Robust Hole Detection in STL Meshes

The digital representation of three-dimensional objects has undergone a paradigm shift from the precise, analytically defined Boundary Representation (B-Rep) models used in CAD kernels to the discrete, faceted approximations characteristic of the STereoLithography (STL) format.1 This transition is frequently achieved through a process of tessellation, where the continuous surfaces defined in formats such as STEP are discretized into a collection of oriented triangular facets.3 While this conversion facilitates universal data exchange and provides the geometric foundation for additive manufacturing and finite element analysis (FEA), it inherently introduces a significant loss of topological and parametric information.1 A fundamental challenge arising from this information loss is the robust identification of functional geometric features, specifically holes, within the resulting mesh.1 Engineers and software developers frequently encounter high rates of false positives—where mesh defects, sampling artifacts, or tiny gaps are incorrectly identified as intentional openings.6 Resolving these inaccuracies requires a comprehensive understanding of discrete differential geometry, algebraic topology, and statistical model fitting.7

The Mathematical Nature of the STEP-to-STL Conversion

The conversion from a STEP file to an STL mesh is not a simple translation but a process of geometric reconstruction and approximation.1 In a STEP model, a circular hole is defined as a cylindrical analytical surface with a precise axis, radius, and boundary curves.1 When this surface is exported to STL, it is approximated by flat triangles whose vertices are constrained to lie on the original surface.10 This discretization is governed by two primary parameters: chordal tolerance and angular tolerance.10

Chordal error, or chord height deviation, represents the maximum distance between the actual mathematical surface and the flat plane of the approximating triangle facet.11 For a circular hole of radius , the vertices of the approximating polygon lie on the circumference, which necessarily places the faces of the triangles on the interior of the circle, effectively cutting chords across the circular cross-section.11 This reduces the effective inner diameter of the hole, typically by approximately 0.1 mm in standard engineering exports.14 If the chordal tolerance is set too high, the result is a visibly faceted surface where the curvature is no longer smooth, creating jagged boundaries that trigger false positives in traditional edge-tracking algorithms.10

Parameter	Mathematical Impact	Engineering Consequence
Chordal Tolerance	Controls maximum distance from mesh facet to true surface	Determines the smoothness of curved features; affects ID accuracy
Angular Tolerance	Limits the angle between adjacent facet normals	Controls the density of triangles in high-curvature regions
Edge Length Limit	Constrains the maximum side length of any single triangle	Prevents “needle” triangles but increases file size and memory load
Chordal Error Ratio	Ratio of chord height to segment length	Ensures consistent refinement across features of varying scales

10

The loss of parametric intent during this process means that an STL file contains no modeling history; cylinders are no longer recognized as cylinders, and fillets are no longer recognized as fillets.1 The “hole” is transformed from a functional feature into a collection of unstructured triangles.1 This lack of explicit topology means that any detection tool must reconstruct the designer’s intent from the discrete geometry alone, a task that is inherently ill-posed due to the potential for noise and sampling errors.17

Topological Foundations of Hole Discovery

The most fundamental level of hole detection is topological, focusing on the connectivity and manifoldness of the mesh.18 A triangular mesh is mathematically defined as a pair , where is a set of vertex positions in and is a simplicial complex defining the connectivity of the vertices into edges and faces.20 In a valid, closed two-manifold mesh, every edge is shared by exactly two triangles, and the neighborhood of every vertex is homeomorphic to a disk.18

Boundary Cycles and the Euler Characteristic

A topological “hole” in a mesh is defined by the presence of “boundary edges,” which are edges connected to only one triangle.18 A hole is formally identified as a closed loop of these boundary edges.18 The global topology of the mesh is described by the Euler characteristic :

where are the counts of vertices, edges, and faces, is the genus, and is the number of boundary components.24 For a simple watertight solid, , , and .24 The presence of functional holes through the solid increases the genus . For example, a block with a single through-hole is topologically equivalent to a torus, resulting in and .24 Conversely, mesh defects—such as missing triangles that do not pass through the entire volume—increase the number of boundary components , thereby decreasing .24

A common source of false positives in topological detection is the “singular vertex,” which has more than two incident boundary edges.19 These occur frequently in poorly converted models where multiple parts meet at a single point or where triangulation artifacts create “bow-tie” configurations.30 Robust detection must utilize an edge-manifold approach that traverses all boundaries while guaranteeing the existence and uniqueness of the cycle associated with each half-edge.30

Data Structures for Efficient Traversal

Efficient hole identification requires advanced data structures that facilitate rapid adjacency queries. The half-edge data structure is widely considered the industry standard.7 It decomposes every edge into two directed “half-edges” pointing in opposite directions.8 Each half-edge stores pointers to its origin vertex, its incident face, its “next” half-edge in the face loop, and its “twin” (or opposite) half-edge.7

In this framework, finding a hole becomes a search for half-edges that lack a twin.8 Once a seed boundary edge is located, the algorithm can trace the entire boundary loop in time, where is the number of edges in the loop, simply by following the connectivity pointers.19 However, simply finding a loop is insufficient for a CAD tool; the tool must classify whether the loop represents a functional opening or a “gap” between surface patches that should have been connected.32

