Friction stir welding control method and system based on binocular vision arc perception

Abstract

The invention discloses a friction stir welding control method and system based on binocular vision arc perception, and relates to the field of friction stir welding. The method comprises the steps that a continuous image sequence of an arc area in the friction stir welding process is obtained through a binocular vision collecting device; extracting a pixel gray value, an edge contour coordinate and a texture feature parameter of the arc region; inputting the extracted characteristic parameters into a binocular arc three-dimensional reconstruction numerical model to generate a form reconstruction result of the arc in a three-dimensional space; calculating to obtain distribution data of a temperature field, a stress field and a flow field of the welding area; correcting abnormal data points in the distributed data by adopting an arc pixel neighborhood repair algorithm, and uploading the corrected data to an FSW process parameter optimization cloud platform; and the platform outputs the adjusted welding parameters, and the welding process is controlled in real time. The system comprises six units which work cooperatively. The accuracy of arc sensing and multi-physical field analysis is improved, and stable welding quality is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of friction stir welding, in particular to a friction stir welding control method and system based on binocular vision arc perception. BACKGROUND

[0002] In industrial manufacturing, friction stir welding is widely used in the fields of aerospace, automobile manufacturing and other fields with strict requirements on welding precision because it can achieve high-quality connection and has less heat effect on materials. With the increasing requirements of manufacturing process on welding quality stability and process controllability, the traditional welding control method relying on manual experience or single sensor monitoring has been difficult to meet the precise perception needs of arc shape and multi-physical field changes in the welding area under complex working conditions. Under this background, the binocular vision technology is introduced into the arc perception of the friction stir welding process, combined with three-dimensional reconstruction, multi-physical field coupling prediction, data repair and cloud platform parameter optimization and other technologies, to build an integrated welding control method and system, which is an important development direction to improve the intelligent level of the welding process and ensure the consistency of the welding quality, and can effectively solve the monitoring problems caused by splashing, reflection and other interference factors in the welding process, and provide technical support for realizing real-time and precise control of the welding process.

[0003] The existing technology has two significant shortcomings in the control of friction stir welding: on the one hand, when processing the arc image data and multi-physical field data collected in the welding process, the existing technology has difficulty in effectively solving the data anomaly problem, and lacks a reliable repair mechanism in the face of image information loss or multi-physical field calculation data deviation caused by arc area splashing, workpiece surface reflection, etc., which leads to insufficient accuracy of arc shape analysis and multi-physical field distribution judgment based on these data, and affects the precise grasp of the welding process state; on the other hand, the process parameter optimization of the existing technology mostly relies on local calculation or single model analysis, which cannot fully integrate multi-source data for comprehensive optimization, and the parameter adjustment lacks deep linkage with real-time arc perception data and multi-physical field change data, making it difficult to output adaptive parameters such as rotational speed, pressure and welding speed in time according to the dynamic changes in the welding process, which leads to the welding quality being easily affected by working condition fluctuations, and makes it difficult to achieve stable high-quality welding. SUMMARY

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present application provides a friction stir welding control method and system based on binocular vision arc perception.

[0005] The technical solution adopted in this invention is a control method for friction stir welding based on binocular vision arc perception, comprising: Step S1, acquiring a continuous image sequence of the arc region during friction stir welding using a binocular vision acquisition device, locating the arc region in the acquired image sequence, filtering out target image frames containing complete arc features, and excluding invalid image frames caused by welding spatter and workpiece surface reflection interference; Step S2, based on the filtered target image frames, extracting the pixel grayscale values, edge contour coordinates, and texture feature parameters of the arc region, establishing the mapping relationship between arc features and image pixels, and determining the basic feature dataset for subsequent 3D reconstruction; Step S3, inputting the basic feature dataset into a binocular arc 3D reconstruction numerical model, and obtaining the coordinate information, morphological parameters, and spatial distribution of the arc in 3D space through model calculation. The process involves several steps: Step S4: 1. Reconstructing the arc's three-dimensional shape; Step S5: 2. Calling the welding multiphysics coupling prediction model, using the arc's three-dimensional shape reconstruction result along with the rotational speed, pressure, and welding speed parameters during friction stir welding as model inputs, and calculating the distribution data of the temperature field, stress field, and flow field in the welding area; Step S6: 3. Identifying abnormal data points in the temperature field, stress field, and flow field distribution data output in Step S4, and using an arc pixel neighborhood repair algorithm to repair the image pixel areas corresponding to the abnormal data points, obtaining corrected multiphysics distribution data; Step S7: 4. Uploading the corrected multiphysics distribution data to the FSW process parameter optimization cloud platform, whereby the cloud platform outputs adjusted friction stir welding rotational speed, pressure, and welding speed parameters based on preset welding quality evaluation standards, enabling real-time control of the welding process.

[0006] Furthermore, the expression for the binocular arc three-dimensional reconstruction numerical model is as follows: ,in, Let be the depth coordinates of a point on the electric arc in three-dimensional space. The focal length of the lens in the binocular vision acquisition device. This is the baseline distance between the two cameras in the binocular vision acquisition device. In capturing images for the left camera The grayscale value of a pixel. In capturing images for the right camera The grayscale value of a pixel. The calculation range of the pixel neighborhood. This refers to the error correction coefficients during the 3D reconstruction process.

[0007] Furthermore, the expression for the welding multiphysics coupling prediction model is as follows: ,in, for welding area at all times Temperature at location, thermal diffusivity of welding material, temperature conduction velocity correction coefficient, welding speed, pressure thermal conversion coefficient, friction stir welding pressure, contact area of the stir head and the workpiece, welding area at time t stress at position x at time t, elastic modulus of welding material, welding area at time t strain at position x at time t, viscosity coefficient of material, thermal stress coefficient.

[0008] Further, the expression of the arc pixel neighborhood repair algorithm is: wherein, gray value of the repaired pixel point, weight coefficient of the pixel point, original gray value of the pixel point, selection range of the neighborhood pixel, Gaussian filter coefficient, neighborhood pixel offset of the pixel point in the direction of x, neighborhood pixel offset of the pixel point in the direction of y, neighborhood pixel offset of the pixel point in the direction of x, neighborhood pixel offset of the pixel point

[0009] Further, the calculation model adopted by the FSW process parameter optimization cloud platform when performing parameter optimization is: wherein, optimized welding speed, initial welding speed, current welding time, standard welding time, temperature at position x at time t in the standard welding process, maximum allowable stress of welding material, stress at position x at time t,

[0010] ​​​​​​​​Further, when acquiring the image sequence by using the binocular vision acquisition device, the image sharpness evaluation model is used to judge the quality of the acquired image, and the model expression is: wherein, is an image sharpness evaluation index, is the number of image width pixels, is the number of image height pixels, is the gray value of a pixel point, is the gray value of a pixel point, is the gray value of a pixel point.

[0011] Further, the step S3 comprises the following sub-steps: S31, classifying and arranging the pixel gray value, edge contour coordinate and texture feature parameter in the basic feature data set extracted in step S2, and dividing the feature data into a plurality of sub-data sets according to the spatial position of the arc region; S32, inputting each sub-data set into different calculation modules of the binocular arc three-dimensional reconstruction numerical model respectively, and each calculation module selects adaptive calculation parameters according to the feature type of the corresponding sub-data set; S33, acquiring the local three-dimensional reconstruction result output by each calculation module, splicing the local three-dimensional reconstruction result according to the spatial position information corresponding to the sub-data set, and forming a preliminary arc three-dimensional shape reconstruction result; S34, performing consistency verification on the spatial coordinate data in the preliminary arc three-dimensional shape reconstruction result, deleting the coordinate conflict points caused by data splicing, and obtaining the final arc three-dimensional shape reconstruction result.

