CN116175035A - Intelligent welding method for steel structure high-altitude welding robot based on deep learning - Google Patents

Abstract

The invention discloses an intelligent welding method of a steel structure high-altitude welding robot based on deep learning, and belongs to the technical field of welding. According to the welding robot disclosed by the invention, the image is shot by carrying the depth camera, the shot image is sent to the industrial computer for recognition and ranging, when a welding line is detected, the robot is controlled to move towards the position of the welding line, the laser scanning is used for accurately measuring and positioning the position, finally, the mechanical arm is controlled to drive the welding gun to move to the starting point of the designated position, the spatial coordinates corresponding to the point on the welding track are calculated by the industrial computer, and the mechanical arm is controlled to drive the welding gun to move to the corresponding position, so that the full-automatic image acquisition, the motion track planning and the automatic welding work are realized, the labor cost of manual welding investment is avoided, and the danger in manual operation is also reduced. The automatic welding method can be applied to steel beam cracks with different shapes, and is wide in application range.

Description

An intelligent welding method for steel structure high-altitude welding robot based on deep learning

technical field

The invention relates to the field of welding technology, in particular to weld seam detection and mechanical arm motion planning technology, and more specifically, to an intelligent welding method for steel structure high-altitude welding robots based on deep learning.

Background technique

High-altitude steel beam maintenance and welding technology is an indispensable step in modern industrial production, mainly involving the detection and welding of steel beam cracks. There are certain dangers in traditional manual operations, and completely standardized operations cannot be achieved; while the welding manipulators used in the industrial field mainly adopt a fixed design mode, which cannot perform corresponding welding work for different types of welds in steel beams, and industrial manipulators It is large in size and cannot be used on steel beams that require constant replacement of welding positions, making it difficult to be widely used in high-altitude welding operations.

After searching, the patent application number is: 201610116804.7, and the name of the invention is: an 8-axis robot space curve welding system and method for laser recognition of welds; in this application, the robot puts the laser sensor in a position and posture that is easy to scan the workpiece weld, and the laser The feature points of the weld seam are obtained by scanning the sensor, and the edge characteristics of the weld seam are obtained by fitting with the least square method. After geometric calculation, the center point of the weld seam in the coordinate system of the laser sensor is obtained; , to get the discrete center point of the whole weld, use the non-uniform rational B-spline curve to fit the discrete center point, get the parameter equation that uniformly describes the whole weld, and then get the welding speed equation of the whole welding process; according to the speed According to the recursive relationship with the parameters, the weld curve is discretized to obtain a series of interpolation points; according to the welding posture of the interpolation points required by the welding process, the inverse kinematics solution of the tilt/rotation two-axis positioner is solved to obtain the interpolation points Points correspond to the tilt axis angle θ7 and rotation axis angle θ8; solve the forward kinematics solution of the tilt/rotation two-axis positioner to obtain the end position and attitude of the welding torch corresponding to the interpolation point; solve the inverse solution of the robot kinematics to obtain the rotation angle of the robot's six axes θ1, θ2, θ3, θ4, θ5, θ6; the tilt/rotation two-axis positioner and the welding robot coordinate and synchronously move for welding; during the welding process, the laser sensor scans the weld seam and performs real-time weld seam tracking to compensate for welding heat Weld seam position deviation caused by deformation and other factors. However, the above application is aimed at the welding application of an 8-axis robot, and its design focuses on determining the rotation angles θ1, θ2, θ3, θ4, θ5, θ6, θ7, and θ8 of the eight axes of the robot. Medium cost is higher.

Contents of the invention

1. The technical problem to be solved by the invention

In order to solve the problem that the traditional manual welding method requires repeated operations and errors in the welding process when welding steel beam cracks, and the existing welding technology cannot automatically detect and weld non-standard steel beam cracks, the invention provides a depth-based Learn the intelligent welding method of steel structure high-altitude welding robot.

2. Technical solution

In order to achieve the above object, the technical scheme provided by the invention is:

A kind of intelligent welding method of steel structure high-altitude welding robot based on deep learning of the present invention, its steps are:

S1. The welding robot is equipped with a depth camera to capture images, and sends the captured images to an industrial computer for identification and processing;

S2. When the weld seam is detected, the robot is controlled to move to the weld seam position;

S3. For the collected depth image, calculate the distance z of the pixel point according to the xy pixel coordinate value, and calculate the corresponding space coordinate [u, v, z] according to the distance in space and the internal parameters of the camera, and fit the space point to obtain A line segment with direction in the space, this line segment is the welding seam position;

S4. According to the calculated spatial position of the line segment, determine the relative position of the line segment in the laser and chassis coordinate system, identify the weld depth according to the laser scanning, and map the weld depth from the welding start end A to the welding stop end B to Go to the weld node from A to B, and calculate the stay time of the welding head at this node to ensure that the weld is welded accurately;

S5. When the robot recognizes that the end is at point B, stop the welding work, turn off the laser scanning at the same time, reduce the frequency of welding seam detection, and continue the next stage of welding work.

