Robot space curve welding seam automatic tracking welding method based on 3D vision

Abstract

The invention relates to the technical field of robot welding, in particular to a robot space curve welding seam automatic tracking welding method based on 3D vision. Comprising the following steps that S1, a 3D visual system extracts workpiece image information; S2, a workpiece model is subjected to three-dimensional reconstruction; S3, a welding path is obtained through a neural network algorithm; S4, adaptive trajectory planning is conducted; the advanced 3D visual system and the high-precision robot control technology are utilized, and automatic recognition, tracking and high-quality welding of complex space curve welding seams are achieved; according to the technology, a 3D image of a welding object is collected in real time, a welding seam path is accurately positioned by applying a deep learning and image processing algorithm, then a robot is commanded to dynamically adjust postures and welding parameters to adapt to changes of a space curve, and it is ensured that the welding process is continuous, stable and efficient; the method is particularly suitable for welding of complex structural parts in the high-precision and advanced fields such as aerospace and automobile manufacturing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot welding, in particular to a robot space curve welding seam automatic tracking welding method based on 3D vision. BACKGROUND

[0002] Traditional space curve welding seam operation faces problems such as difficult positioning, low trajectory control precision, and unstable welding quality. With the integration of 3D vision technology and intelligent robot technology, it is urgent to develop an intelligent system that can automatically adapt to and weld complex space curve welding seams, which is an urgent demand of the industry. The more general algorithm realizes higher precision of welding seam recognition and tracking on the generated model, saves the welding seam information as model information, so that the same workpiece does not need to be modeled again. Reducing manual intervention, suitable for more complex and variable non-standard scenes, so that the welding robot can be effectively applied to intelligent welding in non-standard fields. SUMMARY

[0003] Therefore, the present application provides a robot space curve welding seam automatic tracking welding method based on 3D vision, which improves welding efficiency and quality and reduces manual intervention to solve the above technical problems.

[0004] In order to achieve the above purpose, the present application adopts the following technical scheme: a robot space curve welding seam automatic tracking welding method based on 3D vision, comprising the following steps:

[0005] S1, the 3D vision system extracts workpiece image information:

[0006] By integrating high-resolution 3D cameras and depth sensors, combining advanced image processing, using opencv to pre-process the image;

[0007] S2, three-dimensional reconstruction of workpiece model:

[0008] Using deep learning algorithm, realizing fast and accurate identification and three-dimensional reconstruction of welding seam contour;

[0009] S3, obtaining welding path through neural network algorithm:

[0010] The obtained original data is obtained through the neural network algorithm to obtain the welding point coordinates, and then the welding path is determined;

[0011] S4, adaptive trajectory planning:

[0012] The welding seam data is generated through the algorithm to generate a feasible path after obstacle avoidance, and then the trajectory is generated;

[0013] S5, welding parameter optimization:

[0014] According to the welding seam characteristics and the current position of the robot, the welding current, voltage, speed and other parameters are intelligently adjusted.

[0015] S6、closed loop quality control:

[0016] First, integrate the welding process monitoring and quality evaluation module; then, automatically identify the characteristics of the weld by training the convolutional neural network (CNN) model; detect key indicators such as penetration and weld width.

[0017] By adopting the above technical scheme, high-resolution 3D vision, deep learning and neural network algorithm are integrated to realize workpiece image acquisition, weld three-dimensional reconstruction, intelligent welding path planning, adaptive trajectory generation and welding parameter optimization. Combined with convolutional neural network, real-time monitoring and quality evaluation of the welding process are realized, forming a closed loop control mechanism. This technical scheme can significantly improve the welding precision and automation level, adapt to complex working conditions, ensure the stability and consistency of the welding quality, and greatly improve the production efficiency and intelligent level.

[0018] Preferably, the pre-processing in step S1 specifically includes the following steps:

[0019] S11, convert the color image to a grayscale image to reduce the computational complexity, and use Gaussian filtering to remove noise and smooth the image;

[0020] S12, edge detection, use the Sobel operator to perform convolution operation on the image to extract gradient information;

[0021] S13, feature extraction, match the pre-defined template with the region in the image to extract the texture features of the image;

[0022] S14, weld separation, starting from the seed point, gradually expanding to the adjacent pixels to form a connected region, and extracting the weld region.

