface quality. To mitigate these challenges, we propose a novel method with artificial intelligence to generate virtual surface morphology of Ti-6Al-4V parts by given process parameters. A high-resolution surface morphology image generation system has been developed by optimising conditional generative adversarial networks. The developed virtual surface matches experimental cases well with an Fréchet inception distance score of 174, in the range of accurate matching. Microstructural analysis with parts fabricated with artificial intelligence guidance exhibited less textured microstructural behaviour on the surface which reduces the anisotropy in the columnar structure. This artificial intelligence guidance of virtual surface morphology can help to obtain high-quality parts cost-effectively.

A recent study, jointly led by Professor Im Doo Jung in the Department of Mechanical Engineering at UNIST and Professor Hyokyung Sung from Kookmin University proposed a novel method with artificial intelligence (AI) to generate virtual surface morphology of Ti-6Al-4V parts by given process parameters.

Figure 1. Overview of (a) surface morphology prediction via AI and (b) the surface roughness prediction. A visualisation of the expected virtual surface image suitable for the user input is generated by CGAN.

In this study, the research team developed the CGAN-assisted surface morphology prediction method for Ti-6Al-4V DED parts by optimisation of neural network structures. The experimental results and predictions illustrate the effectiveness of the developed CGAN model for choosing the optimal process conditions for manufacturing metal parts through AM, noted the research team.

According to the research team, with the help of the developed AI process with a high FID score, the virtual surface morphology images of the Ti-6Al-4V DED part were successfully generated from identical process conditions.

Their findings also revealed that the virtual surface of DED parts matches the printed metal surface well in terms of appearance and SEM analysis as well as quantitative study in microstructural analysis near the surface. Furthermore, the CGAN-assisted quick virtual surface generation and its corresponding Ti-6Al-4V DED part had an improved smooth surface with less lack of fusion or surface defects.

“Diverse range of DED process parameters can be quickly checked for their corresponding surface morphology with high accuracy by the developed AI,” noted the research team. “The developed methodology using CGAN can also be used for further studies with side surface, edge, or curved surface morphology.”

This study has been carried out in collaboration with Gyeongsang National University, Gyeongsang National University, and Carnegie Mellon University and supported by the National Research Foundation of Korea (NRF). Their findings have been published in the January 2023 issue of Virtual and Physical Prototyping (IF 10.962, world JCR ranking top 5%).

Journal Reference
Taekyeong Kim, Jung Gi Kim, Sangeun Park, et al., “Virtual surface morphology generation of Ti-6Al-4V directed energy deposition via conditional generative adversarial network,” Virtual Phys. Prototyp., (2023).