Laser vision welding seam tracking method based on deep learning

Abstract

The invention discloses a laser vision welding seam tracking method based on deep learning, and relates to the technical field of industrial welding, a structural light is used for calibrating a welding seam between workpieces, and an image sensor is used for analyzing welding seam data and controlling a welding moving path, and the method comprises the following specific steps that 1, the structural light is emitted to a welding plane; the structured light is captured by the image sensor after being reflected; 2, after an image of the image sensor is read by a system, filtering and denoising processing is carried out, and edge lines on the two sides of a welding seam are obtained; structured light is emitted to a welding plane through a laser emitter, coordinate information of the structured light is captured through an image sensor, continuously-collected image information is integrated to measure an edge line of a welding seam, a median curve of the welding seam is obtained through data fitting, and a welding gun moves along the calculated median curve of the welding seam. The precision of the welding gun moving along the welding seam is improved, and the problem of heat damage caused by the fact that the welding gun is too close to the welding face on one side is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial welding, and in particular to a laser vision weld tracking method based on deep learning. Background Art

[0002] Laser irradiates a moving target and controls the measurement system to point to the target direction based on the deviation angle between the laser signal reflected from the target and the optical axis of the measurement system. Laser tracking systems usually consist of laser illuminators, laser detectors, information processing systems, and tracking systems. Visual sensor calibration is a prerequisite for weld tracking, and the accuracy of the calibration results directly affects the welding accuracy. The measurement model created by structured light calibration and hand-eye calibration has good universality.

[0003] For example, the patent publication number: CN107442900A, "A laser vision weld tracking method", includes a vision tracking module, a robot module and a welding gun module; the vision tracking module fixed on the welding gun is used to obtain the weld image in real time, and the image processing program is used to analyze in real time whether the weld center is offset, and timely feedback is given to the robot module for adjustment.

[0004] The size of the weld between two workpieces is determined by the shape of the gap between the interfaces at the butt joint. In most cases, the welding planes of the workpieces are not flat. When the two workpieces are butt jointed, the gap and width of the weld formed by the two welding planes are inconsistent. In the prior art, when using laser to measure the weld for visual tracking, it is difficult to position the welding gun at the center of the weld in weld gaps of different widths, which can easily lead to different welding energies supplied by the welding gun to the welding planes on both sides. If the welding gun is close to the workpiece on one side, the energy released by the welding gun can easily damage the workpiece, and also make the tightness of the solder combined with the two workpieces different. In addition, when the weld widths are inconsistent, it is also difficult to control the energy size of the welding gun and the capacity of the filling solder, resulting in solder overflow at the narrower weld and solder loss at the wider weld, affecting the welding strength. Summary of the invention

[0005] The purpose of the present invention is to provide a laser vision weld tracking method based on deep learning to solve the problem that it is difficult to position the welding gun at the center of the weld in weld gaps of different widths proposed in the above-mentioned background technology, which easily leads to different welding energies supplied by the welding gun to the welding planes on both sides. If the welding gun is close to the workpiece on one side, the energy released by the welding gun is likely to damage the workpiece, and the degree of tightness of the solder combined with the two workpieces is also different; in addition, when the weld widths are inconsistent, it is also difficult to control the energy size of the welding gun and the capacity of the filling solder, resulting in solder overflow at the narrower weld and solder loss at the wider weld, affecting the welding strength.

[0006] To achieve the above purpose, the present invention provides the following technical solution: a laser vision weld tracking method based on deep learning, which uses a structural cursor to locate the weld between workpieces, analyzes the weld data through an image sensor, and controls the welding movement path. The specific steps are as follows:

[0007] Step 1: emit structured light to the welding plane. The structured light is captured by the image sensor after reflection. By projecting a range structured light laser into a specific area, the structured light is reflected on the three-dimensional object to generate light information that matches the shape of the object. The light information is collected and sorted by the image sensor.

