Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Aagam Shah, Reimar Weissbach, David A. Griggs, A. John Hart, Elif Ertekin, Sameh Tawfick

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.

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

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.
Paper Structure (12 sections, 3 equations, 12 figures, 1 table)

This paper contains 12 sections, 3 equations, 12 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Experiments and Imaging
  4. Generation of Training Data
  5. U-Net Neural Network
  6. Post-Processing
  7. Results and Discussion
  8. Hyperparameter Optimisation
  9. Model Performance
  10. Potential future directions
  11. Conclusion
  12. Acknowledgements

Figures (12)

  • Figure 1: Schematic of the experimental process flow from single layer melting to cross-sectioning. Many parallel single-tracks are typically melted in a single experiment, and subsequently cross-sectioned, and then imaged separately from the sample. The schematic shows sectioning of only single tracks.
  • Figure 2: Example micrographs of melt track cross-sections showing a diversity in the contrast, material, colour profile, and aspect ratio. The images present artefacts such as scale bars of different types and in various locations, excess powder, and an additional plate.
  • Figure 3: Ideal melt track cross-section showing the extracted geometrical metrics: melt track depth, height, width, wetting angle $\alpha$, wall angle $\beta$. The red line shows the baseline for the substrate surface and cyan line shows the border of the melt pool. The scale bar is 100 $\mu$m.
  • Figure 4: Output from the software that uses morphological geodesic active contours (MGAC) to identify the melt pool boundary. The first figure is the raw image. The second figure is a flood fill of the central region after edge detection. Each of the remaining output figures correspond to a different set of hyperparameters for the MGAC based technique. Within each MGAC Option, a small number of the ballooning iterations is depicted using the colors in the legend at the bottom of the figure.
  • Figure 5: Steps while using connected colour thresholding Kulkarni2012Color to segment an image. (a) The raw image; (b) a schematic of the mouse path taken to select the whole region; (c) a snapshot of the region selected halfway through the mouse path in (b); (d) the complete selection.
  • ...and 7 more figures