A laser welding method and equipment

Abstract

The present invention provides a laser welding method and device. The laser welding method includes: controlling a photographing unit to obtain multiple frames of images of a workpiece to be welded according to a photographing trajectory; confirming motion information of the welding area according to the instantaneous pixel movement speed and pixel difference corresponding to the multiple frames of images of the workpiece to be welded; sending the motion information and the image information of the multiple frames of images of the workpiece to be welded into a pre-trained neural network to confirm the welding area of the workpiece to be welded; and controlling a welding unit to weld the welding area according to a target welding trajectory corresponding to the welding area. The present invention improves the accuracy of welding alignment.

Description

Technical Field

[0001] The present invention relates to the technical field of laser welding, and particularly relates to a laser welding method and device. Background Art

[0002] Compared with traditional welding methods, laser welding has the characteristics of high speed, large depth, and small deformation.

[0003] In the automatic welding technology, the user installs the workpiece to be welded on the laser welding machine. The laser welding machine identifies the weld seam and then welds the workpiece to be welded.

[0004] However, in high-precision welding, due to the high precision of workpiece assembly and small weld seams, the laser welding machine cannot accurately identify the weld seams and cannot perform automatic welding of workpieces to be welded with high precision. Summary of the Invention

[0005] The purpose of the present invention is to provide a laser welding method and device, aiming to achieve accurate identification of weld seams.

[0006] The present invention provides a laser welding method, which includes:

[0007] Controlling a photographing unit to acquire multiple frames of images of the workpiece to be welded according to a photographing trajectory;

[0008] Confirming the motion information of the area to be welded according to the instantaneous pixel movement speed and pixel difference corresponding to the multiple frames of images of the workpiece to be welded;

[0009] Feeding the motion information and the image information of the multiple frames of images of the workpiece to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded;

[0010] Controlling a welding unit to weld the area to be welded according to the target welding trajectory corresponding to the area to be welded.

[0011] In some embodiments, the workpiece to be welded is provided with a mark; the laser welding method further includes:

[0012] Controlling the photographing unit to photograph the image of the workpiece to be welded, and the image of the workpiece to be welded includes the

[0013] mark;

[0014] Confirming an initial welding trajectory based on the mark, wherein the initial welding trajectory corresponds to the mark one by one;

[0015] Confirming a photographing trajectory based on the initial welding trajectory.

[0016] In some embodiments, the included angle between the photographing trajectory and the initial welding trajectory is greater than or equal to 80 degrees and less than or equal to 100 degrees.

[0017] In some embodiments, before the step of controlling the photographing unit to acquire multiple frames of images of the workpiece to be welded according to a photographing trajectory, the method further includes:

[0018] Controlling the turntable on which the workpiece to be welded is installed to move to a target area so that the workpiece to be welded is within the field of view of the photographing unit.

[0019] In some embodiments, each frame of the images of the workpiece to be welded includes a complete area to be welded; the method further includes:

[0020] Obtaining the pixel instantaneous movement speed and pixel difference of adjacent frames of images of the workpiece to be welded corresponding to the time from T to T + L to obtain a pixel instantaneous movement speed image and a pixel difference image corresponding to the time from T to T + L;

[0021] Obtaining a pixel instantaneous movement speed image and a pixel difference image corresponding to the time T + L + 1;

[0022] The step of feeding the motion information and the image information of multiple frames of images of the workpiece to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded includes:

[0023] Based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1, confirming the area to be welded at the time T + L + 1; wherein, the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L are pre-stored;

[0024] Repeating the steps: obtaining a pixel instantaneous movement speed image and a pixel difference image corresponding to the time T + L + 1; and the step: based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1, confirming the area to be welded at the time T + L + 1; to obtain the areas to be welded from the time T + L + 1 to the time T + L + N; and fitting the areas to be welded from the time T + L + 1 to the time T + L + N to obtain the final area to be welded.

[0025] In some embodiments, the mark includes: a first part and a second part perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of workpieces to be welded.