[0008] Step 2: After the image of the image sensor is read by the system, it is filtered and denoised to obtain the edge lines on both sides of the weld. When the laser moves along the weld, a large amount of continuous weld surface image information is captured by the laser emitter and the light capturer. The collected image information is integrated to obtain two smooth curves.

[0009] Step 3: Fit the median curve according to the coordinate information of the two edge lines to obtain the moving path of the welding gun. After obtaining the median curve of the weld image fitting, the welding gun moves along the calculated median curve to ensure that the welding energy is evenly distributed on the two workpieces and the solder is evenly combined with the two welding workpieces;

[0010] Step 4. Calculate the weld width based on the coordinate information difference of the two edge lines, regulate the heating amount and the amount of solder injected during the movement of the welding gun, strictly adjust the welding energy according to the width of the weld, control the capacity of the injected solder, and control the solder overflow and defect deviation in the weld.

[0011] Preferably, in step one, a laser weld tracking device is used, and the laser weld tracking device includes a laser calibration mechanism, a laser emitter and a light capturer. The laser emitter is used to emit structured light to the weld of the workpiece, and the light capturer is used to receive the laser light emitted from the weld and form an image on the layer of the image sensor. The laser calibration mechanism is used to correct the viewing angle deviation of the structured light when it is irradiated on the curved surface.

[0012] Preferably, the laser calibration mechanism includes a packaging component and an angle correction component, the packaging component consists of a box body and a rotating base plate, the outer wall of the rotating base plate is slidably connected to the inner wall of the box body, the upper surface of the rotating base plate is fixedly connected to a protective cover, the center position of the protective cover is fixedly connected to a partition plate, rotating plates are respectively provided on both sides of the partition, one end of the rotating plate is rotatably connected to the rotating base plate, the laser emitter is installed on the upper surface of one of the rotating plates, the light catcher is installed on the upper surface of the other rotating plate, and the upper surfaces of the rotating plates are fixedly connected to sub-covers.

[0013] Preferably, the angle correction assembly includes a bottom rod, a side rod one and a side rod two, one end of the bottom rod is rotatably connected to the side rod one, the outer wall of the bottom rod is slidably connected with an insert block three, the insert block three is slidably connected to the outer wall of the side rod two, the other ends of the side rod one and the side rod two are rotatably connected with a middle rod, the outer wall of the middle rod is slidably connected with a cross block, and the cross block is slidably connected to the bottom rod.

[0014] Preferably, the outer wall of the middle rod is slidably connected with an insert block 1, the outer wall of one end of the insert block 1 is rotatably connected to two support rods, and the other ends of the two support rods are rotatably connected with insert blocks 2, one of the insert blocks 2 is slidably connected to the outer wall of the side rod 1, and the other insert block 2 is slidably connected to the outer wall of the side rod 2, a positioning rod is fixedly connected at the center position of the support rod, and a positioning block is fixedly connected to the side wall of the sub-cover, and the positioning rod is slidably connected to the positioning block.

[0015] Preferably, a servo motor 1 is fixedly connected to the side wall of the box body, a transmission belt 1 is fixedly connected to the output end of the servo motor 1, and the servo motor 1 drives the middle rod to rotate through the transmission belt 1.

[0016] Preferably, the output end of the servo motor 2 is fixedly connected to the transmission belt 2, the bottom wall of the box body is fixedly connected to the servo motor 2, the servo motor 2 is connected to the screw rod through the transmission belt 2, and the outer wall of the screw rod is threadedly connected to the slider.

[0017] Preferably, the slider is clamped with the outer wall of the second side rod and rotatably connected to one end of the support rod, the two ends of the bottom rod are respectively slidably connected to the inner wall of the rotating bottom plate, and the two ends of the second side rod are respectively fixedly connected to the inner wall of the box body.

[0018] Preferably, a rangefinder is installed at one end of the rotating base plate, and the detection head of the rangefinder extends below the bottom end of the rotating base plate. A central controller is installed at the other end of the rotating base plate, and the laser emitter and the light capturer are electrically connected to the central controller through signal lines.