[0023] By adopting the above technical scheme, the system first converts the color image to a grayscale image and performs Gaussian filtering to reduce noise and improve image smoothness; then uses the Sobel operator for edge detection to accurately extract image gradient information; then extracts key texture features through template matching to enhance image recognition capability; finally, uses region growing method to separate the weld region from the seed point to form a continuous and clear weld contour, thereby realizing accurate extraction and structured analysis of the weld region, and providing reliable basic data for welding path planning and quality control.

[0024] Preferably, in step S2, the weld image features are automatically extracted through convolutional layers, pooling layers and fully connected layers.

[0025] By adopting the above technical scheme, efficient and accurate identification of weld features can be realized to provide deep feature support for intelligent welding decision-making.

[0026] Preferably, the step S3 specifically includes the following steps:

[0027] S31, first segment scanning of the weld, and obtaining original data;

[0028] S32, using a CNN convolutional neural network algorithm to extract feature points and then obtain weld point coordinates;

[0029] S33, obtaining a welding path by interpolating the weld points.

[0030] By adopting the above technical scheme, the original data is obtained by segment scanning of the weld, the key feature points are extracted and the weld point coordinates are determined in combination with the convolutional neural network (CNN), and then the continuous and smooth welding path is generated through the interpolation algorithm, so that the automatic generation and fine planning of the welding path can be realized, and the welding precision and operation efficiency are improved.

[0031] Preferably, the step S4 comprises the following steps:

[0032] S41, first obtaining the weld data according to the feedback of the 3D vision system;

[0033] S42, then using an RRT probabilistic path planning algorithm to generate a feasible path after obstacle avoidance;

[0034] S43, then generating a smooth and continuous trajectory by adjusting the position and weight of the control points of the B-spline curve.

[0035] By adopting the above technical scheme, the weld data fed back by the 3D vision system is obtained, a feasible path after obstacle avoidance is generated in combination with the RRT probabilistic path planning algorithm, and the trajectory is smoothed and optimized by adjusting the position and weight of the control points by using the B-spline curve, so that a continuous, highly reachable and high-precision welding trajectory can be effectively generated, and the welding head can also stably and accurately run along the optimal path in a complex space environment.

[0036] Advantages of the present application:

[0037] The present application utilizes advanced 3D vision systems and high-precision robot control technology to realize automatic identification, tracking and high-quality welding of complex spatial curve welds. The technology collects 3D images of the welding object in real time, accurately locates the weld path by using deep learning and image processing algorithms, and then commands the robot to dynamically adjust the posture and welding parameters to adapt to the changes of the spatial curve, ensuring continuous, stable and efficient welding process. It is especially suitable for welding complex structural parts in high-precision fields such as aerospace and automobile manufacturing, reduces manual intervention and labor intensity, and is suitable for welding scenes of various complex spatial curve welds. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a method flowchart of the present application;

[0039] Figure 2 is the installation schematic diagram of the 3D vision sensor of the present application;

[0040] Figure 3 is the image processing flow schematic diagram of the present application;

[0041] Figure 4 is the control unit logic schematic diagram of the present application;

[0042] Figure 5 is the installation position and welding area schematic diagram of the 3D vision sensor of the present application;

[0043] Figure 6 is the flow chart schematic diagram of the image processing unit processing the weld feature of the present application;

[0044] Figure 7 is the 3D photographing modeling schematic diagram of different types of workpieces of the present application;

[0045] Figure 8 is the weld angle coverage schematic diagram of the present application;

[0046] Figure 9 is the visual deep groove recognition schematic diagram of the present application. DETAILED DESCRIPTION

[0047] In order to further explain the technical solutions of the present application, combined with the accompanying drawings Figure 1 - the accompanying drawings Figure 9 , the following will be described in detail through specific embodiments.