[0026] In some embodiments, the step of based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1, confirming the area to be welded at the time T + L + 1, includes:

[0027] For each moment from T+1 to T+L+1, the pixel instantaneous movement speed image, the pixel difference image, and the image to be welded are input into a three-way parallel convolutional neural network; the output of the three-way parallel convolutional neural network is connected to a fully connected neural network, and the fully connected neural network outputs the spatial feature information from T+1 to T+L+1.

[0028] In chronological order, the spatial feature information from T+1 to T+L+1 is input into a temporal neural network to obtain spatio-temporal feature information.

[0029] Based on the marking, a coordinate system is established; and according to the spatio-temporal feature information, the coordinate information of the area to be welded is determined.

[0030] In some embodiments, obtaining the pixel instantaneous movement speed image and the pixel difference image corresponding to T+L+1 includes:

[0031] Based on the coordinate information of the area to be welded at the previous moment, the cutting area at the current moment is confirmed, the cutting area includes the area to be welded, and the area of the cutting area is at least twice the area of the area to be welded.

[0032] Based on the cutting area, a cutting image is generated, where the sizes of the cutting images at each moment are the same.

[0033] Based on the cutting image, the corresponding pixel instantaneous movement speed image and pixel difference image are generated.

[0034] In some embodiments, the method further includes:

[0035] Controlling the photographing unit to photograph the image of the welded part; and

[0036] According to the determined area to be welded and the welded image, the welding result is confirmed.

[0037] In some embodiments, the confirming the welding result according to the determined area to be welded and the welded image includes:

[0038] Based on the marking, the welded image is aligned to the coordinate system.

[0039] The image of the area to be welded corresponding to the coordinate information of the area to be welded and the image of the welded area are input into a siamese neural network.

[0040] The siamese neural network outputs the welding result based on the similarity between the image of the area to be welded and the image of the welded area; when the similarity is greater than a preset threshold, the welding result is not welded.

[0041] The present invention also provides a laser welding device, which includes:

[0042] A photographing unit;

[0043] A welding unit; and

[0044] A control unit, which is respectively connected to the photographing unit and the welding unit, and is configured to execute a laser welding program to implement the above-mentioned laser welding method.

[0045] A laser welding method and device proposed by the present invention. The laser welding method controls the photographing unit to move and photograph multiple frames of images of the workpiece to be welded; obtains the motion information of the multiple frames of images of the workpiece to be welded; and then can accurately identify the welding area to be welded of the workpiece to be welded according to the motion information and the image information of the images of the workpiece to be welded; thereby controlling the welding unit to weld the welding area to be welded along the target welding trajectory corresponding to the welding area to be welded. The present invention improves the accuracy of welding alignment. Description of the Drawings

[0046] Figure 1 It is a schematic diagram of the working process of an embodiment of the laser welding method of the present invention;

[0047] Figure 2 It is a schematic diagram of the marking of the workpiece to be welded of the present invention;

[0048] Figure 3 It is a schematic diagram of the working process of another embodiment of the laser welding method of the present invention;

[0049] Figure 4 It is a schematic diagram of the neural network structure of the laser welding method of the present invention.

[0050] Reference numerals in the drawings: Weld seam: 10; Marking: 20. Detailed Embodiments

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] The present invention proposes a laser welding method, which is applicable to a laser welding device. The laser welding device may include: a loading position, a welding position, and an unloading position; wherein, the loading position and the unloading position can be realized by a turntable, and the material is continuously sent to the welding position by the turntable. After the welding is completed and the inspection passes, it is then sent to the unloading position by the turntable. The above process can perform welding during the movement of the product.

[0053] Referring to Figure 1 , in some embodiments, the laser welding method includes:

[0054] S100. Control the photographing unit to obtain multiple frames of images of the workpiece to be welded according to the photographing trajectory;

[0055] S200. Confirm the motion information of the area to be welded according to the instantaneous pixel movement speed and pixel difference corresponding to the multiple frames of images of the workpiece to be welded;

[0056] S300. Send the motion information and the image information of the multiple frames of images of the workpiece to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded;

[0057] S400. Control the welding unit to weld the area to be welded according to the target welding trajectory corresponding to the area to be welded.

