Online Extraction of Pose Information of 3D Zigzag-Line Welding Seams for Welding Seam Tracking

Bo Hong, Aiting Jia, Yuxiang Hong, Xiangwen Li, Jiapeng Gao, Yuanyuan Qu2021 · DOI

TL;DR

This paper presents a novel method for real-time extraction of 3D zigzag-line welding seam pose information using laser displacement sensing and a p-Approximate DBSCAN algorithm for point cloud segmentation. This approach significantly improves the efficiency and accuracy of welding seam tracking, especially for high-speed robotic welding applications.

What Problem Does This Paper Address?

Robotic welding of complex 3D zigzag-line seams, common in industries like marine engineering and heavy lifting, presents a significant challenge for automated systems. Traditional methods for identifying and tracking these seams often suffer from high computational costs and poor real-time performance, making them unsuitable for high-speed welding. This limitation hinders welding efficiency and quality. The paper aims to bridge this gap by developing a system that can accurately and rapidly extract the pose (position and orientation) of 3D zigzag-line welding seams in real-time, enabling precise robotic guidance.

Methods

The proposed method involves several key steps:

  • Fast Acquisition of Point Cloud Data: A laser displacement sensor is used to scan the workpiece, generating a 3D point cloud of the welding seam. This data is processed frame by frame.
  • Point Cloud Segmentation: To handle noisy data and isolate the welding seam from the surrounding surfaces, the authors employ a novel approach. First, they establish a neighborhood normal vector matrix for each point. Then, they use a p-Approximate DBSCAN algorithm to cluster these normal vectors. This algorithm is chosen for its efficiency (O(n) complexity) and ability to handle arbitrary shapes, effectively separating the welding seam from planar surfaces. The number of clusters (c) indicates whether an inflection point is present (c=3) or not (c=2).
  • Extraction of Welding Seam Point Cloud: The Random Sample Consensus (RANSAC) algorithm is applied to the segmented point cloud to robustly extract the actual welding seam points, filtering out remaining noise.
  • Trajectory Fitting: To ensure smooth robot movement and avoid vibration, the extracted welding seam trajectory is fitted using the moving least squares method. This creates a continuous path for the robot to follow.
  • Attitude Estimation: The orientation of the welding seam (direction and normal vectors) is estimated. This is crucial for the robot to adjust its welding torch angle dynamically, ensuring optimal weld quality. The calculation of these vectors depends on whether an inflection point is detected (c=3) or not (c=2).

Datasets

The paper does not explicitly mention the use of pre-existing datasets for training or evaluation. Instead, it relies on experimental data collected from welding experiments performed on custom-made 3D zigzag-line welding seam workpieces with various fold angles. These workpieces were used to generate the point cloud data for testing the proposed algorithm.

Key Insights and Ideas

  • p-Approximate DBSCAN for Efficient Segmentation: The use of p-Approximate DBSCAN is a key innovation. It offers a significant speedup over standard DBSCAN for 3D point clouds, making real-time segmentation feasible.
  • Neighborhood Normal Vectors as a Feature: Establishing neighborhood normal vectors proves to be an effective way to characterize the local surface geometry, aiding in the segmentation of planar surfaces from the welding seam.
  • Unified Approach for Different Seam Geometries: The method's ability to handle both seams with and without inflection points (by adjusting the clustering parameter 'c') demonstrates its adaptability to different 3D zigzag-line configurations.
  • Integration of Sensing and Processing: The seamless integration of laser displacement sensing, point cloud processing, and robot control is crucial for achieving real-time tracking.

Key Results

The experimental results demonstrate the effectiveness of the proposed method:

  • High Accuracy: The welding seam position detection error was less than 0.35 mm, and the welding seam attitude estimation error was less than 2 degrees. The inflection point error was also small, under 0.5 mm.
  • Fast Processing Speed: The main algorithm's running time was within 120 ms, meeting the requirements for high-speed welding (up to 1500 mm/min).
  • Robustness: The method showed good performance across various fold angles (130° to 230°) and was applicable to swing welding.
  • Generalizability: The system's architecture, utilizing a point cloud processing computer for communication with robot controllers, suggests good generalization potential for different robot systems and large structure welding.

Limitations and Future Work

  • Limited Fold Angle Range: The paper notes that very large or very small fold angles can interfere with the laser displacement sensor's ability to acquire point clouds. Future work could involve designing more adaptive sensor installation mechanisms to extend the range of detectable fold angles.
  • Positioning Weld Seam Characteristics: While the method is not significantly affected by positioning weld seams, it currently doesn't provide guidance on their size. Future research could incorporate the identification of positioning weld seam dimensions to enable adaptive welding processes.
  • Further Robot Integration: The authors suggest further research into integrating the system with different robot controllers for broader applications in welding large structure workpieces.

Summary updated

BibTeX
@misc{hong2021online,
  author = {Bo Hong and Aiting Jia and Yuxiang Hong and Xiangwen Li and Jiapeng Gao and Yuanyuan Qu},
  title = {{Online Extraction of Pose Information of 3D Zigzag-Line Welding Seams for Welding Seam Tracking}},
  year = {2021},
  doi = {10.3390/s21020375},
}
Figure