Machine vision-based aluminum alloy friction stir welding control system and control method

Abstract

The application discloses a kind of aluminum alloy friction stir welding control systems based on machine vision, comprising: industrial friction welding robot, tooling fixture, workbench, processor, image collector, temperature collector, distance measurer;Image collector is set to the working end of industrial friction welding robot;Temperature collector is set to the working end of industrial friction welding robot;Tooling fixture has two, two tooling fixtures are oppositely arranged on workbench, and the clamping of being welded is carried out;Two groups of distance measurers are respectively horizontally arranged in two tooling fixtures;The measuring end of two groups of distance measurers is vertically oriented weld, and oppositely arranged;Processor is connected with image collector, temperature collector, distance measurer and industrial friction welding robot respectively.The application realizes completely adaptive friction stir welding control based on machine vision, entire welding process is described by three-dimensional image, can realize automatic identification and eliminate defect, fill the blank in the field.

Description

Technical Field

[0001] This invention relates to the field of friction stir welding technology, and more specifically to a machine vision-based control system and method for aluminum alloy friction stir welding. Background Technology

[0002] Aluminum alloys possess excellent plasticity, corrosion resistance, and low density, making them widely used in aerospace, shipbuilding, and automotive manufacturing. Welding aluminum alloys with dissimilar metals is particularly prevalent; for example, welding aluminum alloys with stainless steel combines the plasticity and low density of aluminum with the high hardness and corrosion resistance of steel. However, due to significant differences in material properties between these two metals, traditional welding methods often result in defects such as cracks and porosity, and some metals cannot even be fused with aluminum. Friction stir welding (FSW) is a novel solid-state welding technology that achieves connection primarily through the plastic flow between high-temperature metals, avoiding the cracks and other defects associated with traditional fusion welding. Therefore, it has become the preferred method for welding dissimilar alloy materials.

[0003] The quality of the surface formation of friction stir welds directly affects the weld strength. In recent years, most domestic and foreign scholars have relied on direct visual inspection to judge the welding quality of friction stir welds, and these judgments are mostly qualitative rather than quantitative. Some have used ultrasonic phased array methods to detect internal defects in the weld, using ultrasonic waves to detect the internal quality of the weld on the equipment. Others have used fractal dimension calculations or neural network algorithm models (CN114418933A: YOLO-based diagnostic method, terminal, and storage medium for friction stir welding). However, these studies are either limited to qualitative judgment of weld quality or only perform delayed quantitative detection after welding is completed, without using the detection of the welded parts during the welding process to make real-time adaptive adjustments to the welding equipment to optimize subsequent welding quality, thus still having certain limitations. In terms of friction stir welding control, the existing robotic friction welding control system technology (publication number: CN104607795A Robotic Friction Stir Welding System and its Force-Position Hybrid Control Method) only uses preset parameters for welding. During the welding process, the vision sensor and displacement sensor only play the role of real-time monitoring and image acquisition. Its control logic is essentially a fixed parameter, fixed pressure, and fixed trajectory closed-loop control, which cannot identify defects and adjust parameters in real time according to the acquired defects. Summary of the Invention

[0004] This invention provides a machine vision-based control system and method for aluminum alloy friction stir welding, which solves the problem that adaptive control processing cannot be achieved in friction stir welding processes in the prior art.

[0005] This invention provides a machine vision-based aluminum alloy friction stir welding control system, comprising: an industrial friction welding robot, tooling fixtures, a worktable, a processor, an image acquisition unit, a temperature acquisition unit, and a distance measuring unit;

[0006] An image acquisition device is installed at the working end of the industrial friction welding robot to acquire images of the surface of the welding point of the workpiece.

[0007] The temperature acquisition device is set at the working end of the industrial friction welding robot, and the temperature acquisition device collects the surface temperature of the welding point of the workpiece to be welded.