Furthermore, the detection method adopted for the identification in step S1 is deep learning, and the detection network is a convolutional neural network.

Furthermore, in step S2, the position and distance information of the robot's movement is roughly positioned based on the pictures detected by the depth camera, and after the weld is determined to be within the weldable range of the welding robot, fine positioning is started to detect the position of the weld.

Furthermore, in step S2, the depth map acquired by the camera is sent to the convolutional neural network at a rate of 5 frames per second, and the result output is (x, y, θ, w), and the detected results of 30 consecutive frames are set When the standard deviation of xy position is less than 0.003, the detected weld position is accurate.

Furthermore, the process of identifying the weld seam in step S3 is as follows:

(1) Crop the collected depth image to obtain the target area, and use high-pass filtering to repair the pixels with infinite depth;

(2) Set the repaired depth map as an image composed of 1 and 0, wherein the identified weld point is set to 1, and other position areas are set to 0, and the evaluation function Q _T is set for the detection results;

(3) Use the deep learning network to identify and detect the weld, and use the evaluation function Q _T in step (2) as the standard, and use the quick sort method to sort all the detection results, and the values of each group are in accordance with [u, v, θ, w] arrangement, use the depth information extraction module to detect the depth value of the detected weld position sequence, and calculate the spatial coordinates [x, y, z] corresponding to the pixel according to the depth value;

(4) Obtain a series of coordinate position sets with directions according to step (2) and step (3), connect these points and fit these points and directions by nonlinear least squares method to obtain a belt in space The line segment with direction is the real-time position of the weld.

Furthermore, the evaluation function Q _T includes an angle evaluation function Φ _T and a width evaluation function W _T , the angle evaluation function Φ _T is used to find the detection result with the smallest difference between the detected angle and the actual angle, and the width evaluation function W _T is used to Find the detection result with the smallest difference between the detected weld width and the actual width.

Furthermore, the angle evaluation function Φ _T and the _width evaluation function W _T are multiplied by the corresponding weights α and β and added to obtain the evaluation function Q _T . accurate image.

Further, the process of welding the weld seam in step S4 is as follows:

1) Convert the position of the weld seam based on the camera coordinate system to the coordinates based on the welding robot coordinate system;

2) According to the spatial coordinates, calculate the vector of the line segment l connected by points AB and each node l ₁ , l ₂ ... l _n of the line segment l, and calculate the node direction θ based on the robot coordinate system from the AB vector;

3) Use the laser sensor to point to the starting point A, scan once from A to B, project the laser to the depth of the weld, and send the weld depth information to the sensor, so as to obtain the actual depth d ₁ , d ₂ …d _n of the weld, and map it to For each weld joint l ₁ , l ₂ ...l _n , the welding time t is calculated according to the depth of the welding spot, the calculation method is λd _x , and λ is determined by the welding speed;

4) Control the movement of the three-axis mechanical arm to point A, start moving from this point along the direction θ, and start the welding machine at the same time for welding operation, the moving speed of the welding head is controlled by the time required for the node to stay during welding according to step 3);

5) When the robot recognizes that the position of the welding head reaches point B, the arc is turned off to stop welding.

Furthermore, in step 4) during the moving process of the welding head, the proportional integral method is used to control the speed of the welding head.

Furthermore, the high-altitude welding robot includes a robot chassis, a welding gun, a robotic arm, an industrial computer, a depth camera, and a laser sensor. The depth camera and the laser sensor are fixed above the robot chassis and fixed with the welding gun clamping mechanism of the robotic arm. For communication, the industrial computer processes and sends the identified welding seam position, and a strong NdFeB magnet is fixed under each crawler track of the robot chassis.

3. Beneficial effect

Compared with the existing known technology, the technical solution provided by the invention has the following remarkable effects:

The invention automatically welds the cracks of high-altitude steel beams by first positioning, then measuring, and then welding. The welding seam can be fully automatically identified and welded without being accompanied by anyone, which saves a lot of manpower and time costs, and reduces errors caused by manual operations. Compared with the existing robotic arm automatic welding technology, it can not only weld any welding seam, but also can be backward compatible with the design of programmable welding; at the same time, the invention combines laser and image technology to make the welding positioning more accurate and the welding effect better. better.