Classification and Taxonomy of Hole Types

Academic research distinguishes between “geometric holes” (intentional features) and “topological defects” (unintended errors).32 To minimize false positives, a functioning tool must classify every detected boundary cycle into a specific geometric category.32

Topological Connectivity Categories

Holes are first divided into two primary categories based on their intersection properties:

1. Single Holes: A unique, closed loop of edges with no self-intersections or shared vertices.32
2. Continuous Holes: Complex missing areas where multiple single holes are connected by one or more intersection points.32 These often arise in areas of high geometric complexity where the triangulation algorithm fails to reconcile adjacent surfaces.32

Continuous holes must be segmented into manageable single units using a “decomposition method of multiply connected domains”.32 This process identifies intersection points and uses a projection-based approach to determine the relative inner and outer relations between separate loops.32

The Five-Type Geometric Taxonomy

Once segmented, single holes are classified using specific mathematical thresholds to identify their origin and appropriate treatment:

• Simple Holes: Standard closed loops found on relatively smooth surfaces. These are typically the primary targets for feature recognition.32
• Crack Holes: Long, narrow openings. They are identified by a very high length-to-width ratio and a small cumulative included angle between consecutive edges.32 Mathematically, if the cumulative angle deviation along the loop is below a set threshold, it is classified as a crack rather than a functional feature.32
• Annular Holes: Ring-shaped gaps characterized by two closed hole lines that are mutually nearest to each other.36 A key mathematical criterion is that the gravity center of one loop must lie within the perimeter of the other when projected onto a common fitting plane.32
• Island Holes: Holes that surround an isolated patch of triangles (an “island”) within the missing area.32 These are common in 3D scanning data where an object’s protrusion occludes the surface behind it.33
• Dislocation Holes: Gaps caused by misaligned mesh sections.32 These are detected when the distance between points on a boundary and their nearest edges on an opposing boundary is below a specified “dislocation distance”.32

Hole Type	Primary Geometric Metric	Classification Rule
Simple	Surface Area / Boundary Length	Area > Threshold AND Circularity > Threshold
Crack	Aspect Ratio (Length/Width)	Width < AND Length/Width >
Annular	Dual-Loop Mutual Proximity	AND Co-axial Centroids
Island	Topological Genus within Loop	Boundary edges found within the area
Dislocation	Point-to-Edge Distance	for separate loops

32

Discrete Differential Geometry and Curvature Metrics

False positives often arise because topological searches identify any surface break, regardless of its geometric context.6 To distinguish between a “mesh tear” and a “machined hole,” the tool must analyze the local curvature of the surface surrounding the hole.7

Discrete Gaussian Curvature

In smooth differential geometry, Gaussian curvature is the product of the principal curvatures and .20 In the discrete domain of a triangle mesh, curvature is approximated at each vertex using the “angle deficit” method.7 The curvature at a vertex is given by:

where is the valence (number of incident edges) of the vertex, are the interior angles of the incident triangles, and is the Voronoi area associated with the vertex.7

This metric is a powerful filter for hole detection. Intentional features in mechanical parts, such as drill holes, are typically located in regions of specific curvature signatures.7 Discrete Gaussian curvature allows the algorithm to remove planar areas—where —from the search space, focusing only on regions where the surface “bends” into an opening.7 Furthermore, curvature analysis can identify the “ridge of maximum inflection” along a feature edge, allowing for the extraction of feature lines even in noisy meshes.39

Principal Curvatures and Mean Curvature

To distinguish a cylindrical hole from a spherical cavity or a saddle-shaped defect, the tool must estimate the principal curvatures and .42 For a perfectly cylindrical surface of radius , one principal curvature is (along the circumferential direction) and the other is (along the axial direction).20 This results in a Gaussian curvature of zero but a non-zero mean curvature .20

Several academic algorithms provide robust estimation of these values:

1. Taubin’s Algorithm: Estimates the curvature tensor by calculating the eigenvalues of a matrix defined by integral formulas over the vertex’s 1-ring neighborhood.42 It is valued for its linear time complexity and robustness to irregular sampling.46
2. Cubic / Quadric Fitting: Locally fits a polynomial surface to the neighborhood of a vertex and derives the curvature tensor analytically from the fit.20 While accurate for smooth data, it can be sensitive to noise in coarse meshes.48
3. Normal Variation Method: Estimates curvature based on the change in unit normals between a vertex and its neighbors.43 The normal curvature in the direction of an edge is approximated as:
   
   where is the estimated vertex normal.43

Primitive Type	Point Type	KG	H	Curvature Sign
Planar	Flat			N/A
Cylindrical	Parabolic			(Concave) / (Convex)
Spherical	Elliptical			Same sign as
Saddle	Hyperbolic		Varies	have opposite signs

7

Advanced Pruning Strategies to Reduce False Positives

The core of the user’s failure—generating too many false positives—can be addressed by applying geometric constraints to the detected loops.7 Academic research proposes several scoring metrics to validate candidate holes.