[0008] There are two tooling fixtures, which are set opposite each other on the worktable to hold the workpiece to be welded;

[0009] There are two sets of distance measuring instruments, which are horizontally set in two tooling fixtures respectively. The measuring range of the distance measuring instruments covers the surface of the workpiece to be welded. The measuring ends of the two sets of distance measuring instruments are vertically oriented towards the weld and are set opposite to each other. The distance measuring instruments measure the height and width of the flash during the welding process.

[0010] The processor is connected to the image acquisition unit, temperature acquisition unit, distance measuring unit, and industrial friction welding robot, respectively. The processor controls the industrial friction welding robot to weld based on the data collected by the image acquisition unit, temperature acquisition unit, and distance measuring unit.

[0011] Furthermore, the distance measuring device is an array-type multi-point infrared ranging sensor.

[0012] This invention also provides a machine vision-based control method for aluminum alloy friction stir welding, applicable to the aforementioned machine vision-based aluminum alloy friction stir welding control system. The control method includes the following steps:

[0013] Step 1: When friction stir welding begins, acquire images of the weld bead at the weld point and the welding temperature;

[0014] Step 2: Preprocess the weld image to generate a grayscale image of the welding temperature;

[0015] Step 3: Combine the preprocessed weld bead image with the grayscale image to create a two-dimensional image containing the welding temperature;

[0016] Step 4: Obtain the height and width of the welding flash;

[0017] Step 5: Combine the flash height and flash width with the 2D image to create a 3D image containing welding texture, welding temperature, and flash data;

[0018] Step 6: Convolve the 3D image using the trained convolutional neural network. When the convolutional neural network determines that a welding defect exists, it outputs the welding defect type and the corresponding quantization parameters.

[0019] Step 7: Use the quantification parameters corresponding to welding defects and the optimized values ​​of welding parameters as input to the trained RBF neural network. The neural network outputs the optimized parameters of each joint of the robot and the corresponding robot control commands.

[0020] The method for obtaining the optimized welding parameters is as follows:

[0021] A three-dimensional surface for rotational speed, moving speed, and axial downward pressure was constructed using orthogonal experimental design. Each point on the three-dimensional surface represents the absence of welding defects.

[0022] The difference between the quantitative parameters corresponding to welding defects and the preset quantitative parameters corresponding to no welding defects is calculated. The difference is then fitted to a three-dimensional surface. During the fitting process, the changes in rotation speed, moving speed, and axial downward pressure are used as the optimization parameters for welding.

[0023] Step 8: Control the industrial friction welding robot to weld using optimized robot control instructions.

[0024] Furthermore, the specific method of step 3 is as follows:

[0025] The weld bead image is converted into a two-dimensional matrix, and the grayscale image of the welding temperature is also converted into a two-dimensional matrix. Both two-dimensional matrices have the same number of rows and columns.

[0026] Construct a two-dimensional matrix from the two-dimensional image, where the number of rows and columns of the two-dimensional matrix from the two-dimensional image is the same as the number of rows and columns of the two-dimensional matrix from the weld bead image.

[0027] Each element in the two-dimensional matrix of the two-dimensional image is a one-dimensional array with two elements formed by the elements in the two-dimensional matrix of the weld bead image and the two-dimensional matrix of the grayscale image, which are in the same row and column.

[0028] Transform a two-dimensional matrix of a two-dimensional image into a two-dimensional image.

[0029] Furthermore, the specific method of step 4 is as follows:

[0030] Based on the two-dimensional matrix data output by the distance measuring device, the columns in the two-dimensional matrix data represent the burr height, and the height of the burr in the pixels of a column is calculated by the number of non-zero data in a column; the values ​​in the two-dimensional matrix data represent the distance from the burr to the distance measuring device, and the burr width is calculated by the difference between the values ​​of the two distance measuring devices on both sides with the same burr height.

[0031] Furthermore, the specific method of step 5 is as follows:

[0032] Construct a two-dimensional matrix of the glitter height and glitter width, where the number of columns in the glitter matrix is ​​the same as the number of rows in the two-dimensional matrix of the two-dimensional image; the number of rows in the glitter matrix represents the glitter height coefficient; and the elements of the glitter matrix represent the glitter width coefficient.

