Arc-shaped steel structure welding system and device

Abstract

The application relates to the technical field of steel structure welding, and discloses an arc-shaped steel structure welding system and device, wherein the welding system comprises an image processing module, a data processing module, a molten pool prediction module and a control module; the control module is electrically connected with the image acquisition module, the data processing module and the molten pool prediction module respectively. In the application, the real-time image of the molten pool is collected through the image processing module, the data processing module and the molten pool prediction module, and the current molten pool state is predicted; when the arc-shaped steel structure with different local thicknesses and butt gaps is welded, the welding current and the welding speed can be automatically adjusted according to the predicted molten pool result; in the welding process, the welding adaptation of the arc-shaped steel structure to the differences in its own parameters is realized, so that the stability of the molten pool during welding is ensured, and the non-penetration or over-penetration is avoided.

Description

Technical Field

[0001] This invention relates to the field of steel structure welding technology, and in particular to an arc-shaped steel structure welding system and apparatus. Background Technology

[0002] In modern manufacturing, welding technology, as a key metal joining process, is widely used in aerospace, shipbuilding, automotive industry, energy and construction and other fields. Among them, curved steel pipes, box girders and arch ribs have been widely used in spatial structures such as long-span bridges, exhibition domes and high-speed railway platform canopies.

[0003] In the welding of curved steel structures, welding quality and efficiency are crucial to the overall performance and reliability of the product. Therefore, the welding process must not only ensure the strength and stability of the welded joint but also maximize production efficiency and reduce labor costs. Firstly, during the welding process, the varying thickness and butt joint gaps within the curved steel structure affect welding parameters such as heat input and welding speed, leading to unstable molten pool conditions and directly impacting the quality of the welded joint. In traditional manual welding, experienced welders observe the molten pool morphology and flow characteristics to adjust welding parameters promptly, ensuring its stability. However, in automated welding systems, dynamically adjusting the welding process based on real-time monitoring of the molten pool condition and replacing manual welders presents challenges. Therefore, this paper proposes a welding system and apparatus for curved steel structures that can adjust the output current and welding speed according to the molten pool condition during welding, thereby improving the welding effect. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing automated welding systems that cannot monitor the molten pool status in real time and dynamically adjust welding parameters during the welding process to ensure the stability of the molten pool. Therefore, this invention proposes an arc-shaped steel structure welding system and apparatus.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: the welding system includes: an image processing module, a data processing module, a molten pool prediction module, and a control module, wherein the control module is electrically connected to the image acquisition module, the data processing module, and the molten pool prediction module respectively;

[0006] The image processing module acquires real-time images of the molten pool during the welding process of the arc-shaped steel structure, distinguishes feature regions in the real-time image of the molten pool by the brightness value, the feature regions include the arc area and the molten pool area, and determines the position of the arc area so that the arc area is located at the center of the real-time image of the molten pool.

[0007] The data processing module includes an arc feature extraction unit and a molten pool feature extraction unit. The arc feature extraction unit extracts a first key feature of the arc region, which includes the arc edge, width, and height features of the arc region. The molten pool feature extraction unit extracts a second key feature of the molten pool region, which includes the molten pool width feature of the molten pool region.

[0008] The molten pool prediction module predicts the molten pool state based on the first key feature and the second key feature, and classifies the molten pool state into: fully molten, not fully molten, and over-molten, and outputs the result.

[0009] The control module responds to the output result and outputs adjustment variables to incrementally adjust the current and speed during welding.

[0010] As a further description of the above technical solution:

[0011] It also includes a welding posture adjustment module. The data processing unit acquires the molten pool state after adjusting the current and speed and outputs a corrected molten pool state. The corrected molten pool state is then classified by the molten pool prediction module. The activation of the welding posture adjustment module includes the following steps:

[0012] Real-time acquisition and correction of the molten pool status;

[0013] When correcting the molten pool state to a fully penetrated state, maintain the current welding posture;

[0014] Adjust the welding posture when the molten pool is in a state of incomplete or excessive penetration.