[0048] The present application provides a robot space curve weld automatic tracking welding method based on 3D vision, comprising the following steps:

[0049] S1, the 3D vision system extracts the workpiece image information:

[0050] By integrating high-resolution 3D camera and depth sensor, combined with advanced image processing, using opencv to pre-process the image;

[0051] S2, three-dimensional reconstruction of workpiece model:

[0052] Using deep learning algorithm, realizing the fast and accurate identification and three-dimensional reconstruction of the weld contour;

[0053] S3, obtaining the welding path through neural network algorithm:

[0054] The obtained original data is obtained through neural network algorithm to get the welding point coordinates, and then the welding path is determined;

[0055] S4, adaptive trajectory planning:

[0056] The weld data is generated through the algorithm to generate the feasible path after obstacle avoidance, and then the trajectory is generated;

[0057] S5, welding parameter optimization:

[0058] According to the weld characteristics and the current position of the robot, the welding current, voltage, speed and other parameters are intelligently adjusted to adapt to different materials, thicknesses and curve curvature changes, and to ensure the welding quality;

[0059] S6, closed-loop quality control:

[0060] First, integrate the welding process monitoring and quality evaluation module; then automatically identify the characteristics of the weld by training the convolutional neural network (CNN) model; detect key indicators such as weld penetration and weld width to achieve closed-loop control and real-time adjustment of welding quality.

[0061] Further, through the above steps, high-resolution 3D vision, deep learning and neural network algorithms are integrated to realize workpiece image acquisition, weld three-dimensional reconstruction, welding path intelligent planning, adaptive trajectory generation and welding parameter optimization. Combined with convolutional neural network for real-time monitoring and quality evaluation of the welding process, a closed-loop control mechanism is formed. This technical solution can significantly improve welding precision and automation level, adapt to complex working conditions, ensure welding quality stability and consistency, and greatly improve production efficiency and intelligent level.

[0062] The preprocessing in step S1 specifically includes the following steps:

[0063] S11, convert the color image to a grayscale image to reduce computational complexity, and use Gaussian filtering to remove noise and smooth the image;

[0064] S12, edge detection, use the Sobel operator to perform convolution operation on the image to extract gradient information;

[0065] S13, feature extraction, match the pre-defined template with the region in the image to extract the texture features of the image;

[0066] S14, weld separation, starting from the seed point, gradually expanding to the adjacent pixels to form a connected region, and extracting the weld region.

[0067] Further, through the above steps, the system first converts the color image to a grayscale image and performs Gaussian filtering to reduce noise and improve image smoothness; then uses the Sobel operator for edge detection to accurately extract image gradient information; then extracts key texture features through template matching to enhance image recognition ability; finally, the region growing method is used to separate the weld region from the seed point to form a continuous and clear weld contour, thereby realizing accurate extraction and structured analysis of the weld region, and providing reliable basic data for welding path planning and quality control.

[0068] The image features of the weld are automatically extracted by the convolution layer, the pooling layer and the full connection layer in step S2, so that efficient and accurate identification of the weld features is realized, and deep feature support is provided for intelligent welding decision.

[0069] The specific steps of step S3 are as follows:

[0070] S31, first, the weld is segmented and scanned to obtain original data;

[0071] S32, the feature points are extracted using the CNN convolution neural network algorithm, and then the welding point coordinates are obtained;

[0072] S33, the welding point is interpolated to obtain the welding path.

[0073] As can be further known through the above steps, by segmenting and scanning the weld to obtain original data, combining the convolution neural network (CNN) to extract key feature points and determine the welding point coordinates, and then generating a continuous and smooth welding path through the interpolation algorithm, the automatic generation and fine planning of the welding path can be realized, and the welding precision and operation efficiency are improved

[0074] The specific steps of step S4 are as follows:

[0075] S41, first, the weld data is obtained according to the feedback of the 3D vision system;

[0076] S42, then, the RRT (Rapidly-exploring Random Tree) probabilistic path planning algorithm is used to generate a feasible path after obstacle avoidance;

[0077] S43, the position and weight of the control point of the B-spline curve are adjusted to generate a smooth and continuous trajectory, so as to ensure that the welding head moves along the optimal path, and even in the face of nonlinear space curves, the precise tracking can be maintained.

[0078] As can be further known through the above steps, by obtaining the weld data fed back by the 3D vision system, combining the RRT probabilistic path planning algorithm to generate a feasible path after obstacle avoidance, and using the B-spline curve to adjust the position and weight of the control point to realize trajectory smoothing and optimization, a continuous, highly reachable and high-precision welding trajectory can be effectively generated, so that the welding head can also stably and accurately run along the optimal path in a complex space environment.