[0012] Further, the step S4 comprises the following sub-steps: S41, extracting the arc center coordinate, arc radius and arc length parameter from the arc three-dimensional shape reconstruction result generated in step S3, integrating these parameters with the friction stir welding rotation speed, pressure and welding speed parameters into a model input data set; S42, inputting the input data set into the preprocessing module of the welding multi-physical field coupling prediction model, the module converts the format of the input data according to the model calculation requirements, and establishes the mapping relationship between the data and the model calculation unit; S43, starting the temperature field calculation unit, stress field calculation unit and flow field calculation unit of the model, and the three calculation units perform parallel calculation on the input data at the same time, and output corresponding temperature field distribution data, stress field distribution data and flow field distribution data respectively; S44, cross- verifying the distribution data output by the three calculation units, judging the rationality of the calculation result by comparing the numerical correlation of different physical field data at the same spatial position, and retaining the distribution data meeting the correlation requirements.

[0013] Further, the step S5 includes the following sub-steps: S51, detecting the temperature field, stress field and flow field distribution data output by step S4 point by point using an abnormal data identification rule, which sets a data threshold range based on the statistical distribution characteristics of similar data, and data outside the threshold range is marked as an abnormal data point; S52, determining the pixel position of each abnormal data point in the corresponding image, and defining a fixed size of neighborhood pixel region centered on the pixel position, recording the gray value and feature parameters of all normal pixel points in the neighborhood; S53, inputting the gray value and feature parameters of the normal pixel points in the neighborhood into the arc pixel neighborhood repair algorithm, which calculates the pixel gray correction value corresponding to the abnormal data point according to the spatial correlation and feature similarity of the neighborhood pixels; S54, substituting the calculated pixel gray correction value into the corresponding multi-physical field distribution data to replace the original value of the abnormal data point, and updating the values of the related data around the data point, and correcting the multi-physical field distribution data as a whole.

[0014] The stirring friction welding control system based on binocular vision arc perception is applied to the stirring friction welding control method based on binocular vision arc perception, and includes a binocular vision arc image acquisition unit, an arc feature data extraction and screening unit, a binocular arc three-dimensional reconstruction calculation unit, a welding multi-physical field coupling prediction unit, an arc pixel neighborhood repair and data correction unit, and an FSW process parameter optimization and control execution unit. The binocular vision arc image acquisition unit is connected with the arc feature data extraction and screening unit, the binocular vision arc image acquisition unit acquires a continuous image sequence of the arc region in the stirring friction welding process and transmits it to the arc feature data extraction and screening unit. The arc feature data extraction and screening unit is connected with the binocular arc three-dimensional reconstruction calculation unit, the arc feature data extraction and screening unit performs arc region positioning on the image sequence, screens target image frames and extracts feature parameters, and transmits the feature parameters to the binocular arc three-dimensional reconstruction calculation unit. The binocular arc three-dimensional reconstruction calculation unit is connected with the welding multi-physical field coupling prediction unit, the binocular arc three-dimensional reconstruction calculation unit generates an arc three-dimensional shape reconstruction result through model calculation and transmits it to the welding multi-physical field coupling prediction unit. The welding multi-physical field coupling prediction unit is connected with the arc pixel neighborhood repair and data correction unit, the welding multi-physical field coupling prediction unit outputs welding region multi-physical field distribution data and transmits it to the arc pixel neighborhood repair and data correction unit. The arc pixel neighborhood repair and data correction unit is connected with the FSW process parameter optimization and control execution unit, the arc pixel neighborhood repair and data correction unit transmits the corrected data to the FSW process parameter optimization and control execution unit after repairing the abnormal data. The FSW process parameter optimization and control execution unit outputs adjusted welding parameters according to the corrected data, and controls the operation state of the stirring friction welding equipment.

[0015] Beneficial effects: The application proposes a friction stir welding control method and system based on binocular vision arc perception, acquires a continuous image sequence of the arc region through a binocular vision acquisition device, screens target image frames, combines an arc pixel neighborhood repair algorithm, can effectively repair the missing image information caused by splashing shielding and workpiece surface reflection and the deviation of multi-physical field calculation data, solves the problems of lack of reliable repair mechanism in data processing of the prior art and insufficient analysis and judgment accuracy, and improves the accurate grasping ability of the welding process state; the three-dimensional reconstruction numerical model of the arc is generated through binocular arc three-dimensional reconstruction, the temperature field, stress field and flow field distribution data of the welding area are calculated through the welding multi-physical field coupling prediction model, and the corrected data is uploaded to the FSW process parameter optimization cloud platform to realize the comprehensive integration and dynamic optimization of multi-source data, break the limitations of the prior art, such as dependence on local calculation or single model, insufficient parameter adjustment and real-time data linkage, can output adaptive speed, pressure and welding speed parameters in time according to the dynamic changes of the welding process, ensure the welding quality from the influence of working condition fluctuation, and through multi-step data processing and multi-unit collaborative control, improve the intelligent level and quality consistency of the welding process, and provide effective support for accurate welding control under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The method of the application is a general flowchart;

[0017] Figure 2 The method of the application is a general flowchart;

[0018] Figure 3 The method of the application is a general flowchart;

[0019] Figure 4 The method of the application is a general flowchart;

[0020] Figure 5 The system of the application is a general flowchart. DETAILED DESCRIPTION

[0021] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict, and the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0022] As Figure 1 shown, the friction stir welding control method based on binocular vision arc perception comprises:

[0023] Step S1, acquire a continuous image sequence of the arc region in the friction stir welding process through a binocular vision acquisition device, locate the arc region of the acquired image sequence, screen out target image frames including complete arc features, and exclude invalid image frames caused by welding splashing and workpiece surface reflection interference;

[0024] Specifically, the specific implementation process of step S1 is as follows: two industrial-grade binocular vision acquisition devices with a resolution of 2560x1440 pixels and a frame rate of 60 frames / s are deployed and fixed on the two sides of the friction stir welding equipment respectively, the angle between the optical axis of the lens and the center line of the welding arc is adjusted to 30°-45°, the vertical distance between the lens and the welding area is controlled to be 800mm-1200mm, and the complete imaging of the arc area is ensured. When collecting, the device continuously acquires a continuous image sequence of the arc area, and the collection time is determined according to the length of the welded workpiece. For example, when the workpiece is 1000mm long and the welding speed is 50mm / min, the collection time is set to 20min, and 72000 frames of image sequence are generated. Then, the image screening module is started, the arc feature gray scale range 120-255 and the edge contour continuity (continuous pixel points ≥ 50) are taken as the judgment standard by using the preset arc area positioning algorithm, the target image frames containing complete arc features are screened out, and the invalid image frames caused by welding spatter (gray value < 50 and area > 100 pixels), workpiece surface reflection (gray value > 250 and area > 200 pixels) are excluded. This step provides high-quality and interference-free image data for subsequent feature extraction, avoids the influence of invalid data on the accuracy of model calculation, guarantees the reliability of the basic data of the control process, and ensures the feasibility and effectiveness of the implementation. All parameter settings are based on the common working conditions of industrial welding scene, which ensures the feasibility and effectiveness of the implementation.