[0033] Match the column number of the non-zero element in the 2D matrix of the flash with the row number of the 2D matrix of the 2D image. Match the value of the non-zero element in the 2D matrix of the flash with the column number of the 2D matrix of the 2D image. Find the one-dimensional array in the 2D matrix of the 2D image corresponding to the non-zero element. Take the difference between the row number of the non-zero element in the 2D matrix of the flash and the total number of rows as the third element of the one-dimensional array in the 2D matrix of the corresponding 2D image.

[0034] The third element is added to the one-dimensional array of the two-dimensional matrix of the two-dimensional image to complete the three-dimensional image synthesis.

[0035] Furthermore, in the two-dimensional matrix of the fly edge, all elements in the last row are set to "0".

[0036] Furthermore, in the two-dimensional matrix of the fly edge, all elements in the last row are set to "1", and the third element in the one-dimensional array of the two-dimensional image corresponding to the element in the last row of the two-dimensional matrix of the fly edge is set to "0".

[0037] Furthermore, in step 6, when convolving the three-dimensional image using the trained convolutional neural network, the inner product of the RGB color channel images in the three-dimensional image is performed respectively, and the values ​​of flash height, flash width, and welding temperature in the three-dimensional image are logically judged according to the threshold.

[0038] Furthermore, in step 6, the threshold in each convolution process is determined based on the new RGB color channel image obtained in each convolution.

[0039] Furthermore, in step 7, the model for constructing the RBF neural network is as follows:

[0040]

[0041] In the formula, S is the defect feature parameter; J is the Jacobian matrix solved in the robot coordinate system and the image coordinate system; q is the joint angle change; and θ is the joint angular velocity.

[0042] The beneficial effects of this invention are:

[0043] This invention achieves fully adaptive machine vision-based friction stir welding control. The entire welding process is described by three-dimensional images. By processing the three-dimensional images, defects can be automatically identified and eliminated. The control scheme fills a gap in the field of automated friction stir welding. It can also intuitively display various defect information to the operator. Finally, an empirical processing surface is established, which can provide a certain reference for subsequent processing when the optimal processing parameters cannot be determined.

[0044] This invention, through image and temperature acquisition, can quickly obtain welding information of the welded parts and display it intuitively to the operator; grayscale conversion reduces the information complexity of subsequent image merging and defect extraction, improving computational efficiency; merging surface defect images with temperature grayscale images can transform them into a set of two-dimensional mathematical matrices containing an inner matrix, converting image information into digital information that can be processed more accurately by the host computer; fusing flash data with two-dimensional images into three-dimensional images achieves dimensionality enhancement of defect features, transforming traditional 2D information into a more intuitive three-dimensional information digital matrix; during CNN convolutional neural network processing, convolution and threshold judgment are performed separately for different information within the matrix based on their weights and representations, thereby maximizing the preservation and extraction of feature information; special processing of the last row of data in the two-dimensional matrix of flash can effectively avoid errors caused by the feedback signal from the weld root when the subsequent convolutional neural network extracts flash defect feature parameters; by establishing a three-dimensional surface and using a fitting method to obtain the optimized amount of welding parameters, the target control quantity can be determined, enabling more accurate and efficient welding control. Attached Figure Description

[0045] The features and advantages of the invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the drawings:

[0046] Figure 1 This is a structural diagram of a specific embodiment of the present invention;

[0047] Figure 2 This is a flowchart of a specific embodiment of the present invention;

[0048] Figure 3 The image is obtained during processing in a specific embodiment of the present invention;

[0049] Figure 4 This is an overall view of infrared measurement in a specific embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the infrared measurement principle in a specific embodiment of the present invention;

[0051] Figure 6 This is a control system diagram of a specific embodiment of the present invention;

[0052] Figure 7 This is an empirical joint surface diagram in a specific embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments, and these embodiments do not constitute a limitation on the embodiments of the present invention.