[0019] Preferably, the laser weld seam tracking device is installed on a base of a machine tool, and a clamp is provided on a workbench of the machine tool.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. In the present invention, structured light is emitted to the welding plane by a laser emitter, and the coordinate information of the structured light is captured by an image sensor. The edge line of the weld is measured by integrating the continuously collected image information, and the median curve of the weld is obtained by data fitting. The welding gun moves along the calculated median curve of the weld, thereby improving the accuracy of the moving position of the welding gun along the weld and avoiding the problem of heat damage caused by the welding gun being too close to one side of the welding surface.

[0022] 2. In the present invention, the width change of the weld is calculated by integrating the coordinate information of the edge line of the weld, and the welding energy and the filling solder capacity released by the welding gun during the movement are controlled to avoid overflow due to excessive solder filling when the weld becomes narrower, and weld defects due to insufficient solder filling when the weld becomes wider.

[0023] 3. In the present invention, the distance change between the welding plane and the laser calibration mechanism is measured by a rangefinder. Since the curvature of the curved surface is related to the radius of curvature, and the change of the radius of curvature will also cause the position of the base and the welding plane to change, the curvature change of the welding plane is obtained, and the normal deflection angle is obtained based on the curvature change, so as to correct the emission angle and capture angle of the laser, thereby avoiding the problem of weld image deformation caused by shooting angle deviation to a great extent.

[0024] 4. In the present invention, the distance change between the welding plane and the laser calibration mechanism is measured by a rangefinder, and the deflection angle of the laser emitter and the light catcher and the weld normal is corrected according to the distance change of the measured line of sight, ensuring that the light catcher can completely capture the optical information of the laser on the welding plane.

[0025] 5. In the present invention, a clamp is provided on the working table of the machine tool, and the clamp is used to position the rod-shaped welding workpiece to achieve welding, thereby satisfying different welding processing functions such as flat plates, curved plates, cylinders and quasi-cylinders. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a structural schematic diagram of a laser calibration mechanism in a laser weld tracking device based on deep learning of the present invention;

[0027] Figure 2 This is a schematic diagram of the internal structure of a box in a laser weld tracking device based on deep learning of the present invention;

[0028] Figure 3 This is a schematic diagram of the front structure of a box in a laser weld tracking device based on deep learning of the present invention;

[0029] Figure 4 This is a schematic diagram of the circuit connections between a laser emitter, a light capturer and a central controller in a laser weld tracking device based on deep learning according to the present invention;

[0030] Figure 5 It is a structural schematic diagram of an angle correction component in a laser weld tracking device based on deep learning of the present invention;

[0031] Figure 6 This is a planar structural diagram of an angle correction component in a laser weld tracking device based on deep learning according to the present invention;

[0032] Figure 7 This is a structural schematic diagram of a deep learning-based laser weld tracking device and a machine tool installation according to the present invention;

[0033] Figure 8 This is a weld image processing effect diagram of a laser vision weld tracking method based on deep learning in the present invention;

[0034] Fig. 9 This is a detection principle diagram of a laser weld tracking device of a laser visual weld tracking method based on deep learning of the present invention;

[0035] Fig.10 The present invention provides a laser weld tracking device for a laser visual weld tracking method based on deep learning and a viewing angle correction principle on a curved surface;

[0036] Fig.11 This is a system block diagram of a laser vision weld tracking method based on deep learning in the present invention.

