AM-SegNet for additive manufacturing in situ X-ray image segmentation and feature quantification

Investigators: Prof Peter Lee, Prof Chu Lan Alex Leung

Researchers: Dr Wei Li

Collaborators: UKRI Research Complex at Harwell, UCL, ESRF

Synchrotron X-ray imaging has been utilised to detect and quantify the dynamic behaviour of molten pools during the metal additive manufacturing (AM) process. Generally, in situ synchrotron experiments are performed at ultra-high temporal and spatial resolutions, thus generating a large volume of X-ray imaging data, making manual data processing time-consuming and impractical. Thus, it becomes essential to propose an efficient and reliable approach to performing image segmentation and analysis.

In this case study, we developed a novel lightweight neural network, AM-SegNet (Figure 1.a), to perform semantic segmentation and feature quantification on X-ray images collected from various AM synchrotron beamtime experiments. We created a large-scale benchmark database comprising more than 10,000 pixel-labelled X-ray images for model training and validation. Results indicated that AM-SegNet outperformed other CNN-based segmentation models available in literature in terms of accuracy, speed and robustness. The trained AM-SegNet model was used to expedite the quantification of critical features (Figure 1.b) and conduct correlation analysis (Figure 1.c). The accuracy and efficiency of AM-SegNet were further validated in different types of AM, and other advanced manufacturing techniques, making it a step closer to achieving real-time segmentation and quantification of X-ray images in high-speed synchrotron experiments. The proposed method will enable end-users in the manufacturing and imaging domains to accelerate the data processing from collection to analytics, and to provide insights into the process’s governing physics.

This study was supported by MAPP-EPSRC Future Manufacturing Hub under the supervision of Prof. Peter Lee and Dr. Alex Leung from University College London. X-ray imaging results were collected from beamtime experiments funded by MAPP. MAPP PDRAs and PhDs made contributions to the establishment of benchmark database for model training and testing.

Figure 1 Design and application of AM-SegNet for semantic segmentation and feature quantification of in situ X-ray images during additive manufacturing.
Figure 1 Design and application of AM-SegNet for semantic segmentation and feature quantification of in situ X-ray images during additive manufacturing.

Publications:

1. Wei Li, Rubén Lambert-Garcia, Anna C.M. Getley, Kwan Kim, Shishira Bhagavath, Marta Majkut, Alexander Rack, Peter D. Lee, Chu Lun Alex Leung, “AM-SegNet for additive manufacturing in situ X-ray image segmentation and feature quantification”, submitted to Virtual and Physical Prototyping.