[0015] As a further description of the above technical solution:

[0016] The image processing module includes a grayscale conversion unit, a binarization processing unit, a region filtering unit, and a selection unit. The grayscale conversion unit converts the real-time image of the color molten pool into a grayscale image. The binarization processing unit sets a brightness threshold to determine the arc area and processes the grayscale image to generate a binary image. The region filtering unit detects white blocks in the binary image and designs an area threshold. The area of ​​the white blocks is compared with the area threshold to further determine the arc area and output the arc area image. After the region filtering unit determines the arc area, the selection unit constructs a centroid coordinate calculation function with the center of the real-time molten pool image as the target and outputs the arc area center position parameter after logical operation. The center position of the real-time molten pool image is determined by the arc area center position parameter.

[0017] As a further description of the above technical solution:

[0018] The arc feature extraction unit converts the arc area image to grayscale and performs noise reduction using Gaussian filtering. Then, it uses convolution operations to perform binarization to detect the arc edge and outputs the arc contour map, thereby obtaining the arc edge, width, and height features of the arc area.

[0019] As a further description of the above technical solution:

[0020] The molten pool feature extraction unit includes a filtering subunit, an enhancement subunit, and a width calculation subunit. The filtering subunit uses Gaussian filtering to reduce noise in the grayscale image. The enhancement subunit changes the brightness distribution of the denoised grayscale image to mark the molten pool boundary and outputs a molten pool boundary map. The width calculation subunit calculates the molten pool width through the molten pool boundary.

[0021] As a further description of the above technical solution:

[0022] The molten pool prediction module includes a preprocessing unit and a model unit. The preprocessing unit tensors the arc contour map, molten pool boundary map, and their combined map to obtain image data tensors, and fuses the weld feature values ​​tensors obtained from image algorithm processing. The model unit processes the fused image data and weld feature value tensors, and includes an input layer, a convolutional layer, a sampling layer, and an output layer. The input layer receives welding current and welding speed process parameters and tensors. The convolutional layer has two sets that overlap with the sampling layer, extracts the process parameters and tensors from the input layer, and outputs the calculation results after passing through an activation function. The output layer classifies the molten pool state based on the calculation results.

[0023] As a further description of the above technical solution:

[0024] The weld feature values ​​are arc width, arc area, molten pool width, molten pool length, input current, and welding speed. Image data and weld feature value tensors are fused using a weighted fusion method. The convolutional layer and sampling layer extract tensor features of image data and weld feature values ​​and combine them with process parameters to form a convolutional layer feature map. Representative features are sampled from the convolutional layer feature map to obtain the combination relationship between representative features. The output layer includes three neurons, corresponding to the penetration state, incomplete penetration state, and over-penetration state, respectively.

[0025] As a further description of the above technical solution:

[0026] The welding posture adjustment module includes a receiving unit and an adjustment unit. The receiving unit receives a start signal from the control module and outputs it to the adjustment unit. The start signal includes an incomplete penetration signal and an over-penetration signal. The adjustment unit controls the robotic arm to adjust the angle between the welding head and the arc-shaped steel structure.

[0027] A welding device for arc-shaped steel structures:

[0028] The device includes a welding mechanism for welding arc-shaped steel structures, a positioner for fixing the arc-shaped steel structures, and a control platform. An industrial camera for acquiring images is installed on the welding mechanism.

[0029] As a further description of the above technical solution:

[0030] The welding mechanism includes a robotic arm and a support platform for supporting and mounting the robotic arm. A welding head is installed at the head of the robotic arm. An industrial camera is mounted on the robotic arm via an adjustment mechanism. The adjustment mechanism includes a first drive component and a second drive component for driving the industrial camera to adjust its angle. The output end of the first drive component is fixedly connected to a mounting bracket for mounting the second drive component. The industrial camera is fixedly connected to the output end of the second drive component.

[0031] The present invention has the following beneficial effects:

[0032] 1. In this invention, real-time images of the molten pool are acquired through an image processing module, a data processing module, and a molten pool prediction module, and the current state of the molten pool is predicted. When welding arc-shaped steel structures with different local thicknesses and butt joint gaps, the welding current and welding speed can be automatically adjusted according to the predicted molten pool results. During the welding process, the welding adaptation to the different parameters of the arc-shaped steel structure itself is realized, thereby ensuring the stability of the molten pool during welding and avoiding incomplete or excessive penetration.