[0037] In the figure: 1. laser calibration mechanism; 11. packaging component; 111. box body; 112. rotating bottom plate; 113. protective cover; 114. sub-cover; 115. partition; 116. positioning block; 117. positioning rod; 118. rotating plate; 12. angle correction component; 121. screw rod; 122. bottom rod; 123. side rod one; 124. side rod two; 125. middle rod; 126. support rod; 127. cross block; 128. slider; 129. insert block one; 13. servo motor one; 14. transmission belt one; 15. servo motor two; 16. transmission belt two; 17. insert block two; 18. insert block three; 2. laser transmitter; 3. light capture device; 4. rangefinder; 5. central controller; 6. machine tool; 7. clamp. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Embodiment 1

[0040] according to Figure 1-11 As shown in the figure: A laser vision weld tracking method based on deep learning uses a structural cursor to locate the weld between workpieces, analyzes the weld data through an image sensor, and controls the welding movement path. The specific steps are as follows:

[0041] 1) The structured light is emitted to the welding plane, and the structured light is captured by the image sensor after reflection;

[0042] 2) The image of the image sensor is read by the system and then filtered and denoised to obtain the edge lines on both sides of the weld;

[0043] 3) Fitting the median curve according to the coordinate information of the two edge lines to obtain the moving path of the welding gun;

[0044] 4) The weld width is calculated based on the coordinate information difference of the two edge lines, and the heating amount and the amount of solder injected during the movement of the welding gun are adjusted.

[0045] In this embodiment, by projecting a range structured light laser into a specific area, the structured light is reflected on the three-dimensional object to generate light information that matches the shape of the object, and the light information is collected and sorted by the image sensor; in the process of the laser moving along the weld, a large amount of continuous weld surface image information is captured by the laser emitter 2 and the light capturer 3, and the actual image information of the weld can be obtained by integrating the collected image information. After filtering and noise reduction, the data with relatively large deviations are deleted, and finally two smooth curves are obtained; after obtaining the median curve of the weld image fitting, the welding gun moves along the calculated median curve to ensure that the welding energy is evenly distributed on the two workpieces, and the solder is evenly combined with the two welding workpieces; the welding energy is strictly adjusted according to the width of the weld, the capacity of the injected solder is controlled, and the solder overflow and defect deviation in the weld are controlled;

[0046] The weld measurement mechanism consists of two parts: the pixel coordinates of the weld image evaluation feature points and the conversion between the measurement position and the position at the machine tool base. The distance change between the welding workpiece and the machine tool base is measured by the rangefinder 4, and then the pixel coordinates of the weld image collected by the light capturer 3 are combined to calculate the three-dimensional coordinate value of the weld, that is, the three-dimensional coordinate value is predicted by geometric consistency.

[0047] The pixel coordinates of the weld image assessment feature points are composed of an inference model and a measurement model. The inference model is used for the coordinate values ​​of the base coordinate system. The measurement model obtains feedback information from the environment to calculate the policy gradient. The obtained gradient value is used in the inference model to guide learning. During testing, the pixel coordinates of the weld image assessment feature points will infer the pixel coordinates of the feature points. The inference model regresses the inferred pixel coordinates to three-dimensional coordinates. Learning and training can simplify the calibration process and create a more accurate conversion relationship.

[0048] Embodiment 2

[0049] according to Figure 1 , Figure 2 and Figure 7As shown, in step 1, a laser weld tracking device is used. The laser weld tracking device includes a laser calibration mechanism 1, a laser emitter 2 and a light capturer 3. The laser emitter 2 is used to emit structured light to the weld of the workpiece, and the light capturer 3 is used to receive the laser light emitted from the weld and form an image on the layer of the image sensor. The laser calibration mechanism 1 is used to correct the viewing angle deviation of the structured light irradiated on the curved surface.

[0050] In this embodiment, the rangefinder 4, the laser emitter 2, the light catcher 3 and the welding gun are arranged quickly in front and back, the rangefinder 4 first passes through the position to be measured at the welding line, then the laser emitter 2 and the light catcher 3 capture the gap shape at the welding position, and finally the welding gun processes;

[0051] As the viewing angle of the structured light collected by the light capturer 3 on the base changes, it is easy to cause the weld image represented by the structured light to be distorted and deviated. Timely correction of the laser collection viewing angle can effectively improve the accuracy of the weld constructed by the laser image. The base is located above the workpiece. The laser emitter 2 and the light capturer 3 on the base emit structured laser and collect images for the weld on the same straight line. The laser emitter 2 and the light capturer 3 are adjusted accordingly to the viewing angle of the weld.