Circularity and Roundness Scores

Intentional drill holes exhibit high degrees of symmetry and circularity that unintended gaps do not.50 The Polsby-Popper score is a widely used measure of compactness:

A perfect circle achieves a score of . In industrial inspection of STL models, a threshold of is frequently used to identify candidate openings for mechanical fasteners.51 Shapes with lower scores are typically classified as cracks, elongated slots, or mesh tears.51

Roundness is a complementary metric that measures the roughness of the boundary rather than its overall elongation.54 It is calculated as the ratio of the average curvature radius of all corners along the boundary to the radius of the maximum inscribed circle.54 An elongated oval may have a low circularity score but a high roundness score, indicating it is a “smooth” feature.54 Conversely, a mesh defect like a “zig-zag” boundary will have low roundness even if its circularity is moderate.15

Torsion and Boundary Smoothness

The boundary of an intentional hole in a smooth CAD surface should itself be smooth and regular.56 Research by Wang et al. suggests analyzing the torsion of the boundary curve.56 Holes generated by occlusion or scanning defects often exhibit high torsion and irregular edge progressions.56 By fitting the boundary edges with a B-spline and calculating the average variation of the torsion along the curve, one can automatically classify “real” holes from “man-made” (mesh artifact) holes.56

Normal Deviation and Feature Angles

In mechanical models, intentional holes are usually machined perpendicular or at a specific angle to the surrounding surface.58 The “feature angle”—the angle between the normal of a boundary triangle and the normal of the adjacent surface triangle—provides a critical discriminator.7 Ideally, this angle should be approximately 90 degrees.7 To ensure a high success rate while accounting for mesh noise, researchers recommend accepting angles between 30 and 120 degrees.7 Edges with feature angles outside this range are likely due to gradual surface curvature or mesh inaccuracies and should be pruned from the detection set.7

Robust Feature Recognition Algorithms

For a CAD tool, identifying a hole is often just the first step toward reconstructing its underlying geometry, such as its axis and radius.48

Hybrid Mesh Segmentation and Conformal Testing

A robust framework for recognizing complex intersecting holes involves partitioning the mesh into meaningful analytical regions.48 This process follows a specific sequence:

1. Preprocessing: Establishing the topology and using machine learning to predict optimal partitioning thresholds.48
2. Segmentation: Clustering facets based on attributes such as normal direction and area rather than noise-sensitive curvature.48
3. Primitive Identification: Subjecting each cluster to “conformal tests” to verify if it represents a plane, cylinder, cone, or torus.48

To identify a cylindrical feature from a cluster of triangles, the following mathematical routine is employed:

• Axis Estimation: The facet normals of the input triangles are fitted to a plane using Singular Value Decomposition (SVD). The normal to this plane (the eigenvector of the least eigenvalue) is the cylinder’s axis.48
• Rotation and Circle Fitting: The axis is rotated to be perpendicular to the XY plane, and all vertices are projected into 2D.48 A circle is then fitted to these 2D points using a total least squares or 3-point strategy.48
• Validation: Triangles are accepted as part of a cylindrical primitive only if more than 90% of them pass a test against the calculated axis and radius.48

This approach is superior to slice-based methods because it can separate interacting features, such as a hole intersecting a slot, where the topology is highly irregular.48

RANSAC for Outlier-Resistant Fitting

The Random Sample Consensus (RANSAC) algorithm is a powerful tool for fitting geometric primitives in the presence of noise and outliers.62 In the context of hole detection, RANSAC iteratively selects a random subset of triangles, fits a model (e.g., a cylinder), and counts the number of “inliers”—triangles that fall within a distance threshold of the model.63

An improved methodology, “PC-RANSAC,” incorporates principal curvature constraints.62 By filtering out points that do not match the expected cylindrical curvature signature before beginning the iterative sampling, the algorithm significantly increases its detection speed and accuracy.62 Once a cylinder is detected, Euclidean clustering can be used to distinguish between multiple holes of the same diameter.65

Fitting Method	Robustness Metric	Computational Cost	Advantage
Least Squares	Sum of squared residuals	Low	Fast, but sensitive to outliers
Total Least Squares	Perpendicular distance minimization	Medium	Handles errors in all directions ()
RANSAC	Number of inliers within threshold	High (Iterative)	Extremely robust to high noise/outliers
SVD (Axis extraction)	Eigenvalue distribution	Medium	Provides precise orientation for co-axial features

61

Persistent Homology: The Multi-Scale Solution to Noise

One of the most promising areas of academic research for minimizing false positives is Topological Data Analysis (TDA), specifically Persistent Homology (PH).28 Standard topological checks fail because they operate at a single fixed scale, meaning any tiny gap in the mesh is seen as a “hole”.69 PH addresses this by analyzing the “persistence” of features across multiple spatial resolutions.29

Filtration and Feature Lifespans

The PH process begins by building a “filtration”—a nested sequence of simplicial complexes indexed by a scale parameter (such as the radius of balls centered at each vertex).67 As increases, components merge and holes (1-cycles) are created and eventually filled.28 The moments of creation (birth, ) and annihilation (death, ) are recorded.67

