Automatic welding inspection system and method for fine thermocouple wire

Abstract

The invention provides an automatic welding inspection system and method for a fine thermocouple wire, and the method comprises the steps: after an automatic welding inspection starting instruction is sent out, industrial cameras A and B start to collect real-time images of a welding wire butt joint scene at the same time, carry out the three-dimensional positioning of a welding needle and welding wire end points, calculate the position deviation between the end points, generate a control instruction, and carry out the automatic welding inspection of the fine thermocouple wire; four motors in the driving execution mechanism control four motion platforms to act, thermocouple wires are driven to adjust the position, the relative position of a welding needle and a welding wire is continuously corrected till butt joint of the thermocouple wires is completed, and then the welding procedure is started immediately. And after welding is completed, the industrial cameras A, B and C are synchronously triggered to synchronously collect 360-degree images of the welding spots, the welding spot quality detection algorithm is used for processing, and the appearance size parameters of the welding spots, whether defects exist on the surfaces or not and the specific types are output. According to the technical scheme, the technical problems that in the prior art, the automatic welding inspection quality of the thermocouple wire is low, and the strict requirements of a temperature measurement system in the spaceflight field for measurement precision, stability and reliability cannot be met are solved.

Description

Technical Field

[0001] This invention relates to the field of thermocouple manufacturing technology in the aerospace industry, and in particular to an automatic welding and inspection system and method for micro thermocouple wires. Background Technology

[0002] Thermocouples occupy an important position in industrial temperature measurement due to their wide temperature measurement range, rapid dynamic response characteristics, and excellent measurement accuracy. Their core working principle is based on the Seebeck effect, which involves welding two thermocouple wires of different materials to form a closed circuit, utilizing the definite relationship between thermoelectric potential and temperature difference to achieve temperature signal conversion. Therefore, the welding quality directly affects the temperature measurement stability, accuracy, and long-term reliability of the thermocouple.

[0003] In the aerospace field, Type B thermocouples are the preferred choice for temperature measurement equipment due to their ultra-wide high-temperature measurement range and extremely strong stability. However, their material is brittle and has a high melting point. During welding, the high temperature at the melting point can easily cause overheating damage to the welding wire itself. Coupled with the high material cost, this places extremely stringent requirements on the welding process. Engineering practice has verified that the "I"-shaped welding method, with its small heat-affected zone and stable weld structure, minimizes interference with the thermoelectric properties of the thermocouple wire and has become the optimal welding method for Type B thermocouple wires. Currently, the "I"-shaped welding of Type B thermocouples in the aerospace field mainly relies on manual operation. The operator visually aligns the two ends of the thermocouple wire, uses their foot to trigger the argon arc welding machine to complete the welding, and then uses a microscope for quality inspection and data recording. This method is highly dependent on the operator's experience and skill level, easily leading to material loss and product performance fluctuations due to human factors. Furthermore, manual inspection is inefficient and makes it difficult to guarantee the consistency of batch products.

[0004] Existing automated thermocouple wire welding and inspection systems, both domestically and internationally, rely on fixed mechanical structures to butt and weld thermocouple wires. The welded wires are then moved from the welding area to the inspection area for quality control. However, due to the long tail wires of thermocouples used in aerospace applications, the fine thermocouple wires are prone to bending or slight deformation. Therefore, existing systems rely solely on static correction through mechanical structures, lacking a feedback mechanism. This makes it difficult to address the slight deformations caused by wire vibrations during transport, leading to butt joint deviations and affecting welding quality. Furthermore, after multiple welding operations, the welding needle of an argon arc welding machine may experience tip spheroidization due to high-temperature ablation, resulting in decreased welding quality or welding failure. Disassembly and reassembly can cause slight positional changes. Relying solely on mechanical mechanisms to butt the thermocouple wires can lead to misalignment between the arc center and the welding wire butt joint point during welding, causing defects such as misaligned welds and weld beads, thus affecting the overall quality of thermocouple welding. Some systems use mechanical stretching to inspect the welding quality of weld points, but stretching changes the diameter of the thermocouple wire, affecting the stability of the thermoelectric potential output and the consistency of the thermal response characteristics. This cannot meet the stringent requirements of aerospace temperature measurement systems for measurement accuracy, stability, and reliability, and is only suitable for quality spot checks. Summary of the Invention

[0005] This invention provides an automatic welding and inspection system and method for micro thermocouple wires, which can solve the technical problem that the automatic welding and inspection quality of thermocouple wires in the prior art is low and cannot meet the stringent requirements of temperature measurement systems in the aerospace field for measurement accuracy, stability and reliability.

[0006] According to one aspect of the present invention, an automatic welding and inspection system for micro-thermocouple wires is provided. The system includes a computer, a control box, actuators, and an argon arc welding machine. The computer serves as the system control center, controlling an industrial camera in the actuator to acquire images and generating control commands through a built-in software processing algorithm, which are then sent to the control box. The control box receives displacement and speed commands from the computer for each motion platform, drives the actuator to perform endpoint docking, and controls the argon arc welding machine to perform welding. The actuator includes industrial camera A, industrial camera B, and industrial camera C. The system consists of a left adjustment assembly, a right adjustment assembly, an argon arc welding torch, a grounding contact, and a slag tray. Industrial camera A is vertically aligned with the welding area to capture a top-down view. Industrial camera B is positioned horizontally from the side, working together to achieve 3D positioning of the welding needle and wire tips. Industrial camera C is positioned at a 45° downward angle to the horizontal plane, supplementing the observation view from behind the weld point. Synchronous imaging by the three cameras achieves 360° coverage of the weld surface. The slag tray is located at the bottom of the argon arc welding torch for slag recovery. The left adjustment assembly... The component includes a left thermocouple wire clamp, a left-side left-right motion platform and a first motor, a front-back motion platform and a second motor, and a vertical motion platform and a third motor, allowing the left welding wire tip to be adjusted in three directions. The left thermocouple wire clamp is used to hold the left welding wire. The first motor drives the left-side left-right motion platform to move in the left-right direction, the second motor drives the front-back motion platform to move in the front-back direction, and the third motor drives the vertical motion platform to move in the vertical direction. The left thermocouple wire clamp is set on the left-side left-right motion platform, the left-right motion platform is set on the front-back motion platform, and the front-back motion platform is set on the vertical motion platform. The right adjustment component includes a right thermocouple wire clamp, a right-side left-right motion platform, and a fourth motor, allowing the right welding wire to be adjusted only in the left-right direction. The fourth motor drives the left-right motion platform to move in the left-right direction. The thermocouple wire clamp is set on the left-right motion platform, and the right thermocouple wire clamp is used to hold the right welding wire. The argon arc welding torch is arranged at a 45° upward angle to the horizontal plane and is positioned opposite the grounding contact, so that the arc is stably formed between the welding needle and the grounding contact, quickly melting the thermocouple wire joint to form a weld point.

[0007] According to another aspect of the present invention, an automatic welding inspection method for micro thermocouple wires is provided, wherein the automatic welding inspection method for micro thermocouple wires uses the automatic welding inspection system for micro thermocouple wires as described above to perform automatic welding inspection of micro thermocouple wires.

[0008] Furthermore, the automatic welding inspection method for micro thermocouple wires includes: after the computer issues an automatic welding inspection start command, industrial cameras A and B simultaneously begin acquiring real-time images of the welding wire docking scene. The welding needle, left welding wire end point, right welding wire end point, and ideal welding point are then three-dimensionally positioned using a processing algorithm. Based on the coordinates of the welding needle, left welding wire end point, right welding wire end point, and ideal welding point, the positional deviation between the end points is calculated to generate control commands, which are transmitted to the control box. After parsing, the first, second, third, and fourth motors in the actuator control the left left-right, front-back, and up-down motion platforms, and the right left-right motion platform, causing the thermocouple wires to adjust their position. After adjustment, images are acquired and fed back to the computer, forming a complete closed-loop control. By continuously correcting the relative position of the welding needle and welding wire until the left and right welding wires are docked, the welding process is immediately started. After welding is completed, the computer immediately triggers industrial cameras A, B, and C to simultaneously acquire 360° images of the weld point. These images are then processed by a weld point quality detection algorithm to output the weld point's appearance and size parameters, the presence of surface defects, and their specific types.

