Abstract
Non-circular-cross-section tubes are widely used as heat exchange components throughout the industry, including in fields such as aerospace, automobiles, and vessels. Tube manufacturing in these fields is a highly complex process; to manufacture the needed tubes and guarantee a precise and stress-free assembly, the geometric parameters should be measured during manufacturing and before assembly. However, current studies and manufacturing instrumentation focus on tubes of circular cross section. This study proposes an extraction method for the tubular 3D skeleton of non-circular-cross-section tubes based on multi-view vision. Then, non-circular-cross-section tubes can be measured based on the proposed method. The experimental results show that the average deviation does not exceed 0.3 mm with $3\sigma $ reliability and the measuring time is fewer than 5 s. The accuracy and efficiency of this method are sufficient to satisfy the requirements for industrial applications.
The Challenge of Tube Inspection
In high-performance industries like aerospace and automotive manufacturing, bend tubes (ducts) are critical components for fuel lines and heat exchangers. While circular tubes are standard, many applications require non-circular cross-sections—such as rectangular or square ducts—to optimize spatial constraints or thermal efficiency.
Ensuring these components fit perfectly during assembly requires high-precision geometric measurement. However, existing inspection methodologies face significant trade-offs:
- Contact Methods (CMM/Mechanical Gauges): Highly accurate but extremely slow, often requiring manual operation that bottlenecks production.
- Non-Contact Scanning: Laser scanning is faster but struggles with reflective metallic surfaces and generates massive, computationally expensive point clouds.
- Traditional Vision Systems: Most current algorithms are mathematically optimized for circular cross-sections, rendering them ineffective for ducts or free-form shapes.
Proposed Methodology: 3D Skeleton Extraction
The researchers developed a multi-view vision system using eight cameras to reconstruct the 3D skeleton (centerline) of a tube rather than its entire surface. This approach significantly reduces computational overhead while maintaining high accuracy.
The framework involves four key innovations:
- Iterative Skeleton Extraction: The algorithm calculates "skeleton units" (a 3D point and tangent vector) iteratively along the tube's length.
- Shape-Independent Tangent Reconstruction: By utilizing the 2D edges of the tube across multiple views, the system determines the 3D tangent vector. This geometric relationship remains consistent regardless of the cross-section shape.
- Dynamic Thresholding: To mitigate issues like self-occlusion or uneven lighting, the system automatically filters out unreliable camera views for each specific skeleton unit.
- Plücker Coordinate Clustering: The extracted skeleton is segmented into straight and curved sections using Fuzzy C-means (FCM) clustering in a 6-dimensional Plücker coordinate space.
Experimental Validation
To validate the "universal" nature of the system, the authors tested it against a variety of tube geometries.
The validation set included physical specimens with rectangular, square, compound, and circular cross-sections. Ground truth was established using a high-precision Coordinate Measuring Machine (CMM) with 0.003 mm accuracy. The study also evaluated tubes with complex topologies, including varying bend counts and orientations.
Key Results and Performance
The system demonstrated industrial-grade performance across all test cases:
- Accuracy: Average deviation from CMM ground truth remained under 0.3 mm.
- Reliability: The method achieved 3σ reliability, corresponding to a 99.9937% qualification rate.
- Efficiency: The end-to-end measurement process—from image acquisition to parameter calculation—took less than 5 seconds, a 10x speed improvement over traditional surface reconstruction.
Limitations and Future Outlook
Despite its robustness, the system has two primary limitations. First, complete self-occlusion (e.g., tight coils) prevents reconstruction if a segment is hidden from all eight cameras. Second, small bend angles (under 5 degrees) are difficult for the clustering algorithm to distinguish from straight segments due to subtle changes in Plücker coordinates. Future research will focus on optimizing camera placement and developing interactive methods to handle these extreme topological cases.
Summary updated