[0052] Taking the normal line of the measured weld as the standard, the laser emitter 2 and the light capturer 3 are respectively located on both sides of the normal line of the measured weld, and the inclination angles to the normal line are equal. When welding a planar workpiece such as a plate, the welded workpiece is placed flat on the workbench of the machine tool 6, and the laser weld tracking device always moves at the same height and viewing angle, so as to capture a complete weld image;

[0053] When measuring popular welding workpieces such as automobile outer shells or column-like objects such as curved rods, the laser weld tracking device is not located on the normal of the weld directly below. At this time, if the detection angle of the laser weld tracking device is still aimed at the weld directly below, the measured coordinate data will deviate, and even cause the light capturer 3 to be unable to normally collect the weld lighting information. At this time, it is necessary to correct the angle of view measured by the laser emitter 2 and the light capturer 3 so that the angle of view of the laser emitter 2 and the light capturer 3 is adjusted to be aligned with the measurement plane.

[0054] Embodiment 3

[0055] according to Figure 2 , Figure 3 , Figure 5 and Figure 6As shown, the laser calibration mechanism 1 includes a packaging component 11 and an angle correction component 12. The packaging component 11 consists of a box body 111 and a rotating base plate 112. The outer wall of the rotating base plate 112 is slidably connected to the inner wall of the box body 111. A protective cover 113 is fixedly connected to the upper surface of the rotating base plate 112. A partition 115 is fixedly connected to the center of the protective cover 113. Rotating plates 118 are respectively arranged on both sides of the partition 115. One end of the rotating plate 118 is rotatably connected to the rotating base plate 112. The laser emitter 2 is installed on the upper surface of one of the rotating plates 118, and the light catcher 3 is installed on the upper surface of the other rotating plate 118. The upper surfaces of the rotating plates 118 are all fixedly connected with sub-covers 114.

[0056] The angle correction assembly 12 includes a bottom rod 122, a side rod 123 and a side rod 2 124. One end of the bottom rod 122 is rotatably connected to the side rod 1 123. The outer wall of the bottom rod 122 is slidably connected to an insert block 3 18. The insert block 3 18 is slidably connected to the outer wall of the side rod 2 124. The other ends of the side rods 123 and 124 are rotatably connected to a middle rod 125. The outer wall of the middle rod 125 is slidably connected to a cross block 127. The cross block 127 is slidably connected to the bottom rod 122. The outer wall of the middle rod 125 is slidably connected to the It is rotatably connected with an insert block 129, and the outer side wall of one end of the insert block 129 is rotatably connected to two support rods 126. The other ends of the two support rods 126 are rotatably connected with insert blocks 17, one of which is slidably connected to the outer side wall of the side rod 123, and the other insert block 17 is slidably connected to the outer side wall of the side rod 124. A positioning rod 117 is fixedly connected at the center position of the support rod 126, and a positioning block 116 is fixedly connected to the side wall of the sub-cover 114, and the positioning rod 117 is slidably connected to the positioning block 116.

[0057] A servo motor 13 is fixedly connected to the side wall of the box body 111, and a transmission belt 14 is fixedly connected to the output end of the servo motor 13. The servo motor 13 drives the middle rod 125 to rotate through the transmission belt 14. A servo motor 15 is fixedly connected to the bottom wall of the box body 111, and a transmission belt 16 is fixedly connected to the output end of the servo motor 15. The servo motor 15 is transmission-connected to the screw rod 121 through the transmission belt 16. The outer wall of the screw rod 121 is threadedly connected with a slider 128. The slider 128 is clamped with the outer wall of the side rod 124 and is rotationally connected to one end of the support rod 126. The two ends of the bottom rod 122 are respectively slidably connected to the inner wall of the rotating bottom plate 112, and the two ends of the side rod 124 are respectively fixedly connected to the inner wall of the box body 111.