The lifespan of a feature, , provides a definitive measure of its geometric significance.28 Features that persist for a long duration across the filtration are considered “true” topological characteristics of the object, while short-lived features are classified as sampling noise or mesh artifacts.69 For a CAD tool, this means that an intentional drill hole will produce a persistent 1-cycle that survives significant filtration, whereas a tiny gap caused by a chordal error will “die” almost immediately.28

Stability and Invariance

A critical advantage of PH is its proven stability: small perturbations in the input data (such as vertex noise or coordinates-based rounding errors in STL) only result in small changes to the persistence diagram.67 This makes PH an ideal “noise filter” for hole detection in converted STEP files, where the exact locations of vertices are subject to approximation errors.68 Furthermore, PH is coordinate-free and invariant to rotation and translation, ensuring consistent results regardless of the part’s orientation in the CAD workspace.67

Industrial Implementation Pipeline

Creating a functioning tool that satisfies engineering requirements requires integrating these mathematical concepts into a unified pipeline. Research suggests that a robust detection engine should follow a multi-stage approach.48

Stage 1: Pre-Processing and Mesh Healing

Before detection can begin, the “raw” STL must be prepared to ensure topological consistency.77 This includes:

• Vertex Welding: Merging coincident vertices that have the same coordinates but separate indices, a common artifact of the STL format’s redundant vertex listing.79
• Normal Unification: Ensuring all facet normals point outward from the solid volume.81 Flipped normals are a major cause of slicing errors and can confuse inside/outside classification.84
• Non-Manifold Edge Repair: Resolving edges shared by more than two triangles, which break the fundamental assumptions of boundary-tracking algorithms.81

Stage 2: Feature-Aware Boundary Search

The search for holes should not be exhaustive but targeted.7 Using discrete Gaussian curvature as a filter, the tool can focus only on “non-planar” regions.7 Within these regions, boundary edges are traced to form closed loops using a winged-edge or half-edge structure.19

Stage 3: Geometric Pruning and Scoring

Each detected loop is subjected to the classification taxonomy discussed previously.32 Loops are scored based on Circularity, Roundness, and feature angles.7

• If , the loop is likely a crack and is discarded or flagged for stitching.32
• If the feature angle deviates from 90 degrees by more than 60 degrees, the loop is classified as surface noise.7
• Persistent Homology is applied to the remaining candidates to ensure they are geometrically stable features rather than artifacts of local triangulation density.28

Stage 4: SVD Axial Verification and Parameterization

For confirmed holes, an SVD-based fit of the surrounding normals extracts the axis and determines if the hole is “Through” or “Blind”.48 A “Through” hole is confirmed if two co-axial loops are found on opposite sides of the part, while a “Blind” hole is confirmed by the presence of a co-axial planar or conical “bottom” face.48

The Role of Mesh Resolution and Quality

The performance of any detection algorithm is fundamentally limited by the resolution of the input STL.13 Over-refined meshes, while appearing smooth, increase computational complexity and can introduce “needle triangles” that cause numerical instability in curvature calculations.57 Conversely, low-resolution meshes fail to capture the curvature of small holes, leading to missed detections.10

Research indicates that a “regular” mesh—where triangles are approximately equal-sized and equilateral—is essential for accurate discrete curvature approximation.57 Mesh regularity can be monitored by checking if the number of facets across successive concentric rings of the mesh follows an arithmetic progression.57 If the mesh is irregular or poorly sampled, a pre-processing remeshing step using quadratic edge collapse or vertex clustering may be required to homogenize the triangulation before hole detection is attempted.77

Future Directions: Geometric Deep Learning

Traditional rule-based and heuristic systems, while mathematically grounded, can be brittle when faced with the diverse modeling styles and feature interactions found in real-world industrial models.90 The frontier of academic research is currently exploring Geometric Deep Learning (GDL) to provide more generalizable solutions.90

Architectures such as MeshCNN treat the mesh as a graph and use edge-based convolutions to learn geometric and topological features directly from the data.9 These models have demonstrated high precision in surface-defect localization and can be trained to recognize “repeatable geometric patterns” that characterize functional features.38 More advanced Topo-Geom fusion networks integrate attribute adjacency graphs with geometric relationship graphs, allowing the AI to account for both local topological details and global geometric constraints simultaneously.91 This technology promises to bridge the “semantic link” between a raw mesh and the manufacturer’s intent, potentially automating the entire feature recognition process with zero-shot generalization.91

Substantive Conclusions and Practical Recommendations

The difficulty in creating a robust hole detection tool for STL files derived from STEP conversion lies in the inherent conflict between the designer’s parametric intent and the lossy nature of discretization. A functioning tool cannot rely on a single mathematical criterion but must instead utilize a layered methodology that integrates topological verification, differential geometric analysis, and statistical model fitting.

To minimize false positives, developers should prioritize the following implementation strategies:

1. Topological Validation: Use a half-edge data structure to robustly identify boundary cycles while handling singular vertices and singular edges.
2. Curvature-Based Filtering: Apply discrete Gaussian curvature (angle deficit) to prune planar regions and noise from the initial search space.
3. Geometric Scoring: Implement circularity (Polsby-Popper) and roundness scores to distinguish between functional machined holes and elongated cracks or jagged mesh artifacts.
4. Robust Fit Verification: Utilize SVD for axis extraction and RANSAC for model fitting to ensure that candidate holes align with known geometric primitives (cylinders and cones).
5. Multiscale Discrimination: For high-precision requirements, integrate Persistent Homology to discriminate between true geometric features and short-lived sampling noise.

