**Summary:**
This paper presents a method for real-time tracking of pipe weld grooves under natural lighting conditions. It employs two-dimensional wavelet transform and binarization techniques to process the initial weld groove image, extracting texture information as a reference template. The same processing is applied to subsequent images, allowing pattern recognition methods to identify the basic position and centerline of the weld groove. This approach provides a theoretical foundation for real-time tracking of weld grooves in pipeline welding applications.
**Keywords:** 2D wavelet transform, template matching, pattern recognition
**Foreword**
Pipeline transportation is a safe, economical, and environmentally friendly method of moving oil and gas. Over the next decade, China plans to build 14 major oil and gas pipelines, forming a network covering over 10,000 kilometers. These pipelines will span harsh environments and operate at high pressures (up to 7.5 MPa) and large diameters (up to 1,420 mm), which increases the complexity of welding processes. The quality of circumferential welds is crucial for both project safety and construction efficiency. To meet these challenges, many companies are investing in automated welding technologies that improve accuracy, reduce labor, and speed up construction.
Automatic tracking of the weld groove is essential for successful automation. This study focuses on identifying circular weld grooves using image processing and pattern recognition techniques. Traditional methods based on gray-level changes often fail in noisy or low-contrast conditions. Pattern recognition offers a more robust solution, especially when combined with advanced signal processing like wavelet transforms.
In this work, we first capture the initial weld groove image using a CCD sensor. Then, we apply a two-dimensional wavelet transform to extract texture features, followed by binarization to create a reference template. Subsequent images are processed similarly, and template matching is used to locate the groove and determine its centerline. This four-step process ensures accurate and efficient real-time tracking of weld grooves.
**1. Image Acquisition of the Groove**
To capture the weld groove, an area array CCD sensor is used. The acquired analog image is converted into digital format using a Matrox Meteor-II/Standard image acquisition card. An example of a captured image is shown in Figure 1, where the torch is properly aligned with the groove.
**Figure 1: One frame of the weld groove image**
**2. Creating the Groove Template**
The template is built using the initial image, taking into account the geometric properties of the groove and the alignment of the camera and torch. A B-spline wavelet transform is applied to extract texture features, and the optimal scale is determined through an optimization algorithm. The resulting binary image serves as the template for future comparisons.
**3. Identifying the Circular Groove Using Template Matching**
The template is matched against subsequent images using cross-correlation. When a peak is detected in the correlation result, it indicates the location of the groove. This allows for precise identification of the groove's centerline and position.
**4. Outputting the Position Data**
Once the groove is located, the horizontal deviation between the torch and the groove center is calculated. This data is then converted into serial output, providing real-time feedback for the welding system.
**5. Conclusion**
This study demonstrates that two-dimensional wavelet transform and pattern recognition can effectively extract and track weld groove features in real time. By creating dynamic templates and using template matching, the system achieves accurate and reliable results, even in challenging environments. This method lays the groundwork for further advancements in automated pipeline welding systems.
Nantong Zhongyi Electronic Technology Co., Ltd. , https://www.apl-pickleball.com