[0025] Step S2, based on the target image frames screened out, the pixel gray value, edge contour coordinate and texture feature parameter of the arc area are extracted, the mapping relationship between the arc feature and the image pixel is established, and the basic feature data set for subsequent three-dimensional reconstruction is determined;

[0026] Specifically, the specific implementation process of step S2 is: calling image feature extraction software to process the target image frames screened in step S1 frame by frame, for each arc region, using gray histogram analysis technology to extract pixel gray value, and counting each pixel gray data to form a 2560x1440-dimensional gray value matrix; using an edge detection algorithm to identify the arc edge contour with a gradient value > 30 as the threshold, record the contour pixel coordinates to form a coordinate set, and the coordinate accuracy is 1 pixel; through a texture feature analysis algorithm to calculate the arc region texture contrast (0-100), texture entropy (0-8), texture correlation (-1-1) and other parameters, which are one-to-one corresponding to the image frames. Then a mapping relationship between the arc features and the image pixels is established, the pixel gray value, edge contour coordinates and texture feature parameters are associated and stored according to the image frame number to form a basic feature data set, each data set contains three types of feature data of the corresponding frame, and the storage format is binary to reduce the storage space and improve the subsequent calling speed. This step converts the image information into structured data that can be processed by the model, provides accurate and comprehensive input data for the binocular arc three-dimensional reconstruction numerical model, and ensures that the model can accurately calculate the three-dimensional morphological parameters of the arc. The data storage format and processing precision setting are adapted to the subsequent model calculation requirements.

[0027] Step S3, input the basic feature data set into the binocular arc three-dimensional reconstruction numerical model, obtain the coordinate information, morphological parameters and spatial distribution characteristics of the arc in the three-dimensional space through model calculation, and generate the arc three-dimensional morphological reconstruction result;

[0028] Specifically, the specific implementation process of step S3 is: on an industrial computer configured with an Intel Core i9-13900K CPU, an NVIDIA RTX 4090 GPU, and 64 GB of memory, start the binocular arc three-dimensional reconstruction numerical model calculation program, and set the calculation precision to 6 decimal places. Import the basic feature data set of step S2 into the model. The model first normalizes the pixel gray value (0-1 after processing), and converts the edge contour coordinates to the coordinate system. The midpoint of the baseline of the binocular device is taken as the origin, the X-axis is parallel to the baseline, the Y-axis is perpendicular to the baseline, and the Z-axis is perpendicular to the XY plane. The image pixel coordinate system is converted to the world coordinate system. Then, according to the binocular vision principle, combined with the 500 mm baseline distance and 16 mm lens focal length of the two devices, as well as the gray value difference and edge contour coordinate deviation of the basic feature data, the X, Y, and Z coordinates of each feature point in the arc three-dimensional space are calculated. The coordinate range is set as X∈[-250mm, 250mm], Y∈[-200mm, 200mm], and Z∈[0mm, 500mm]. At the same time, the arc length (distance between two end points in Z-axis direction), diameter (average value of maximum diameter of cross section), volume (three-dimensional space integral calculation), and spatial distribution characteristics (distribution density of feature point coordinate interval) are calculated. Finally, the coordinate information, morphological parameters, and distribution characteristics are integrated to generate the arc three-dimensional morphological reconstruction results output in the form of three-dimensional point cloud graph and data table. This step converts two-dimensional images into three-dimensional spatial information, intuitively presents the actual shape and position of the arc, provides real state input data for the welding multi-physical field coupling prediction model, and ensures the rationality of the multi-physical field calculation.

[0029] Step S4, call the welding multi-physical field coupling prediction model, take the arc three-dimensional morphological reconstruction results and the rotation speed, pressure, and welding speed parameters in the friction stir welding process as model inputs, and calculate the distribution data of the temperature field, stress field, and flow field in the welding area.

[0030] Specifically, the specific implementation process of step S4 is: starting the welding multi-physical field coupling prediction model on the industrial computer, the model containing temperature field, stress field, and flow field three calculation modules, all using finite element analysis method, grid division precision 1mmx1mmx1mm, time step 0.01s. From the arc three-dimensional shape reconstruction result of step S3, the core parameters such as arc center coordinates (feature point X, Y, Z coordinate average), length, diameter, volume are extracted, and real-time welding parameters are obtained from the welding equipment control system, welding speed 800-2000r / min (1200r / min for aluminum alloy workpiece), pressure 10-50kN (25kN for aluminum alloy workpiece), speed 30-100mm / min (50mm / min for aluminum alloy workpiece). The arc parameters and welding parameters are input into the model, and the model is first converted into the node load and boundary condition data required for finite element calculation. The temperature field module takes the arc heat radiation and heat conduction as the heat source, combined with the aluminum alloy specific heat capacity of 900J / (kg·K), thermal conductivity of 237W / (m·K), and melting point of 660℃, to calculate the temperature value of each grid node at different time steps; the stress field module is based on the temperature field result, considering the aluminum alloy elastic modulus of 70GPa, Poisson's ratio of 0.33, and thermal expansion coefficient of 23x10^-6 / ℃, to calculate the node stress value; the flow field module calculates the flow rate and direction of the molten pool metal according to the shear force generated by the welding speed and pressure, combined with the aluminum alloy molten state viscosity of 0.001Pa·s, density of 2700kg / m³. The three modules are calculated synchronously, and the multi-physical field distribution results containing the temperature, stress, and flow rate data of each node are output and stored in the form of cloud pictures and data files. This step comprehensively masters the changes of heat, force, and flow state in the welding area through multi-physical field coupling calculation, provides data support for judging welding quality and discovering potential defects, and lays a foundation for subsequent data repair and parameter optimization.

[0031] Step S5, identifying the abnormal data points in the temperature field, stress field, and flow field distribution data output by step S4, using the arc pixel neighborhood repair algorithm to repair the image pixel region corresponding to the abnormal data points, and obtaining the corrected multi-physical field distribution data;

[0032] Specifically, the specific implementation process of step S5 is: starting the data anomaly recognition program, loading the CSV format multi-physical field distribution data (including grid node number, coordinate, temperature value, stress value, flow rate value) output by step S4. The program uses statistical analysis method to process each kind of physical field data separately, first calculates the average value and standard deviation of the data, sets the abnormal threshold value as the average value ± 3 times the standard deviation, and marks the data beyond the range as abnormal data points. Then through the inverse conversion of the coordinate system, the abnormal node world coordinates are converted into image pixel coordinates (x, y), and a 5x5 pixel neighborhood area is defined with the coordinates as the center. The gray value and edge contour features of all normal pixel points (not corresponding to abnormal data) in the neighborhood are read, and the gray average value, contour curvature and other parameters are calculated. Start the arc pixel neighborhood repair algorithm, input the neighborhood normal pixel feature parameters, the algorithm calculates the pixel gray correction value corresponding to the abnormal data points according to the neighborhood pixel spatial position relationship (the closer to the center, the greater the weight), the feature similarity (the smaller the difference between the gray value and the contour curvature and the center, the greater the weight), and then calculates the physical field correction value of the abnormal data points through the gray value and physical field data corresponding curve established in the previous experiment. Finally, the physical field correction value is used to replace the original abnormal value, and the physical field data in the 10x10 grid range around the abnormal point is smoothed to avoid data mutation, and the corrected multi-physical field distribution data consistent with the original format is obtained. This step eliminates the abnormal deviation of multi-physical field data, ensures that the data truly reflects the state of the welding area, provides accurate input data for the FSW process parameter optimization cloud platform, and avoids data anomalies leading to parameter optimization deviation.

[0033] Step S6, upload the corrected multi-physical field distribution data to the FSW process parameter optimization cloud platform, and according to the preset welding quality evaluation standard, output the adjusted friction stir welding speed, pressure and welding speed parameters, and control the welding process in real time.