[0055] like Figure 1 As shown in the figure, a specific embodiment of the present invention provides a machine vision-based aluminum alloy friction stir welding control system, which is mainly aimed at the control of vision-based friction stir welding robots. As can be seen from the image, the control system design includes: a six-degree-of-freedom industrial friction welding robot 1, an image acquisition device 2, welding tools 3, a temperature acquisition device 4, a host computer processing system 5, an infrared distance measuring device 6, a workpiece to be welded 7, a worktable 8, and tooling fixtures 9.

[0056] Image acquisition device 2 is set at the working end of the six-degree-of-freedom industrial friction welding robot 1. The image acquisition device acquires surface images of the welding points of the workpiece 7 to be welded.

[0057] A temperature acquisition device is set at the working end of the six-degree-of-freedom industrial friction welding robot 1, and the temperature acquisition device collects the surface temperature of the welding point of the workpiece 7 to be welded.

[0058] There are two tooling fixtures 9, which are set opposite to each other on the worktable 8 to hold the workpiece 7 to be welded;

[0059] Infrared distance measuring device 6 is an array-type multi-point infrared ranging sensor, and its output data is a matrix, the size of which is the same as the resolution of infrared distance measuring device 6. For example... Figure 5As shown, there are two sets of infrared distance measuring devices 6, which are horizontally set in two tooling fixtures. The measuring range of the infrared distance measuring devices 6 covers the surface of the workpiece 7 to be welded. The measuring ends of the two sets of infrared distance measuring devices 6 are vertically facing the weld seam on the water surface and are set opposite to each other. The infrared distance measuring devices 6 emit ranging infrared light towards the weld seam. When the infrared light hits the flash, it will be reflected back to the receiving end of the infrared distance measuring devices 6. At this time, the distance between the flash and the infrared distance measuring devices 6 can be obtained by simple calculation. Since it is an array-type multi-point infrared ranging sensor, the height of the flash can be easily obtained by the height of the infrared reflection received.

[0060] The host computer processing system 5 is connected to the image acquisition device 2, the temperature acquisition device 4, the infrared distance measuring device 6, and the six-degree-of-freedom industrial friction welding robot 1. The host computer processing system 5 controls the six-degree-of-freedom industrial friction welding robot 1 to weld based on the data collected by the image acquisition device 2, the temperature acquisition device 4, and the infrared distance measuring device 6.

[0061] This invention also provides a machine vision-based control method for aluminum alloy friction stir welding, which, as shown in the embodiments, Figure 2 As shown, the overall control system and method will be explained in detail by taking the welding of a 10mm thick 6061-T6 aluminum alloy sheet as an example.

[0062] The control method includes the following steps:

[0063] Step 1: When friction stir welding begins, acquire images of the weld bead at the weld point and the welding temperature;

[0064] When the welding system is started, the robotic welding system will begin welding with the initially preset welding parameters (spindle speed ω, travel speed v, and axial downward pressure S). Simultaneously, the image acquisition unit and temperature acquisition unit will start working, acquiring images of surface defects in the weld bead behind the tool and its temperature distribution, such as... Figure 3 As shown;

[0065] Step 2: Preprocess the weld image to generate a grayscale image of the welding temperature;

[0066] The host computer processing system preprocesses the acquired temperature and weld surface images. Its main tasks include preprocessing such as filtering and noise reduction of the acquired images, as well as the conversion and calibration of the temperature images. For example, for... Figure 3The temperature image acquired at a location is a color-coded image, and the actual temperature corresponding to it cannot be identified. Therefore, grayscale conversion is required. The specific conversion method is as follows: First, the color temperature field image is calibrated, with the melting temperature of 6061-T6 aluminum alloy (600 degrees Celsius) as the upper limit and 0 degrees Celsius as the lower limit. The actual measured temperature T is matched with the corresponding chromaticity Q in the acquired image, solving for the correspondence between image chromaticity and actual temperature: T = KQ, where K is the correlation coefficient. Simultaneously, the grayscale values ​​of 0-255 in the grayscale image are linearly mapped to the maximum temperature variation range of 0-600 degrees Celsius in actual welding, obtaining the relationship between grayscale value W and temperature T: W = 0.425T. At this point, the conversion relationship from the temperature acquisition image to the grayscale image is obtained: W = 0.425KQ. Using this relationship, the temperature image can be converted into a grayscale image, and the grayscale value at (x, y) of each location in the image can represent the actual temperature at that location. After processing, the appearance defect image matrix and temperature grayscale matrix are represented by L1 and L2 matrices respectively:

[0067]

[0068]

[0069] In the above formula, the number of rows and columns of the matrix is ​​equal to the number of rows and columns of the pixels in the image. In this example, for the sake of simplicity, we will use a 6*4 matrix for illustration.

[0070] Step 3: Combine the preprocessed weld bead image with the grayscale image to create a two-dimensional image containing weld bead texture and welding temperature. Specifically:

[0071] The weld bead image is converted into a two-dimensional matrix, and the grayscale image of the welding temperature is also converted into a two-dimensional matrix. Both two-dimensional matrices have the same number of rows and columns.

[0072] Construct a two-dimensional matrix from the two-dimensional image, where the number of rows and columns of the two-dimensional matrix from the two-dimensional image is the same as the number of rows and columns of the two-dimensional matrix from the weld bead image.

[0073] Each element in the two-dimensional matrix of the two-dimensional image is a one-dimensional array with two elements formed by the elements in the two-dimensional matrix of the weld bead image and the two-dimensional matrix of the grayscale image, which are in the same row and column.

[0074] Transform a two-dimensional matrix of a two-dimensional image into a two-dimensional image;

[0075] The newly acquired two-dimensional matrix L3 is shown below:

[0076]

[0077] Step 4: Obtain the height and width of the welding flash;

[0078] Based on the two-dimensional matrix data output by the distance measuring device, the columns in the two-dimensional matrix data represent the burr height, and the height of the burr in the pixels of a column is calculated by the number of non-zero data in a column; the values ​​in the two-dimensional matrix data represent the distance from the burr to the distance measuring device, and the burr width is calculated by the difference between the values ​​of the two distance measuring devices with the same burr height.

[0079] Specifically, its transmitting end emits multiple infrared beams horizontally, which are received by the receiving part on the opposite side of the weld, such as Figure 4 and Figure 5 As shown, due to the presence of defects, some signals are blocked and reflected back to the transmitting end. At this time, the change of defects in the height direction of the measurement position can be determined by the design logic. For the part received by the other side, 0 elements are assigned when generating the matrix. For the part reflected back, the horizontal and vertical positions from the transmitting position are solved based on the reflected information. The measurement results are shown in matrix L4:

[0080]

[0081] Step 5: Combine the flash height with the 2D image to create a 3D image containing flash height, flash width, and welding temperature. Specifically:

[0082] Construct a two-dimensional matrix of the glitter height and glitter width, where the number of columns in the glitter matrix is ​​the same as the number of rows in the two-dimensional matrix of the two-dimensional image; the number of rows in the glitter matrix represents the glitter height coefficient; and the elements of the glitter matrix represent the glitter width coefficient.

[0083] Match the column number of the non-zero element in the 2D matrix of the flash with the row number of the 2D matrix of the 2D image. Match the value of the non-zero element in the 2D matrix of the flash with the column number of the 2D matrix of the 2D image. Find the one-dimensional array in the 2D matrix of the 2D image corresponding to the non-zero element. Take the difference between the row number of the non-zero element in the 2D matrix of the flash and the total number of rows as the third element of the one-dimensional array in the 2D matrix of the corresponding 2D image.

[0084] To synthesize a three-dimensional image, the third element is added to the one-dimensional array of the two-dimensional matrix of the two-dimensional image that is missing the third element.