Furthermore, the quality of the detection is deeply coupled with the quality of the input mesh. Software tools should advise users on optimal STL export parameters—specifically chordal and angular tolerances—to ensure that curved features are represented with sufficient fidelity for mathematical reconstruction. By adopting this hybrid, feature-centric approach, CAD developers can create tools that not only detect holes but understand their functional context, providing the accuracy necessary for automated manufacturing and engineering workflows.

Works cited

1. STL to STEP: Converting Mesh Files to Editable CAD Models – Rapid-Protos, accessed April 1, 2026, https://www.rapid-protos.com/stl-to-step/
2. STP vs STL for FEA and CFD: Why Geometry Matters – Joseph Webb, accessed April 1, 2026, https://josephwebb.net/why_geometry_matters.html
3. Understanding the STL File: The Standard 3D Mesh Format for 3D Printing, accessed April 1, 2026, https://www.3dmag.com/3d-wikipedia/stl-file-standard-3d-mesh-format/
4. Exploring the Effect of Preprocessing on Real-world STL Model Classification Performance – MS Word Template for CAD Conference Papers, accessed April 1, 2026, https://www.cad-journal.net/files/vol_23/CAD_23(5)_2026_626-645.pdf
5. A method for identifying grid features on wall panel surfaces for laser nondestructive detection systems | Rapid Prototyping Journal | Emerald Publishing, accessed April 1, 2026, https://www.emerald.com/rpj/article/32/3/605/1304982/A-method-for-identifying-grid-features-on-wall
6. We Found Six STL Repair Tools That Really Work | All3DP Pro, accessed April 1, 2026, https://all3dp.com/1/for-models-your-slicer-cant-fix-these-free-stl-repair-tools-really-worked/
7. A Simple and Fast Hole Detection Algorithm for Triangulated Surfaces, accessed April 1, 2026, https://diamhomes.ewi.tudelft.nl/~kvuik/papers/Kum21V.pdf
8. A Simple and Fast Hole Detection Algorithm for Triangulated Surfaces | J. Comput. Inf. Sci. Eng. | ASME Digital Collection, accessed April 1, 2026, https://asmedigitalcollection.asme.org/computingengineering/article/21/4/044502/1090537/A-Simple-and-Fast-Hole-Detection-Algorithm-for
9. Comparative Evaluation of Voxel and Mesh Representations for Digital Defect Detection in Construction-Scale Additive Manufacturing – MDPI, accessed April 1, 2026, https://www.mdpi.com/2075-5309/16/4/805
10. Guide to STL 3D Print Files for Better 3D Printing – Markforged, accessed April 1, 2026, https://markforged.com/resources/blog/how-to-create-high-quality-stl-files-for-3d-prints
11. STL File Tolerance: A Short Explanation of Faceting and Chord Height – PADT, Inc., accessed April 1, 2026, https://www.padtinc.com/2012/11/13/stl-file-tolerance-a-short-explanation-of-faceting-and-chord-height/
12. (a) Chordal error, (b) Error in contour generated from STL model. – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/figure/a-Chordal-error-b-Error-in-contour-generated-from-STL-model_fig2_318298128
13. Performance Evaluation of Various STL File Mesh Refining Algorithms Applied for FDM-RP Process | Request PDF – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/304028408_Performance_Evaluation_of_Various_STL_File_Mesh_Refining_Algorithms_Applied_for_FDM-RP_Process
14. Is there any way to get small holes of a certain radius? : r/3Dprinting – Reddit, accessed April 1, 2026, https://www.reddit.com/r/3Dprinting/comments/htjdkr/is_there_any_way_to_get_small_holes_of_a_certain/
15. The 5 Mistakes Not to Make in 3D Printing STL Files, accessed April 1, 2026, https://polyd.com/en/blog/errors_stl_file
16. Tips for mesh generation – Using the Gradation Factor and Chordal Error Ratio Parameters to Control the Mesh Density – MotionPort, accessed April 1, 2026, https://motionport.com/tips-for-mesh-generation-using-the-gradation-factor-and-chordal-error-ratio-parameters-to-control-the-mesh-density/
17. ROBUST FEATURE LINE EXTRACTION ON CAD TRIANGULAR MESHES – Semantic Scholar, accessed April 1, 2026, https://pdfs.semanticscholar.org/6790/2cd6b2f4e7ffee6edd03781720e8adfcd6f8.pdf
18. (PDF) A Robust Hole-Filling Algorithm for Triangular Mesh – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/4301214_A_Robust_Hole-Filling_Algorithm_for_Triangular_Mesh
19. Filling Holes in Triangular Meshes in Engineering – Journal of Software, accessed April 1, 2026, https://www.jsoftware.us/vol7/jsw0701-20.pdf
20. Curvatures estimation on triangular mesh* – Journal of Zhejiang University, accessed April 1, 2026, https://jzus.zju.edu.cn/oldversion/opentxt.php?doi=10.1631/jzus.2005.AS0128
21. Edgebreaker: Connectivity compression for triangle meshes – People @EECS, accessed April 1, 2026, https://people.eecs.berkeley.edu/~jrs/meshpapers/Rossignac.pdf