[0034] Specifically, the implementation process of step S6 is as follows: Establish a communication connection between the industrial computer and the FSW process parameter optimization cloud platform, using 5G or fiber optic communication (rate ≥100Mbps), with data transmission secured using the AES-256 encryption protocol. Package the corrected multiphysics distribution data from step S5 into a preset format (including acquisition time, workpiece number, and CSV format data file), and upload it to the cloud platform via communication. The cloud platform consists of a computing cluster composed of multiple Intel Xeon Platinum 8480+ CPUs and 256GB memory servers, with a storage capacity of 100TB. It has a built-in welding quality assessment standard database, storing assessment indicators for materials such as aluminum alloys, steel, and titanium alloys, as well as scenarios such as butt welding and fillet welding, including temperature field uniformity (fluctuation ≤50℃), stress field peak value (≤80% of material yield strength), and flow field stability (flow velocity fluctuation ≤0.1m / s). After receiving the data, the cloud platform initiates the analysis module to extract multi-physics field data and compare it with database standards. It analyzes temperature field uniformity, stress levels, and flow field stability. If the temperature fluctuates by 60℃, the stress reaches 90% of the material's yield strength, or the flow velocity fluctuates by 0.15m / s, it determines that parameters need adjustment. Subsequently, the parameter optimization module, combining the comparison results, arc 3D morphology data, and over 10,000 historical successful cases, uses a multi-objective optimization algorithm to calculate the adjusted parameters, such as adjusting the rotation speed from 1200r / min to 1300r / min, the pressure from 25kN to 23kN, and the speed from 50mm / min to 45mm / min. After optimization, the cloud platform feeds the parameters back to the industrial computer, which then transmits them to the welding equipment control system. The control system adjusts the motor speed, pressure actuator output, and welding platform movement speed according to the new parameters, achieving real-time control of the welding process. This step, through the cloud platform's computing power and data resources, enables dynamic optimization and real-time adjustment of welding parameters, ensuring the welding process meets quality standards, improving welding quality stability and consistency, while also achieving remote control and data sharing, and enhancing the intelligence and automation level of welding production.

[0035] Preferably, the expression for the binocular arc three-dimensional reconstruction numerical model is: ,in, Let be the depth coordinates of a point on the electric arc in three-dimensional space. The focal length of the lens in the binocular vision acquisition device. This is the baseline distance between the two cameras in the binocular vision acquisition device. In capturing images for the left camera The grayscale value of a pixel. In capturing images for the right camera The grayscale value of a pixel. The calculation range of the pixel neighborhood. This refers to the error correction coefficients during the 3D reconstruction process.

[0036] Specifically, during step S3, when performing binocular arc 3D reconstruction, a binocular arc 3D reconstruction numerical model is activated. Before implementing this model, core technical parameters need to be determined. The focal length of the lens of the binocular vision acquisition device is set to 16mm. This parameter is calibrated based on the imaging range of the welding area and the device installation distance (800mm-1200mm) to ensure clear capture of arc details. The baseline distance between the two cameras is fixed at 500mm. This distance was determined through multiple tests to avoid insufficient 3D reconstruction accuracy due to excessive distance, while also preventing increased deployment difficulty due to excessive distance. During model calculation, the grayscale values ​​of corresponding pixels in the images acquired by the left and right cameras are first obtained. The grayscale values ​​are directly read from the image sensor with a precision of 16 bits, capable of distinguishing subtle grayscale differences. The calculation range of the pixel neighborhood is set to 5 pixel units. This range is determined based on the texture density of the arc image to balance computational efficiency and reconstruction accuracy. The error correction coefficient is set to 0.001 to compensate for calculation deviations caused by lens distortion and ambient light interference. This was obtained through testing and calibration of 100 sets of standard arc models in the early stage. During implementation, the model first calculates the grayscale difference between corresponding pixels in the left and right camera images, then combines the neighborhood calculation range to statistically summarize the total grayscale difference, and finally substitutes it into the model to complete the depth coordinate calculation, generating the three-dimensional spatial position data of the electric arc. By clarifying the model parameters and calculation logic, the accuracy of the three-dimensional reconstruction of the electric arc is improved, and the reconstruction result error is controlled within 0.1mm. This provides accurate electric arc spatial morphology data for subsequent multiphysics coupling prediction, avoiding errors in welding status judgment due to reconstruction deviations.

[0037] Preferably, the expression for the welding multiphysics coupling prediction model is: ,in, for welding area at all times Temperature at location The thermal diffusivity of the welding material, This is a correction factor for the temperature conduction velocity. For welding speed, The pressure heat conversion coefficient, For friction stir welding pressure, This refers to the contact area between the stirring head and the workpiece. for welding area at all times Stress at location, The elastic modulus of the welding material, for welding area at all times Strain at position The viscosity coefficient of the material. This is the thermal stress coefficient.

[0038] Specifically, in step S4, the welding multi-physical field coupling prediction model is called when the welding multi-physical field coupling prediction is performed. The model includes two core calculation modules of temperature field and stress field, and the material and process related technical parameters are set in advance. If the welding material is aluminum alloy, the thermal diffusion coefficient is set to 7.6*10^-5 m² / s, which is measured by material thermal physical property experiment and reflects the heat conduction capacity of the material; the temperature conduction speed correction coefficient is set to 0.85, which is used to correct the influence of heat convection on the conduction speed in the welding process, and is calibrated by 20 groups of different welding speeds (30 mm / min-100 mm / min); the pressure heat conversion coefficient is set to 0.02, which is determined according to the correlation experiment of friction stir welding pressure (10 kN-50 kN) and heat output, and can accurately quantify the influence of pressure on temperature field; the contact area of the stirring head and the workpiece is set to 200 mm², which is calculated according to the stirring head diameter (15 mm) and the pressing amount (2 mm). In terms of material mechanics parameters, the elastic modulus of aluminum alloy is set to 70 GPa, the viscosity coefficient is set to 1.2*10^6 Pa・s, and the thermal stress coefficient is set to 15 MPa / ℃, which are all obtained by material mechanics performance test to ensure that they meet the actual material characteristics. When implemented, the model first inputs the welding speed, pressure and other parameters, starts the temperature field and stress field calculation modules respectively, and synchronously outputs the temperature and stress distribution data. By defining the multi-physical field calculation parameters and coupling logic, the temperature field calculation error is controlled within 5℃, the stress field calculation error is controlled within 2 MPa, and the thermal and force states of the welding area are fully reflected, providing reliable multi-physical field basic data for subsequent data repair and parameter optimization.

[0039] Preferably, the expression of the arc pixel neighborhood repair algorithm is: wherein, is the gray value of the repaired pixel point, is the weight coefficient of the pixel point, is the original gray value of the pixel point, is the selection range of the neighborhood pixels, is the Gaussian filter coefficient, is the neighborhood pixel offset in the direction of the pixel point, is the neighborhood pixel offset in the direction of the pixel point.

[0040] ​​​​​​​Specifically, in the step S5 of processing the abnormal data, the arc pixel neighborhood repair algorithm is applied, and in the implementation, the selection range of the neighborhood pixels is set as a 5*5 pixel matrix, which is determined according to the common influence radius of the abnormal data points in the arc image, can cover the effective pixel information around the abnormal points, and can avoid introducing too many irrelevant pixels to interfere with the repair result; the Gaussian filter coefficient is set as 1.2, which is calibrated through 100 groups of arc images containing abnormal data, can effectively smooth the repaired data, and does not blur the details of the arc characteristics; the weight coefficient is set according to the distance between the pixel and the abnormal point, the weight of the pixel unit 1 away from the abnormal point is 0.3, the weight of the pixel unit 2 is 0.2, and the weight of the pixel unit 3-5 is 0.1, which is determined according to the contribution degree experiment of the pixel to the characteristics of the abnormal point, and ensures that the near-neighbor effective pixels play a major repair role. In the implementation, the algorithm first defines the neighborhood range of the abnormal pixel points, reads the gray values of all normal pixel points in the neighborhood, then combines the weight coefficient and the Gaussian filter coefficient to calculate the repaired gray value, and finally converts the repaired gray value into corresponding multi-physical field data. By defining the algorithm parameters and the repair logic, the accuracy of the abnormal data repair can reach more than 98%, the multi-physical field data deviation caused by splashing and reflection can be eliminated, and the data uploaded to the cloud platform can truly reflect the welding state, so that the process parameter optimization failure caused by abnormal data can be avoided.