[0033] 2. In this invention, by using the welding posture adjustment module, if the molten pool state is still unqualified after adjusting the current and welding speed, the angle between the welding head and the arc-shaped steel structure can be adjusted according to the current molten pool state, which can further ensure the state of the molten pool during welding and ensure the welding effect.

[0034] 3. In this invention, the image processing module distinguishes between the arc area and the molten pool area based on the brightness value, and further processes the arc area and the image to filter and determine the center position of the real-time image of the molten pool. This ensures that the industrial camera always follows the real-time image of the molten pool, improving the certainty of the acquired image and making the subsequent prediction of the molten pool state more accurate, thus avoiding errors in judging the molten pool state due to image offset.

[0035] 4. In this invention, the data processing module uses different methods to extract the first key feature of the arc zone and the second key feature of the molten pool zone to obtain the edge, width and height features of the arc zone and the width of the molten pool. This allows for a more accurate determination of the current molten pool state. The molten pool prediction module then classifies the molten pool state and adjusts the welding current and welding speed with the goal of stabilizing the molten pool state, thereby ensuring the strength of the welded joint. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the workflow of an arc-shaped steel structure welding system proposed in this invention.

[0037] Figure 2 This is a schematic diagram of the arc-shaped steel structure welding system proposed in this invention;

[0038] Figure 3 Here is a structural flow diagram of an arc-shaped steel structure welding system proposed in this invention;

[0039] Figure 4 This is a flowchart of the image processing module of an arc-shaped steel structure welding system proposed in this invention.

[0040] Figure 5 This is a flowchart of the data processing module of an arc-shaped steel structure welding system proposed in this invention;

[0041] Figure 6 This is a three-dimensional structural schematic diagram of an arc-shaped steel structure welding device proposed in this invention;

[0042] Figure 7 This is a partial enlarged view of the arc-shaped steel structure welding device proposed in this invention.

[0043] Legend:

[0044] 1. Welding mechanism; 101. Support platform; 102. Robotic arm; 103. Welding head; 2. Positioner; 3. Industrial camera; 4. Adjustment mechanism; 401. First drive component; 402. Mounting bracket; 403. Second drive component; 5. Control platform. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0046] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0047] To address the issue that in the automated welding of curved steel structures, due to variations in local thickness and butt joint gaps, the molten pool becomes unstable when using the same current and welding speed. Furthermore, automated welding systems cannot dynamically adjust welding parameters based on real-time molten pool conditions to ensure stability. Therefore, this application proposes a welding system for curved steel structures. Figures 1-7 As shown, it includes an image processing module, a data processing module, a melt pool prediction module, and a control module. The control module is electrically connected to the image acquisition module, the data processing module, and the melt pool prediction module.

[0048] The image processing module acquires real-time images of the molten pool during the welding process of the arc-shaped steel structure. It distinguishes feature regions in the real-time image of the molten pool by using brightness values. The feature regions include the arc area and the molten pool area. It also determines the position of the arc area so that the arc area is located at the center of the real-time image of the molten pool.

[0049] The data processing module includes an arc feature extraction unit and a molten pool feature extraction unit. The arc feature extraction unit extracts the first key feature of the arc region, which includes the arc edge, width and height features of the arc region. The molten pool feature extraction unit extracts the second key feature of the molten pool region, which includes the molten pool width feature of the molten pool region.

[0050] The molten pool prediction module predicts the molten pool state based on the first key feature and the second key feature, classifies the molten pool state into three categories: fully molten, not fully molten, and over-molten, and outputs the results.

[0051] The control module responds to the output results and outputs adjustment variables to incrementally adjust the current and speed during welding.

[0052] In this way, the molten pool characteristics can be obtained through the image processing module and the image acquisition module during the welding of the arc-shaped steel structure. Based on the relationship between the molten pool characteristics and the current and welding speed during welding, the penetration state of the weld can be predicted and judged in real time. Since the thickness of the welded steel structure and the butt gap are different, the state of the molten pool changes. By adjusting the current and welding speed in real time, the molten pool is always kept in a stable state. Even after adjusting the current and speed, if the molten pool is still incomplete or over-penetrated, the state of the molten pool can be predicted and judged again. Based on the state of the molten pool, the angle between the welding head 103 and the arc-shaped steel structure can be adjusted to ensure the state of the molten pool and thus obtain a good welding result.

