Weld seam recognition and tracking technology is an important direction for the future development of welding robots, and plays a key role in improving the automation and intelligence level of welding robots. This paper systematically explains the relevant technical characteristics, domestic and foreign development status and future development trends from the aspects of robot sensing technology, weld seam recognition and feature extraction technology, and weld seam tracking control technology. Real-time recognition and feature extraction technology of weld seams is the core of weld seam recognition and tracking control system, and effective noise processing is the key to improving recognition and extraction accuracy. Active sensing based on laser vision, advanced weld seam feature extraction algorithm, image denoising technology and stable and reliable tracking control system are important guarantees for realizing efficient and stable operation of weld seam recognition and tracking system. The ability to recognize multiple types of weld seams and have good adaptability and anti-interference ability are important foundations for promoting the widespread application of weld seam recognition and tracking technology; multi-weld seam recognition technology and multi-level feature extraction intelligent learning algorithm are the key directions for future development.

1. Introduction

As an important connection process in the field of modern manufacturing, welding is widely used in the processing of multiple varieties and types of materials in various industries. Traditional manual welding has high requirements for operators and low efficiency, which is difficult to meet the efficient and high-quality production needs of modern industry. With the development of semi-automatic welding technology, the level of welding automation has improved, but it still relies on a lot of manual intervention and is difficult to adapt to the challenges of diversified and complex welding products.

The emergence of welding robots has greatly improved welding efficiency and flexibility, reduced production costs, and promoted the rapid development of the welding field. When the weld form is simple and the workpiece position is fixed, the traditional robot teaching programming method can still meet the needs. However, in the face of complex and changeable weld trajectories, ordinary teaching methods require a lot of manual teaching, which is difficult to meet the needs of small-batch and diversified welding production, limiting the further popularization of robot welding technology.

In recent years, the emergence of weld recognition and tracking technology has significantly promoted the development of robot welding. This technology can actively identify different weld characteristics, realize robot autonomous teaching welding, greatly improve the stability and efficiency of welding, and promote the widespread application and development of robot welding technology.

Because weld autonomous recognition and tracking has high requirements for detection accuracy and trajectory planning, and involves multidisciplinary cross-disciplinary knowledge, the technology is not yet fully mature and needs further in-depth research. Based on the above background, this paper elaborates on the development of weld recognition and tracking technology from the aspects of sensor technology, weld recognition and feature extraction, and weld tracking control.

2. Characteristics of weld seam recognition and tracking technology

Weld seam recognition and tracking technology mainly covers three aspects: weld seam detection and identification, weld seam feature extraction and weld seam tracking control. First, the weld seam feature type is identified through a specific sensor; second, the weld seam feature is converted into three-dimensional coordinate information using an image processing algorithm; finally, the robot establishes a mathematical model based on the real-time weld seam information, dynamically adjusts the welding gun position, and achieves efficient and high-quality welding.

Real-time and accurate weld seam feature recognition and extraction is the core of this technology. Only by quickly and accurately obtaining the weld seam position and converting different types of weld seam features into robot-recognizable data can the robot welding process be adjusted autonomously without manual intervention, meeting the welding needs of multiple occasions and multiple tasks. Therefore, weld seam tracking sensors with simple structure, stable process, high sensitivity and efficient feature extraction algorithms are the key to ensuring weld seam recognition and tracking accuracy and welding quality.

3. Research status of weld seam recognition and tracking technology

3.1 Research progress of sensor technology

(1) Sensor classification

Common sensors for welding robots are divided into arc sensors, contact sensors and non-contact sensors according to the contact method. In traditional applications, arc and contact sensors are widely used due to their fast response speed and low cost, but their accuracy is limited and it is difficult to meet the needs of complex welds. Ultrasonic, infrared and visual sensing technologies have gradually developed. In particular, visual sensing has become the mainstream technology for weld tracking due to its rich information, strong anti-interference ability, high sensitivity and non-contact advantages.