[0009] This invention provides an automated welding and inspection system for micro-thermocouple wires. This system, through the coordinated operation of three cameras and four motion platforms, automates the entire process of micro-thermocouple wire connection, welding, and weld quality inspection. Two industrial cameras acquire images to achieve three-dimensional positioning of the welding needle, wire tip, and ideal welding point. The four motion platforms then work in tandem to complete the wire connection. Immediately after the welding process, the weld point's appearance and dimensions are measured, and surface defects are detected, promptly removing defective products and effectively improving the efficiency and consistency of thermocouple product quality control. The layout design of the industrial cameras, motion platforms, and related components simultaneously meets the requirements for high-precision connection and 360° weld point inspection of micro-thermocouple wires. To achieve high connection accuracy, two industrial cameras are vertically positioned to acquire images of the wire connection, determining the three-dimensional coordinates of the welding needle, wire tip, and ideal welding point. Combined with the left three-axis motion platform and the right single-axis motion platform, the left welding wire can be precisely connected to the right welding wire through multi-dimensional adjustments, simplifying the system structure. To ensure 360° inspection of weld joints without blind spots, a third industrial camera is deployed to collaboratively capture details of different slopes and corners of the weld joints, effectively reducing missed detections caused by irregular weld joint shapes and local structural obstructions. Furthermore, the argon arc welding torch and its opposing bent-angle grounding contact are positioned at a 45° angle to the camera, geometrically avoiding obstruction of the welding needle and wire, ensuring uninterrupted endpoint positioning and weld joint inspection. In the endpoint 3D positioning method, error correction is used to improve endpoint positioning accuracy. Based on the spatial position association constructed using dual-camera vision, an angle compensation mechanism is introduced to correct potential positional deviations during installation. Simultaneously, the coordinates are corrected using reprojection errors obtained from calibration, further improving positioning accuracy. Therefore, the automatic welding and inspection system for micro-thermocouple wires provided by this invention can achieve high-quality welding of micro-thermocouple wires, while simultaneously measuring weld joint dimensions and detecting surface defects, meeting the stringent requirements of aerospace temperature measurement systems for measurement accuracy, stability, and reliability. Attached Figure Description

[0010] The accompanying drawings, which form part of this specification, are provided to further illustrate embodiments of the invention and, together with the textual description, explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0011] Figure 1 A structural block diagram of an automatic welding and inspection system for micro thermocouple wires according to a specific embodiment of the present invention is shown;

[0012] Figure 2 A schematic diagram of the actuator structure of the automatic welding and inspection system for micro thermocouple wires provided according to a specific embodiment of the present invention is shown;

[0013] Figure 3 A right view of an industrial camera assembly, argon arc welding torch, and grounding contact structure according to a specific embodiment of the present invention is shown.

[0014] Figure 4 A block diagram of closed-loop control for the welding stage according to a specific embodiment of the present invention is shown;

[0015] Figure 5 A schematic diagram of the coordinate calculation principle provided by a specific embodiment of the present invention is shown in the right-view direction.

[0016] The above figures include the following reference numerals:

[0017] 1. Industrial camera A; 2. Industrial camera B; 3. Industrial camera C; 4. Left adjustment assembly; 41. Left thermocouple wire clamp; 42. Left left-right motion platform; 43. First motor; 44. Front-back motion platform; 45. Second motor; 46. Up-down motion platform; 47. Third motor; 5. Right adjustment assembly; 51. Right thermocouple wire clamp; 52. Right left-right motion platform; 53. Fourth motor; 6. Argon arc welding torch; 7. Grounding contact; 8. Welding slag tray. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. 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 a part of the embodiments of the present invention, and not all of them. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. 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.

[0019] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0020] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0021] like Figures 1 to 5As shown, an automatic welding and inspection system for micro-thermocouple wires is provided according to a specific embodiment of the present invention. This system includes a computer, a control box, an actuator, and an argon arc welding machine. The computer acts as the system control center, controlling the industrial camera in the actuator to acquire images and generating control commands through a built-in software processing algorithm, which are then sent to the control box. The control box receives displacement and speed commands from the computer for each motion platform, drives the actuator to perform endpoint docking, and controls the argon arc welding machine to perform welding. The actuator includes industrial camera A1, industrial camera B2, industrial camera C3, a left adjustment component 4, and a right... The system includes a side adjustment assembly 5, an argon arc welding torch 6, a grounding contact 7, and a slag tray 8. Industrial camera A1 is vertically aligned with the welding area to capture a top-down view. Industrial camera B2 is positioned horizontally from the side, working together to achieve three-dimensional positioning of the welding needle and wire tips. Industrial camera C3 is positioned at a 45° downward angle to the horizontal plane, supplementing the observation view from behind the weld point. Synchronous imaging by the three cameras achieves 360° full coverage of the weld point surface. The slag tray is located below the argon arc welding torch 6 for slag recovery. The left adjustment assembly includes a left thermocouple wire clamp 41 and a left-right movement mechanism. Platform 42, along with the first motor 43, the front-to-back motion platform 44, the second motor 45, the up-and-down motion platform 46, and the third motor 47, enables the left welding wire tip to be adjusted in three directions. The left thermocouple wire clamp 41 is used to hold the left welding wire. The first motor 43 drives the left-to-right motion platform 42 to move left and right, the second motor 45 drives the front-to-back motion platform 44 to move forward and backward, and the third motor 47 drives the up-and-down motion platform 46 to move up and down. The left thermocouple wire clamp 41 is mounted on the left-to-right motion platform 42, and the left-to-right motion platform 42 is mounted on the front-to-back motion platform 44. The front and rear motion platform 44 is set on the upper and lower motion platform 46; the right adjustment component includes the right thermocouple wire clamp 51, the right left and right motion platform 52 and the fourth motor 53, so that the right welding wire can only be adjusted in the left and right direction. The fourth motor 53 is used to drive the left and right motion platform 52 to move in the left and right direction. The thermocouple wire clamp 51 is set on the left and right motion platform 52. The right thermocouple wire clamp 51 is used to clamp the right welding wire. The argon arc welding gun 6 is arranged at a 45° angle upward to the horizontal plane and is placed opposite the grounding contact 7, so that the electric arc is stably formed between the welding needle and the grounding contact, and the thermocouple wire docking part is quickly melted to form a weld point.