22. How to detect a hole in the triangular mesh? – Stack Overflow, accessed April 1, 2026, https://stackoverflow.com/questions/12968478/how-to-detect-a-hole-in-the-triangular-mesh
23. (PDF) Hole-Filling Techniques for 3D-Objects: A Comparative Study – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/259502921_Hole-Filling_Techniques_for_3D-Objects_A_Comparative_Study
24. How to detect a hole in the meshed body? – Mathematics Stack Exchange, accessed April 1, 2026, https://math.stackexchange.com/questions/3338863/how-to-detect-a-hole-in-the-meshed-body
25. An Easy Way of Detecting Subdivision Connectivity in a Triangle Mesh, accessed April 1, 2026, https://www.inf.usi.ch/hormann/papers/Hormann.2002.AEW.pdf
26. Filling Holes in Triangular Meshes by Curve Unfolding – DSpace@MIT, accessed April 1, 2026, https://dspace.mit.edu/bitstream/handle/1721.1/53749/Brunton-2009-Filling%20holes%20in%20triangular%20meshes%20by%20curve%20unfolding.pdf?sequence=1&isAllowed=y
27. Detect hole in geometry – algorithm – Stack Overflow, accessed April 1, 2026, https://stackoverflow.com/questions/30810431/detect-hole-in-geometry
28. Persistent Homology in Medical Image Processing: A Literature Review – medRxiv, accessed April 1, 2026, https://www.medrxiv.org/content/10.1101/2025.02.21.25322669v1.full-text
29. On the Effectiveness of Persistent Homology – NSF Public Access Repository, accessed April 1, 2026, https://par.nsf.gov/servlets/purl/10399582
30. [2311.12466] Robust Hole-Detection in Triangular Meshes Irrespective of the Presence of Singular Vertices – arXiv, accessed April 1, 2026, https://arxiv.org/abs/2311.12466
31. Robust Hole-Detection in Triangular Meshes Irrespective of the Presence of Singular Vertices – SINTEF, accessed April 1, 2026, https://www.sintef.no/en/publications/publication/2231197/
32. Research on an Intelligent Identification and Classification Method of Complex Holes in Triangle Meshes for 3D Printing – PMC, accessed April 1, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC9173961/
33. a comparison of hole–filling methods in 3d, accessed April 1, 2026, http://pldml.icm.edu.pl/pldml/element/bwmeta1.element.bwnjournal-article-amcv26i4p885bwm/c/amc26413.pdf
34. Minimum Surface Area Based Complex Hole Filling Algorithm of 3D Mesh, accessed April 1, 2026, https://www.art-science.org/journal/v14n2/v14n2pp26/artsci-v14n2pp26.pdf
35. Research on an Intelligent Identification and Classification Method of Complex Holes in Triangle Meshes for 3D Printing – PubMed, accessed April 1, 2026, https://pubmed.ncbi.nlm.nih.gov/35685141/
36. An Intelligent Identification and Repair Method for Annular Holes in 3D Printing – PMC, accessed April 1, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC9098285/
37. Filling holes in meshes using a mechanical model to simulate the curvature variation minimization | Request PDF – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/223432137_Filling_holes_in_meshes_using_a_mechanical_model_to_simulate_the_curvature_variation_minimization
38. Comparative Evaluation of Voxel and Mesh Representations for Digital Defect Detection in Construction-Scale Additive Manufacturing – Preprints.org, accessed April 1, 2026, https://www.preprints.org/manuscript/202601.0140
39. Extraction of feature lines on triangulated surfaces using morphological operators – Computer Graphics, accessed April 1, 2026, https://graphics.rwth-aachen.de/media/papers/morpho.pdf
40. Curvature of a triangle mesh, definition and computation. – Rodolphe Vaillant’s homepage, accessed April 1, 2026, https://rodolphe-vaillant.fr/entry/33/curvature-of-a-triangle-mesh-definition-and-computation
41. Multiscale Curvature Estimation for Segmenting a Triangle Mesh into Shape Features – Institute for Applied Maths and Information Technologies, accessed April 1, 2026, http://pers.ge.imati.cnr.it/patane/Publications_files/patane-EGIT2003.pdf
42. A Novel Computational Paradigm for Reconstructing Solid CAD Features from a Segmented Manifold Triangular Mesh – MDPI, accessed April 1, 2026, https://www.mdpi.com/2076-3417/14/14/6183
43. Estimating Curvatures and Their Derivatives on Triangle Meshes – Princeton Graphics Group, accessed April 1, 2026, https://gfx.cs.princeton.edu/pubs/Rusinkiewicz_2004_ECA/curvpaper.pdf
44. A CAD-based Methodology for Planar and Cylindrical Feature Recognition – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/profile/Pasquale-Franciosa-2/publication/274832813_A_CAD-based_Methodology_for_Planar_and_Cylindrical_Feature_Recognition/links/552a66f60cf2779ab791028b/A-CAD-based-Methodology-for-Planar-and-Cylindrical-Feature-Recognition.pdf
