Image Segmentation for Defect Analysis in Laser Powder Bed Fusion: Deep Data Mining of X-Ray Photography from Recent Literature

Zhang, Jiahui1,2; Lyu, Tianyi1; Hua, Yujie1; Shen, Zeren1; Sun, Qiang3; Rong, Yiming2; Zou, Yu1

2022-09-01

10.1007/s40192-022-00272-5

Integrating Materials and Manufacturing Innovation   Impact Factor & Quartile

2193-9764

2193-9772

The in situ X-ray imaging method has attracted significant attention in the metal additive manufacturing community for characterizing keyhole dynamics and defect generation during laser-material interaction processes, particularly for laser powder bed fusion. Due to a high temporal and spatial resolution in this method, a vast volume of data are generated and collected, leading to a challenge for data processing and analysis. In this study, we present an accurate, robust, and powerful image analytical approach that can identify the high-fidelity automated features and extract important information from X-ray images. In total, we train six semantic segmentation models and six object detection models using 628 X-ray images obtained from two recent literature. Our study demonstrates that the U net + MobileNet model is the overall best choice among 12 models to recognize and extract desired regions, in terms of accuracy, time consumption, and dataset sensitivity. Using this model, we have collected and summarized geometric features and dynamic behaviors of the keyholes and generated bubbles. The image segmentation approach may pave the path for unveiling new mechanisms that might not be easily identified using conventional analysis methods in additive manufacturing processes.

Laser powder bed fusion Machine learning Image segmentation Geometric features Dynamic behaviors

SCI ; EI

English

Corresponding

Natural Sciences and Engineering Research Council of Canada (NSERC)[RGPIN-2018-05731] ; Centre for Analytics and Artificial Intelligence Engineering (CARTE)[NFRFE-2019-00603] ; NSERC Alliance Grants-Missions[ALLRP 570708-2021]

Engineering ; Materials Science

Engineering, Manufacturing ; Materials Science, Multidisciplinary

WOS:000849277400001

SPRINGER HEIDELBERG

20223612696940

3D printers ; Additives ; Data handling ; Data mining ; Deep learning ; Defects ; Learning systems ; Object detection ; Semantic Segmentation

Ergonomics and Human Factors Engineering:461.4 ; Data Processing and Image Processing:723.2 ; Artificial Intelligence:723.4 ; Printing Equipment:745.1.1 ; Chemical Agents and Basic Industrial Chemicals:803 ; Materials Science:951

Web of Science

1.Univ Toronto, Dept Mat Sci & Engn, Toronto, ON M5S 3E4, Canada
2.Southern Univ Sci & Technol, Dept Mech & Energy Engn, Shenzhen 518055, Guangdong, Peoples R China
3.Univ Toronto, Dept Stat Sci, Toronto, ON M5S 3G3, Canada

Zhang, Jiahui,Lyu, Tianyi,Hua, Yujie,et al. Image Segmentation for Defect Analysis in Laser Powder Bed Fusion: Deep Data Mining of X-Ray Photography from Recent Literature[J]. Integrating Materials and Manufacturing Innovation,2022,11:418-432.

Zhang, Jiahui.,Lyu, Tianyi.,Hua, Yujie.,Shen, Zeren.,Sun, Qiang.,...&Zou, Yu.(2022).Image Segmentation for Defect Analysis in Laser Powder Bed Fusion: Deep Data Mining of X-Ray Photography from Recent Literature.Integrating Materials and Manufacturing Innovation,11,418-432.

Zhang, Jiahui,et al."Image Segmentation for Defect Analysis in Laser Powder Bed Fusion: Deep Data Mining of X-Ray Photography from Recent Literature".Integrating Materials and Manufacturing Innovation 11(2022):418-432.