[0022] This configuration provides an automated welding and inspection system for micro-thermocouple wires. The system utilizes three cameras and four motion platforms working in tandem to automate the entire process from wire docking and welding to weld quality inspection. Two industrial cameras acquire images to achieve 3D positioning of the welding needle, wire tip, and ideal welding point. These are then linked to the four motion platforms to complete the wire docking. Immediately after welding, the weld's appearance and dimensions are measured, and surface defects are detected, promptly removing defective products and effectively improving the efficiency and consistency of thermocouple product quality control. The layout design of the industrial cameras, motion platforms, and related components simultaneously meets the requirements for high-precision docking and 360° weld inspection of micro-thermocouple wires. To achieve high docking accuracy, two industrial cameras are vertically positioned to acquire images of the wire docking, determining the 3D coordinates of the welding needle, wire tip, and ideal welding point. Combined with the left three-axis motion platform and the right single-axis motion platform, the left welding wire can be precisely docked with the right welding wire through multi-dimensional adjustments, simplifying the system structure. To ensure 360° inspection of weld joints without blind spots, a third industrial camera is deployed to collaboratively capture details of different slopes and corners of the weld joints, effectively reducing missed detections caused by irregular weld joint shapes and local structural obstructions. Furthermore, the argon arc welding torch and its opposing bent-angle grounding contact are positioned at a 45° angle to the camera, geometrically avoiding obstruction of the welding needle and wire, ensuring uninterrupted endpoint positioning and weld joint inspection. In the endpoint 3D positioning method, error correction is used to improve endpoint positioning accuracy. Based on the spatial position association constructed using dual-camera vision, an angle compensation mechanism is introduced to correct potential positional deviations during installation. Simultaneously, the coordinates are corrected using reprojection errors obtained from calibration, further improving positioning accuracy. Therefore, the automatic welding and inspection system for micro-thermocouple wires provided by this invention can achieve high-quality welding of micro-thermocouple wires, while simultaneously measuring weld joint dimensions and detecting surface defects, meeting the stringent requirements of aerospace temperature measurement systems for measurement accuracy, stability, and reliability.

[0023] According to another aspect of the present invention, an automatic welding inspection method for micro-thermocouple wires is provided. This method uses the automatic welding inspection system for micro-thermocouple wires as described above to perform automatic welding inspection of micro-thermocouple wires. Specifically, the automatic welding inspection method for micro-thermocouple wires includes: after the computer issues an automatic welding inspection start command, industrial cameras A1 and B2 simultaneously begin to acquire real-time images of the welding wire docking scene; the welding needle, the left welding wire end point, the right welding wire end point, and the ideal welding point are three-dimensionally positioned using a processing algorithm; the positional deviation between the end points is calculated based on the coordinates of the welding needle, the left welding wire end point, the right welding wire end point, and the ideal welding point to generate a control command, which is transmitted to the control box. After parsing, the first motor 43, the second motor 45, the third motor 47, and the fourth motor 53 in the drive actuator control the left-side left and right movement. Platform 42, forward and backward movement platform 44, up and down movement platform 46, and right-side left and right movement platform 52 move to adjust the position of the thermocouple wires. After adjustment, images are collected and fed back to the computer to form a complete closed-loop control. By continuously correcting the relative position of the welding needle and welding wire until the left and right welding wires are connected, the welding process is started. After welding is completed, the computer immediately triggers industrial cameras A1, B2, and C3 to simultaneously collect 360° images of the weld point. After processing by the weld point quality detection algorithm, the appearance and size parameters of the weld point, whether there are surface defects, and their specific types are output.

[0024] Specifically, in this invention, the computer acts as the system control center, controlling the industrial camera in the actuator to acquire images, and generating control commands through built-in software processing algorithms, which are then sent to the controller in the control box to drive the actuator.

[0025] The control box receives displacement and speed commands from the computer for each motion platform, drives the actuators to perform endpoint docking, and controls the argon arc welding machine to perform welding.

[0026] The actuator includes industrial camera A (1), industrial camera B (2), industrial camera C (3), left adjustment component (4), right adjustment component (5), argon arc welding torch (6), grounding contact (7), and slag tray (8) (the slag tray is located at the lower part of the argon arc welding torch (6) and is used to collect welding slag). Industrial camera A (1) is vertically aligned with the welding area to capture a top-down view image. Industrial camera B (2) is arranged horizontally from the side. The two industrial cameras work together to complete the three-dimensional positioning of the welding needle and welding wire ends. Industrial camera C (3) is arranged at a 45° angle downward to the horizontal plane to complete the observation view from the oblique rear of the weld point. The synchronous imaging of the three cameras realizes 360° full coverage of the weld point surface. The left adjustment assembly includes a thermocouple wire clamp (41), a left-right motion platform (42) and a motor (43), a front-back motion platform (44) and a motor (45), and a vertical motion platform (46) and a motor (47), allowing the left welding wire tip to be adjusted in three directions; the right adjustment assembly includes a thermocouple wire clamp (51), a left-right motion platform (52) and a motor (53), allowing the right welding wire to be adjusted only in the left-right direction. The argon arc welding torch (6) is arranged at a 45° angle upward to the horizontal plane and is positioned opposite the grounding contact (7), so that the electric arc is stably formed between the welding torch and the grounding contact, rapidly melting the thermocouple wire joint to form a weld point.

[0027] Automated welding inspection process:

[0028] After the operator issues an automatic welding inspection start command through the host computer, industrial cameras A(1) and B(2) simultaneously begin to acquire real-time images of the welding wire docking scene. The processing algorithm performs three-dimensional positioning of the welding needle and welding wire endpoints in the argon arc welding gun, calculates the positional deviation between the endpoints to generate control commands, and transmits them to the control box. After parsing, the controller drives four motors (43), (45), (47), and (53) in the actuator to control the four motion platforms (42), (44), (46), and (52) to adjust the position of the thermocouple wire. After adjustment, the image is acquired and fed back to the computer to form a complete closed-loop control. By continuously correcting the relative position of the welding needle and welding wire until the thermocouple wire docking is completed, the welding process is started. After welding is completed, the computer immediately triggers industrial cameras A(1), B(2), and C(3) to simultaneously acquire 360° images of the weld point. After processing by the weld point quality detection algorithm, the appearance and size parameters of the weld point, whether there are surface defects, and their specific types are output.

[0029] Furthermore, in this invention, the three-dimensional localization of the welding needle, the left welding wire endpoint, the right welding wire endpoint, and the ideal welding point through the processing algorithm specifically includes: obtaining distortion correction parameters through single-camera intrinsic parameter calibration, and obtaining reprojection error parameters through dual-camera stereo calibration; inputting the thermocouple wire docking scene images synchronously acquired by industrial cameras A1 and B2 into a semantic segmentation network to segment the welding needle and welding wire regions and output a mask image; based on the distortion correction parameters, locating the welding needle, the left and right welding wire endpoints, and the ideal welding point in the mask image, and obtaining the pixel coordinates of the endpoints in the two images; establishing a spatial coordinate system, and combining the reprojection error parameters, pixel coordinates, and spatial geometric relationships to infer the three-dimensional coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point.

[0030] Specifically, the thermocouple wire docking scene images synchronously acquired by industrial cameras A1 and B2 are input into a semantic segmentation network to segment the welding pin and welding wire regions, and output a mask image. This process includes: introducing a global adaptive brightness adjustment image enhancement module into the U-net network, decomposing the image brightness L into an illumination component I and a reflection component R, i.e., L = I × R, and extracting the brightness information L through RGB weighted summation. w This simulates the human eye's sensitivity to different colors; it statistically models global brightness, calculating the logarithmic mean brightness value and maximum brightness value of image pixels, based on the logarithmic mean brightness value, the maximum brightness value of the input image, and the original image brightness L. w The image dynamic range is compressed using a logarithmic function to improve image contrast. The depthwise separable convolution of the MobileNetV3 network replaces the standard convolution in the U-net network. Depthwise separable convolution decomposes traditional convolution into depthwise convolution and pointwise convolution. Depthwise convolution focuses on spatial feature extraction, while pointwise convolution handles inter-channel information fusion. This significantly reduces the number of model parameters and improves inference speed while ensuring effective representation of key edge features in welding needle and wire images. The segmentation network model is then trained to segment the welding needle and wire regions in the image and output a mask image.