[0058] In this embodiment, the laser emitter 2, the light catcher 3, the rangefinder 4 and the central controller 5 are all installed inside the box body 111, and are all located on the upper layer of the rotating bottom plate 112. The box body 111 and the rotating bottom plate 112 together constitute a sealed whole, wherein a rotating shaft is embedded in the center of the rotating bottom plate 112. The rangefinder 4 is used to measure the distance between the workpiece surface and the base in the viewing direction, and the curvature change of the workpiece surface is measured by the continuous change of the workpiece surface. According to the change value of the curvature of the curved surface on the monitored welding workpiece, the pulse amount of the input servo motor 13 is controlled, and then the servo motor 13 is used to control the angle of the middle rod 125;

[0059] The middle rod 125 and the bottom rod 122 are always kept in a vertical state, and the two are connected to each other by a cross block 127, so the position of the bottom rod 122 is also adjusted with the control of the servo motor 13, and the viewing direction of the laser emitter 2 and the light catcher 3 and the weld is corrected by correcting the direction of the bottom rod 122. As the curvature of the measured workpiece plane changes, the visual distance between the laser emitter 2 and the light catcher 3 and the measurement point also changes accordingly;

[0060] As the distance of the measuring line of sight increases, the machine screw 121 is controlled to rotate by the servo motor 15 to control the slider 128 to move downward, so that the slider 128 drives the support rod 126 to open outward. Since the two support rods 126 are connected by an insert block 129, the viewing directions of the support rods 126 on both sides are equal. The two support rods 126 respectively drive the positioning rod 117 to swing in the vertical direction. When the positioning rod 117 flips in the vertical direction, the positioning block 116 driven by the positioning rod 117 will also flip in the vertical direction, thereby reducing the angle between the laser irradiation direction and the laser capture direction and the normal line, and obtaining a better laser measurement angle.

[0061] Vice versa, when the distance of the measuring time is shortened, the control slider 128 moves upward, so that the slider 128 drives the support rod 126 to fit in the vertical direction, allowing the positioning rod 117 to swing in the horizontal direction, and the positioning rod 117 drives the positioning block 116 to lift the sub-cover 114, thereby increasing the angle between the laser irradiation direction and the laser capture direction and the normal line, and obtaining a better laser measurement angle.

[0062] Embodiment 4

[0063] according to Figure 4 , Figure 5 , Figure 6 and Figure 7As shown, a rangefinder 4 is installed at one end of the rotating base plate 112, and the detection head of the rangefinder 4 extends below the bottom end of the rotating base plate 112. A central controller 5 is installed at the other end of the rotating base plate 112. The laser emitter 2 and the light catcher 3 are both electrically connected to the central controller 5 through a signal line. The laser weld tracking device is installed on the base of the machine tool 6, and a clamp 7 is provided on the workbench of the machine tool 6.

[0064] In this embodiment, according to the shape of the welded workpiece, the laser tracking of the weld is divided into four categories. One is for a flat workpiece with a flat surface. At this time, two welded workpieces are spliced ​​and placed flat on a workbench, and the base of the machine tool 6 is controlled to move along the track of the weld. At this time, there is no need to adjust the detection angle of the laser weld tracking device on the base, and the accurate data of the weld can be measured by always maintaining a single angle of view;

[0065] The second is a cylindrical workpiece with a perfect circular outer surface. In this case, the two welded workpieces are assembled and respectively mounted on the clamp 7. The clamp 7 is controlled to rotate on the axis, and the base of the machine tool 6 is controlled to move linearly according to the trajectory of the weld. At this time, there is still no need to adjust the detection angle of the laser weld tracking device on the base.