[0041] Preferably, the calculation model used by the FSW process parameter optimization cloud platform when performing parameter optimization is: wherein, is the optimized welding speed, is the initial welding speed, is the current welding time, is the standard welding time, is the temperature at the moment at the position in the standard welding process, is the maximum allowable stress of the welding material, is the stress at the moment at the position .

[0042] Specifically, when the FSW process parameter optimization is performed in step S6, the parameter optimization model of the FSW process parameter optimization cloud platform is enabled, and in implementation, the initial welding speed is determined according to the welding material and the workpiece thickness, the initial welding speed is set to 50 mm / min when the aluminum alloy workpiece thickness is 5 mm, and the parameter is determined by referring to the industry standard welding process library; the standard welding time is set to 20 min, which is calculated according to the workpiece length (1000 mm) and the initial welding speed, and represents the welding time under the standard working condition; the temperature data in the standard welding process is collected through 100 standard welding experiments to establish a standard temperature database, including the temperature distribution at different welding time nodes; the maximum allowable stress of the welding material is set to 120 MPa, which is calculated according to the yield strength (150 MPa) of the aluminum alloy material and the safety factor (1.25), to ensure that the material does not deform plastically during welding. In implementation, the cloud platform first acquires the temperature integral data and stress data of the current welding process, and then compares them with the standard temperature integral and the maximum allowable stress to calculate the parameter adjustment coefficient, and finally outputs the optimized welding speed. By defining the parameters and calculation logic of the cloud platform optimization model, the welding speed optimization response time is controlled within 1s, the parameter adjustment accuracy is 1mm / min, the welding quality qualified rate is improved to more than 99%, and the welding quality fluctuation caused by fixed parameters is avoided.

[0043] Preferably, when the image sequence is acquired by the binocular vision acquisition device, an image sharpness evaluation model is used to judge the quality of the acquired image, and the model expression is: wherein, is an image sharpness evaluation index, is the number of image width pixels, is the number of image height pixels, is the gray value of a pixel point, is the gray value of a pixel point, is the gray value of a pixel point.

[0044] Specifically, in step S1, when acquiring the arc image, an image sharpness evaluation model is used to determine the image quality. During implementation, the image width in pixels is set to 2560 pixels and the height in pixels to 1440 pixels, consistent with the resolution of the binocular vision acquisition device, ensuring complete coverage of all image pixels. During model calculation, the grayscale values ​​of image pixels are read row by row and column by column, with a grayscale value reading precision of 16 bits, capable of capturing subtle grayscale changes. During implementation, the model first calculates the grayscale difference between adjacent horizontal pixels (adjacent pixels in the same row), then calculates the grayscale difference between adjacent vertical pixels (adjacent pixels in the same column), then sums all horizontal and vertical grayscale differences, and finally divides by the total number of image pixels (2560×1440) to obtain the sharpness evaluation index. This index is set to a value range of 0-200. A value ≥120 indicates a sharp image, meeting the requirements for subsequent processing; a value <120 indicates a blurry image, requiring re-acquisition. By defining the image sharpness evaluation parameters and judgment criteria, the image quality screening accuracy rate reaches over 99.5%, effectively eliminating blurry images caused by lens contamination and focal length shift. This ensures that the target image frames screened in step S1 all possess high-quality arc features, laying a reliable foundation for subsequent feature extraction and model calculation, and avoiding deviations in subsequent data processing due to image blur.

[0045] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, classifying and organizing the pixel grayscale values, edge contour coordinates, and texture feature parameters in the basic feature dataset extracted in step S2, and dividing the feature data into multiple subsets according to the spatial location of the arc region; S32, inputting each subset into different calculation modules of the binocular arc 3D reconstruction numerical model, with each calculation module selecting appropriate calculation parameters for the feature type of the corresponding subset; S33, obtaining the local 3D reconstruction results output by each calculation module, and stitching the local 3D reconstruction results according to the spatial location information corresponding to the subsets to form a preliminary arc 3D morphology reconstruction result; S34, performing consistency verification on the spatial coordinate data in the preliminary arc 3D morphology reconstruction result, deleting coordinate conflict points caused by data stitching, and obtaining the final arc 3D morphology reconstruction result.

[0046] Specifically, in step S3, the binocular arc 3D reconstruction first involves S31, which involves data classification and organization. Data processing software is used to classify the pixel grayscale values, edge contour coordinates, and texture feature parameters in the basic feature dataset extracted in step S2. The dataset is divided into sub-datasets based on the spatial location of the arc region, using the quadrant distribution of the arc in the image. The image is divided into four quadrants using a 2×2 grid, with each quadrant corresponding to a sub-dataset. This ensures that the arc feature data range included in each sub-dataset does not overlap and covers the entire arc region. In step S32, the multi-module model calculation is initiated. The four sub-datasets are imported into the four calculation modules of the binocular arc 3D reconstruction numerical model. Each module adapts its calculation parameters according to the feature type of the sub-dataset and processes the pixel grayscale values. The module sets the grayscale resolution to 16 bits, the module processing edge contour coordinates to a coordinate precision of 0.01 pixels, and the module processing texture features to a texture parameter calculation step size of 5 pixels, ensuring compatibility with different types of data. S33 performs local result stitching, acquiring the local 3D reconstruction results output by each module. Based on the quadrant spatial location information corresponding to the subset dataset, a stitching coordinate system is established with the image center as the origin. The local results are stitched together according to their coordinate correspondence to form a preliminary reconstruction result. S34 performs consistency verification, using a coordinate comparison algorithm to compare all spatial coordinate data in the preliminary result pairwise. Coordinate points with a deviation exceeding 0.05 mm are identified as conflict points and deleted, ultimately obtaining an accurate 3D arc morphology reconstruction result. By refining the reconstruction process step by step, the reconstruction accuracy is improved by 15%, avoiding calculation deviations caused by data contamination, and providing more reliable arc morphology data for subsequent multiphysics prediction.

[0047] Preferred, such as Figure 3 As shown, step S4 includes the following sub-steps: S41, extracting the arc center coordinates, arc radius, and arc length parameters from the arc three-dimensional morphology reconstruction results generated in step S3, and integrating these parameters with the friction stir welding rotation speed, pressure, and welding speed parameters into a model input dataset; S42, importing the input dataset into the preprocessing module of the welding multiphysics coupling prediction model, which converts the format of the input data according to the model calculation requirements and establishes a mapping relationship between the data and the model calculation unit; S43, starting the model's temperature field calculation unit, stress field calculation unit, and flow field calculation unit, which simultaneously perform parallel calculations on the input data and output the corresponding temperature field distribution data, stress field distribution data, and flow field distribution data respectively; S44, cross-validating the distribution data output by the three calculation units, judging the rationality of the calculation results by comparing the numerical correlation of different physical field data at the same spatial location, and retaining the distribution data that meets the correlation requirements.

