TL;DR
This paper introduces a novel vision-based method for accurately detecting narrow butt joints and their 3D orientation, crucial for robotic welding. By combining a crosshair laser projection with an auxiliary light source and intelligent image processing, the system achieves high precision. This method is then integrated into a robotic seam tracking system, where a Kalman filter significantly improves tracking accuracy by smoothing noisy sensor data, outperforming a simple moving average.
What Problem Does This Paper Address?
Precise alignment of a welding torch with a joint is paramount for ensuring weld quality. While robotic welding offers automation, actual joint trajectories can deviate from programmed paths due to manufacturing and assembly errors. Existing methods for detecting narrow butt joints (less than 0.2 mm wide) often struggle with accuracy or only provide 2D information. However, for narrow joints with 3D trajectories, knowing the joint's 3D position and the workpiece surface's normal vector is essential for maintaining the correct torch orientation and achieving high-quality welds. This paper addresses this gap by proposing a robust vision-based system capable of extracting both 3D position and orientation information for narrow butt joints, and integrating it into a real-time seam tracking system.
Methods
The core of the proposed method involves a vision-based detection system using a crosshair laser and an LED auxiliary light source. Here's a breakdown of the key techniques:
- Image Acquisition: The system captures images with an auto-exposure function to ensure appropriate grayscale distribution for both the joint and laser stripe regions. The illumination intensity of the auxiliary light is carefully controlled to be weaker than the laser to ensure clear differentiation.
- Image Processing:
- Thresholding: Histograms of the Region of Interest (ROI) are analyzed to determine thresholds for binarizing the image into laser stripe and joint regions. A Gaussian filter smooths the histogram to eliminate false valleys.
- Morphology Operations: A "close" operation is used to connect disconnected small regions in the binarized images.
- Connected Domain Extraction and Selection: Connected components are extracted and filtered based on area and circularity ratio to retain only relevant features (laser stripes and joint lines). This is performed only for the first frame to improve speed.
- Valid Point Extraction: Line segments longer than a threshold are identified, and their midpoints are marked as valid points for subsequent line extraction.
- Line Extraction: A Hough transform is employed to detect straight lines representing the laser stripes and the joint. A "rough" Hough transform is used for the first frame to quickly estimate line positions, while a "precise" Hough transform with reduced parameter ranges is used for subsequent frames to improve accuracy and speed.
- 3D Coordinate and Normal Vector Calculation: By combining the 2D information from the joint region with the 3D information from the laser stripe (which represents the workpiece surface), the system calculates the 3D position of the joint and the normal vector of the workpiece surface. This involves camera calibration and the use of light plane equations.
- Robotic Seam Tracking: The detected 3D position and orientation in the camera coordinate system are transformed into the robot's base coordinate system. For real-time seam tracking, the detected results are smoothed to reduce dithering. Two smoothing methods are compared:
- Moving Average (MA): A simple averaging of recent detected results.
- Kalman Filter (KF): A more sophisticated recursive filter that uses a state-space model to predict and update estimates, effectively smoothing noisy measurements.
Datasets
The paper does not explicitly mention using pre-existing datasets for training or evaluation. Instead, it describes experiments conducted using custom-prepared plane and curved workpieces with narrow butt joints (width < 0.2 mm). The theoretical results for comparison were obtained from drawings of these workpieces.
Key Insights and Ideas
- Unified 2D/3D Detection: The method's strength lies in its ability to extract both 2D joint information and 3D surface orientation simultaneously, which is critical for complex welding scenarios.
- Robustness to Narrow Joints: The approach is specifically designed to handle narrow butt joints where traditional structured light methods might fail due to the lack of significant stripe deformation.
- Adaptive Hough Transform: Using a rough Hough transform for the initial frame and a precise one for subsequent frames is an effective strategy for balancing speed and accuracy in real-time applications.
- Kalman Filter for Dithering Reduction: The significant improvement in tracking accuracy achieved by the Kalman filter over a simple moving average highlights the importance of robust filtering for real-time robotic control systems dealing with noisy sensor data.
- Illumination Control is Key: The paper emphasizes the importance of carefully controlling the auxiliary light source's intensity relative to the laser to ensure clear image segmentation, even in challenging lighting conditions.
Key Results
- Joint Detection Accuracy: The proposed method achieved a position error not exceeding ±0.15 mm and an angle error not exceeding ±1.5° for narrow butt joints.
- Seam Tracking Improvement: The Kalman filter significantly reduced robot dithering and improved tracking accuracy compared to the moving average method.
- Processing Speed: The system's processing speed (around 44.7 ms on average for detection and smoothing) meets the real-time requirements for welding speeds up to 10 mm/s.
Limitations and Future Work
- Lighting Conditions: While the method reduces requirements compared to some previous approaches, it still relies on specific lighting conditions and may be sensitive to extreme variations or strong ambient light. The paper notes that the method requires images with "little disturbance," making it best suited for welding processes with minimal spatter, like GTAW.
- Diffuse Reflection Surfaces: The paper mentions that achieving desired high grayscale for specular reflection workpieces is easier than for diffuse reflection surfaces, where illumination intensity might be insufficient.
- Future Work: The paper implicitly suggests that further improvements in robustness to varying lighting conditions and different workpiece materials could be explored. The successful application in GTAW also opens avenues for adapting this system to other welding processes.
Summary updated