[0031] In this invention, the endpoint three-dimensional localization method is specifically as follows:

[0032] Input the thermocouple wire docking scene images synchronously acquired by industrial cameras A(1) and B(2), and segment the welding needle and welding wire regions through a semantic segmentation network; the center line of the welding needle region is determined by taking the mean of the extreme points, and the center line of the welding wire region is determined by fitting a polynomial curve. The intersection of each center line with its edge is the endpoint. The intersection of the welding needle center line and the right welding wire center line is defined as the ideal welding point, and the pixel coordinates of the endpoints in the two images are obtained; then, a spatial coordinate system is established, and the three-dimensional coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point are inferred by combining the calibration parameters, pixel coordinates, and spatial geometric relationships. Among them, the pixel coordinates of a point in space (if the three-dimensional coordinates of the welding needle are required, then the point in space refers to the welding needle endpoint) in the images captured by cameras A and B are (x A ,y A ), (x B ,y B After distortion correction, it becomes (x′). A ,y′ A ), (x′ B ,y′ B If ), then its y and z coordinates in space are:

[0033]

[0034] In the formula, h is the distance from the optical center of both camera lenses to the ideal welding point, ε1 and ε2 are the angles between the optical axes of cameras A and B and the z and y axes, respectively. Under ideal installation conditions, the camera optical axis coincides with the y and z axes, so ε1 and ε2 are both 0. k1 and k2 are the slopes of the total direction angle of the imaging ray (optical axis deviation angle + ray offset angle) along the y and z axes, respectively, calculated by the following formula:

[0035]

[0036] In the formula, l a Given the diameter of the thermocouple wire, n m is the number of pixels occupied by the thermocouple wire in the image, and f is the focal length of the camera lens.

[0037] Then, the endpoint x-axis coordinates are corrected using the reprojection error obtained from calibration.

[0038]

[0039] In the formula, These are the error variances calculated from the reprojection errors obtained from camera calibration. Let L be the straight-line distances from the optical centers of cameras A and B to their endpoints, respectively. (Assuming we want to calculate the three-dimensional coordinates of the welding pin, L...) A This refers to the straight-line distance in space from the optical center of camera A to the welding pin.

[0040] From equations (1), (2), and (3), we can deduce that the three-dimensional coordinates of the three endpoints of the welding needle, the left welding wire, the right welding wire, and the ideal welding point are (xt, yt, zt), (xl, yl, zl), (xr, yr, zr), and (xd, yd, zd).

[0041] Solder joint quality inspection algorithm:

[0042] Weld joint quality inspection was performed on weld joint images simultaneously acquired by industrial cameras A(1), B(2), and C(3), including surface defect detection and appearance dimension measurement. For surface defect detection, an improved semantic segmentation model was used to process the images acquired by the three industrial cameras to segment the weld joint area and potential surface defects such as deformities, cracks, and pits. The images acquired by the three industrial cameras were inspected separately, and all detected defects were identified as actual defects on the weld joint surface. For appearance dimension measurement, the coordinates of extreme points in the x and y axes were extracted from the weld joint area segmented from each image. After distortion correction, the difference in extreme point coordinates was calculated. Combined with the known ratio of thermocouple wire diameter to the number of pixels, the actual distance between extreme points in the x-axis direction was the weld joint length, and the actual distance between extreme points in the y-axis direction was the weld joint diameter. The length and diameter of the weld joint were measured from the images captured by the three cameras, and the larger value was taken as the final length and diameter of the weld joint.

[0043] In this invention, the displacement required for each motion platform to reach the docking position of the left and right welding wires is: the left-side motion platform 42 needs to move dx1 = x l -x d The right-side horizontal motion platform 52 needs to move dx2 = x r -x d The front and rear motion platform 44 needs to move dy = y l -y d The vertical movement platform 46 needs to move dz = z l -z d The three-dimensional coordinates of the three endpoints of the welding needle, the left welding wire, the right welding wire, and the ideal welding point are respectively (x... t ,y t ,z t) 、(x l ,y l ,z l ), (x r ,y r ,z r ), (x d ,yd,z d ).

[0044] Furthermore, after processing by the weld joint quality detection algorithm, the output of weld joint appearance size parameters, whether there are surface defects and their specific types includes: labeling weld joint images acquired simultaneously by industrial cameras A, B, and C to construct a dataset containing weld joints and defects such as deformities, cracks, and pits; using the Mask2Former network as a basis, employing Swim Transformer as its backbone layer to identify various features of thermocouple wire weld joints pixel by pixel; and making targeted improvements to various features of thermocouple wire weld joints to achieve surface defect detection of thermocouple wire weld joints.

[0045] Specifically, targeted improvements to various features of thermocouple wire welding points include: embedding a low-illuminance region enhancement network RRDNet before the backbone. This network consists of two stages: decomposition and restoration. In the decomposition stage, RRDNet processes the reflection component R, illumination component I, and noise component N of the input image independently through a three-branch fully convolutional neural network. Then, it adjusts the illumination component through Gamma transformation and fuses the adjusted illumination component with the noise-free reflectivity to obtain the final restoration result. After extracting feature maps of different levels from the backbone, an adaptive spatial feature fusion module ASFF is added to adaptively learn the fusion spatial weights of the feature maps at each scale, which consists of two steps: feature scaling and adaptive fusion. Feature scaling adjusts each sampling layer to the same scale, then adaptively fuses the features. The last three layers of the Swim Transformer (x1, x2, x3) are multiplied by weights α3, β3, and γ3, and summed with features from different layers to obtain new fused features. The original loss function of Mask2Former consists of classification loss and mask loss. The standard cross-entropy loss of the original classification loss is replaced with Focal Loss. By dynamically adjusting the weights of easy and difficult samples, the model focuses more on samples with cracks, dents, or other defects. The segmentation network model trained with these improvements processes the images captured by the three industrial cameras separately, segmenting the weld area and its potential surface defects such as deformities, cracks, and dents. All defects detected by the images captured by the three industrial cameras are considered actual defects on the surface of the weld.

[0046] The output solder joint appearance dimension measurement parameters specifically include: from the solder joint region segmented from each image, firstly, extracting the extreme point coordinates in the x-axis and y-axis directions; then, introducing the distortion correction parameters obtained from the single-camera intrinsic parameter calibration to correct the extracted coordinate data and eliminate geometric deviations caused by the optical characteristics of the imaging system; then, combining the known ratio of thermocouple wire diameter to the number of pixels occupied, converting the difference between the two extreme coordinates in the x-axis direction into the actual length of the solder joint, and converting the difference between the two extreme coordinates in the y-axis direction into the actual diameter of the solder joint; the solder joint length and diameter detected by the three industrial cameras A, B, and C are respectively taken as the final length and diameter of the solder joint.

[0047] To gain a further understanding of the present invention, the following description is provided in conjunction with... Figures 1 to 5 The automatic welding and inspection device and method for micro thermocouple wires provided by the present invention will be described in detail.

[0048] like Figures 1 to 5 As shown, the automatic welding and inspection system for micro thermocouple wires provided by this invention includes four parts: a computer, a control box, an actuator, and an argon arc welding machine. The system structure block diagram is shown below. Figure 1 As shown.

[0049] The computer acts as the system control center, controlling the industrial camera in the actuator to acquire images, and generating control commands through built-in software processing algorithms, which are then sent to the controller in the control box to drive the actuator.

[0050] The control box receives displacement and speed commands from the computer for each motion platform, drives the actuators to perform endpoint docking, and controls the argon arc welding machine to perform welding.

[0051] The structure of the execution mechanism is as follows Figure 2 As shown, the assembly includes industrial camera A (1), industrial camera B (2), industrial camera C (3), left adjustment assembly (4), right adjustment assembly (5), argon arc welding torch (6), grounding contact (7), and slag tray (8) (located at the lower part of the argon arc welding torch to receive slag). The left adjustment assembly includes a thermocouple wire clamp (41), a left-right movement platform (42) and a motor (43), a front-back movement platform (44) and a motor (45), and a vertical movement platform (46) and a motor (47), allowing the left welding wire tip to be adjusted in three directions. The right adjustment assembly includes a thermocouple wire clamp (51), a left-right movement platform (52), and a motor (53), allowing the right welding wire to be adjusted only in the left-right direction. The left-right movement direction is the y-direction, the front-back movement direction is the z-direction, and the vertical movement direction is the x-direction.