[0048] Specifically, the welding multi-physics field coupling prediction of step S4, S41, integrates parameters, extracts the arc center coordinates (accuracy 0.01 mm), arc radius (measurement error ≤0.02 mm), arc length (measurement error ≤0.05 mm) parameters from the arc three-dimensional form reconstruction result of step S3, integrates with the friction stir welding rotation speed (measurement accuracy ±5 r / min), pressure (measurement accuracy ±0.1 kN), welding speed (measurement accuracy ±0.1 mm / min) parameters, forms a model input data set of 12 columns and 300 rows, and each row corresponds to a calculation node; S42 performs data preprocessing, imports the input data set into the model preprocessing module, the module converts the data format to floating point type (accuracy 64 bits) according to the requirements of the calculation unit, establishes the mapping relationship between the data and the temperature field, stress field and flow field calculation unit, and ensures that each parameter is accurately corresponding to the input port of the corresponding calculation unit; S43 starts parallel calculation, activates three calculation units at the same time, sets the calculation step length of the temperature field unit to 0.01 s and the grid density to 1 mm×1 mm×1 mm, sets the elastic modulus calculation accuracy of the stress field unit to ±0.1 GPa and the Poisson's ratio to 0.33 (standard value of aluminum alloy), sets the viscosity coefficient of the flow field unit to 0.001 Pa・s (melting state value of aluminum alloy), and the three units are calculated and output corresponding distribution data at the same time; S44 carries out cross-validation, calculates the numerical correlation of different physical fields at the same space node, the correlation coefficient of temperature and stress needs to be ≥0.8, and the correlation coefficient of temperature and flow rate needs to be ≥0.7, the data meeting the requirements are reserved to ensure the logical consistency of the multi-physical field data. Through step-by-step optimization of the calculation process, the multi-physical field calculation efficiency is improved by 20%, and the data consistency is more than 98%, which provides high-quality basic data for subsequent data repair.

[0049] Preferably, as shown in Figure 4 S5 includes the following steps: S51, using an abnormal data identification rule to detect the temperature field, stress field and flow field distribution data output by step S4 point by point, the rule sets a data threshold range based on the statistical distribution characteristics of similar data, and data outside the threshold range is marked as an abnormal data point; S52, determine the pixel position of each abnormal data point in the corresponding image, and define a fixed size of neighborhood pixel region centered on the pixel position, and record the gray value and feature parameters of all normal pixel points in the neighborhood; S53, input the gray value and feature parameters of the normal pixel points in the neighborhood into the arc pixel neighborhood repair algorithm, and the algorithm calculates the pixel gray correction value corresponding to the abnormal data point according to the spatial correlation and feature similarity of the neighborhood pixels; S54, substitute the calculated pixel gray correction value into the corresponding multi-physical field distribution data, replace the original value of the abnormal data point, and update the values of the related data around the data point, and correct the whole multi-physical field distribution data.

[0050] Specifically, the arc pixel neighborhood repair of step S5, S51 performs abnormal data identification, adopts statistical analysis algorithm to detect the multi-physical field data output by step S4 point by point, takes the average value ± 3 times the standard deviation of the same kind of data as the threshold (temperature field standard deviation ≤5℃, stress field standard deviation ≤2MPa, flow field standard deviation ≤0.05m / s), the data exceeding the threshold is marked as an abnormal point, and the marking accuracy needs to be ≥99%; S52 determines the repair range, converts the world coordinates (accuracy 0.01mm) of the abnormal data points into image pixel coordinates (accuracy 1 pixel) through coordinate system inversion, and marks the 5×5 pixel neighborhood area with the coordinates as the center. The neighborhood range is calibrated through 100 groups of abnormal repair experiments to ensure that the effective repair pixels are covered; S53 calculates the repair value, inputs the gray value (accuracy 16 bits), texture parameters (contrast 0-100, entropy 0-8) of the normal pixels in the neighborhood into the arc pixel neighborhood repair algorithm, and the algorithm calculates the abnormal point gray correction value according to the pixel space distance (the weight decreases with the increase of distance, the weight of the near neighbor is 0.3, and the weight of the far neighbor is 0.1) and the feature similarity (the pixel weight is doubled when the gray difference is ≤10), and the correction accuracy reaches ±1 gray level; S54 updates the data, converts the gray correction value into multi-physical field correction value (temperature correction accuracy ±1℃, stress correction accuracy ±0.5MPa, flow velocity correction accuracy ±0.01m / s), replaces the original abnormal value, and carries out smoothing processing (smoothing coefficient 0.8) on the data in the 10×10 grid range around the abnormal point to avoid data mutation. Through the step-by-step refinement of the repair process, the accuracy of abnormal data repair reaches 99%, the error of multi-physical field data is controlled within 5%, accurate data support is provided for cloud platform parameter optimization, and parameter adjustment deviation caused by abnormal data is avoided.

[0051] The binocular arc three-dimensional reconstruction numerical model in the application is a calculation model for converting the two-dimensional image of the arc collected by binocular vision into three-dimensional spatial form data. The implementation process needs to deploy two binocular vision collection devices with a resolution of 2560*1440 pixels and a frame rate of 60 frames / s, fix them on both sides of the welding equipment, set the lens distance from the welding area to be 800-1200 mm and the angle between the lens and the center line of the arc to be 30-45°, collect continuous image sequences of the arc and select effective frames; then extract the pixel gray value, edge contour coordinates and texture feature parameters of the arc in the image to form a basic feature data set; then the model calls preset parameters (lens focal length 16 mm, baseline distance 500 mm, pixel neighborhood calculation range 5 pixels, error correction coefficient 0.001), generates the three-dimensional spatial coordinates, shape parameters (length, diameter, volume) and distribution characteristics of the arc by calculating the corresponding pixel gray difference of the left and right camera images, combining neighborhood statistics and error compensation, and outputs the three-dimensional point cloud diagram and data table. The model converts two-dimensional image information into three-dimensional spatial information, accurately presents the actual shape and position of the arc, provides real and accurate spatial data of the arc for subsequent welding multi-physical field coupling prediction, avoids the deviation of the welding state judgment caused by the insufficient two-dimensional image information, controls the arc shape reconstruction error within 0.1 mm, and lays a reliable spatial data foundation for the entire welding control process.

[0052] The welding multi-physical field coupling prediction model in the application is a multi-module coupling calculation model for calculating the distribution data of the temperature field, stress field and flow field of the welding area. The implementation process needs to extract the arc center coordinates (accuracy 0.01 mm), radius (error ≤0.02 mm) and length (error ≤0.05 mm) parameters from the binocular arc three-dimensional reconstruction results, combine the real-time parameters of the welding equipment (speed ±5 r / min, pressure ±0.1 kN, speed ±0.1 mm / min) to integrate them into an input data set of 12 columns and 300 rows; then import the data set into the preprocessing module, convert it into a 64-bit floating-point format, and establish a mapping relationship with the temperature field, stress field and flow field calculation unit; then activate the three parallel calculation units, the temperature field unit takes the arc thermal radiation / conduction as the heat source, designs the calculation step length to be 0.01 s and the grid density to be 1 mm -5The temperature distribution is calculated; the stress field unit is provided with an elastic modulus of ±0.1 GPa and a Poisson's ratio of 0.33, and the stress distribution is calculated in combination with a thermal stress coefficient; the flow field unit is provided with a molten state viscosity of 0.001 Pa·s, and the flow velocity distribution is calculated in combination with shear force; and finally, the data consistency is ensured through cross verification (temperature-stress correlation coefficient is greater than or equal to 0.8, and temperature-flow velocity correlation coefficient is greater than or equal to 0.7). The model comprehensively obtains the dynamic change data of the multi-physical field in the welding area. In order to judge the welding quality and find potential defects, the temperature field error is less than or equal to 5 DEG C, and the stress field error is less than or equal to 2 MPa, so as to avoid blind control of the welding process due to lack of multi-physical field data, and to provide high-quality basic data for subsequent data repair and parameter optimization.