[0053] Furthermore, different image processing techniques are used to extract features from the arc zone and the molten pool zone, accurately obtaining the first and second key features of the arc zone and the molten pool zone. Based on the first and second key features, the current molten pool state is obtained. Then, the output of the welding machine is adaptively adjusted according to the output of the prediction module, thereby realizing real-time control of the molten pool state and keeping the molten pool state stable.

[0054] In addition, by automatically identifying the arc zone and the molten pool zone and then automatically and accurately selecting the center position of the real-time image of the molten pool, the dynamic changes of the arc zone and the molten pool zone that need to be acquired with the changes of time and working conditions can be avoided, which would affect the image acquisition effect. This enables the industrial camera 3 to automatically track the arc zone and the molten pool zone, ensuring the continuity and stability of image acquisition.

[0055] The arc-shaped steel structure welding system and apparatus described in this application can be applied to the welding of arc-shaped steel structures, specifically but not limited to the welding of arched, grid shell, and grid frame types of arc-shaped steel structures.

[0056] In one embodiment, see Figures 2-3 It also includes a welding posture adjustment module. The data processing unit acquires the molten pool state after adjusting the current and speed and outputs the corrected molten pool state. The corrected molten pool state is then classified by the molten pool prediction module. The start-up of the welding posture adjustment module includes the following steps:

[0057] Real-time acquisition and correction of the molten pool status;

[0058] When correcting the molten pool state to a fully penetrated state, maintain the current welding posture;

[0059] When the molten pool is in a state of incomplete or excessive penetration, adjust the welding posture;

[0060] The welding posture adjustment module includes a receiving unit and an adjustment unit. The receiving unit receives the start signal from the control module and outputs it to the adjustment unit. The start signal includes an incomplete penetration signal and an over-penetration signal. The adjustment unit controls the robotic arm 102 to adjust the angle between the welding head 103 and the curved steel structure. When the start signal is an incomplete penetration signal, the angle between the welding head 103 and the surface of the curved steel structure is increased to reduce the contact area between the arc and the curved steel structure, making the heat input more concentrated and improving the fusion depth. When the start signal is an over-penetration signal, the angle between the welding head 103 and the surface of the curved steel structure is decreased to increase the contact area between the arc and the curved steel structure, making the heat input more evenly distributed and reducing the risk of local overheating.

[0061] In one embodiment, see Figures 1-5The image processing module includes a grayscale conversion unit, a binarization processing unit, a region filtering unit, and a selection unit. The grayscale conversion unit converts the real-time image of the color molten pool into a grayscale image. The binarization processing unit sets a brightness threshold to determine the arc area and processes the grayscale image to generate a binary image. The region filtering unit detects white blocks in the binary image and designs an area threshold. It compares the area of ​​the white blocks with the area threshold to further determine the arc area and outputs the arc area image. After the region filtering unit determines the arc area, the selection unit constructs a centroid coordinate calculation function with the center of the real-time image of the molten pool as the target and outputs the arc area center position parameter after logical operation. The center position of the real-time image of the molten pool is determined by the arc area center position parameter.

[0062] The formula for calculating the centroid coordinates of the real-time image of the molten pool is as follows:

[0063] , ,

[0064] Where CX and CY are the centroid coordinates of the arc region, representing the center position of the region, M00 is the zeroth moment, M10 is the first moment of the X-axis, and M01 is the first moment of the Y-axis.

[0065] To ensure accurate tracking of the arc region's location, consecutive images are divided into 2-second intervals. The average centroid coordinates of the white blocks across all video images within a given period are used as the center of the arc region within that period. The calculation formula is as follows:

[0066] ,

[0067] Where n is the number of sampling points recorded within 2 seconds. and The center position of the real-time image of the molten pool is the current position. When generating the center position of the real-time image in the current period, the center position of the real-time image of the molten pool in the previous period is used as a reference. The center position of the real-time image of the molten pool is periodically updated according to this rule to determine the continuity and stability.