[0052] The right view of the industrial camera assembly, argon arc welding torch, and grounding contact structure is shown below. Figure 3As shown in the diagram. Industrial camera A (1) is vertically aligned with the welding area, while industrial camera B (2) is positioned horizontally from the side. Both cameras acquire images synchronously, providing visual data support for the three-dimensional positioning and docking of the welding needle and wire ends. Industrial camera C (3) is positioned at a 45° angle downwards from the horizontal plane, capturing images of the weld joint from below the welding area. By using three cameras to collaboratively capture images, the 360° area of ​​the weld joint is fully covered, providing visual basis for the 360° defect detection of the weld joint surface. The argon arc welding torch (6) is installed at a 45° angle to industrial camera A (1), with a bent-angle grounding contact (7) positioned opposite it. This ensures that the welding arc is stably formed between the welding needle and the grounding contact, and the welding heat is concentrated and radiated to the thermocouple wire docking area, ensuring the stable formation of the weld joint.

[0053] Automatic welding inspection process: After the operator issues the automatic welding inspection start command through the host computer, industrial cameras A (1) and B (2) simultaneously start to collect real-time images of the welding wire docking scene. The processing algorithm performs three-dimensional positioning of the welding needle, the left and right welding wire ends and the ideal welding point, and then calculates the position deviation between the ends to generate control commands, which are transmitted to the control box. After parsing, the driver drives the four motors (43), (45), (47), and (53) in the actuator to control the four motion platforms (42), (44), (46), and (52) to move, thereby driving the thermocouple wire to adjust its position. After the adjustment is completed, the image is collected and fed back to the computer to form a complete closed-loop control. The control block diagram is as follows. Figure 4 As shown. By continuously correcting the relative position of the welding needle and the welding wire until the wires are connected, the welding process is started. After welding is completed, the computer immediately triggers the industrial cameras A (1), B (2), and C (3) to synchronously acquire 360° images of the weld point. After processing by the weld point quality detection algorithm, the appearance size parameters of the weld point, whether there are defects on the surface and their specific types (including deformities, cracks, and pits) are output.

[0054] Specifically, the endpoint three-dimensional positioning method first obtains distortion correction parameters through single-camera intrinsic parameter calibration and reprojection error parameters through dual-camera stereo calibration; then, the thermocouple wire docking scene images synchronously acquired by industrial cameras A (1) and B (2) are input into a semantic segmentation network to segment the welding needle and welding wire regions; subsequently, the welding needle and the left and right welding wire endpoints and the ideal welding point in the image are located, and the pixel coordinates of the endpoints in the two images are obtained (the y and z coordinates are calculated first); finally, a spatial coordinate system is established, and the three-dimensional coordinates of the welding needle and the left and right welding wire endpoints and the ideal welding point are inferred by combining the calibration parameters, pixel coordinates and spatial geometric relationships (then the x coordinates are calculated, and the three-dimensional coordinates can be obtained by combining the y and z coordinates).

[0055] Further, the segmentation of the welding pin and welding wire areas includes: introducing a global adaptive brightness adjustment image enhancement module into the U-net network, decomposing the image brightness L into an illumination component I and a reflection component R, i.e., L = I × R, and extracting the brightness information L through RGB weighted summation. w This simulates the human eye's sensitivity to different colors. Then, a global brightness statistical model is developed to calculate the logarithmic average brightness value of the image pixels (obtained according to Formula 4) and the maximum brightness value (i.e., L). wmax This function compresses the dynamic range of an image using a logarithmic function, thereby improving image contrast. The logarithmic mean of the brightness values ​​of the input image is:

[0056]

[0057] Where m×n represents the size of the image, and σ is a very small value to avoid numerical overflow when performing log calculations on pure black pixels.

[0058] Then, a nonlinear brightness remapping process is performed to remap the original brightness L. w Mapped to the new brightness L g To enhance local contrast, the calculation formula is:

[0059]

[0060] Among them, L wmax This represents the maximum brightness of the input image.

[0061] With the new brightness L g Compared with the original brightness L w The ratio of L to G is used as a gain coefficient to adjust the bright and dark areas of the image. w If the gain is 0, the gain is 0. When gain is greater than 1, the brightness of the area is increased; when gain is less than 1, the brightness of the area is suppressed (when gain is equal to 1, no action is taken). This is applied to the RGB channels, and each channel is adjusted synchronously.

[0062] Furthermore, the depthwise separable convolution of the MobileNetV3 network was used to replace the standard convolution in the U-net network. Depthwise separable convolution decomposes traditional convolution into depthwise convolution and pointwise convolution. Depthwise convolution focuses on spatial feature extraction, while pointwise convolution is responsible for inter-channel information fusion. This significantly reduces the number of model parameters and improves inference speed while ensuring effective representation of key edge features in welding needle and wire images.

[0063] After training and convergence, the above segmentation network model can segment the welding needle and welding wire regions in the input image and output a mask image.

[0064] The positioning of the welding needle, welding wire endpoints, and ideal welding point is as follows: The approximately symmetrical welding needle region contour is extracted from the mask image. The left and right endpoints of this contour are determined, and a vertical center line is drawn. The lower intersection of this center line and the welding needle contour is defined as the welding needle endpoint. For the left and right welding wires, local points are selected on the x-axis at 20% of the total span of their respective contours closest to their endpoints. A quadratic polynomial curve of x with respect to y is fitted based on these local points. The center line of this curve is denoted as the welding wire center line, and the intersection of this center line and the corresponding edge of the welding wire is denoted as the welding wire endpoint. Since the welding needle position is fixed, the right welding wire only moves in the left-right direction, while the left welding wire can move flexibly in the left-right, up-down, and front-back directions. Therefore, the intersection of the welding needle center line and the right welding wire center line is the ideal welding point. Both the left and right welding wire endpoints approach the ideal welding point until the connection is completed. In the images captured by cameras A and B, the pixel coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point are respectively (x...). At ,y At ), (x Bt ,y Bt );(x Al ,y Al ), (x Bl ,y Bl );(x Ar ,y Ar ), (x Br ,y Br ) and (x Ad ,y Ad ), (x Bd ,y Bd ).

[0065] The establishment of the spatial coordinate system is as follows: Cameras A and B are mounted on the same vertical plane as the argon arc welding needle, with their optical axes intersecting at point O, which is designated as the origin. The positive x-axis is defined as the extension of the right welding wire centerline to the right, the positive z-axis is defined as the direction perpendicular to the operating table and upwards, and the positive y-axis is defined as the direction perpendicular to the xOz plane and outwards. This constructs the spatial coordinate system. The yOz plane in the right viewing direction is as follows: Figure 5 As shown. The optical centers of the two camera lenses are O1 and O2, respectively, with an average distance of h from point O, and the focal length of the lenses is uniformly f. Due to potential deviations during actual camera installation, this invention sets the angles between the optical axes O′1O″1 and O′2O″2 of cameras A and B and the z and y axes to ε1 and ε2, respectively. Therefore, the coordinates of the optical centers of the two cameras are (0, -hsinε1, hcosε1) and (0, hcosε2, -hsinε2), respectively.