[0053] The arc pixel neighborhood repair algorithm in the application is a processing algorithm for correcting abnormal values in welding multi-physical field data and corresponding image pixel deviation. The implementation process needs to detect the welding multi-physical field distribution data (temperature, stress, flow velocity) point by point, take the average value of the same data ± 3 times the standard deviation as the threshold (temperature standard deviation is less than or equal to 5 DEG C, stress is less than or equal to 2 MPa, and flow velocity is less than or equal to 0.05 m / s), mark the abnormal data points, and the accuracy is greater than or equal to 99%; then the world coordinates (accuracy 0.01 mm) of the abnormal points are converted into image pixel coordinates (accuracy 1 pixel) through coordinate system inversion, and a 5*5 pixel neighborhood area (calibrated through 100 groups of experiments) is defined with the coordinates as the center; then the gray value (16-bit accuracy) and texture parameters (contrast 0-100, entropy 0-8) of the normal pixels in the neighborhood are read, the abnormal point gray correction value (accuracy ±1 gray level) is calculated according to the spatial distance (near neighbor weight 0.3, far neighbor 0.1) and feature similarity (gray difference is less than or equal to 10, weight is doubled); finally, the gray correction value is converted into multi-physical field correction value (temperature ±1 DEG C, stress ±0.5 MPa, flow velocity ±0.01 m / s), the original abnormal value is replaced, and the surrounding 10*10 grid data is processed with a smoothing coefficient of 0.8 to avoid data mutation. The algorithm eliminates the abnormal deviation of the multi-physical field data caused by splashing and reflection, and repairs the image pixel defects. It is ensured that the data uploaded to the cloud platform truly reflects the welding state, the repair accuracy is 99%, the data error is less than or equal to 5%, the FSW process parameter optimization caused by data abnormality is avoided, and the accuracy of the subsequent control process is ensured.

[0054] The FSW process parameter optimization cloud platform in the application is a cloud computing platform for integrating multi-source data, optimizing friction stir welding parameters and realizing real-time control. FSW is Friction Stir Welding. The implementation process needs to first establish a communication connection (5G / optical fiber, rate ≥ 100 Mbps, AES-256 encryption) between the industrial computer and the platform, and upload the corrected multi-physical field data (including collection time, workpiece number, CSV format data) to the platform; the platform is composed of multiple Intel Xeon Platinum 8480+ CPUs, 256 GB memory servers, and a computing cluster with a storage capacity of 100 TB, and a built-in welding quality evaluation standard database (including aluminum alloy, steel materials, and other materials, butt welding, corner welding, etc., temperature fluctuation ≤50℃, stress ≤80% of the material yield strength, flow velocity fluctuation ≤0.1m / s); then the platform analyzes the data and compares it with the standard to judge the uniformity of the temperature field, the stress level, and the stability of the flow field. If the parameters exceed the standard, the optimization module is started; the optimization module combines arc three-dimensional shape data and 10,000+ groups of historical successful cases to calculate the adjusted parameters (rotation speed ±5r / min, pressure ±0.1kN, speed ±0.1mm / min) using a multi-objective optimization algorithm; finally, the optimized parameters are fed back to the industrial computer and then transmitted to the welding equipment control system to adjust the equipment operating state. The platform realizes multi-source data comprehensive analysis, welding parameter dynamic optimization and real-time control. It breaks through the limitation of local computing resources, the parameter optimization response time is ≤1s, the qualified rate is improved to more than 99%, the welding quality fluctuation caused by fixed parameters is avoided, remote control and data sharing are realized, and the intelligentization and automation level of welding production is improved.

[0055] As Figure 5As shown, the arc perception based on binocular vision friction stir welding control system is applied to the arc perception based on binocular vision friction stir welding control method, and includes a binocular vision arc image acquisition unit, an arc feature data extraction and screening unit, a binocular arc three-dimensional reconstruction calculation unit, a welding multi-physical field coupling prediction unit, an arc pixel neighborhood repair and data correction unit, and an FSW process parameter optimization and control execution unit. The binocular vision arc image acquisition unit is connected with the arc feature data extraction and screening unit. The binocular vision arc image acquisition unit acquires continuous image sequences of the arc region in the friction stir welding process and transmits them to the arc feature data extraction and screening unit. The arc feature data extraction and screening unit is connected with the binocular arc three-dimensional reconstruction calculation unit. The arc feature data extraction and screening unit performs arc region positioning on the image sequences, screens target image frames, and extracts feature parameters, which are transmitted to the binocular arc three-dimensional reconstruction calculation unit. The binocular arc three-dimensional reconstruction calculation unit is connected with the welding multi-physical field coupling prediction unit. The binocular arc three-dimensional reconstruction calculation unit generates arc three-dimensional morphology reconstruction results through model calculation and transmits them to the welding multi-physical field coupling prediction unit. The welding multi-physical field coupling prediction unit is connected with the arc pixel neighborhood repair and data correction unit. The welding multi-physical field coupling prediction unit outputs welding region multi-physical field distribution data and transmits them to the arc pixel neighborhood repair and data correction unit. The arc pixel neighborhood repair and data correction unit is connected with the FSW process parameter optimization and control execution unit. The arc pixel neighborhood repair and data correction unit transmits the corrected data to the FSW process parameter optimization and control execution unit after repairing the abnormal data. The FSW process parameter optimization and control execution unit outputs adjusted welding parameters according to the corrected data and controls the operation state of the friction stir welding equipment.

[0056] The arc perception based on binocular vision friction stir welding control method and system acquire continuous image sequences of the arc region through the binocular vision acquisition device, screen out target image frames including complete arc features, exclude invalid interference images, and lay a high-quality foundation for subsequent data processing. Meanwhile, the arc pixel neighborhood repair algorithm is introduced, which can accurately repair abnormal data points and corresponding image pixel regions in the multi-physical field distribution data caused by welding spatter and surface reflection, solve the problem of lack of reliable data repair mechanism in the prior art, leading to insufficient analysis and judgment accuracy, and greatly improve the accurate perception ability of the arc morphology and multi-physical field distribution state in the welding process, providing accurate data support for subsequent control.

[0057] In the process parameter optimization and control synergy, the method and system are also outstanding, which can completely break through the limitations of the prior art: through the three-dimensional reconstruction numerical model of binocular arc to generate the three-dimensional shape reconstruction result of arc, and the precise calculation of the temperature field, stress field and flow field distribution data of the welding area by the welding multi-physical field coupling prediction model, the comprehensive grasp of the multi-dimensional state of the welding process is realized; then the corrected multi-physical field data is uploaded to the FSW process parameter optimization cloud platform, relying on the cloud platform to realize the comprehensive integration and dynamic optimization of multi-source data, avoiding the problems of relying on local calculation or single model, parameter adjustment and real-time data linkage in the prior art, and timely outputting the speed, pressure and welding speed parameters suitable for the dynamic change of the welding process, so as to ensure the welding quality from the influence of working condition fluctuation, and through multi-step data processing and multi-unit collaborative control, the intelligent level and quality consistency of the welding process are improved.

[0058] In the description of the present application, it should be pointed out that, unless otherwise specified and limited, the terms "arrangement", "installation", "connection", "connection", "fixing" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For ordinary skilled in the art, the specific meaning of the above terms in the present application can be understood through specific circumstances.

[0059] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various equivalent changes, modifications, replacements and variations of the embodiments can be made without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalent scope.

Claims

1. A friction stir welding control method based on binocular vision arc perception, characterized by, The method comprises the following steps: Step S1, acquiring a continuous image sequence of an arc region in a friction stir welding process through a binocular vision acquisition device, positioning the arc region of the acquired image sequence, screening out target image frames including complete arc features, and excluding invalid image frames caused by welding spatters and surface reflection interference; Step S2, based on the screened target image frames, extracting pixel gray values, edge contour coordinates and texture feature parameters of the arc region, establishing a mapping relationship between the arc features and image pixels, and determining a basic feature data set for subsequent three-dimensional reconstruction; Step S3, inputting the basic feature data set into a binocular arc three-dimensional reconstruction numerical model, obtaining coordinate information, shape parameters and spatial distribution characteristics of the arc in a three-dimensional space through model calculation, and generating an arc three-dimensional shape reconstruction result; Step S4, calling a welding multi-physical field coupling prediction model, taking the arc three-dimensional shape reconstruction result and the rotation speed, pressure and welding speed parameters in the friction stir welding process as model inputs, and calculating distribution data of a temperature field, a stress field and a flow field in a welding area; Step S5, identifying abnormal data points in the temperature field, stress field and flow field distribution data output by step S4, repairing image pixel regions corresponding to the abnormal data points by using an arc pixel neighborhood repair algorithm, and obtaining corrected multi-physical field distribution data; Step S6, uploading the corrected multi-physical field distribution data to an FSW process parameter optimization cloud platform, outputting adjusted friction stir welding rotation speed, pressure and welding speed parameters according to a preset welding quality evaluation standard, and performing real-time control on the welding process.