Description of drawings

Fig. 1 is a schematic diagram of the overall structure of the welding robot of the present invention.

Figure 2 is a schematic diagram of the weld seam.

Explanation of the labels in the schematic diagram:

1. Z-axis guide rail; 2. Welding torch clamping mechanism; 3. Robot chassis; 4. Y-axis guide rail motor; 5. X-axis guide rail motor; 6. Z-axis guide rail motor; 7. Z-axis guide rail platform; 8. Y-axis Rail platform; 9. Depth camera.

Detailed ways

In order to make the functions, features and advantages of the present invention more comprehensible, the following describes specific embodiments of the present invention in conjunction with the accompanying drawings. Apparently, what is described is a part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without making creative efforts shall belong to the protection scope of the present invention.

In the following description, a lot of specific details are set forth in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described here, and those skilled in the art can do it without departing from the meaning of the present invention. By analogy, the present invention is therefore not limited to the specific examples disclosed below. The three-axis coordinate direction targeted by the present invention is the positive direction of the robot in the direction of the robot, and the robot motion plane as the reference, and the positive direction of the Y axis rotated 90° counterclockwise in the direction of the X axis is the positive direction of the Y axis, and the direction perpendicular to the XY direction is the Z axis. Positive direction.

Referring to FIG. 1 , the present invention builds a fully automatic high-altitude steel beam welding robot, including a robot chassis 3, a welding torch, a three-degree-of-freedom gantry-type mechanical arm, an industrial computer, a depth camera, and a laser sensor. Among them, the Z-axis guide rail 1 and the Z-axis guide rail motor 6 of the three-degree-of-freedom gantry-type mechanical arm drive the Z-axis guide rail platform 7 to move in the Z-axis direction, the welding torch clamping mechanism 2 is arranged on the Z-axis guide rail platform 7, and the Y-axis guide rail and The Y-axis guide rail motor 4 drives the Y-axis guide rail platform 8 to move in the Y-axis direction. The Z-axis guide rail 1 is set on the Y-axis guide rail platform 8. The X-axis guide rail and the X-axis guide rail motor 5 drive the Z-axis guide rail 1 and the welding torch clamping mechanism. 2 and the Y-axis guide rail move in the X-axis direction. The motors all use servo motors to control the motion of each axis joint, so that the welding torch can move freely in multiple directions, and the overall welding mechanism is driven by the robot chassis 3 to control the overall welding robot to achieve movement. The depth camera 9 collects images and transmits the collected information to the mechanical arm. The depth camera 9 and the laser sensor are fixed above the robot chassis 3 and fixed to the welding torch clamping mechanism 2 of the robotic arm. The depth measurement accuracy of the depth camera 9 is 10mm, the accuracy of the laser sensor is 0.01mm, and the welding accuracy of the robotic arm can reach 1mm. The robot communicates through TCP/IP Ethernet, processes and sends the identified welding seam position through the computer in it, and controls the mechanical arm to drive the welding machine to perform welding work. The welding robot can automatically move repeatedly on the steel beam. A 40mm*20mm*20mm NdFeB strong magnet is fixed under each track of the welding robot chassis to ensure that the welding robot can be adsorbed on the steel beam.

The detection and welding method of high-altitude steel girder welds involved in the present invention includes steps such as weld identification, positioning, path identification, and robot arm path planning. The entire identification and welding process is fully automated by robots. The specific process is as follows:

Taking the weld shown in Figure 2 as an example, set one end of the weld as point A and the other end as point B. The depth camera mounted on the welding robot is kept turned on, and the captured image is sent to the computer for recognition and processing every 0.5s. The image size is 640*480 and the frame rate is 30fps. Save each frame of image as the detection input, and use the serial number starting from 1 to save the picture. The detection method used for recognition is deep learning, and the detection network is a convolutional neural network, which is mainly composed of convolutional layers, pooling layers, and nonlinear activation layers. When using a pre-trained model for detection, it can Reach 99.5% weld detection rate.