[0058] Before step S100, it further includes: controlling the turntable on which the workpiece to be welded is installed to move to the target area so that the workpiece to be welded is within the field of view of the photographing unit. During the rotation of the turntable, when the turntable sends the workpiece to be welded into the target area, control the turntable to stop rotating, and then the photographing unit continuously photographs the target area, and multiple frames of images of the workpiece to be welded can be obtained. In order to accurately identify the position of the weld 10, the photographing device can be controlled to dynamically photograph the workpiece to be welded. At this time, in the multiple frames of images of the workpiece to be welded obtained by photographing, the weld 10 is in a moving state. Then, in the images of the workpiece to be welded in adjacent frames, the pixel difference represents the movement of the weld 10, and at the same time, the instantaneous pixel movement speed also represents the position change of the weld 10.

[0059] In this embodiment, the photographing unit can be implemented using an imaging device and a three-axis gimbal or a two-axis gimbal. The control unit controls the gimbal to rotate the imaging device along a preset trajectory to photograph multiple frames of images of the workpiece to be welded.

[0060] In step S200, the instantaneous pixel movement speed refers to the instantaneous movement speed of the same pixel point or pixel block between two frames of images of the workpiece to be welded, and can be calculated from the position change amount of the pixel point or pixel block and the photographing time interval between the two frames of images of the workpiece to be welded. The pixel difference refers to the difference in pixel values between the corresponding pixels between two frames of images of the workpiece to be welded.

[0061] In practical applications, the motion information can be represented by an instantaneous pixel movement speed image and a pixel difference image, and the image information of the workpiece to be welded image can be represented by the workpiece to be welded image.

[0062] In step S300, the pixel instantaneous movement speed image, the pixel difference image, and the image of the workpiece to be welded can be fed into a parallel convolutional neural network to output three-way feature vectors. Then, a fully connected neural network fuses the three-way feature vectors to obtain spatial feature information. Then, in chronological order, the spatial feature information is fed into a temporal neural network to obtain the area to be welded, that is, the weld seam 10.

[0063] In S400, after accurately identifying the weld seam 10 and confirming the trajectory and coordinate information of the weld seam 10, the robotic arm equipped with the laser welding device is controlled to work, so that the laser welding device moves along the weld seam 10, and then the weld seam 10 is welded. Compared with the initial welding trajectory, the target welding trajectory eliminates the position offset caused by the placement of the workpiece to be welded.

[0064] The laser welding device proposed by the present invention controls the shooting unit to move and shoot multiple images of the workpiece to be welded to obtain the motion information of the multiple images of the workpiece to be welded. Furthermore, the area to be welded of the workpiece to be welded can be accurately identified according to the motion information and the image information of the image of the workpiece to be welded. Thus, the welding unit can be controlled to weld the area to be welded along the target welding trajectory corresponding to the area to be welded. The present invention improves the accuracy of welding alignment.

[0065] Refer to Figure 2 , in some embodiments, the workpiece to be welded is provided with a mark 20; the laser welding method further includes: controlling the shooting unit to shoot an image of the workpiece to be welded, and the image of the workpiece to be welded includes the mark 20;

[0066] Based on the mark 20, the initial welding trajectory is confirmed, wherein the initial welding trajectory corresponds to the mark 20 one by one;

[0067] Based on the initial welding trajectory, the shooting trajectory is confirmed.

[0068] It should be noted that the weld seam 10 is slender. In order to better identify the weld seam 10, the shooting trajectory is preferably to move along the vertical direction of the weld seam 10. In practical applications, the included angle between the shooting trajectory and the initial welding trajectory can be set to be greater than or equal to 80 degrees and less than or equal to 100 degrees. Therefore, in order to determine the shooting trajectory, in this embodiment, an image of the workpiece to be welded is first shot, and then the mark 20 is identified. The mark 20 can indicate the initial welding trajectory, and then the shooting trajectory can be confirmed.

[0069] Exemplarily, the mark 20 can represent the placement direction of the workpiece to be welded and the type of the workpiece to be welded, and the laser welding device can pre-store the initial welding trajectory corresponding to the type of the workpiece to be welded.

[0070] Refer to Figure 2, the marker 20 includes a first part and a second part that are perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of workpieces to be welded.