[0068] In one embodiment, see Figures 1-5 The arc feature extraction unit converts the arc area image to grayscale and uses Gaussian filtering for noise reduction. Then, it uses convolution operation to perform binarization to detect the arc edge and outputs the arc contour map, thereby obtaining the arc edge, width and height features of the arc area.

[0069] Binarization defines a black convolutional kernel and a white convolutional kernel, used to detect low-brightness and high-brightness regions respectively. The center value of the black convolutional kernel is 0, and the other 8 positions are all 1. The center value of the white convolutional kernel is 1, and the other 8 positions are all 0. The calculation formula is as follows:

[0070] ,

[0071] because Multiplying by 0 will not affect the final result, but The image is binarized by multiplying the values ​​by 255 using only 0 or 1. The outermost edge is then extracted using the following formula:

[0072] ,

[0073] in This represents the image after the dilation operation. The image after the erosion operation is used to determine the difference area before and after expansion, and to extract the outermost edge. The outermost edge is marked as 255, and the non-edge area is marked as 0, thus obtaining the arc edge extraction result.

[0074] The molten pool feature extraction unit includes a filtering subunit, an enhancement subunit, and a width calculation subunit. The filtering subunit uses Gaussian filtering to reduce noise in the grayscale image. The enhancement subunit changes the brightness distribution of the denoised grayscale image to mark the molten pool boundary and outputs the molten pool boundary map. The width calculation subunit calculates the molten pool width through the molten pool boundary.

[0075] After converting the molten pool image to grayscale, Gaussian filtering is used for noise reduction to remove textures caused by molten pool surface fluctuations, further reducing noise in the molten pool. The Sigmoid transform algorithm is then used to alter the brightness distribution of the molten pool image, enhancing the contrast of the molten pool boundaries. The Sigmoid transform formula is as follows:

[0076] ,

[0077] Where x is the gray value of the image, α is a parameter to adjust the image contrast. A larger α value will enhance the image contrast, making the difference between bright and dark areas more obvious. β is a parameter to control the image brightness. Increasing β will make the overall brightness of the image darker, while decreasing β will make the image brighter. The Sigmoid function maps the input gray value to the range [0, 255].

[0078] After applying the Sigmoid transform, the details in areas with low brightness and weak contrast in the molten pool image are effectively enhanced, the molten pool boundary is presented more clearly, and the image features of the boundary of the molten pool area and the interior of the molten pool are distinguished. A binary image of the width information of the molten pool and the image features of the middle part of the molten pool is obtained. When the binary image is displayed on the original image, the difference between the horizontal coordinates of the pixels on the left and right sides is the width of the molten pool.

[0079] In one embodiment, see Figures 1-4The molten pool prediction module includes a preprocessing unit and a model unit. The preprocessing unit tensors the arc contour map, molten pool boundary map, and their combined map to obtain image data tensors, and fuses the weld feature values ​​tensors obtained from image algorithm processing. The model unit processes the fused image data and weld feature values ​​tensors, and includes an input layer, a convolutional layer, a sampling layer, and an output layer. The input layer receives welding current and welding speed process parameters and tensors. The convolutional layer has two sets that overlap with the sampling layer to extract the process parameters and tensors from the input layer. After passing through an activation function, the calculation results are output. The output layer classifies the molten pool state based on the calculation results.

[0080] The input arc contour map, molten pool boundary map, and their combined map are scaled to 224x224. This method can ensure the integrity of key image features while adjusting the image size. In order to make full use of image features, the three binary images are stacked along the channel dimension to obtain the image tensor of each sample. The fused image algorithm includes, but is not limited to, traditional image processing algorithms such as edge detection algorithm and fixed threshold segmentation algorithm. The input weld feature values ​​are obtained by the image processing algorithm, and the extracted weld feature values ​​are mapped to a fixed-dimensional tensor using a fully connected layer.