[0085] The previously obtained two-dimensional image matrix L3 and infrared matrix L4 are merged again to obtain a three-dimensional matrix L5 containing all defect information. Figure 3 Taking a location image as an example, its matrix is ​​shown below:

[0086]

[0087] The inner 1x3 matrix represents the texture, temperature, and height information of the weld bead location. For the third element of the 1x3 inner matrix, if it is zero, it indicates that in the (n) i ,m i The height of the burr defect at position ) is 0, meaning there is no burr.

[0088] Step 6: Convolve the 3D image using the trained convolutional neural network. When the convolutional neural network determines that a welding defect exists, it outputs the welding defect type and the corresponding quantization parameters.

[0089] Specifically, when convolving a 3D image using a trained convolutional neural network, the inner product of the RGB color channel images in the 3D image is performed, and the values ​​of flash height, flash width, and welding temperature in the 3D image are logically judged according to thresholds; the thresholds in each convolution process are determined based on the new RGB color channel images obtained in each convolution.

[0090] Using this image matrix L5 as input to a CNN (Convolutional Neural Network), image convolution is performed to extract its feature information. The final output is a set of defect feature parameters S, S = [S1, S2, ..., S...]. p ] T Among them, each of the S items i In this example, i ∈ (1, p) represents various types of defect information, such as S1 representing the flash height, S2 representing the temperature difference, S3 representing the groove depth, etc. The processing part of this image matrix L5 is the control chart. Figure 6 The feature extraction module in the middle compares it with the preset ideal defect feature parameters S. d =[S d1 .S d2 ,....S dp ] T The parameters that are not up to standard under the current welding condition are compared and the information is sent to the vision control processing module.

[0091] Step 7: Use the quantification parameters corresponding to welding defects and the optimized values ​​of welding parameters as input to the trained RBF neural network. The neural network outputs the optimized parameters of each joint of the robot and the corresponding robot control commands.

[0092] The method for obtaining the optimized welding parameters is as follows:

[0093] A three-dimensional surface for rotational speed, traverse speed, and axial downward pressure was constructed using orthogonal experimental design. Figure 7 As shown, each point on the three-dimensional surface represents the absence of welding defects.

[0094] The creation of a three-dimensional surface requires a pre-defined acceptable S-shape. dThe parameters are determined by manually controlling two processing parameter variables and changing a third parameter during actual processing. The evaluation considers whether the weld defect characteristic parameter S after processing conforms to the specified parameters. d The method is used to obtain the points of the three-dimensional surface in space. If the condition is not met, the set of parameters is discarded.

[0095] The difference between the quantitative parameters corresponding to welding defects and the preset quantitative parameters corresponding to no welding defects is calculated. The difference is then fitted to a three-dimensional surface. During the fitting process, the changes in rotation speed, moving speed, and axial downward pressure are used as the optimization parameters for welding.

[0096] Taking the S obtained from the L5 matrix as an example, the extracted defect feature parameters S = [1, 68, 4, 0]. T The preset ideal feature parameter is S. d =[1,20,2,0] T The error at this point is E = [0, 48, 2, 0]. T E = [0,0,0,0] T That is, the control target, in Figure 4 Find the nearest ideal E=0 surface in the empirical processing diagram, and calculate the required axial movement distance (△X, △Y, △Z) = (△ω, △v, △S) in the empirical processing three-dimensional space. This information is the target parameter that needs to be controlled by the RBF sliding mode controller.

[0097] The model for constructing the RBF neural network is as follows:

[0098] S = Jq

[0099]

[0100]

[0101] e1 = qq d

[0102]

[0103] S represents the defect feature, J is the Jacobian matrix obtained in the robot coordinate and image coordinate systems, and q represents the joint angle variation.

[0104] M represents the inertia matrix, C is the Coriolis force and centripetal force matrix, G is the gravity matrix, and τ is the joint torque. d External interference error compensation

[0105] e1 and e2 are sliding mode control laws, q d Expected value

[0106] [A,B,C] represent texture information, temperature information, and height information, respectively.