2. The twin-vision arc perception based control method for friction stir welding of claim 1, wherein, The expression for the three-dimensional reconstruction numerical model of the binocular electric arc is: ,in, Let be the depth coordinates of a point on the electric arc in three-dimensional space. The focal length of the lens in the binocular vision acquisition device. This is the baseline distance between the two cameras in the binocular vision acquisition device. In capturing images for the left camera The grayscale value of a pixel. In capturing images for the right camera The grayscale value of a pixel. The calculation range of the pixel neighborhood. This refers to the error correction coefficients during the 3D reconstruction process.

3. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The expression of the welding multi-physical field coupling prediction model is: wherein, is the temperature of the welding area at the moment , is the thermal diffusion coefficient of the welding material, is the temperature conduction velocity correction coefficient, is the welding speed, is the pressure thermal conversion coefficient, is the friction stir welding pressure, is the contact area of the stirring head and the workpiece, is the stress of the welding area at the moment , is the elastic modulus of the welding material, is the strain of the welding area at the moment , is the material viscosity coefficient, is the thermal stress coefficient.

4. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The expression of the electric arc pixel neighborhood repair algorithm is: wherein, is the gray value of the pixel point after repair is the gray value of the pixel point after repair is the gray value of the pixel point after repair is the weight coefficient of the pixel point is the weight coefficient of the pixel point is the original gray value of the pixel point is the selection range of the neighborhood pixels is the Gaussian filter coefficient is the neighborhood pixel offset of the pixel point in the direction is the neighborhood pixel offset of the pixel point in the direction is the neighborhood pixel offset of the pixel point in the direction 5. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The calculation model used by the FSW process parameter optimization cloud platform when performing parameter optimization is: wherein, is the optimized welding speed, is the initial welding speed, is the current welding time, is the standard welding time, is the temperature of the position at the moment in the standard welding process, is the moment, is the position, is the maximum allowable stress of the welding material, is the current stress, is the moment, is the position.

6. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, When the image sequence is acquired by using the binocular vision acquisition device, the image definition evaluation model is used to judge the quality of the acquired image, and the model expression is: Wherein, is an image definition evaluation index, is the number of image width pixels, is the number of image height pixels, is the gray value of a pixel point, is the gray value of a pixel point, is the gray value of a pixel point.

7. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The step S3 comprises the following sub-steps: S31, classifying and arranging the pixel gray values, edge contour coordinates and texture feature parameters in the basic feature data set extracted in step S2, and dividing the feature data into a plurality of sub-data sets according to the spatial positions of the arc regions; S32, inputting each sub-data set into different calculation modules of the binocular arc three-dimensional reconstruction numerical model, each calculation module selecting adaptive calculation parameters according to the feature types of the corresponding sub-data set; S33, obtaining local three-dimensional reconstruction results output by each calculation module, splicing the local three-dimensional reconstruction results according to the spatial position information of the sub-data sets, and forming a preliminary arc three-dimensional shape reconstruction result; S34, performing consistency verification on the spatial coordinate data in the preliminary arc three-dimensional shape reconstruction result, deleting coordinate conflict points caused by data splicing, and obtaining a final arc three-dimensional shape reconstruction result.

8. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The step S4 comprises the following sub-steps: S41, extracting the arc center coordinates, arc radius, and arc length parameters from the arc three-dimensional morphology reconstruction result generated in step S3, and integrating these parameters with the friction stir welding rotation speed, pressure, and welding speed parameters into a model input data set; S42, inputting the input data set into a preprocessing module of the welding multi-physical field coupling prediction model, which converts the format of the input data according to the model calculation requirements, and establishes a mapping relationship between the data and the model calculation unit; S43, starting the temperature field calculation unit, stress field calculation unit, and flow field calculation unit of the model, and performing parallel calculation on the input data by the three calculation units, and respectively outputting the corresponding temperature field distribution data, stress field distribution data, and flow field distribution data; S44, cross- verifying the distribution data output by the three calculation units, judging the rationality of the calculation results by comparing the numerical correlation of different physical field data at the same spatial position, and retaining the distribution data meeting the correlation requirements.

9. The dual-vision arc perception based control method for friction stir welding of claim 1, wherein, The step S5 comprises the following sub-steps: S51, performing point-by-point detection on the temperature field, stress field, and flow field distribution data output in step S4 by using an abnormal data identification rule, which sets a data threshold range based on the statistical distribution characteristics of similar data, and marks the data beyond the threshold range as abnormal data points; S52, determining the pixel position of each abnormal data point in the corresponding image, and defining a neighborhood pixel region of a fixed size with the pixel position as the center, and recording the gray values and feature parameters of all normal pixel points in the neighborhood; S53, inputting the gray values and feature parameters of the normal pixel points in the neighborhood into the arc pixel neighborhood repair algorithm, and calculating the pixel gray correction value corresponding to the abnormal data point according to the spatial correlation and feature similarity of the neighborhood pixels; S54, substituting the calculated pixel gray correction value into the corresponding multi-physical field distribution data, replacing the original value of the abnormal data point, simultaneously updating the values of the related data around the data point, and overall correcting the multi-physical field distribution data.

10. A friction stir welding control system based on arc perception using binocular vision, characterized in that, The system is applied to the friction stir welding control method based on binocular vision arc perception of claim 1, comprising: a binocular vision arc image acquisition unit, an arc feature data extraction and screening unit, a binocular arc three-dimensional reconstruction calculation unit, a welding multi-physical field coupling prediction unit, an arc pixel neighborhood repair and data correction unit, and an FSW process parameter optimization and control execution unit; the binocular vision arc image acquisition unit is connected with the arc feature data extraction and screening unit, the binocular vision arc image acquisition unit acquires the continuous image sequence of the arc region in the friction stir welding process and transmits it to the arc feature data extraction and screening unit; the arc feature data extraction and screening unit is connected with the binocular arc three-dimensional reconstruction calculation unit, the arc feature data extraction and screening unit locates the arc region, screens the target image frame and extracts the characteristic parameters from the image sequence, and transmits the characteristic parameters to the binocular arc three-dimensional reconstruction calculation unit; the binocular arc three-dimensional reconstruction calculation unit is connected with the welding multi-physical field coupling prediction unit, the binocular arc three-dimensional reconstruction calculation unit generates the arc three-dimensional shape reconstruction result through model calculation and transmits it to the welding multi-physical field coupling prediction unit; the welding multi-physical field coupling prediction unit is connected with the arc pixel neighborhood repair and data correction unit, the welding multi-physical field coupling prediction unit outputs the welding area multi-physical field distribution data and transmits it to the arc pixel neighborhood repair and data correction unit; the arc pixel neighborhood repair and data correction unit is connected with the FSW process parameter optimization and control execution unit, the arc pixel neighborhood repair and data correction unit transmits the corrected data to the FSW process parameter optimization and control execution unit after repairing the abnormal data; the FSW process parameter optimization and control execution unit outputs the adjusted welding parameters according to the corrected data, and controls the operation state of the friction stir welding equipment.

Images