PSI - Issue 77

Pawel Madejski et al. / Procedia Structural Integrity 77 (2026) 357–364 Author name / Structural Integrity Procedia 00 (2026) 000–000

360

4

To analyze CT images of 3D printed polymer samples, a technology stack, “PorePolymer3D,” was developed by Isyna Izzal Muna, utilizing Python package libraries such as scikit for image segmentation algorithms and OpenCV-python for advanced image processing. The porosity classification framework employs a dual porosity approach that automatically distinguishes between structural porosity (pores originating from the intended infill geometry) and defect porosity (pores arising from manufacturing irregularities). This classification ensures that pore analysis is not biased by the inherent design of additive manufacturing infills, allowing for a targeted assessment of unintended defects. To achieve robust pore detection, the methodology integrates multiple image segmentation techniques, including OTSU thresholding, adaptive thresholding, watershed segmentation, and region growing. OTSU and adaptive thresholding provide reliable intensity-based separations of pore regions from background material, while watershed segmentation helps to resolve overlapping pore structures. Region growing further refines segmentation by expanding pore boundaries based on pixel connectivity and intensity similarity. The combination of these methods enhances accuracy and adaptability across a wide range of image conditions. In this work, the framework incorporates infill patterns such as line and triangular structures. Following 2D pore segmentation, the system performs 3D volume estimation by reconstructing pore distributions from serial image slice analysis using ImageJ software. This enables the calculation of porosity across the entire sample volume, rather than being limited to single cross-sectional observations. For each pattern, a volume sample was sliced into 260 slices along the X-Z section. In this work, a subset of the image sequence was selected from slices 50 to 199 with an increment of 10. The total number of sliced images used for porosity analyses are 15. The CT scan intensity range values are categorized into different definitions of material types and their expected porosity, as shown in Table 1. OTSU thresholding automatically determines the optimal threshold to separate voids from material by minimizing intra-class variance. We apply this to the intensity range of 30 to 120 to identify structural voids. Adaptive thresholding computes local thresholds for different regions and uses Gaussian-weighted neighborhood to determine local threshold. Watershed segmentation treats the image as a topographic surface and finds watershed boundaries. For infill voids, we identify void regions (with intensities ranging from 30 to 120) and use a distance transform to find the void centers as markers. Region growing begins with seed points in infill void regions (intensity range: 30-120) and expands regions based on intensity similarity. This method is particularly effective for connected void structures in infill patterns. Table 1. Material types category with respect to their intensity range

Intensity Range

Material type

Expected porosity

0-30

Background or air Infill voids Solid material

Up to 95% (outside sample) Up to 40% (structural porosity) Up to 5% (material matrix)

30-120 120-255

2.4. Porosity and Density Through various image segmentations that consider structural and defect porosity, an estimated total porosity percentage of a 3D printed sample can be obtained. The estimated porosity will be used to calculate an estimated density (see Equations 1 and 2). To verify the accuracy of the percentage of total porosity, the estimated sample density will be compared with the experimental sample density. (%) = (%) + (%) (1) = = ( 100% − % ) (2)