Zhuojin Pipe Fitting Co., LtdParametric Reconstruction of Tee Pipe Models Based on Overlapping Point Cloud Registration for Robotic Welding
Literature Overview
This paper by Jiang Linjun, Huang Yunbao, Song Zikang, Li Haiyan, and He Yong, published in Modular Machine Tool and Automatic Manufacturing Technique in 2026 (Issue 7, pp. 50-56), presents a novel algorithm for the parametric reconstruction of tee pipe models based on overlapping point cloud registration. The research was supported by the National Natural Science Foundation of China (Project No. 12127810). The study addresses the challenges of low efficiency, unstable quality, and poor multi-view point cloud registration accuracy in robotic welding applications. The proposed algorithm uses laser scanning to acquire multi-view point clouds of tee pipes, applies feature-based registration for coarse alignment, and uses iterative closest point (ICP) optimization for fine alignment, achieving a 65.46% improvement in registration accuracy compared to other algorithms.
Core Algorithm and Technical Approach
The proposed algorithm consists of four main stages, each addressing a specific aspect of the point cloud registration and model reconstruction problem:
- Multi-view point cloud acquisition and preprocessing: A laser camera scans the tee pipe from multiple viewpoints to obtain point cloud data. The point clouds are preprocessed and transformed to a unified coordinate system to achieve a good initial pose.
- Key point extraction and feature matching: The Intrinsic Shape Signatures (ISS) algorithm and a normal vector angle threshold are used to extract key points from the point clouds. A novel Local Angles and Curvature Histogram (LACH) feature descriptor is constructed for matching and extracting overlapping point clouds.
- Fine registration: A linear least squares optimization point-to-plane ICP algorithm is used for precise registration to achieve point cloud stitching.
- Model reconstruction and weld seam extraction: The stitched complete point cloud is registered with the discrete point cloud of the workpiece CAD model through secondary registration to complete the tee pipe model reconstruction. The weld seams are extracted from the reconstructed model for robotic welding.
The following table summarizes the key technical parameters and results:
| Parameter | Value |
|---|---|
| Registration Accuracy Improvement | 65.46% compared to other algorithms |
| Average Weld Seam Extraction Error | Less than 1 mm |
| Feature Descriptor | LACH (Local Angles and Curvature Histogram) |
| Coarse Registration | ISS + Normal Vector Angle Threshold |
| Fine Registration | Point-to-Plane ICP with Linear Least Squares Optimization |
| Welding Result | Meets process requirements |
Technical Analysis of the LACH Feature Descriptor
The LACH feature descriptor is a key innovation in this study. Unlike traditional feature descriptors such as Fast Point Feature Histogram (FPFH) or 3D Histogram of Oriented Gradients (3D-HOG), the LACH descriptor is specifically designed for tee pipe geometry. The descriptor captures local angular and curvature information, which is particularly relevant for the complex intersection geometry of tee pipes. The use of ISS for key point extraction ensures that the selected points are stable and distinctive across different viewpoints, which is critical for reliable matching.
The point-to-plane ICP algorithm, as opposed to the conventional point-to-point ICP, provides better convergence properties and higher accuracy for the fine registration stage. The linear least squares optimization formulation allows for efficient computation and robust convergence, which is essential for real-time or near-real-time application in robotic welding systems.
Engineering Practice and Robotic Welding Integration
The integration of the proposed algorithm with robotic welding systems has several practical implications:
- The parametric reconstruction of the tee pipe model provides accurate geometric information for welding path planning, ensuring that the welding torch follows the correct trajectory along the weld seam.
- The weld seam extraction with an average error of less than 1 mm is critical for achieving consistent weld quality, as even small deviations in the welding path can lead to weld defects such as undercuts, excessive reinforcement, or incomplete fusion.
- The algorithm's ability to handle multi-view point clouds with improved registration accuracy addresses a key challenge in robotic welding of complex geometries, where single-view scanning is often insufficient to capture the complete geometry of the workpiece.
The following table compares the proposed method with conventional approaches for robotic welding of tee pipes:
| Aspect | Conventional Approach | Proposed Method |
|---|---|---|
| Point Cloud Registration Accuracy | Lower | 65.46% improvement |
| Weld Seam Extraction Error | Larger | Less than 1 mm |
| Geometry Coverage | Limited to single or few views | Complete multi-view stitching |
| Welding Quality | Variable | Meets process requirements |
| Efficiency | Lower | Improved |
Key Questions and Reflections
This study raises an important question about the generalizability of the proposed algorithm to other pipe fitting geometries. While the LACH feature descriptor is designed for tee pipe geometry, its applicability to other complex geometries such as cross pipes, reducers, and bends needs to be investigated. The algorithm's performance may degrade for geometries with less distinctive features or with more complex curvature variations.
Another reflection is the role of sensor accuracy and point cloud density in the overall performance of the algorithm. The results reported in the study are likely dependent on the quality of the laser scanner used, the scanning distance, and the point cloud density. In practical applications, the scanner settings may need to be optimized for different workpiece sizes and geometries, and the algorithm's robustness to variations in point cloud quality should be evaluated.
Study Insights and Implications
This literature provides a valuable contribution to the field of robotic welding by addressing the critical challenge of accurate point cloud registration and model reconstruction for complex pipe fitting geometries. The fundamental lesson is that accurate geometric reconstruction is essential for achieving consistent weld quality in robotic welding of complex components. For engineers working on robotic welding systems, this study reinforces the importance of:
- Developing feature descriptors that are tailored to the specific geometry being processed.
- Using appropriate registration algorithms that balance accuracy and computational efficiency.
- Integrating geometric reconstruction with welding path planning to achieve end-to-end automation.
- Validating the algorithm's performance through experimental testing on actual workpieces.
The paper is a useful reference for robotic welding researchers and engineers who are developing or implementing automated welding systems for complex pipe fitting geometries. It demonstrates that a combination of feature-based coarse registration and optimization-based fine registration can achieve high accuracy in point cloud registration, which is critical for reliable robotic welding. The 65.46% improvement in registration accuracy and the sub-millimeter weld seam extraction error are significant achievements that demonstrate the practical value of the proposed approach.