Abstract:With the increasing adoption of metal additive manufacturing (AM), researchers and practitioners are turning to data-driven approaches to optimise printing conditions. Cross-sectional images of melt tracks provide valuable information for tuning process parameters, developing parameter scaling data, and identifying defects. Here we present an image segmentation neural network that automatically identifies and measures melt track dimensions from a cross-section image. We use a U-Net architecture to train on a data set of 62 pre-labelled images obtained from different labs, machines, and materials coupled with image augmentation. When neural network hyperparameters such as batch size and learning rate are properly tuned, the learned model shows an accuracy for classification of over 99% and an F1 score over 90%. The neural network exhibits robustness when tested on images captured by various users, printed on different machines, and acquired using different microscopes. A post-processing module extracts the height and width of the melt pool, and the wetting angles. We discuss opportunities to improve model performance and avenues for transfer learning, such as extension to other AM processes such as directed energy deposition.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: With the increasing popularity of metal additive manufacturing (AM) technology, especially laser powder bed fusion (LPBF), researchers and practitioners hope to optimize printing conditions through data - driven methods. The bead information in cross - sectional images is of great value for adjusting process parameters, developing parameter scaling data, and identifying defects. However, currently, the analysis of these images mainly relies on manual operations, which are time - consuming and error - prone. Specifically, the paper aims to solve the problem in the following ways: 1. **Automated Segmentation and Analysis**: Use deep - learning techniques, especially the convolutional neural network of the U - Net architecture, to automatically identify and measure the bead size in the cross - sectional images of the bead. 2. **Improve Efficiency and Accuracy**: Reduce the time consumption of manual analysis through automated methods and improve the accuracy of the results, avoiding human errors. 3. **Adapt to Multiple Experimental Conditions**: Ensure that the model can handle images obtained from different laboratories, different machines, and different microscopes, enhancing its robustness and universality. 4. **Extract Key Geometric Features**: Extract key geometric features such as the height, width, and wetting angle of the molten pool from the segmented images, providing quantitative data support for subsequent research. In summary, the goal of this paper is to develop an efficient and accurate automated tool for analyzing the cross - sectional images of LPBF beads, thereby accelerating and optimizing the research and application of metal additive manufacturing processes.

Automated Segmentation and Analysis of Microscopy Images of Laser Powder Bed Fusion Melt Tracks