[0079] As shown in Figure 7 Without importing a model, the model is autonomously generated, and the equal-section and variable-section H-shaped steel can be modeled and identified. Different models of welding robots can only autonomously identify the equal-section H-shaped steel, and through accurate trajectory planning, adaptive path adjustment, and perfect weld angle realization, as shown in Figure 8 .

[0080] AsFigure 9 As shown, the present application performs multi-layer multi-pass welding through visual deep groove recognition, relies on laser 3D vision technology to solve the problem of difficult detection in narrow grooves, visually recognizes and plans intermittent welding, autonomously plans intermittent welding paths and welding sequences for workpieces with strict heat deformation requirements, can reduce welding heat deformation, establishes a reasoning model of welding deformation and welding sequence, and the robot autonomously learns to give an optimized shooting and welding sequence workflow through AI algorithms, which not only eliminates position deviation caused by welding deformation and poor point fixation, but also optimizes the overall efficiency and deformation of the workpiece.

[0081] The present application provides a kind of robot space curve weld automatic tracking welding method based on 3D vision, utilizes advanced 3D vision system and high-precision robot control technology, realizes the automatic identification, tracking and high-quality welding of complex space curve weld.The technology acquires the 3D image of welding object in real time, accurately locates weld path using deep learning and image processing algorithm, then commands robot to dynamically adjust posture and welding parameters to adapt to the change of space curve, ensures continuous, stable and efficient welding process, especially suitable for complex structure welding in high-precision fields such as aerospace and automobile manufacturing, reduces manual intervention and labor intensity, and is suitable for welding of various complex space curve welds.

[0082] The above is only the preferred example of the present application and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or make equivalent replacement for part of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A 3D vision-based robot space curve weld automatic tracking welding method, characterized in that: The following steps are involved: S1. 3D vision system extracts workpiece image information: By integrating high-resolution 3D cameras and depth sensors, combined with advanced image processing, and using OpenCV to pre-process images; S2. 3D reconstruction of workpiece model: Using deep learning algorithms to achieve fast and accurate recognition and 3D reconstruction of weld contours; S3. Obtain welding path through neural network algorithm: The acquired raw data is used through a neural network algorithm to obtain the coordinates of the welding points and then determine the welding path; S4, Adaptive trajectory planning: The weld data is used through an algorithm to generate a feasible path after obstacle avoidance, and then a trajectory is generated; S5. Welding parameter optimization: Intelligently adjust welding current, voltage, speed and other parameters according to the weld characteristics and the current position of the robot; S6. Closed-loop quality control: First, the welding process monitoring and quality assessment module is integrated; then, by training the convolutional neural network (CNN) model, the characteristics of the weld are automatically identified; and key indicators such as penetration depth and weld width are detected.

2. The 3D vision-based robot space curve weld automatic tracking welding method according to claim 1, characterized in that: The pre-processing in step S1 specifically includes the following steps: S11, convert the color image into a grayscale image to reduce the computational complexity, use Gaussian filtering to remove noise, and smooth the image; S12, edge detection, use the Sobel operator to perform convolution operation on the image and extract gradient information; S13, feature extraction, matching a predefined template with an area in the image to extract texture features of the image; S14, weld segmentation, starting from the seed point, gradually expanding to the adjacent pixels to form a connected area, and extracting the weld area.

3. The 3D vision-based robot space curve weld automatic tracking welding method according to claim 1, characterized in that: In step S2, weld image features are automatically extracted through convolutional layers, pooling layers, and fully connected layers.

4. The 3D vision-based robot space curve weld automatic tracking welding method according to claim 1, characterized in that: The specific steps of step S3 are: S31, first scan the weld in sections to obtain original data; S32, using CNN convolutional neural network algorithm to extract feature points and then obtain welding point coordinates; S33, interpolating the welding points to obtain a welding path.

5. The 3D vision-based robot space curve weld automatic tracking welding method according to claim 1, characterized in that: The specific steps of step S4 are: S41, first obtain weld data based on feedback from the 3D vision system; S42, then using the RRT probabilistic path planning algorithm to generate a feasible path after obstacle avoidance; S43 then generates a smooth and continuous trajectory by adjusting the positions and weights of the B-spline curve control points.

Images