[0081] The weld feature values ​​are arc width, arc area, molten pool width, molten pool length, input current, and welding speed. Image data and weld feature value tensors are fused using a weighted fusion method. Convolutional layers and sampling layers extract tensor features from image data and weld feature values ​​and combine them with process parameters to form convolutional layer feature maps. Representative features are sampled from the convolutional layer feature maps to obtain the combination relationship between representative features. The output layer includes three neurons, corresponding to the penetration state, incomplete penetration state, and over-penetration state, respectively.

[0082] When fusing the tensors of image data and weld feature values, a learnable weight alpha is added to adjust the contribution of image data and weld feature values. The formula for weighted fusion is as follows:

[0083] ,

[0084] in For the fused feature tensor, For image data tensors, The extended numerical tensor enables dynamic adjustment of the contribution of image data and weld feature values, balancing their importance and making the information complementary to improve recognition ability and accuracy.

[0085] The convolutional layer receives image data and weld feature values ​​through ten input channels, three from image data and seven from weld feature values. The output layer uses the Softmax activation function to transform the unnormalized score of each category into a probability distribution, thereby ensuring that the output can be interpreted as the confidence level of each state.

[0086] The model unit uses two convolutional layers to model the molten pool state. Based on the prediction results of the model unit, the welding machine output is adaptively adjusted to achieve real-time control of the molten pool state. When the prediction result is a full penetration state, the current welding parameters are maintained to avoid molten pool oscillations caused by parameter disturbances, ensuring weld formation stability, while reducing control command redundancy and reducing actuator mechanical wear. When the prediction result is an over-penetration state, the welding current is linearly reduced at a certain rate. The reduced current reduces heat input and alleviates the risk of collapse caused by molten pool overheating. At the same time, the welding speed is increased at a certain rate. The increased speed shortens the heat source's contact time and inhibits excessive molten pool diffusion. When the prediction result is an incomplete penetration state, the welding current is increased at a certain rate. The increased current increases the heat input to the molten pool and promotes full melting of the base material. At the same time, the welding speed is reduced at a certain rate. The reduced speed prolongs the heat source's residence time and improves the fusion line morphology.

[0087] In a specific embodiment: During the welding process, the image processing module acquires real-time images of the molten pool, the output processing module analyzes and processes the real-time images of the molten pool, and the molten pool prediction module predicts and outputs the current state of the molten pool. When the welding position changes and the measured state of the molten pool is incomplete, the control module increases the welding current at a rate of 5A / s and decreases the welding speed at a rate of 0.005m / s². After adjusting the welding current and welding speed, the corrected molten pool state is obtained. If the corrected molten pool state is still incomplete, the welding posture adjustment module controls the welding head 103 to increase the angle with the surface of the arc-shaped steel structure, and after adjustment, the real-time image of the molten pool is acquired again and processed.

[0088] In one embodiment, see Figures 6-7 A welding device for arc-shaped steel structures includes a welding mechanism 1 for welding arc-shaped steel structures and a positioner 2 for fixing the arc-shaped steel structures. The welding mechanism 1 is equipped with an industrial camera 3 for acquiring images and a control platform 5. The industrial camera 3 acquires real-time images of the molten pool and sends them to an image processing module for processing. The control platform 5 is electrically connected to the welding mechanism 1. The positioner 2 clamps and positions the arc-shaped steel structure after welding and inputs the material characteristics of the arc-shaped steel structure, such as its size, shape, thickness, and welding position, into the control platform 5. The clamping angle is preset and maintained during the welding process.

[0089] The welding mechanism 1 includes a robotic arm 102 and a support platform 101 for supporting and mounting the robotic arm 102. A welding head 103 is mounted on the head of the robotic arm 102. An industrial camera 3 is mounted on the robotic arm 102 via an adjustment mechanism 4. The control platform 5 presets the travel path of the robotic arm 102 and the angle of the welding head 103 based on material characteristics, and adjusts the angle of the industrial camera 3 via the adjustment mechanism 4 so that the lens of the industrial camera 3 is aligned with the position of the welding head 103 and the arc-shaped steel structure.

