TL;DR
This paper introduces a novel system for automatically guiding welding torches on complex, spatially intermittent welding seams, a significant challenge in automated manufacturing. By combining an adaptive laser scanning displacement sensor with a unique shape-based trajectory recognition algorithm, the system achieves 100% classification accuracy and significantly boosts welding efficiency without manual programming.
What Problem Does This Paper Address?
Automating the welding of complex structures, particularly those with "skip welding" requirements (where the welding torch jumps between points), is hindered by the difficulty of precisely identifying and tracking these discontinuous seams in real-time. Traditional methods often rely on time-consuming manual programming or complex calibration, limiting flexibility and efficiency, especially for small-batch production of large components like those found in heavy machinery, shipbuilding, and transportation. This paper tackles the challenge of robustly and efficiently recognizing these spatially intermittent welding seam trajectories to enable fully automated skip welding.
Methods
The core of this work lies in two main components:
- Adaptive Laser Scanning Displacement Sensing System: This system uses a point laser optical triangulation sensor combined with a rotary mirror. This setup allows the sensor to dynamically change its measurement angle without altering its physical position, enabling it to adapt to varying corner geometries and improve its "degree of freedom" for measurement. This is crucial for capturing the 3D information of complex weld seams.
- Weld Trajectory Recognition Algorithm:
- Shape Feature Extraction: The algorithm first extracts the welding seam trajectory using a difference method. Then, it constructs a "characteristic triangle" from the contour curve and extracts two key shape features: the inscribed radius (representing global shape) and the concavity-convexity (representing local details).
- Fourier Analysis: These shape features are then processed using discrete Fourier analysis to create a "Fourier shape descriptor." This descriptor is invariant to the starting point of the sampling, making it robust.
- Euclidean Distance Classification: Finally, the system classifies the weld trajectory by calculating the Euclidean distance between the test sample's Fourier shape descriptor and a pre-defined "class matching library." This library is built using a least-squares method on training samples.
Datasets
The paper does not explicitly mention pre-existing datasets for training or evaluation. Instead, the authors generated their own data:
- Training Data: Ten sets of welding seam trajectory data for Categories 1, 2, and 3 were randomly selected and used as training samples to build the classification and matching library.
- Test Data: Fifty sets of data for each category (totaling 150 sets) were used as test samples to evaluate the classification accuracy.
- Field Test Data: Data from a real-world foldable special container production line was used for validation.
Key Insights and Ideas
- Adaptive Sensing is Key: The adaptive angle adjustment of the laser sensor is a critical innovation, allowing it to handle diverse geometries without recalibration, a major bottleneck in current systems.
- Shape Descriptors for Robustness: Using shape features (inscribed radius and concavity-convexity) and then transforming them with Fourier analysis provides a robust way to represent weld trajectories, making the system less sensitive to noise and variations.
- Euclidean Distance for Simplicity and Speed: Employing Euclidean distance for classification is computationally efficient, enabling real-time processing and achieving a low overall processing time.
- 100% Accuracy on Tested Categories: The algorithm's ability to achieve perfect classification on the tested categories of spatially intermittent welds is a strong indicator of its effectiveness.
Key Results
- Classification Accuracy: The proposed method achieved a 100% classification accuracy rate for four typical spatial discontinuous welds in complex box girder structures.
- Processing Time: The overall processing time for weld trajectory detection and classification did not exceed 65 ms, making it suitable for real-time applications.
- Field Test Performance: The system successfully identified discontinuous welds during high-speed MAG welding (1.2 m/min) and guided the welding torch to automatically complete skip welding, significantly improving manufacturing efficiency and quality stability.
- No Pre-Welding Teaching Required: The system eliminates the need for time-consuming pre-welding teaching programming and visual inspection system calibration.
Limitations and Future Work
- Limited Categories Tested: While 100% accuracy was achieved for the tested categories, the system's performance on a wider variety of weld seam types and complex structures needs further investigation.
- Classification Library Scalability: The paper notes that the processing time increases with the number of classes in the matching library.
- Future Work: The authors propose to improve the intelligent classification library using advanced machine learning algorithms to handle more types of weld trajectories and adapt to a broader range of complex welding structures. They envision its application in heavy industries like port cranes, large logistics equipment, and rail transit.
Summary updated