45. A Fast Hole-Filling Method for Triangular Mesh in Additive Repair – MDPI, accessed April 1, 2026, https://www.mdpi.com/2076-3417/10/3/969
46. Improved Curvature Estimation on Triangular Meshes – Washington University in St. Louis Scholarly Repository, accessed April 1, 2026, https://openscholarship.wustl.edu/cgi/viewcontent.cgi?article=2057&context=cse_research
47. Estimating Curvature on Triangular Meshes – College of Engineering | Oregon State University, accessed April 1, 2026, https://web.engr.oregonstate.edu/~grimmc/Curvature/cagd_curvature_gatzke_grimm_submit.pdf
48. Complex Hole Recognition from CAD Mesh Models – CCTech, accessed April 1, 2026, https://www.cctech.co.in/downloads/publications/2018/complex-hole-recognition-from-cad-mesh-models-ICETSET-2018.pdf
49. (PDF) Estimating Curvature on Triangular Meshes. – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/220539554_Estimating_Curvature_on_Triangular_Meshes
50. Calculating roundness/compactness of polygon? – GIS StackExchange, accessed April 1, 2026, https://gis.stackexchange.com/questions/85812/calculating-roundness-compactness-of-polygon
51. Circularity Calculations with the CircularityCalculator – FME Support Center – Safe Software, accessed April 1, 2026, https://support.safe.com/hc/en-us/articles/25407493686669-Circularity-Calculations-with-the-CircularityCalculator
52. Circularity – GD&T Basics, accessed April 1, 2026, https://www.gdandtbasics.com/circularity/
53. Comparison of circularity (C) and proposed shape regularity criteria… – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/figure/Comparison-of-circularity-C-and-proposed-shape-regularity-criteria-SRC-on-three_fig6_318600573
54. Fast Sphericity and Roundness approximation in 2D and 3D using Local Thickness – arXiv, accessed April 1, 2026, https://arxiv.org/html/2504.05808v1
55. An Effective Hole Detection Method for 3D Models – EURASIP, accessed April 1, 2026, https://new.eurasip.org/Proceedings/Eusipco/Eusipco2012/Conference/papers/1569581583.pdf
56. A Hole-Filling Algorithm for Triangular Meshes Using Local Radial Basis Function – Department of Computing Science, accessed April 1, 2026, https://webdocs.cs.ualberta.ca/~wfb/ammi/publications/C-2006-MR.pdf
57. Assessing the Regularity of 3D Triangular Mesh Tessellation Using a Topological Structured Pattern – CAD Journal, accessed April 1, 2026, https://www.cad-journal.net/files/vol_8/CAD_8(5)_2011_633-648.pdf
58. Detect Holes, accessed April 1, 2026, https://2021.help.altair.com/2021/hwdesktop/hm/topics/pre_processing/meshing/holes_detect_t.htm
59. Circular hole recognition for STL-based toolpath generation | Request PDF – ResearchGate, accessed April 1, 2026, https://www.researchgate.net/publication/242347889_Circular_hole_recognition_for_STL-based_toolpath_generation
60. Mathematical Approaches in Evaluation of Circularity Measurement – IJERT, accessed April 1, 2026, https://www.ijert.org/research/mathematical-approaches-in-evaluation-of-circularity-measurement-IJERTV5IS100428.pdf
61. Evaluation of Circularity from Coordinate data using Maximum Distance Point Strategy (MDPS) – International Journal of Engineering Research & Technology, accessed April 1, 2026, https://www.ijert.org/research/evaluation-of-circularity-from-coordinate-data-using-maximum-distance-point-strategy-mdps-IJERTV1IS4150.pdf
62. Improved RANSAC Point Cloud Spherical Target Detection and Parameter Estimation Method Based on Principal Curvature Constraint – MDPI, accessed April 1, 2026, https://www.mdpi.com/1424-8220/22/15/5850
63. 3D Shape Detection with RANSAC and Python (Sphere and Plane) – 3D Geodata Academy, accessed April 1, 2026, https://learngeodata.eu/3d-shape-detection-with-ransac-and-python-sphere-and-plane/
64. Robust cylinder detection and pose estimation using 3D point cloud information – Vislab@ISR, accessed April 1, 2026, https://vislab.isr.tecnico.ulisboa.pt/wp-content/uploads/2017/05/rfigueiredo-icarsc2017.pdf
65. Cylindrical feature hole recognition for automatic repair, accessed April 1, 2026, http://www.txxb.com.cn/EN/10.11996/JG.j.2095-302X.2021030511
66. Evaluation of Circularity from Coordinate data using Maximum Distance Point Strategy (MDPS) – International Journal of Engineering Research & Technology, accessed April 1, 2026, https://www.ijert.org/evaluation-of-circularity-from-coordinate-data-using-maximum-distance-point-strategy-mdps
67. Persistent Homology in TDA – Emergent Mind, accessed April 1, 2026, https://www.emergentmind.com/topics/persistent-homology-ph
68. Topological Data Analysis and Topological Deep Learning Beyond Persistent Homology–A Review – arXiv, accessed April 1, 2026, https://arxiv.org/html/2507.19504v1