[0071] For example, the first part is in the shape of a "one", indicating that the initial welding trajectory of the workpiece to be welded is a straight line. The first part is in the shape of an "o", indicating that the initial welding trajectory of the workpiece to be welded is a ring, that is, welding is performed around the target area.

[0072] Exemplarily, the number of the markers 20 is 4, which are respectively located at the four corners of the welded part to meet the requirements for establishing the coordinate system.

[0073] Referring to Figure 3 , the present application also proposes a method for extracting motion information applicable to a laser welding device, which can greatly reduce the amount of calculation while meeting the accuracy.

[0074] Specifically, each frame of the workpiece image to be welded includes a complete welding area to be welded; the method further includes:

[0075] S500. Obtain the pixel instantaneous movement speed and pixel difference of the workpiece images to be welded in adjacent frames corresponding to the time from T to T + L, and obtain the pixel instantaneous movement speed image and pixel difference image corresponding to the time from T to T + L;

[0076] S600. Obtain the pixel instantaneous movement speed image and pixel difference image corresponding to the time T + L + 1;

[0077] Step S300. Send the motion information and the image information of multiple frames of workpiece images to be welded into a pre-trained neural network to confirm the welding area to be welded of the workpiece to be welded, including:

[0078] S301. Based on the pixel instantaneous movement speed image and pixel difference image from the time T + 1 to the time T + L + 1, confirm the welding area to be welded at the time T + L + 1; wherein, the pixel instantaneous movement speed image and pixel difference image from the time T + 1 to the time T + L are pre-stored;

[0079] Repeat the steps: obtain the pixel instantaneous movement speed image and pixel difference image corresponding to the time T + L + 1; and the step: based on the pixel instantaneous movement speed image and pixel difference image from the time T + 1 to the time T + L + 1, confirm the welding area to be welded at the time T + L + 1; to obtain the welding areas to be welded from the time T + L + 1 to the time T + L + N; and fit the welding areas to be welded from the time T + L + 1 to the time T + L + N to obtain the final welding area to be welded.

[0080] In this embodiment, each pixel has an instantaneous movement speed and a pixel difference with numerical values. By replacing them with pixel values, the corresponding image can be obtained.

[0081] In this embodiment, in order to obtain the area to be welded at the current moment, the pixel instantaneous movement speed image and the pixel difference image at the previous L moments are used. To reduce the computational complexity, the pixel instantaneous movement speed image and the pixel difference image at the previous L moments are pre-stored instead of being calculated in real time.

[0082] In other words, as time progresses, in the process of continuously generating the area to be welded at the current moment, only the pixel instantaneous movement speed image and the pixel difference image at the current moment need to be calculated at each moment, and then the pixel instantaneous movement speed image and the pixel difference image calculated at the previous L moments are reused.

[0083] Referring to Figure 3 , further, the marker 20 includes: a first part and a second part that are perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of workpieces to be welded; Step S301, based on the pixel instantaneous movement speed image and the pixel difference image from the (T + 1)-th moment to the (T + L + 1)-th moment, confirming the area to be welded at the (T + L + 1)-th moment, includes:

[0084] S3011. Feed the pixel instantaneous movement speed image, the pixel difference image, and the image of the area to be welded at each moment from the (T + 1)-th moment to the (T + L + 1)-th moment into a three-way parallel convolutional neural network; the output of the three-way parallel convolutional neural network is connected to a fully connected neural network, and the fully connected neural network outputs the spatial feature information from the (T + 1)-th moment to the (T + L + 1)-th moment;

[0085] S3012. Feed the spatial feature information from the (T + 1)-th moment to the (T + L + 1)-th moment into a temporal neural network in chronological order to obtain spatio-temporal feature information;

[0086] S3013. Establish a coordinate system based on the marker 20; and determine the coordinate information of the area to be welded according to the spatio-temporal feature information.

[0087] The neural network architecture proposed in this application is as Figure 4 shown. The pixel instantaneous movement speed image, the pixel difference image, and the image of the area to be welded at each moment are fed into their respective convolutional neural networks for feature extraction to obtain three feature maps. Then, the flattening layer flattens the feature maps into feature vectors. Finally, the three feature vectors are concatenated and then fed into a fully connected neural network to output the spatial feature information. In this embodiment, the convolutional kernel sizes of the convolutional neural networks decrease sequentially.