[0090] The adjustment mechanism 4 includes a first drive member 401 and a second drive member 403 for driving the industrial camera 3 to adjust its angle. The output end of the first drive member 401 is fixedly connected to a mounting bracket 402 for mounting the second drive member 403. The industrial camera 3 is fixedly connected to the output end of the second drive member 403. During the welding process, the angle of the industrial camera 3 is adjusted by the first drive member 401 and the second drive member 403 so that the arc zone is located at the center of the real-time image of the molten pool. The output current and speed during welding are controlled by the control platform 5 to ensure the welding effect.

[0091] To better understand the working process of an arc-shaped steel structure welding system and apparatus according to an embodiment of this application, refer to... Figures 1-7 The following is a specific embodiment:

[0092] The tack welded arc-shaped steel structure is placed on the positioner 2. The material characteristics and welding position of the steel structure are pre-input into the control platform 5. The steel structure is clamped and the angle is adjusted by the preset feature positioner 2. The movement path of the robotic arm 102 driving the welding head 103, as well as the current and speed during welding, are pre-planned. During the welding process, the industrial camera 3 acquires real-time images of the molten pool. The image processing module then divides the real-time images of the molten pool into the arc area and the molten pool area. The area filtering unit further determines the arc area and outputs the arc area image. At this time, according to the center position of the arc area image, the first drive unit 401 and the second drive unit 403 are activated to adjust the position of the industrial camera 3 so that the lens of the industrial camera 3 is directly facing the center position of the arc area. This ensures that the center position of the acquired image is the center position of the arc area, thus ensuring the accuracy of the acquired image and laying a good foundation for subsequent feature analysis.

[0093] In this process, features are extracted from the arc zone and the molten pool zone to obtain the arc edge, width, and height features of the arc zone and the width features of the molten pool zone. Based on the obtained features, the molten pool prediction module analyzes and calculates to obtain the current state of the molten pool and outputs it. The molten pool state is output to the control platform 5. Based on the input molten pool state, the control platform 5 adjusts the welding current and welding speed. In the penetration state, the current input current is maintained to obtain the welding speed. In the over-penetration state, the input current is reduced and the welding speed is increased. In the incomplete penetration state, the input current is increased and the welding speed is reduced to keep the molten pool in a stable state and ensure the welding effect.

[0094] After adjusting the input current and welding speed, and analyzing the real-time image of the molten pool captured by the industrial camera 3, the control platform 5 determines whether the current molten pool state is over-penetrated or under-penetrated. The control platform then uses the robotic arm 102 to adjust the angle of the welding head 103. In the under-penetrated state, the angle between the welding head 103 and the surface of the curved steel structure is increased, reducing the contact area between the arc and the curved steel structure, thus concentrating the heat input and increasing the fusion depth. In the over-penetrated state, the angle between the welding head 103 and the surface of the curved steel structure is decreased, increasing the contact area between the arc and the curved steel structure, thus distributing the heat input more evenly and reducing the risk of localized overheating. This ensures the stability of the molten pool state and improves the welding quality.

[0095] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0097] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A welding system for arc-shaped steel structures, characterized in that, The welding system includes: an image processing module, a data processing module, a molten pool prediction module, and a control module. The control module is electrically connected to the image acquisition module, the data processing module, and the molten pool prediction module. The image processing module acquires real-time images of the molten pool during the welding process of the arc-shaped steel structure, distinguishes feature regions in the real-time image of the molten pool by the brightness value, the feature regions include the arc area and the molten pool area, and determines the position of the arc area so that the arc area is located at the center of the real-time image of the molten pool. The data processing module includes an arc feature extraction unit and a molten pool feature extraction unit. The arc feature extraction unit extracts a first key feature of the arc region, which includes the arc edge, width, and height features of the arc region. The molten pool feature extraction unit extracts a second key feature of the molten pool region, which includes the molten pool width feature of the molten pool region. The molten pool prediction module predicts the molten pool state based on the first key feature and the second key feature, and classifies the molten pool state into: fully molten, not fully molten, and over-molten, and outputs the result. The control module responds to the output result and outputs adjustment variables to incrementally adjust the current and speed during welding. It also includes a welding posture adjustment module. The data processing module acquires the molten pool state after adjusting the current and speed and outputs a corrected molten pool state. The corrected molten pool state is then classified by the molten pool prediction module. The activation of the welding posture adjustment module includes the following steps: Real-time acquisition and correction of the molten pool status; When correcting the molten pool state to a fully penetrated state, maintain the current welding posture; When the molten pool is in a state of incomplete or excessive penetration, adjust the welding posture; The image processing module includes a grayscale conversion unit, a binarization processing unit, a region filtering unit, and a selection unit. The grayscale conversion unit converts the real-time image of the color molten pool into a grayscale image. The binarization processing unit sets a brightness threshold to determine the arc area and processes the grayscale image to generate a binary image. The region filtering unit detects white blocks in the binary image and designs an area threshold. The area of ​​the white blocks is compared with the area threshold to further determine the arc area and output the arc area image. After the region filtering unit determines the arc area, the selection unit constructs a centroid coordinate calculation function with the center of the real-time molten pool image as the target and outputs the arc area center position parameter after logical operation. The center position of the real-time molten pool image is determined by the arc area center position parameter. The welding posture adjustment module includes a receiving unit and an adjustment unit. The receiving unit receives a start signal from the control module and outputs it to the adjustment unit. The start signal includes an incomplete penetration signal and an over-penetration signal. The adjustment unit controls the robotic arm to adjust the angle between the welding head and the arc-shaped steel structure.