When the welding seam is detected, the robot is controlled to move to the welding seam position. At this time, the crawler chassis can be controlled to rotate through the differential speed. When it is necessary to shift to the left, the left wheel motor is controlled to rotate forward, and the right wheel motor is turned backward. Rotation; when the weld seam is detected to be on the right side, it is necessary to control the robot to move to the right front. At this time, the left wheel motor is controlled to rotate backward, and the right wheel motor is rotated forward; when the detected weld seam is in front of the robot, control the left and right motors of the robot Simultaneously turn backwards.

Control the robot to move to a distance of 40cm from the welding seam, and the position and distance information can be roughly positioned by the pictures detected by the depth camera. After confirming that the weld seam is within the weldable range of the welding robot, start fine positioning to detect the weld seam position. At this time, the robot chassis needs to be stabilized. Send the depth map acquired by the camera to the grasping neural network at a speed of 5 frames per second, and the result is output (x, y, θ, w). If the xy position standard deviation of the detected results for 30 consecutive frames is less than 0.003, The detected weld position is considered to be accurate.

Calculate the specific spatial position of the detected sequence point value according to the principle of pinhole imaging. The specific operation is to calculate the distance z of the point according to the xy pixel coordinate value, and calculate the corresponding spatial coordinate [u, v according to the distance in space and the camera internal parameters. ,z]. Fit these space points according to the nonlinear optimization method, and the result is a directional line segment in space, which can be regarded as the weld position.

According to the calculated spatial position of the line segment, and at the same time, determine the relative position of the line segment in the laser and chassis coordinate system according to the known coordinate relationship between the camera, the laser and the robot chassis. Select the detected end A of the welding seam close to the robot as the welding start end, and the far end B as the welding stop end. Weld depth identification based on laser scanning, mapping the weld depth from A to B to the weld node from A to B, and calculating the length of stay of the welding head at this node according to the formula λd _x to ensure that the weld can be welded correctly stand up.

When the mechanical arm recognizes that the end is at point B, the welding work is stopped. At this time, the computer controls the arc extinguishing operation of the welding head, and at the same time raises the mechanical arm to the Z axis by 5cm, and returns the XY axis to the original position. At the same time, laser scanning is turned off, the frequency of weld detection is reduced, and the welding work of the next stage is continued.

The image processing process of the present invention for weld seam recognition will be described in detail below in conjunction with Embodiment 1, and the welding process of the present invention will be described in detail in conjunction with Embodiment 2.

Example 1

The process of image processing for weld seam recognition in this embodiment is as follows:

Step 1: Use the depth camera to collect images, use the host computer in the welding robot to process the collected images, and cut each frame of the image into a 300*300 square target area. Among them, in order to ensure that the obtained sparse image has distance measurability, high-pass filtering is used to repair the pixels with infinite depth, and the repaired dense image is saved.

Step 2: Set the repaired depth map as an image composed of 1 and 0, where the identified weld points are set to 1, and other position areas are set to 0. Set the evaluation function Q _T for the detection results, divide it into angle evaluation function Φ _T and width evaluation function W _T , multiply the angle evaluation function Φ _T and width evaluation function W _T by the corresponding weights α, β and add them to obtain the final Evaluation function Q _T . The angle evaluation function Φ _T is used to find the detection result with the smallest difference between the detection angle and the actual angle, that is, the evaluation is within the range of [-π/2,π/2], the difference between the detection angle result and the actual angle, using the network learning distribution and Evaluate the corresponding value, the result should be distributed on [-π/2,π/2]. The width evaluation function is used to find the detection result with the smallest difference between the detected weld width and the actual width, and at the same time map the result to the range of 0-1. The final evaluation function is Q _T =αΦ _T+ βW _T , and the most accurate weld width and angle can be obtained according to the evaluation function of each image.

Step 3: Use the deep learning network to identify and detect the weld seam, and obtain the difference between the width and angle of the weld seam of each image and the actual value from step 2, use the difference as a standard, and use the quick sort method to sort all the detection results Sorting, the values of each group are data arranged according to [u, v, θ, w], which respectively represent the width and height pixel values of the weld position in the image, the deflection angle, and the width of the weld.

Use the depth information extraction module in the computer vision library to detect the depth value of the detected weld position sequence, and calculate the spatial coordinates [x, y, z] corresponding to the pixel according to the depth value. The specific method is to convert the depth value z, Add to the corresponding pixel value [u, v], use Eigen to define the camera internal parameter as a two-dimensional array K, the specific parameters are [[f _x ,0,c _x ],[0,f _y ,c _y ],[0 ,0,1]], according to the principle of pinhole imaging, the x value is calculated ((uc _x )*z/f _x ), the y value is ((vc _y )*z/f _y ), and the z value is calculated depth value.