[0088] Then, in chronological order, the spatial feature information is fed into a temporal neural network. At this time, the features output at each moment contain both temporal information and spatial information, so it is called spatio-temporal feature information. Exemplarily, the temporal neural network in this embodiment is preferably an LSTM network or a GRU network.

[0089] Finally, a coordinate system is established based on the marker 20, and then the coordinates of the weld 10 are confirmed based on the spatio-temporal feature information.

[0090] In this embodiment, a coordinate system is established based on the marker 20, so that the workpiece to be welded can be placed arbitrarily, and the position of the weld 10 in the coordinate system can be recognized.

[0091] In some embodiments, step S600, obtaining the pixel instantaneous movement speed image and the pixel difference image corresponding to the T+L+1 moment, includes:

[0092] Based on the coordinate information of the workpiece to be welded in the previous moment, the cutting area at the current moment is confirmed, and the area of the cutting area is at least twice the area of the workpiece to be welded area.

[0093] A cutting image is generated based on the cutting area, where the size of the cutting image is the same at each moment;

[0094] Based on the cutting image, the corresponding pixel instantaneous movement speed image and pixel difference image are generated.

[0095] In this embodiment, since the size of the cutting image is the same at each moment, the pixel instantaneous movement speed image and the pixel difference image extraction model can be trained based on the cutting images of the training data first. Then, in actual application, only the cutting image needs to be fed into the extraction model to obtain the pixel instantaneous movement speed image and the pixel difference image, effectively reducing the calculation amount.

[0096] In some embodiments, the laser welding method further includes: controlling the photographing unit to photograph the image of the welded workpiece; and confirming the welding result according to the determined workpiece to be welded area and the welded image.

[0097] In this embodiment, confirming the welding result according to the determined workpiece to be welded area and the welded image can be achieved by a traditional neural network. However, in order to reduce the calculation amount, and since the difference between the workpiece to be welded image and the welded image lies only in the welding area, a siamese network is used to judge the welding effect, with a small calculation amount and sufficient accuracy.

[0098] Therefore, the confirming the welding result according to the determined workpiece to be welded area and the welded image includes:

[0099] Based on the marker 20, align the welded image to the coordinate system;

[0100] Send the image of the area to be welded corresponding to the coordinate information of the area to be welded and the image of the welded area into the siamese neural network;

[0101] The siamese neural network outputs a welding result based on the similarity between the image of the area to be welded and the image of the welded area; when the similarity is greater than a preset threshold, the welding result is unwelded.

[0102] In practical applications, the siamese network can calculate the similarity between two input images (the image of the area to be welded and the image of the welded area). When the similarity is greater than the preset threshold, the siamese network can determine that it is unwelded. When the similarity is less than the preset threshold, the siamese network can determine that it is welded. Output the welding results: unwelded and welded.

[0103] Embodiment 2

[0104] This application also proposes a laser welding device, which includes: a photographing unit, a welding unit, and a control unit. The control unit is respectively connected to the photographing unit and the welding unit. The control unit is configured to execute a laser welding program to implement the above laser welding method. In some embodiments, the laser welding method includes:

[0105] S100. Control the photographing unit to obtain multiple frames of images of the workpiece to be welded according to the photographing trajectory;

[0106] S200. Confirm the motion information of the area to be welded according to the instantaneous pixel movement speed and pixel difference corresponding to multiple frames of the images of the workpiece to be welded;

[0107] S300. Send the motion information and the image information of multiple frames of the images of the workpiece to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded;

[0108] S400. Control the welding unit to weld the area to be welded according to the target welding trajectory corresponding to the area to be welded.

[0109] In some embodiments, the workpiece to be welded is provided with a mark 20, and the laser welding method further includes:

[0110] Control the photographing unit to photograph the image of the workpiece to be welded, and the image of the workpiece to be welded includes the mark 20;

[0111] Confirm the initial welding trajectory based on the mark 20, where the initial welding trajectory corresponds to the mark 20 one by one;

[0112] Confirm the photographing trajectory based on the initial welding trajectory.