[0107] Step 8: Control the industrial friction welding robot to weld using optimized robot control instructions.

[0108] The output of the RBF sliding mode controller is the angle q and the speed θ of the joint changes of each axis of the robot's motion.

[0109] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A machine vision-based control system for aluminum alloy friction stir welding, characterized in that, include: Industrial friction welding robots, tooling fixtures, worktables, processors, image acquisition devices, temperature acquisition devices, and distance measuring devices; An image acquisition device is installed at the working end of the industrial friction welding robot to acquire images of the surface of the welding point of the workpiece. The temperature acquisition device is set at the working end of the industrial friction welding robot, and the temperature acquisition device collects the surface temperature of the welding point of the workpiece to be welded. There are two tooling fixtures, which are set opposite each other on the worktable to hold the workpiece to be welded; There are two sets of distance measuring devices, each set horizontally positioned within a separate fixture. The measuring range of each device covers the surface of the workpiece to be welded. The measuring ends of both sets of devices face vertically toward the weld seam and are positioned opposite each other. The distance measuring devices measure the height and width of the burr during welding. Specifically, based on the two-dimensional matrix data output by the distance measuring devices, the columns in the two-dimensional matrix data represent the burr height, and the height of the burr in a column is calculated by counting the number of non-zero data points in that column. The numerical values ​​in the two-dimensional matrix data represent the distance from the burr to the distance measuring device, and the burr width is calculated by the difference between the numerical values ​​of the two distance measuring devices with the same burr height. The processor is connected to an image acquisition unit, a temperature acquisition unit, a distance measuring unit, and an industrial friction welding robot. The processor controls the industrial friction welding robot's welding process based on data acquired by the image acquisition unit, temperature acquisition unit, and distance measuring unit. Specifically: Construct a two-dimensional matrix of the glitter height and glitter width, where the number of columns in the glitter matrix is ​​the same as the number of rows in the two-dimensional matrix of the two-dimensional image; the number of rows in the glitter matrix represents the glitter height coefficient; and the elements of the glitter matrix represent the glitter width coefficient. Match the column number of the non-zero element in the 2D matrix of the flash with the row number of the 2D matrix of the 2D image. Match the value of the non-zero element in the 2D matrix of the flash with the column number of the 2D matrix of the 2D image. Find the one-dimensional array in the 2D matrix of the 2D image corresponding to the non-zero element. Take the difference between the row number of the non-zero element in the 2D matrix of the flash and the total number of rows as the third element of the one-dimensional array in the 2D matrix of the corresponding 2D image. The third element is added to the one-dimensional array of the two-dimensional matrix of the two-dimensional image to complete the three-dimensional image synthesis, and the welding of the industrial friction welding robot is controlled based on the three-dimensional image.

2. The machine vision-based aluminum alloy friction stir welding control system as described in claim 1, characterized in that, The distance measuring device is an array-type multi-point infrared ranging sensor.