2. The arc-shaped steel structure welding system according to claim 1, characterized in that, The arc feature extraction unit converts the arc area image to grayscale and performs noise reduction using Gaussian filtering. Then, it uses convolution operations to perform binarization to detect the arc edge and outputs the arc contour map, thereby obtaining the arc edge, width, and height features of the arc area.

3. The arc-shaped steel structure welding system according to claim 2, characterized in that, The molten pool feature extraction unit includes a filtering subunit, an enhancement subunit, and a width calculation subunit. The filtering subunit uses Gaussian filtering to reduce noise in the grayscale image. The enhancement subunit changes the brightness distribution of the denoised grayscale image to mark the molten pool boundary and outputs a molten pool boundary map. The width calculation subunit calculates the molten pool width through the molten pool boundary.

4. The arc-shaped steel structure welding system according to claim 3, characterized in that, The molten pool prediction module includes a preprocessing unit and a model unit. The preprocessing unit tensors the arc contour map, molten pool boundary map, and their combined map to obtain image data tensors, and fuses the weld feature values ​​tensors obtained from image algorithm processing. The model unit processes the fused image data and weld feature value tensors, and includes an input layer, a convolutional layer, a sampling layer, and an output layer. The input layer receives welding current and welding speed process parameters and tensors. The convolutional layer has two sets that overlap with the sampling layer, extracts the process parameters and tensors from the input layer, and outputs the calculation results after passing through an activation function. The output layer classifies the molten pool state based on the calculation results.

5. The arc-shaped steel structure welding system according to claim 4, characterized in that, The weld feature values ​​are arc width, arc area, molten pool width, molten pool length, input current, and welding speed. Image data and weld feature value tensors are fused using a weighted fusion method. The convolutional layer and sampling layer extract tensor features of image data and weld feature values ​​and combine them with process parameters to form a convolutional layer feature map. Representative features are sampled from the convolutional layer feature map to obtain the combination relationship between representative features. The output layer includes three neurons, corresponding to the penetration state, incomplete penetration state, and over-penetration state, respectively.

6. A welding device for arc-shaped steel structures, comprising the arc-shaped steel structure welding system according to any one of claims 1-5, characterized in that, The device includes a welding mechanism for welding arc-shaped steel structures, a positioner for fixing the arc-shaped steel structures, and a control platform. An industrial camera for acquiring images is installed on the welding mechanism.

7. The arc-shaped steel structure welding device according to claim 6, characterized in that, The welding mechanism includes a robotic arm and a support platform for supporting and mounting the robotic arm. A welding head is installed at the head of the robotic arm. An industrial camera is mounted on the robotic arm via an adjustment mechanism. The adjustment mechanism includes a first drive component and a second drive component for driving the industrial camera to adjust its angle. The output end of the first drive component is fixedly connected to a mounting bracket for mounting the second drive component. The industrial camera is fixedly connected to the output end of the second drive component.

Images