69. Topological inference via meshing – Inria, accessed April 1, 2026, https://geometrica.saclay.inria.fr/team/Steve.Oudot/papers/hmos-tim-10/hmos-tim-10.pdf
70. Noise robustness of persistent homology on greyscale images, across filtrations and signatures – PMC, accessed April 1, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC8462731/
71. Topological Data Analysis for Multivariate Time Series Data – MDPI, accessed April 1, 2026, https://www.mdpi.com/1099-4300/25/11/1509
72. RETHINKING PERSISTENT HOMOLOGY FOR VISUAL RECOGNITION – Proceedings of Machine Learning Research, accessed April 1, 2026, https://proceedings.mlr.press/v196/khramtsova22a/khramtsova22a.pdf
73. Robust Model Reconstruction Based on the Topological Understanding of Point Clouds Using Persistent Homology – arXiv, accessed April 1, 2026, https://arxiv.org/html/2508.00251v2
74. Topological data analysis: What is persistent homology? – Broad Institute, accessed April 1, 2026, https://www.broadinstitute.org/talks/topological-data-analysis-what-persistent-homology
75. Flow estimation solely from image data through persistent homology analysis – PMC, accessed April 1, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC8429714/
76. Topological data analysis – Wikipedia, accessed April 1, 2026, https://en.wikipedia.org/wiki/Topological_data_analysis
77. How to Repair STL Files for 3D Printing With the 5 Best (Free) STL Repair Tools | Formlabs, accessed April 1, 2026, https://formlabs.com/blog/best-stl-file-repair-software-tools/
78. Mesh Healer Guide: A Comprehensive Tutorial for Effective 3D Model Inspection and Repair, accessed April 1, 2026, https://meshinspector.com/knowledge-base/mesh-repair/mesh-healer-guide-a-comprehensive-tutorial-for-effective-3d-model-inspection-and-repair/
79. Defect Analysis in STL Files | PDF | Triangle | Vertex (Graph Theory) – Scribd, accessed April 1, 2026, https://www.scribd.com/document/931774301/1-s2-0-S0895717703900793-main
80. Fixing Common 3D Geometry Issues – SimplyPrint Helpdesk, accessed April 1, 2026, https://help.simplyprint.io/en/article/fixing-common-3d-geometry-issues-5wqwh7/
81. Identifying and Repairing Common Mesh Errors – Simplify 3D, accessed April 1, 2026, https://www.simplify3d.com/resources/articles/identifying-and-repairing-common-mesh-errors/
82. Repair Any STL: Inspector + Separate Shells + Zipper Edges – Meshmixer, accessed April 1, 2026, https://meshmixer.org/repair-any-stl-inspector-separate-shells-zipper-edges/
83. The Most Common STL File Errors You Will Meet in 3D Printing Process – FacFox Docs, accessed April 1, 2026, https://facfox.com/docs/kb/the-most-common-stl-file-errors-you-will-meet-in-3d-printing-process
84. The Top 5 STL File Errors You Should Know Before You Print – AMFG, accessed April 1, 2026, https://www.amfg.ai/post/top-5-stl-file-errors-you-should-know
85. What are the top STL file errors? Here’s how to fix them | Protolabs Network, accessed April 1, 2026, https://www.hubs.com/knowledge-base/fixing-most-common-stl-file-errors/
86. Mesh Stitching: Techniques for Repairing Gaps and Cracks in 3D Models, accessed April 1, 2026, https://alexsyniakov.com/2025/01/07/mesh-stitching/
87. Analytical Detection of a Mesh Vertex Being on the Inside of a Round Through-Hole, accessed April 1, 2026, https://www.researchgate.net/publication/358468118_Analytical_Detection_of_a_Mesh_Vertex_Being_on_the_Inside_of_a_Round_Through-Hole
88. We need to talk about STEP vs. STL files. There is a significant misconception floating around. : r/3Dprinting – Reddit, accessed April 1, 2026, https://www.reddit.com/r/3Dprinting/comments/1j936w1/we_need_to_talk_about_step_vs_stl_files_there_is/
89. Robust and Efficient Local Repair for Intersecting Triangle Meshes, accessed April 1, 2026, https://internationalmeshingroundtable.com/assets/papers/2025/1001-compressed.pdf
90. A Survey of AI Methods for Geometry Preparation and Mesh Generation in Engineering Simulation – arXiv.org, accessed April 1, 2026, https://arxiv.org/html/2512.23719v2
91. Topo-Geom DualGNN: A Dual-Graph Fusion Network for Machining Feature Recognition, accessed April 1, 2026, https://www.mdpi.com/2075-1702/14/4/362
92. Autonomous Geometry Processing Using Machine Learning and Forge | by Autodesk University – Medium, accessed April 1, 2026, https://medium.com/autodesk-university/autonomous-geometry-processing-using-machine-learning-and-forge-3b89b40e97cb
93. Hybrid Rule-Based Classification and Defect Detection System Using Insert Steel Multi-3D Matching – MDPI, accessed April 1, 2026, https://www.mdpi.com/2079-9292/14/23/4701