Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attenuation-adjusted deep learning of pore defects in 2D radiographs of additive manufacturing powders

Andreas Bjerregaard, David Schumacher, Jon Sporring

TL;DR

The study tackles high-throughput porosity analysis in additive manufacturing by using single 2D radiographs of feedstock powders. It combines physics-based attenuation modeling with four segmentation pipelines, including a Combined UNet approach trained on attenuation-subtracted inputs; this yields an $11.4$ percentage-point improvement over a baseline UNet and achieves a best F1-score of $0.872$ with an AUC-ROC of $0.977$, while runtimes range from $0.014s$ to $0.291s$ per particle. The approach leverages synthetic radiographs generated by aRTist and a distance-to-boundary attenuation map to pretrain and iteratively refine pore predictions, enabling scalable, inline porosity analysis. The work demonstrates material-invariant segmentation potential and lays groundwork for higher-throughput quality control in AM powder feeds, with future directions including more realistic particle shapes and multi-view radiography.

Abstract

The presence of gas pores in metal feedstock powder for additive manufacturing greatly affects the final AM product. Since current porosity analysis often involves lengthy X-ray computed tomography (XCT) scans with a full rotation around the sample, motivation exists to explore methods that allow for high throughput -- possibly enabling in-line porosity analysis during manufacturing. Through labelling pore pixels on single 2D radiographs of powders, this work seeks to simulate such future efficient setups. High segmentation accuracy is achieved by combining a model of X-ray attenuation through particles with a variant of the widely applied UNet architecture; notably, F1-score increases by $11.4\%$ compared to the baseline UNet. The proposed pore segmentation is enabled by: 1) pretraining on synthetic data, 2) making tight particle cutouts, and 3) subtracting an ideal particle without pores generated from a distance map inspired by Lambert-Beers law. This paper explores four image processing methods, where the fastest (yet still unoptimized) segments a particle in mean $0.014s$ time with F1-score $0.78$, and the most accurate in $0.291s$ with F1-score $0.87$. Due to their scalable nature, these strategies can be involved in making high throughput porosity analysis of metal feedstock powder for additive manufacturing.

Attenuation-adjusted deep learning of pore defects in 2D radiographs of additive manufacturing powders

TL;DR

The study tackles high-throughput porosity analysis in additive manufacturing by using single 2D radiographs of feedstock powders. It combines physics-based attenuation modeling with four segmentation pipelines, including a Combined UNet approach trained on attenuation-subtracted inputs; this yields an percentage-point improvement over a baseline UNet and achieves a best F1-score of with an AUC-ROC of , while runtimes range from to per particle. The approach leverages synthetic radiographs generated by aRTist and a distance-to-boundary attenuation map to pretrain and iteratively refine pore predictions, enabling scalable, inline porosity analysis. The work demonstrates material-invariant segmentation potential and lays groundwork for higher-throughput quality control in AM powder feeds, with future directions including more realistic particle shapes and multi-view radiography.

Abstract

The presence of gas pores in metal feedstock powder for additive manufacturing greatly affects the final AM product. Since current porosity analysis often involves lengthy X-ray computed tomography (XCT) scans with a full rotation around the sample, motivation exists to explore methods that allow for high throughput -- possibly enabling in-line porosity analysis during manufacturing. Through labelling pore pixels on single 2D radiographs of powders, this work seeks to simulate such future efficient setups. High segmentation accuracy is achieved by combining a model of X-ray attenuation through particles with a variant of the widely applied UNet architecture; notably, F1-score increases by compared to the baseline UNet. The proposed pore segmentation is enabled by: 1) pretraining on synthetic data, 2) making tight particle cutouts, and 3) subtracting an ideal particle without pores generated from a distance map inspired by Lambert-Beers law. This paper explores four image processing methods, where the fastest (yet still unoptimized) segments a particle in mean time with F1-score , and the most accurate in with F1-score . Due to their scalable nature, these strategies can be involved in making high throughput porosity analysis of metal feedstock powder for additive manufacturing.
Paper Structure (17 sections, 3 equations, 10 figures, 2 tables)

This paper contains 17 sections, 3 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. X-ray computed tomography data
  4. Simulated X-ray computed tomography data
  5. Preprocessing, labelling and data split
  6. Highlighting pores by distance to the boundary
  7. Models for pore segmentation
  8. The local threshold model
  9. The attenuation-adjusted threshold model
  10. The UNet model
  11. The combined model
  12. Results and discussions
  13. Conclusions
  14. Data availability
  15. Large render of the 3D powder particle volumes; hue encodes equivalent pore diameter
  16. ...and 2 more sections

Figures (10)

  • Figure 1: A single, stitched 2D projection (\ref{['fig:stitch']}) of the 801 projections used for 3D reconstruction, and (\ref{['fig:helper']}) the same view over a 3D rendering of the segmented volume; hue encodes the equivalent diameter of pores. Appendix \ref{['app:render']} shows the full-size image.
  • Figure 2: Particles from a single projection. (\ref{['fig:cutout3']}) The normalized original; (\ref{['fig:cutout4']}) the cutout with a faint single particle mask outline in white; (\ref{['fig:real_attenuation']}) the intensity profile of a preprocessed particle indicated by the blue line in (\ref{['fig:cutout3']}); (\ref{['fig:dia_hist']}) a normalized histogram of the diameter of the particles observed fitted with a log-normal distribution; (\ref{['fig:simulated']}) a simulated radiograph example.
  • Figure 3: Cutout subtraction and pores for the particle shown in Figure \ref{['fig:cutout4']}. (\ref{['fig:edt']}) The signed distance field; (\ref{['fig:dist_original']}) a scatter plot of observed intensities vs distance to boundary for distances greater than 0; (\ref{['fig:edt2']}) the distance field overlaid with the ground-truth pores; (\ref{['fig:sdfmedian']}) similar scatter but without assumed pore pixels, with aggregated points in red and a fit on these to Eq. \ref{['eq:fit']} in green; (\ref{['fig:cutoutsubtracted']}) the image of the particle with the resulting attenuation model subtracted.
  • Figure 4: (\ref{['fig:local_thr']}) The result of local thresholding; (\ref{['fig:local_thr_post']}) the denoised image; (\ref{['fig:gridsearch']}) a gridsearch over the validation set showing F1-score plotted versus hyperparameter values.
  • Figure 5: Evolution of training and validation loss for (\ref{['fig:pretrained_real_history_raw']}) UNet, and (\ref{['fig:pretrained_real_history_sdf']}) combined model. Training is done with first 7 epochs of artificial aRTist data, and then 25 epochs of real data. After augmentation, the training set sizes are 200 and 24 images, respectively. The jump in loss value is due to swapping to real data --- and despite the large peak, pretraining was found useful for convergence and for decreasing the training time.
  • ...and 5 more figures