Visual sensing can be divided into two categories: active vision and passive vision. Active vision uses structured light such as lasers and halogen tungsten lamps to extract three-dimensional features, effectively overcoming the interference of natural light and becoming the mainstream of weld tracking; passive vision directly uses the characteristics of welding light sources, which are easily affected by spatter and arc light, resulting in poor imaging quality and difficulty in post-image processing.

(2) Laser sensor

Laser vision sensors project lasers onto the surface of welds to form stripes, and the visual sensor captures the image, extracts weld feature points, and assists the welding robot in real-time calibration of the welding position. Laser stripes include single-line, multi-line, cross, and ring, and different structures are suitable for different weld shapes. Multi-line laser structured light sensors increase the number and accuracy of feature points. Ring lasers are suitable for complex trajectory welds, and cross lasers increase the number of feature points to improve tracking effects.

At present, laser vision sensors are widely used in welding robot weld tracking systems due to their high precision and stability. Although there are related product developments in China, there is still a certain gap with the international advanced level.

3.2 Weld seam recognition and extraction technology

Weld seam recognition and extraction is based on images obtained by visual sensing. Recognizing effective weld seam features through image processing algorithms is the key to automatic tracking of welding robots. Efficient, real-time, and anti-interference image processing algorithms are crucial to improving the accuracy of weld seam recognition.

In recent years, a variety of weld seam recognition algorithms have been proposed, including methods based on hidden Markov models, three-line structured light recognition, laser scanning displacement sensor recognition, etc., which are designed for different weld seam features and complex environments. At the same time, recognition algorithms for multiple types of welds are also constantly developing to adapt to complex and changeable welding scenes.

There are various methods for extracting weld seam stripes, such as extreme value method, threshold method, grayscale centroid method, Hough transform, etc., combined with adaptive threshold, random Hough transform and other technologies, which effectively improve the accurate extraction of laser stripe centerline. In addition, for the weld groove structure, feature point extraction methods of different geometric shapes, such as slope analysis and straight line intersection method, are also widely used.

Strong arc light and spatter in complex welding environment will seriously affect the image quality, and denoising technology becomes the key. Commonly used image filters include median filtering, mean filtering, and Gaussian filtering, among which median filtering takes into account both denoising and edge protection. Wavelet analysis and deep learning technology have also been introduced into weld image denoising and feature extraction, improving anti-interference ability and recognition accuracy.

Deep learning-based weld recognition technology has made significant progress in recent years. The automatic positioning and tracking of feature points are realized through convolutional neural networks, which improves the intelligence and adaptability of the system.

3.3 Robot tracking control technology

The welding robot adjusts the welding gun position in real time according to the identified weld feature information to ensure the accuracy of the welding trajectory. The deviation between the weld recognition information and the expected trajectory is compensated by the control system, which is an important link in improving welding quality.

Traditional control methods include iterative learning control, task space division and robust control. Although the structure is simple, the adaptability to external disturbances such as additional loads on the welding gun is limited. Trajectory replanning and end position posture adjustment have become research hotspots.

Visual target tracking technology has been introduced into weld tracking, and the accuracy and stability of tracking have been improved through methods such as particle filtering and Bayesian framework. Intelligent control modules such as Cartesian spatial position correction system combined with PID and adaptive PID control algorithms have significantly improved the dynamic response and tracking accuracy of the system.

Existing weld tracking systems have been widely used in industry and have the functions of multi-weld recognition and real-time adaptive adjustment, but their robustness in complex environments still needs to be further enhanced.

4. Future development trends

(1) Sensor integration and intelligence

Integrate multimodal sensing technologies such as laser, infrared, vision and force feedback sensors to improve the accuracy and robustness of weld detection.

(2) Weld recognition based on deep learning

Use deep neural networks trained with large-scale welding data to achieve accurate recognition and real-time tracking of multiple types of welds.

(3) Intelligent tracking control algorithm

Combining machine learning and predictive control technology, it realizes intelligent planning and adaptive adjustment of welding paths, and improves the flexibility and reliability of robot welding.

(4) Overall system optimization design

Promote software and hardware collaborative design, improve system response speed and stability, and promote intelligent welding technology to move closer to Industry 4.0 standards.