[0066] The third is a plate-like workpiece with a streamlined surface. At this time, the welded workpiece still needs to be spliced ​​and placed flat on the workbench, and the base of the machine tool 6 is controlled to move along the trajectory of the weld. At the same time, the distance between the welding plane and the line of sight of the laser on the base is measured by the rangefinder 4. Since the rangefinder 4 is set in front of the laser emitter 2 and the light catcher 3, the distance change information of the to-be-detected position can always be obtained first, and the curvature change information of the workpiece surface can also be obtained, and the viewing angle direction can be adjusted accordingly to correct the test viewing angle deviation problem caused by the change of the surface curvature;

[0067] The fourth is a quasi-cylinder. After assembling two welded workpieces, they are respectively mounted on the clamp 7. The clamp 7 is controlled to rotate on the axis, and the base of the machine tool 6 is controlled to move linearly according to the trajectory of the weld. The rangefinder 4 is used to detect the line of sight length of the welding point to obtain the curvature change information of the outer surface of the quasi-cylinder, and the viewing angle direction is adjusted accordingly to correct the viewing angle deviation problem caused by the curvature change of the outer surface of the quasi-cylinder.

[0068] Embodiment 5

[0069] according to Figure 8-11As shown: the directions of the normal and tangent at a certain point on the curve are always perpendicular, and its average curvature is the change angle of the tangent between the two end points of the unit arc length. If the curvature is always consistent, the change angle of the tangent between the two end points of the unit arc length is equal. If the curvature increases or decreases, the change angle of the tangent between the two end points of the unit arc length will also increase or decrease accordingly. The difference in the deflection of the normal angle can be obtained from the curvature change value of the curved surface. The line of sight length of the welding point is measured by the rangefinder 4, and the curvature change rate of the arc surface of the welding workpiece is obtained by integrated calculation to adjust the position of the laser emission viewing angle in real time.

[0070] In this embodiment, the unit tangent vector is guided by the arc length The vector of the original curve Orthogonal, its magnitude is the curvature Indicates the curvature of the curve, that is, the degree of deviation from the tangent vector at a certain point. If k≠0, we can define is the unit principal normal vector, which indicates the bending direction of the curve, that is, the direction of the tangent vector at a certain point. Under general parameters, the formula is:

[0071]

[0072] The plane where the unit tangent vector and the unit principal normal vector lie is called the osculating plane and can be viewed as and The determined plane is the limiting case where Δs approaches 0. The osculating plane can also be regarded as the plane where the osculating circle lies, and the radius of the circle is the radius of curvature. The position vector of the center of the circle is exist tangent to the curve at , with the same tangent vector

[0073] In the case of a plane curve, the unit tangent vector Taking the derivative about the arc length s, we get It is customary to define directly This is not the direction in which the curve bends; It constitutes a basis for positive orientation, and the bending direction is reflected in the curvature In the symbol, k>0 means bending in the counterclockwise direction, that is, the measurement line of sight of the laser needs to be bent in the counterclockwise direction, and k<0 means bending in the clockwise direction, that is, the measurement line of sight of the laser needs to be bent in the counterclockwise direction.

[0074] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A laser vision weld tracking method based on deep learning, characterized in that: Use the structural cursor to locate the weld between the workpieces, analyze the weld data through the image sensor, and control the welding movement path. The specific steps are as follows: S1, emits structured light to the welding plane, and the structured light is captured by the image sensor after reflection; S2, the image of the image sensor is read by the system and then filtered and denoised to obtain the edge lines on both sides of the weld; S3, fitting a median curve according to the coordinate information of the two edge lines to obtain a moving path of the welding gun; S4. Calculate the weld width according to the coordinate information difference of the two edge lines, and adjust the heating amount and the amount of solder injected during the movement of the welding gun.

2. According to the deep learning-based laser vision weld tracking method of claim 1, it is characterized in that: In step S1, a laser weld seam tracking device is used, and the laser weld seam tracking device includes a laser calibration mechanism (1), a laser emitter (2) and a light capturer (3). The laser emitter (2) is used to emit structured light to the weld seam of the workpiece, and the light capturer (3) is used to receive the laser light emitted from the weld seam and form an image on a layer of an image sensor. The laser calibration mechanism (1) is used to correct the viewing angle deviation of the structured light irradiated on the curved surface.