[0066] Let P be a point in space projected perpendicularly onto the yOz plane, and its pixel coordinates in the two images be (x, y, z) respectively. A ,y A ), (x B ,y BFirst, distortion correction parameters obtained from single-camera intrinsic parameter calibration are used for distortion correction. The corrected coordinates are (x') A ,y' A ), (x' B ,y' B Then, convert the pixel coordinates to physical coordinates using the following formula:

[0067]

[0068] Among them, l a Given the diameter of the thermocouple wire, n m This represents the number of pixels occupied by the thermocouple wire in the image. (x”) A ,y” A ), (x” B ,y” B ) represents the physical coordinates of P in the images captured by cameras A and B.

[0069] (1) Solve for the y and z axis coordinates

[0070] The ratio of the distance from the physical coordinate plane to the optical axis to the focal length reflects the offset angle of the imaging ray from the optical axis. In camera A, the offset angle δ1 of the imaging ray P from the optical axis is given by the following formula:

[0071]

[0072] Similarly, in camera B, the offset angle δ2 between the imaging ray from the left welding wire tip P and the optical axis has the following relationship:

[0073]

[0074] The total direction angle of the imaging rays is "optical axis deviation angle + ray offset angle". Therefore, the total direction angles of the imaging rays from cameras A and B are ε1+δ1 and ε2+δ2, respectively. The projection slopes of the lines from their optical centers to P onto the yOz plane are tan(ε1+δ1) and tan(ε2+δ2), respectively, and their equations are:

[0075] y+hsinε1=tan(ε1+δ1)·(z-hcosε1) (8)

[0076] z+hsinε2=tan(ε2+δ2)·(y-hcosε2) (9)

[0077] Expanding the slope term using trigonometric and angle formulas, we define the total slope as k1 = tan(ε1 + δ1) and k2 = tan(ε2 + δ2), and then expand to obtain:

[0078]

[0079] Substituting equations (6) and (7) into (10) and (11), and then solving equations (8) and (9) simultaneously to eliminate variables, we finally obtain the solution:

[0080]

[0081] (2) Solve for the x-axis coordinates

[0082] The linear distances from the optical centers of cameras A and B to point P are respectively:

[0083]

[0084] Introducing a spatial scale magnification factor L A / f、L B The / f function converts the physical x-axis coordinates to three-dimensional spatial x-axis coordinates. The three-dimensional spatial x-axis coordinates of P in the images captured by cameras A and B are as follows:

[0085]

[0086] Suppose the calibration board contains m feature points, and the reprojection errors of cameras A and B are ΔP1 and ΔP2, respectively. Then the error variance is:

[0087]

[0088] To reduce the measurement error between the two cameras, an objective function is constructed to minimize the total error:

[0089]

[0090] Taking the derivative with respect to x and setting it to 0, we obtain the optimal x as:

[0091]

[0092] According to equations (12), (13), and (21), the three-dimensional coordinates of P are (x, y, z). Substituting the pixel coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point in the images captured by cameras A and B, the three-dimensional coordinates of each point can be obtained, which are (x, y, z). t ,y t ,z t ), (x l ,y l ,z l ), (x r ,y r ,z r ) and (x d ,y d ,z d ).

[0093] Therefore, the displacement required for each motion platform to reach the docking position of the left and right welding wires can be determined: the left and right motion platforms (44) need to move dx1 = xl -x d The left and right motion platform (54) needs to move dx2 = x r -x d The front and rear motion platforms (46) need to move dy = y l -y d The vertical movement platform (48) needs to move dz = z l -z d .

[0094] The distance between the tip of the welding needle and the ideal welding point is According to the requirements of the argon arc welding process, in order to ensure the stability of the arc, the distance between the welding needle and the welding position must be strictly controlled within a certain range. If the distance l is within the specified range, the subsequent welding process can continue; if the distance l exceeds the range, an alarm will be triggered and the process will be paused. Manual adjustment is required before the welding program can be restarted.

[0095] After the welding wires are joined, the argon arc welding machine starts welding. The left and right motion platforms (44) and (54) maintain a speed of 6 mm / s and perform a precision compensation feed of 0.5 mm to simulate the "extrusion welding" process of manual welding (which is the optimal welding motion control parameter obtained through multiple tests). By slightly squeezing the welding wire, it avoids "empty welding" or incomplete penetration caused by insufficient welding material, and at the same time replicates the "full and rounded" forming effect of manual welding. The welding stops immediately when the preset compensation amount is reached.

[0096] The aforementioned solder joint quality detection algorithm inputs industrial cameras A(1), B(2), and C(3) to simultaneously acquire solder joint images, and performs solder joint surface defect detection and appearance dimension measurement respectively.

[0097] Further, it includes weld joint surface defect detection: First, weld joint images are simultaneously acquired by industrial cameras A, B, and C and labeled to construct a dataset containing weld joints and defects such as deformities, cracks, and pits. Then, based on the Mask2Former network (Mask2Former is an existing network), Swim Transformer (Swim Transformer is a deep learning model architecture) is used as its backbone layer (backbone is the main network) to identify various features of thermocouple wire weld joints pixel by pixel; then, it is further improved to achieve surface defect detection of thermocouple wire weld joints. SwinTransformer is a window-based hierarchical visual transformer architecture in the existing technology. Its core design aims to handle local features and hierarchical feature extraction in image recognition tasks, belonging to the model architecture innovation in the field of deep learning. This patent uses PyCharm as the code compilation and development tool. The architecture logic of the Swin Transformer model is not a function built into PyCharm or other software, but an algorithm model independently implemented based on deep learning theory. The specific steps for targeted improvement of various features of thermocouple wire weld joints are as follows:

[0098] (1) A low-light region enhancement network, RRDNet, is embedded before the backbone. This RRDNet network consists of two stages: Decomposition and Restoration. In the decomposition stage, RRDNet uses a three-branch fully convolutional neural network to process the reflection component R, illumination component I, and noise component N of the input image independently. The illumination component is then adjusted using a Gamma transform:

[0099]

[0100] γ is a preset parameter, and S(x) is the original illumination component. Dark area illumination enhancement is achieved through nonlinear transformation. The noise-free reflectance is (in this invention, the noise-free reflectance is derived directly from the input image I, noise component N, and illumination component S, rather than directly using the reflectance component R obtained from the initial decomposition (because the initial reflectance component may still contain noise or incompletely separated interference)):

[0101]

[0102] By fusing the adjusted illuminance with the noise-free reflectance, the final restoration result is obtained as follows:

[0103]

[0104] (2) After extracting feature maps from different levels in the backbone, an adaptive spatial feature fusion module (ASFF) is added. This module adaptively learns the fusion spatial weights for each scale of feature map, and the process consists of two steps: feature scaling and adaptive fusion. Feature scaling adjusts each sampling layer to the same scale, and then adaptive fusion is performed on each feature layer. The last three layers of features from the SwimTransformer (a window-based hierarchical visual transformer architecture) are then fused. 1 x 2 x 3 Multiply the features from different layers by the weight parameter α 3 ,β 3 and γ 3 The two features are then added together to obtain a new fusion feature. The fusion process is as follows:

[0105]

[0106] in, Let l be the feature value of the fused feature map at position (i,j), where l represents the scale of the fused feature map (i.e., the target scale). The spatial weight parameters for adaptive learning correspond to the weights of three different layers of feature maps. They are obtained by performing a 1×1 convolution on the three layers of feature maps after feature scaling. After normalization, their values ​​are in the range of [0,1] and satisfy the condition that the sum is 1. They are used to adaptively allocate the contribution of different features. For the first layer feature map x 1 The feature value at position (i,j) after scaling to scale l; For the first layer feature map x 2 The feature value at position (i,j) after scaling to scale l; For the third layer feature map x 3 The feature value at position (i,j) after scaling to scale l. The purpose of step (2): to adaptively fuse features of different levels and scales (such as the last three layers of SwinTransformer features x1, x2, x3) through the adaptive spatial feature fusion module (ASFF) to obtain new fused features.