[0113] In some embodiments, the mark 20 includes a first part and a second part that are perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of workpieces to be welded.

[0114] In some embodiments, the number of the marks 20 is four, which are respectively located at the four corners of the workpiece to be welded to meet the requirements for establishing a coordinate system.

[0115] In some embodiments, the method further includes:

[0116] S500. Obtain the pixel instantaneous movement speed and the pixel difference of the adjacent-frame workpiece images to be welded corresponding to the time from T to T + L, and obtain the pixel instantaneous movement speed image and the pixel difference image corresponding to the time from T to T + L;

[0117] S600. Obtain the pixel instantaneous movement speed image and the pixel difference image corresponding to the time T + L + 1;

[0118] Step S300. Send the motion information and the image information of multiple frames of workpiece images to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded, including:

[0119] S301. Based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1, confirm the area to be welded at the time T + L + 1; wherein, the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L are pre-stored;

[0120] Repeat the steps: obtain the pixel instantaneous movement speed image and the pixel difference image corresponding to the time T + L + 1; and the step: based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1, confirm the area to be welded at the time T + L + 1; to obtain the areas to be welded from the time T + L + 1 to the time T + L + N; and fit the areas to be welded from the time T + L + 1 to the time T + L + N to obtain the final area to be welded.

[0121] In some embodiments, the mark 20 includes: a first part and a second part that are perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of workpieces to be welded; step S301. The step of confirming the area to be welded at the time T + L + 1 based on the pixel instantaneous movement speed image and the pixel difference image from the time T + 1 to the time T + L + 1 includes:

[0122] S3011. Feed the pixel instantaneous movement speed image, pixel difference image, and the image to be welded at each moment from time T+1 to time T+L+1 into a three-way parallel convolutional neural network; the output of the three-way parallel convolutional neural network is connected to a fully connected neural network, and the fully connected neural network outputs the spatial feature information from time T+1 to time T+L+1.

[0123] S3012. Feed the spatial feature information from time T+1 to time T+L+1 into a temporal neural network in chronological order to obtain spatio-temporal feature information.

[0124] S3013. Establish a coordinate system based on the marker 20; and determine the coordinate information of the area to be welded according to the spatio-temporal feature information.

[0125] In some embodiments, step S600 of obtaining the pixel instantaneous movement speed image and the pixel difference image corresponding to time T+L+1 includes:

[0126] Based on the coordinate information of the area to be welded at the previous moment, confirm the cutting area at the current moment, and the area of the cutting area is at least twice the area of the area to be welded.

[0127] Generate a cutting image based on the cutting area, where the size of the cutting image is the same at each moment.

[0128] Generate the corresponding pixel instantaneous movement speed image and pixel difference image based on the cutting image.

[0129] In some embodiments, the control unit is further configured to:

[0130] Control the imaging unit to capture an image of the welded part; and

[0131] Confirm the welding result according to the determined area to be welded and the welded image.

[0132] In some embodiments, the confirming the welding result according to the determined area to be welded and the welded image includes:

[0133] Align the welded image to the coordinate system based on the marker 20;

[0134] Feed the image of the area to be welded corresponding to the coordinate information of the area to be welded and the image of the welded area into a siamese neural network;

[0135] The siamese neural network outputs a welding result based on the similarity between the image of the area to be welded and the image of the welded area; when the similarity is greater than a preset threshold, the welding result is unwelded.