3. The laser vision weld tracking method based on deep learning according to claim 2 is characterized in that: The laser calibration mechanism (1) comprises a packaging component (11) and an angle correction component (12), wherein the packaging component (11) is composed of a box body (111) and a rotating base plate (112), wherein the outer wall of the rotating base plate (112) is slidably connected to the inner wall of the box body (111), a protective cover (113) is fixedly connected to the upper surface of the rotating base plate (112), a partition (115) is fixedly connected at the center of the protective cover (113), rotating plates (118) are respectively arranged on both sides of the partition (115), one end of the rotating plate (118) is rotatably connected to the rotating base plate (112), the laser emitter (2) is mounted on the upper surface of one of the rotating plates (118), the light catcher (3) is mounted on the upper surface of the other rotating plate (118), and the upper surfaces of the rotating plates (118) are fixedly connected to sub-covers (114).

4. The laser vision weld tracking method based on deep learning according to claim 3 is characterized in that: The angle correction assembly (12) comprises a bottom rod (122), a side rod one (123) and a side rod two (124); one end of the bottom rod (122) is rotatably connected to the side rod one (123); an outer wall of the bottom rod (122) is slidably connected to an insert block three (18); the insert block three (18) is slidably connected to the outer wall of the side rod two (124); the other ends of the side rod one (123) and the side rod two (124) are rotatably connected to a middle rod (125); the outer wall of the middle rod (125) is slidably connected to a cross block (127); the cross block (127) is slidably connected to the bottom rod (122).

5. The laser vision weld seam tracking method based on deep learning according to claim 4, characterized in that: The outer wall of the middle rod (125) is slidably connected to an insert block 1 (129), and the outer wall of one end of the insert block 1 (129) is rotatably connected to two support rods (126), and the other ends of the two support rods (126) are both rotatably connected to insert blocks 2 (17), one of which is slidably connected to the outer wall of the side rod 1 (123), and the other insert block 2 (17) is slidably connected to the outer wall of the side rod 2 (124), and a positioning rod (117) is fixedly connected at the center position of the support rod (126), and a positioning block (116) is fixedly connected to the side wall of the sub-cover (114), and the positioning rod (117) is slidably connected to the positioning block (116).

6. The laser vision weld tracking method based on deep learning according to claim 3 is characterized in that: A servo motor (13) is fixedly connected to the side wall of the box body (111), and a transmission belt (14) is fixedly connected to the output end of the servo motor (13). The servo motor (13) drives the middle rod (125) to rotate through the transmission belt (14).

7. The laser vision weld tracking method based on deep learning according to claim 3 is characterized in that: The bottom wall of the box body (111) is fixedly connected to a servo motor 2 (15); the output end of the servo motor 2 (15) is fixedly connected to a transmission belt 2 (16); the servo motor 2 (15) is transmission-connected to a screw rod (121) via the transmission belt 2 (16); and the outer wall of the screw rod (121) is threadedly connected to a slider (128).

8. The laser vision weld seam tracking method based on deep learning according to claim 7, characterized in that: The slide block (128) is clamped with the outer wall of the second side rod (124) and is rotatably connected to one end of the support rod (126); the two ends of the bottom rod (122) are respectively slidably connected to the inner wall of the rotating bottom plate (112); and the two ends of the second side rod (124) are respectively fixedly connected to the inner wall of the box body (111).

9. The laser vision weld tracking method based on deep learning according to claim 3 is characterized in that: A rangefinder (4) is installed at one end of the rotating base plate (112), and a detection head of the rangefinder (4) extends below the bottom end of the rotating base plate (112). A central controller (5) is installed at the other end of the rotating base plate (112), and the laser emitter (2) and the light capturer (3) are both electrically connected to the central controller (5) via a signal line.

10. The laser vision weld tracking method based on deep learning according to claim 1, characterized in that: The laser weld seam tracking device is installed on a base of a machine tool (6), and a clamp (7) is provided on a workbench of the machine tool (6).

Images