[0107] (3) The original loss function of Mask2Former (in Mask2Former, Mask stands for "mask", 2 represents that it is the second-generation iterative model of "MaskFormer", and "Former" originates from the "Transformer" architecture in deep learning) is divided into classification loss and Mask loss. This invention replaces the standard cross-entropy loss of the original classification loss with Focal Loss, and dynamically adjusts the weights of easy and difficult samples to make the model pay more attention to samples with defects such as cracks and pits. The formula is as follows:

[0108] L Focal =-α t (1-p t ) γ log(p t (26)

[0109] Formula (26) is the formula for Focal Loss. It uses the prediction confidence p of the true class. t Category balancing weight α t The focusing parameter γ is used for calculation. Where p t α represents the confidence level of the model's prediction of the true class. t For class-balanced weights, γ is a focusing parameter that controls the weight decay of easy and difficult samples. When γ = 0, it degenerates into standard cross-entropy; when γ > 0, the loss of easy-to-classify samples is significantly reduced, and difficult-to-classify samples dominate the optimization. In this invention, γ is set to 2.0, and α... t The calculation is based on the reciprocal of the category frequency.

[0110] The improved segmentation network model, after training, processes the images acquired by the three industrial cameras separately, segmenting the solder joint areas and identifying potential surface defects such as deformities, cracks, and pits. All defects detected by the images from the three industrial cameras are recognized as actual defects existing on the surface of the solder joint.

[0111] The measurement of the solder joint's external dimensions involves first extracting the extreme point coordinates along the x-axis and y-axis from the segmented solder joint region of each image. Then, distortion correction parameters obtained from single-camera intrinsic calibration (distortion correction parameters obtained during camera calibration after installation) are introduced to correct the extracted coordinate data, eliminating geometric deviations caused by the optical characteristics of the imaging system. Next, combining the known ratio of thermocouple wire diameter to its occupied pixels, the difference between the two extreme coordinates along the x-axis is converted into the actual length of the solder joint, and the difference between the two extreme coordinates along the y-axis is converted into the actual diameter of the solder joint. The larger value among the solder joint length and diameter detected by the three cameras is taken as the final length and diameter of the solder joint.

[0112] Compared with the prior art, the advantages and effects of the present invention are as follows:

[0113] 1. By employing three cameras and four motion platforms working in concert, the entire process of micro-thermocouple wire assembly, from docking and welding to weld quality inspection, has been automated. Two industrial cameras acquire images to achieve 3D positioning of the welding needle, wire tip, and ideal welding point. These are then linked to the four motion platforms to complete the wire docking. Immediately after the welding process, the weld's appearance and dimensions are measured, and surface defects are detected, allowing for the timely removal of defective products and effectively improving the efficiency and consistency of thermocouple product quality control.

[0114] 2. Through the layout design of industrial cameras, motion platforms, and related components, the system can simultaneously meet the requirements for high-precision docking of micro-thermocouple wires and 360° inspection of weld joints. To achieve high docking accuracy, two industrial cameras are placed vertically to capture images of the welding wire docking, determining the three-dimensional coordinates of the welding needle, welding wire endpoints, and the ideal welding point. Combined with the left-side three-axis motion platform and the right-side single-axis motion platform, the left welding wire can be precisely docked with the right welding wire through multi-dimensional adjustments, simplifying the system structure. To ensure 360° inspection of weld joints without blind spots, a third industrial camera is deployed to collaboratively capture details of different slopes and corners of the weld joint, effectively reducing missed detections caused by irregular shapes of three-dimensional weld joints and local structural obstructions. Furthermore, the argon arc welding torch and its opposing bent-angle grounding contact are arranged at a 45° angle to the camera, using geometric avoidance to prevent obstruction of the welding needle and welding wire, ensuring that endpoint positioning and weld joint inspection are not interfered with.

[0115] 3. In the endpoint 3D positioning method, endpoint positioning accuracy is improved by correcting errors. Based on the spatial position association constructed using dual-camera vision, an angle compensation mechanism is introduced to correct potential positional deviations during installation. Simultaneously, the coordinates are corrected using reprojection errors obtained from calibration, thereby further improving positioning accuracy.

Claims

1. An automatic welding and inspection system for micro thermocouple wires, characterized in that, The automatic welding and inspection system for micro thermocouple wires includes a computer, a control box, an actuator, and an argon arc welding machine. The computer serves as the system control center, controlling the industrial camera in the actuator to acquire images and generating control commands through built-in software processing algorithms, which are then sent to the control box. The control box receives displacement and speed commands from each motion platform sent by the computer, drives the actuator to perform endpoint docking, and controls the argon arc welding machine to perform welding. The actuator includes an industrial camera A (1), an industrial camera B (2), an industrial camera C (3), a left adjustment component (4), a right adjustment component (5), an argon arc welding torch (6), a grounding contact (7), and a slag tray (8). The industrial camera A (1) is vertically aligned with the welding area to capture a top-down view image. The industrial camera B (2) is arranged horizontally from the side. The two industrial cameras work together to complete the three-dimensional positioning of the welding needle and welding wire ends. The industrial camera C (3) is arranged at a 45° angle downwards from the horizontal plane to complete the observation view from the oblique rear of the welding point. The synchronous imaging of the three cameras achieves 360° full coverage of the welding point surface. The slag tray is set at the lower part of the argon arc welding torch (6) for recycling welding slag. The left adjustment assembly includes a left thermocouple wire clamp (41), a left left-right movement platform (42) and a first motor (43), a front-back movement platform (44) and a second motor (45), an up-down movement platform (46) and a third motor (47), so that the left welding wire end point can be adjusted in three directions. The left thermocouple wire clamp (41) is used to clamp the left welding wire. The first motor (43) is used to drive the left left-right movement platform (42) to move in the left-right direction. The second motor (45) is used to drive the front-back movement platform (44) to move in the front-back direction. The third motor (47) is used to drive the up-down movement platform (46) to move in the up-down direction. The left thermocouple wire clamp (41) is set on the left left-right movement platform (42). The left-right movement platform (42) is set on the front-back movement platform (44). The front-back movement platform (44) is set on the up-down movement platform (46). The right adjustment assembly includes a right thermocouple wire clamp (51), a right left and right motion platform (52), and a fourth motor (53), so that the right welding wire can be adjusted only in the left and right directions. The fourth motor (53) is used to drive the left and right motion platform (52) to move in the left and right directions. The thermocouple wire clamp (51) is set on the left and right motion platform (52). The right thermocouple wire clamp (51) is used to clamp the right welding wire. The argon arc welding gun (6) is arranged at a 45° angle upward to the horizontal plane and is placed opposite the grounding contact (7), so that the electric arc is stably formed between the welding needle and the grounding contact, and the thermocouple wire docking part is quickly melted to form a weld point.

2. An automatic welding inspection method for micro thermocouple wires, characterized in that, The automatic welding and inspection method for micro thermocouple wires uses the automatic welding and inspection system for micro thermocouple wires as described in claim 1 to perform automatic welding and inspection of micro thermocouple wires.