3. A machine vision-based control method for aluminum alloy friction stir welding, applicable to the machine vision-based aluminum alloy friction stir welding control system as described in claim 1 or 2, characterized in that, The control method includes the following steps: Step 1: When friction stir welding begins, acquire images of the weld bead at the weld point and the welding temperature; Step 2: Preprocess the weld image to generate a grayscale image of the welding temperature; Step 3: Combine the preprocessed weld bead image with the grayscale image to create a two-dimensional image containing the welding temperature; Step 4: Obtain the height and width of the welding burr. The specific method is as follows: Based on the two-dimensional matrix data output by the distance measuring device, the columns in the two-dimensional matrix data represent the burr height. The height of the burr in the pixels of a column is calculated by the number of non-zero data in a column. The values ​​in the two-dimensional matrix data represent the distance from the burr to the distance measuring device. The burr width is calculated by the difference between the values ​​of the distance measuring devices on both sides with the same burr height. Step 5: Combine the flash height and flash width with the 2D image to create a 3D image containing welding texture, welding temperature, and flash data. The specific method is as follows: Construct a two-dimensional matrix of the glitter height and glitter width, where the number of columns in the glitter matrix is ​​the same as the number of rows in the two-dimensional matrix of the two-dimensional image; the number of rows in the glitter matrix represents the glitter height coefficient; and the elements of the glitter matrix represent the glitter width coefficient. Match the column number of the non-zero element in the 2D matrix of the flash with the row number of the 2D matrix of the 2D image. Match the value of the non-zero element in the 2D matrix of the flash with the column number of the 2D matrix of the 2D image. Find the one-dimensional array in the 2D matrix of the 2D image corresponding to the non-zero element. Take the difference between the row number of the non-zero element in the 2D matrix of the flash and the total number of rows as the third element of the one-dimensional array in the 2D matrix of the corresponding 2D image. To synthesize a three-dimensional image, the third element is added to the one-dimensional array of the two-dimensional matrix of the two-dimensional image that is missing the third element. Step 6: Convolve the 3D image using the trained convolutional neural network. When the convolutional neural network determines that a welding defect exists, it outputs the welding defect type and the corresponding quantization parameters. Step 7: Use the quantification parameters corresponding to welding defects and the optimized values ​​of welding parameters as input to the trained RBF neural network. The neural network outputs the optimized parameters of each joint of the robot and the corresponding robot control commands. The method for obtaining the optimized welding parameters is as follows: A three-dimensional surface for rotational speed, moving speed, and axial downward pressure was constructed using orthogonal experimental design. Each point on the three-dimensional surface represents the absence of welding defects. The difference between the quantitative parameters corresponding to welding defects and the preset quantitative parameters corresponding to no welding defects is calculated. The difference is then fitted to a three-dimensional surface. During the fitting process, the changes in rotation speed, moving speed, and axial downward pressure are used as the optimization parameters for welding. Step 8: Control the industrial friction welding robot to weld using optimized robot control instructions.

4. The machine vision-based control method for aluminum alloy friction stir welding as described in claim 3, characterized in that, The specific method for step 3 is as follows: The weld bead image is converted into a two-dimensional matrix, and the grayscale image of the welding temperature is also converted into a two-dimensional matrix. Both two-dimensional matrices have the same number of rows and columns. Construct a two-dimensional matrix from the two-dimensional image, where the number of rows and columns of the two-dimensional matrix from the two-dimensional image is the same as the number of rows and columns of the two-dimensional matrix from the weld bead image. Each element in the two-dimensional matrix of the two-dimensional image is a one-dimensional array with two elements formed by the elements in the two-dimensional matrix of the weld bead image and the two-dimensional matrix of the grayscale image, which are in the same row and column. Transform a two-dimensional matrix of a two-dimensional image into a two-dimensional image.

5. The machine vision-based control method for aluminum alloy friction stir welding as described in claim 3, characterized in that, In the two-dimensional matrix of the fringe, all elements in the last row are set to "0". Alternatively, in the two-dimensional matrix of the fringe, all elements in the last row are set to "1", and the third element in the one-dimensional array of the two-dimensional image corresponding to the element in the last row of the two-dimensional matrix of the fringe is set to "0".

6. The machine vision-based control method for aluminum alloy friction stir welding as described in claim 3, characterized in that, In step 6, when convolving the 3D image using the trained convolutional neural network, the inner product of the RGB color channel images in the 3D image is performed, and the values ​​of flash height, flash width, and welding temperature in the 3D image are logically judged according to the threshold.

7. The machine vision-based control method for aluminum alloy friction stir welding as described in claim 6, characterized in that, In step 6, the threshold in each convolution process is determined based on the new RGB color channel image obtained in each convolution.

8. The machine vision-based control method for aluminum alloy friction stir welding as described in claim 3, characterized in that, In step 7, the model for constructing the RBF neural network is as follows: In the formula, S is the defect feature parameter; J is the Jacobian matrix solved in the robot coordinate system and the image coordinate system; q is the joint angle change; and θ is the joint angular velocity.

Images