Step 4: According to Step 2 and Step 3, a series of coordinate position sets with directions can be obtained, and each point has a certain width, which is equivalent to a point with width. Connect these points and fit these points and directions through the nonlinear least square method of the Ceres library. The result is a line segment with direction in space, and each point on the line segment has a certain width corresponding to the actual weld width. In this way, any coordinate in the obtained coordinate position set can be detected, and the detection result can basically be output as an actual value, so that the real-time position of the weld can be obtained.

Example 2

In this embodiment, the process of using a welding robot to automatically weld the weld seam is as follows:

Step 1: Select the end of the detected weld line segment close to the robot as the starting point A, and the end far away from the robot as the end point B, and convert the position of the weld based on the camera coordinate system to the coordinates based on the welding robot coordinate system. Measure the relative position T ₁ of the camera and the robot base, the relative position of the weld in the camera coordinate system is T ₂ , and the position of the weld in the robot coordinate system is T ₃ = T ₁ * according to the transformation relationship of the rigid body space coordinates T ₂ , wherein T ₃ is a 4*4 transformation matrix, which is obtained by combining a 3*3 rotation matrix and a 3*1 translation matrix. In this way, the position of the welding seam is transformed into coordinates based on the robot coordinate system.

Step 2: Set the starting point in step 1 as A and the end point as B, and calculate the vector of the line segment l connected by the points AB according to its space coordinates, each node of the line segment l is l ₁ , l ₂ ...l _n , from the AB vector The calculation is based on the node direction θ=(r,p,y) of the robot coordinate system, and the weld is expressed as a directed line segment with A as the starting point, B as the end point, and the direction as θ. When welding to point B, it can be turned off. Arc operation stops welding.

Step 3: Use the laser sensor on the robotic arm to point to the starting point A calculated in step 2, scan from A to B, project the laser to the depth of the weld, and send the weld depth information to the sensor to obtain the actual depth d of the weld ₁ ,d ₂ ...d _n , mapped to each weld node l ₁ ,l ₂ ...l _n , the welding time t needs to be calculated according to the depth of the welding spot, the calculation method is λd _x , and λ is determined by the welding speed.

Step 4: Control the movement of the three-axis mechanical arm to a position 8.5mm from point A, start moving from this point along the direction θ, and start the welding machine at the same time to perform welding operations. The moving speed of the welding head is determined according to the node in step 3 during welding The time required to stay is controlled; at the same time, in order to prevent the welding head from being damaged by the instantaneous acceleration and deceleration of the welding head, the proportional integral method is used to control the speed of the welding head to ensure that the welding head moves with a smooth variable speed movement.

Step 5: When the robot recognizes that the position of the welding head is at a height of 8.5mm above point B, the arc is turned off to stop welding. Raise the position of the welding head, and at the same time move the position of the XY axis of the robot arm to the initial position of the center to complete the welding.

Step 6: After the welding of the robot arm is completed, the welding seam identification and positioning are performed again to ensure that the robot can automatically weld the remaining welding seams.

The present invention can fully automatically identify and weld the welding seam without being accompanied by anyone, saving manpower and time costs, and simultaneously reducing errors caused by manual operations. Compared with the existing robotic arm automatic welding technology, the present invention combines laser and image technology to make welding positioning more accurate and welding effect better.

Claims (10)