3. The automatic welding and inspection method for micro thermocouple wires according to claim 2, characterized in that, The automatic welding and inspection method for micro thermocouple wires includes: After the computer issues the automatic welding inspection start command, industrial camera A (1) and industrial camera B (2) simultaneously start to acquire real-time images of the welding wire docking scene. The processing algorithm performs three-dimensional positioning of the welding needle, the left welding wire end point, the right welding wire end point and the ideal welding point. Based on the coordinates of the welding needle, the left welding wire end point, the right welding wire end point and the ideal welding point, the position deviation between the endpoints is calculated to generate a control command, which is transmitted to the control box. After analysis, the first motor (43), the second motor (45), the third motor (47) and the fourth motor (53) in the drive actuator control the left left and right motion platform (42), the front and back motion platform (44), the up and down motion platform (46) and the right left and right motion platform (52) to move, thereby driving the thermocouple wire to adjust its position. After adjustment, images are collected and fed back to the computer to form a complete closed-loop control; the relative positions of the welding needle and welding wire are continuously corrected until the left and right welding wires are connected, and then the welding process is started. After welding is completed, the computer immediately triggers industrial cameras A(1), B(2), and C(3) to simultaneously acquire 360° images of the weld joint. After processing by the weld joint quality detection algorithm, the computer outputs the appearance and size parameters of the weld joint, whether there are surface defects, and their specific types.

4. The automatic welding and inspection method for micro thermocouple wires according to claim 3, characterized in that, The processing algorithm performs three-dimensional positioning of the welding needle, the left welding wire tip, the right welding wire tip, and the ideal welding point, specifically including: Distortion correction parameters are obtained through single-camera intrinsic parameter calibration, and reprojection error parameters are obtained through dual-camera stereo calibration. The thermocouple wire docking scene images synchronously acquired by industrial camera A(1) and industrial camera B(2) are input into the semantic segmentation network to segment the welding needle and welding wire regions and output the mask image. Based on the distortion correction parameters, the endpoints of the welding pin and the left and right welding wires, as well as the ideal welding point, are located in the mask image, and the pixel coordinates of the endpoints in the two images are obtained. A spatial coordinate system is established, and the three-dimensional coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point are deduced by combining the reprojection error parameters, pixel coordinates, and spatial geometric relationships.

5. The automatic welding and inspection method for micro thermocouple wires according to claim 4, characterized in that, The thermocouple wire docking scene images synchronously acquired by industrial camera A(1) and industrial camera B(2) are input into the semantic segmentation network to segment the welding pin and welding wire regions, and the output mask image specifically includes: A global adaptive brightness adjustment image enhancement module is introduced into the U-net network, which decomposes the image brightness L into an illumination component I and a reflection component R, i.e., L = I × R. Brightness information L is extracted by RGB weighted sum. w It simulates the human eye's sensitivity to different colors; Global brightness statistical modeling is performed to calculate the logarithmic mean brightness value and maximum brightness value of image pixels. Based on the logarithmic mean brightness value, the maximum brightness value of the input image, and the original image brightness L, w This compresses the dynamic range of an image using a logarithmic function, thereby improving image contrast. We replace the standard convolution in the U-net network with the depthwise separable convolution of the MobileNetV3 network. The depthwise separable convolution decomposes the traditional convolution into depthwise convolution and pointwise convolution. The depthwise convolution focuses on spatial feature extraction, while the pointwise convolution is responsible for the fusion of information between channels. While ensuring the effective expression of key edge features of welding needle and welding wire images, we significantly compress the number of model parameters and improve the inference speed. Complete the training of the segmentation network model, segment the welding needle and welding wire regions in the image, and output the mask image.

6. The automatic welding inspection method for micro thermocouple wires according to any one of claims 1 to 5, characterized in that, The three-dimensional coordinates of the welding needle, the left and right welding wire endpoints, and the ideal welding point can be determined according to... The values ​​are calculated and obtained as follows: h is the distance from the optical center of both camera lenses to the ideal welding point; ε1 and ε2 are the angles between the optical axes of cameras A and B and the z and y axes, respectively; k1 is the slope of the total direction angle of the imaging ray along the y-direction; and k2 is the slope of the total direction angle of the imaging ray along the z-direction. The error variance is calculated from the reprojection error obtained from the calibration of camera A. L is the error variance calculated from the reprojection error obtained from the calibration of camera B. A Let L be the straight-line distance in space from the optical center of camera A to the endpoint to be determined. B Let x be the straight-line distance from the optical center of camera B to the endpoint. The pixel coordinates of the endpoint in the images captured by cameras A and B are (x, y). A ,y A ), (x B ,y B After distortion correction, it becomes (x') A ,y' A ), (x' B ,y' B ).

7. The automatic welding inspection method for micro thermocouple wires according to claim 6, characterized in that, The displacement required for each motion platform to reach the docking position of the left and right welding wires: The left and right motion platforms (42) need to move dx1 = x l -x d The right-side left and right motion platform (52) needs to move dx2 = x r -x d The front and rear motion platforms (44) need to move dy = y l -y d The vertical movement platform (46) needs to move dz = z l -z d The three-dimensional coordinates of the three endpoints of the welding needle, the left welding wire, the right welding wire, and the ideal welding point are respectively (x... t ,y t ,z t) 、(x l ,y l ,z l ), (x r ,y r ,z r ), (x d ,yd,z d ).

8. The automatic welding inspection method for micro thermocouple wires according to claim 7, characterized in that, After processing by the solder joint quality inspection algorithm, the output includes the solder joint appearance and size parameters, the presence and type of surface defects, and more: Images of weld points acquired simultaneously by industrial cameras A, B, and C are labeled to construct a dataset containing weld points and defects such as deformities, cracks, and pits. Based on the Mask2Former network, the Swim Transformer is used as its backbone layer to identify various features of thermocouple wire solder joints pixel by pixel. Targeted improvements were made to various characteristics of thermocouple wire weld joints to enable the detection of surface defects in thermocouple wire weld joints.

9. The automatic welding and inspection method for micro thermocouple wires according to claim 8, characterized in that, Targeted improvements were made to various characteristics of thermocouple wire solder joints, including: A low-light region enhancement network, RRDNet, is embedded before the backbone. This network consists of two stages: Decomposition and Restoration. In the decomposition stage, RRDNet processes the reflection component R, illumination component I, and noise component N of the input image in parallel through a three-branch fully convolutional neural network. Then, the illumination component is adjusted through Gamma transformation, and the adjusted illumination component is fused with the noise-free reflectance to obtain the final restoration result. After extracting feature maps from different levels in the backbone, an adaptive spatial feature fusion module (ASFF) is added. ASFF adaptively learns the fusion spatial weights for each scale of feature map, and the process consists of two steps: feature scaling and adaptive fusion. Feature scaling adjusts each sampling layer to the same scale. Then, adaptive fusion is performed on each feature layer. The last three layers of the Swim Transformer's features x1, x2, and x3 are multiplied by the features from different layers, weighted by parameters α3, β3, and γ3, and then summed to obtain new fused features. The original loss function of Mask2Former is divided into classification loss and Mask loss. The standard cross-entropy loss of the original classification loss is replaced with Focal Loss. By dynamically adjusting the weights of easy and difficult samples, the model pays more attention to samples with defects such as cracks and pits. The improved segmentation network model, after training, processes the images acquired by the three industrial cameras separately, segmenting the solder joint areas and identifying potential surface defects such as deformities, cracks, and pits. All defects detected by the images from the three industrial cameras are recognized as actual defects existing on the surface of the solder joint.

10. The automatic welding inspection method for micro thermocouple wires according to claim 9, characterized in that, The output solder joint appearance size measurement parameters specifically include: firstly, extracting the extreme point coordinates in the x-axis and y-axis directions from the segmented solder joint region of each image; then, introducing the distortion correction parameters obtained from the single camera intrinsic parameter calibration to correct the extracted coordinate data and eliminate geometric deviations caused by the optical characteristics of the imaging system; Combining the known relationship between the diameter of the thermocouple wire and the proportion of pixels occupied, the difference between the two extreme coordinates in the x-axis direction is converted into the actual length of the solder joint, and the difference between the two extreme coordinates in the y-axis direction is converted into the actual diameter of the solder joint. The length and diameter of the solder joint detected by the three industrial cameras A, B, and C are taken as the final length and diameter of the solder joint.

Images