[0136] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0137] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A laser welding method, characterized in that: The method comprises: Controlling the shooting unit to obtain multiple frames of images of the parts to be welded according to the shooting trajectory; before the step of controlling the shooting unit to obtain multiple frames of images of the parts to be welded according to the shooting trajectory, it also includes: controlling a turntable on which the parts to be welded are installed to move to a target area so that the parts to be welded are located within the field of view of the shooting unit; Confirming the motion information of the area to be welded according to the instantaneous moving speed of pixels and pixel differences corresponding to the multiple frames of images of the workpiece to be welded; Sending the motion information and the image information of the multiple frames of images of the workpiece to be welded into a pre-trained neural network to confirm the area to be welded of the workpiece to be welded; Controlling the welding unit to weld the area to be welded with a target welding trajectory corresponding to the area to be welded; wherein the part to be welded is provided with a mark; and the laser welding method further comprises: Control the shooting unit to shoot the image of the workpiece to be welded, the image of the workpiece to be welded includes the mark; confirming an initial welding trajectory based on the mark, wherein the initial welding trajectory corresponds to the mark one by one; Based on the initial welding trajectory, confirming the shooting trajectory; Each frame of the image of the workpiece to be welded includes a complete area to be welded; the method further includes: Obtain the instantaneous moving speed and pixel difference of the image of the workpiece to be welded in adjacent frames corresponding to the time from time T to time T+L, and obtain the instantaneous moving speed image and pixel difference image of the pixel corresponding to the time from time T to time T+L; Obtain the pixel instantaneous moving speed image and pixel difference image corresponding to time T+L+1; The step of sending the motion information and the image information of the multiple frames of the images of the parts to be welded into a pre-trained neural network to confirm the areas to be welded of the parts to be welded includes: Based on the pixel instantaneous moving speed image and the pixel difference image from time T+1 to time T+L+1, confirm the area to be welded at time T+L+1; wherein the pixel instantaneous moving speed image and the pixel difference image from time T+1 to time T+L in this step are pre-stored; Repeat the steps of: obtaining the pixel instantaneous moving speed image and the pixel difference image corresponding to the time T+L+1; and the steps of: confirming the area to be welded at the time T+L+1 based on the pixel instantaneous moving speed image and the pixel difference image from the time T+1 to the time T+L+1; obtaining the area to be welded from the time T+L+1 to the time T+L+N; and fitting the final area to be welded according to the area to be welded from the time T+L+1 to the time T+L+N; The mark includes: a first part and a second part that are perpendicular to each other to align the horizontal axis and the vertical axis of the coordinate system; and the shape of the first part corresponds to different types of parts to be welded; the pixel instantaneous movement speed image and the pixel difference image from time T+1 to time T+L+1 are used to confirm the area to be welded at time T+L+1, including: The instantaneous pixel moving speed image, the pixel difference image and the image to be welded at each moment from time T+1 to time T+L+1 are sent to a three-way parallel convolutional neural network; the outputs of the three-way parallel convolutional neural network are connected to a fully connected neural network, and the fully connected neural network outputs the spatial feature information from time T+1 to time T+L+1; In chronological order, the spatial feature information from time T+1 to time T+L+1 is sent to the time series neural network to obtain the spatiotemporal feature information; Based on the marks, a coordinate system is established; and according to the spatiotemporal characteristic information, the coordinate information of the area to be welded is determined.

2. The laser welding method according to claim 1, characterized in that: The angle between the shooting track and the initial welding track is greater than or equal to 80 degrees and less than or equal to 100 degrees.

3. The laser welding method according to claim 1, characterized in that: Obtain the instantaneous pixel moving speed image and pixel difference image corresponding to the time T+L+1, including: Based on the coordinate information of the area to be welded at the previous moment, determining the cutting area at the current moment, the cutting area includes the area to be welded, and the area of ​​the cutting area is at least twice the area of ​​the area to be welded; Generate a cropped image based on the cropped area, wherein the size of the cropped image at each moment is consistent; Based on the cropped image, the corresponding pixel instantaneous movement speed image and pixel difference image are generated.

4. The laser welding method according to claim 1, characterized in that: The method further comprises: Controlling the photographing unit to photograph an image of the welded part; and The welding result is confirmed based on the determined area to be welded and the welded image.

5. The laser welding method according to claim 4, characterized in that: The step of confirming the welding result based on the determined area to be welded and the welded image includes: aligning the welded image to the coordinate system based on the marking; Sending the image of the area to be welded and the image of the welded area corresponding to the coordinate information of the area to be welded into the twin neural network; The twin neural network outputs a welding result based on the similarity between the image of the area to be welded and the image of the welded area; when the similarity is greater than a preset threshold, the welding result is unwelded.

6. A laser welding device, characterized in that: include: Filming unit; Welding unit; as well as A control unit is connected to the shooting unit and the welding unit respectively, and the control unit is configured to execute a laser welding program to implement the laser welding method according to any one of claims 1 to 5.

Images