1. A steel structure high-altitude welding robot intelligent welding method based on deep learning, is characterized in that, its steps are: S1. The welding robot is equipped with a depth camera to capture images, and sends the captured images to an industrial computer for identification and processing; S2. When the weld seam is detected, the robot is controlled to move to the weld seam position; S3. For the collected depth image, calculate the distance z of the pixel point according to the xy pixel coordinate value, and calculate the corresponding space coordinate [u, v, z] according to the distance in space and the internal parameters of the camera, and fit the space point to obtain A line segment with direction in the space, this line segment is the welding seam position; S4. According to the calculated spatial position of the line segment, determine the relative position of the line segment in the laser and chassis coordinate system, identify the weld depth according to the laser scanning, and map the weld depth from the welding start end A to the welding stop end B to Go to the weld node from A to B, and calculate the stay time of the welding head at this node to ensure that the weld is welded accurately; S5. When the robot recognizes that the end is at point B, stop the welding work, turn off the laser scanning at the same time, reduce the frequency of welding seam detection, and continue the next stage of welding work. 2. A deep learning-based intelligent welding method for steel structure high-altitude welding robots according to claim 1, characterized in that: the detection method used for identification in step S1 is deep learning, and the detection network is a convolutional neural network. 3. A deep learning-based high-altitude welding robot intelligent welding method for steel structures according to claim 2, characterized in that: in step S2, the position and distance information of the robot's movement is roughly positioned by the picture detected by the depth camera, and determined After the weld seam is within the weldable range of the welding robot, fine positioning is started to detect the weld seam position. 4. A deep learning-based intelligent welding method for steel structure high-altitude welding robots according to claim 3, characterized in that: in step S2, the depth map acquired by the camera is sent to the convolutional neural network at a speed of 5 frames per second Network, the result output bits (x, y, θ, w), set the detected position of the weld seam to be accurate when the standard deviation of the xy position in the detected results of 30 consecutive frames is less than 0.003. 5. A deep learning-based high-altitude welding robot intelligent welding method for steel structures according to any one of claims 1-4, wherein the process of identifying the weld seam in step S3 is as follows: (1) Crop the collected depth image to obtain the target area, and use high-pass filtering to repair the pixels with infinite depth; (2) Set the repaired depth map as an image composed of 1 and 0, wherein the identified weld point is set to 1, and other position areas are set to 0, and the evaluation function Q _T is set for the detection results; (3) Use the deep learning network to identify and detect the weld, and use the evaluation function Q _T in step (2) as the standard, and use the quick sort method to sort all the detection results, and the values of each group are in accordance with [u, v, θ, w] arrangement, use the depth information extraction module to detect the depth value of the detected weld position sequence, and calculate the spatial coordinates [x, y, z] corresponding to the pixel according to the depth value; (4) Obtain a series of coordinate position sets with directions according to step (2) and step (3), connect these points and fit these points and directions by nonlinear least squares method to obtain a belt in space The line segment with direction is the real-time position of the weld. 6. A kind of intelligent welding method of high-altitude welding robot for steel structure based on deep learning according to claim 5, characterized in that: described evaluation function Q _T includes angle evaluation function Φ _T and width evaluation function W _T , angle evaluation The function Φ _T is used to find the detection result with the smallest difference between the detected angle and the actual angle, and the width evaluation function W _T is used to find the detection result with the smallest difference between the detected weld width and the actual width. 7. A deep learning-based intelligent welding method for steel structure high-altitude welding robots according to claim 6, characterized in that: the angle evaluation function Φ _T and the width evaluation function W _T are multiplied by corresponding weights α, β and added The evaluation function Q _T is obtained, and the evaluation function Q _T can select the most accurate image of the angle and weld width through the linear weighted sum method. 8. A kind of deep learning-based high-altitude welding robot intelligent welding method for steel structure according to claim 7, characterized in that: the process of step S4 welding the weld seam is as follows: 1) Convert the position of the weld seam based on the camera coordinate system to the coordinates based on the welding robot coordinate system; 2) According to the spatial coordinates, calculate the vector of the line segment l connected by points AB and each node l ₁ , l ₂ ... l _n of the line segment l, and calculate the node direction θ based on the robot coordinate system from the AB vector; 3) Use the laser sensor to point to the starting point A, scan once from A to B, project the laser to the depth of the weld, and send the weld depth information to the sensor, so as to obtain the actual depth d ₁ , d ₂ …d _n of the weld, and map it to For each weld joint l ₁ , l ₂ ...l _n , the welding time t is calculated according to the depth of the welding spot, the calculation method is λd _x , and λ is determined by the welding speed; 4) Control the movement of the three-axis mechanical arm to point A, start moving from this point along the direction θ, and start the welding machine at the same time for welding operation, the moving speed of the welding head is controlled by the time required for the node to stay during welding according to step 3); 5) When the robot recognizes that the position of the welding head reaches point B, the arc is turned off to stop welding. 9. A deep learning-based intelligent welding method for steel structure high-altitude welding robots according to claim 8, characterized in that: in step 4) during the movement of the welding head, the proportional integral method is used to control the speed of the welding head. 10. A deep learning-based high-altitude welding robot intelligent welding method for steel structures according to claim 9, characterized in that: the high-altitude welding robot includes a robot chassis, a welding torch, a mechanical arm, an industrial computer, a depth camera, a laser sensor, a depth The camera and laser sensor are fixed above the robot chassis and fixed with the welding gun clamping mechanism of the robotic arm. The robot communicates through Ethernet, processes and sends the recognized weld position through the industrial computer, and fixes NdFeB under each track of the robot chassis. strong magnet.

Images