AI-Enhanced Casting Porosity Analysis in Dragonfly 3D World

January 12, 2026

Casting quality depends heavily on understanding the internal porosity of metal parts. Hidden voids and shrinkage pores can compromise integrity, lead to early failures, and increase costs - especially in safety-critical applications. While traditional thresholding methods like Otsu segmentation have proven effective for fast evaluations, they sometimes miss fine or low-contrast defects. To overcome these challenges, Dragonfly 3D World introduces AI-driven segmentation tools that take porosity analysis to the next level.

From Thresholding to AI-Based Insight

X-ray computed tomography (CT) is widely used to visualize porosity in cast components. Conventional thresholding techniques such as the Otsu method can successfully segment most gas pores and larger voids. However, these methods may underestimate true porosity by missing narrow crack-like voids, clusters of small pores, or low-contrast shrinkage defects near the voxel limit of the scan.

The AI tools in 3D World address these limitations by learning directly from image data. A model trained on one dataset can identify subtle features such as shrinkage pores or narrow voids that traditional thresholding overlooks, even when applied to entirely new samples.

Proving the Difference

Using high-resolution CT scans from a Comet Yxlon system, the application note demonstrates the improvement AI brings to casting inspection. Below examples confirm that artificial intelligence can reliably detect finer porosity details while maintaining consistency across varying samples and conditions.

Sample 1: AI segmentation identified narrow void regions and shrinkage pores that the Otsu method failed to capture.

Examples of improved segmentation using AI tools (right) compared to thresholding (left), in this case for narrow void regions and some shrinkage porosity regions in particular.
Examples of improved segmentation using AI tools (right) compared to thresholding (left), in this case especially visible for shrinkage porosity (middle).

Sample 2: The same AI model, trained only on the first dataset, accurately segmented a different casting - even in areas affected by imaging artifacts where thresholding made errors.

An example of improved segmentation using AI tools (right) compared to thresholding (left), in this case for a different sample using the model trained on the previous data only.
An example of improved segmentation using AI tools (right) compared to thresholding (left), in this case for a section of the second sample with artifacts (dark area near top) where thresholding makes mistakes and AI can still segment it properly.

Advantages for Casting Engineers

Dragonfly 3D World’s AI segmentation gives engineers a more detailed, quantitative understanding of casting quality. It enables:

  • 3D visualization of pore geometry and size distribution.
  • Improved detection of small or irregular shrinkage features.
  • Automated workflows using macros for faster, repeatable analysis.
Visualization of porosity in second sample, segmented using basic Otsu thresholding
Visualization of porosity in second sample, segmented using AI segmentation based on training data of first sample only.

The result is a more complete assessment of defects and a clearer picture of process performance, helping manufacturers refine parameters, ensure safety, and reduce variability in production.

Why It Matters

Non-destructive CT inspection combined with 3D World’s AI tools allows casting engineers to move beyond simple threshold-based evaluation. The technology not only enhances defect detection but also provides the foundation for data-driven process optimization and long-term quality control.

See the Complete Study

Learn how AI segmentation improves the accuracy and efficiency of casting porosity analysis.

Download the full application note
Back to overview