GIS shell ring seam welding deviation real-time correction method

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a GIS shell ring seam welding deviation real-time correction method, which comprises the following steps: obtaining a target image and a target image sequence; positioning a laser region according to the pixel point gray value interval and the pixel point brightness concentration condition in the target image, and obtaining a laser region sequence; taking the exponential power of the gray value of the pixel point in the laser region as a weight to form a laser centroid sequence; performing smoothing processing on the laser centroid sequence to obtain a smooth trajectory point and generate a laser actual motion trajectory; performing smoothing on the laser actual motion trajectory to generate a reference welding trajectory; calculating a real-time correction vector according to the reference welding trajectory and the laser actual motion trajectory; calculating a new target position according to the real-time correction vector and the welding reference position in the reference welding trajectory, and adjusting the welding head position to realize real-time correction of the welding deviation. The present application effectively solves the problem that the GIS shell ring seam welding process is disturbed by the laser, resulting in inaccurate deviation correction.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method for real-time correction of welding deviation in the circumferential seam of a GIS shell. Background Technology

[0002] As a key structure housing high-voltage power components, the GIS shell is mostly cylindrical or tubular, requiring extremely high sealing performance, strength, and geometric precision. When welding the circumferential seam formed by joining two sections of the shell, positional deviations, geometrical deviations, and weld formation deviations can easily occur due to factors such as temperature changes and the weight of the workpiece. These deviations directly affect equipment quality and operational safety. To accurately control the entire GIS shell circumferential seam welding process and ensure that the sealing performance, strength, and geometric precision of the welded shell fully meet design standards, a real-time correction method is required. This involves real-time monitoring, analysis, and dynamic adjustments to promptly eliminate various deviations that occur during the welding process.

[0003] Traditional welding correction systems typically generate fixed reference trajectories in advance through offline programming. During the welding process, the welding head moves along these trajectories, attempting to ensure welding accuracy through a pre-set path. Some technologies use industrial cameras to acquire weld seam images, identify the weld seam position using simple algorithms such as fixed threshold segmentation and traditional centroid methods, and then make adjustments based on the detected deviations.

[0004] However, the fixed reference trajectory generated by offline programming is difficult to cope with dynamic interferences such as temperature changes, thermal deformation caused by workpiece weight, and mechanical vibration of the GIS shell during welding. It is easy for the overall trajectory to deviate. In addition, there are interferences such as molten pool reflection, spatter, and strong light during welding. Traditional technology is prone to misjudging the interference area as the target area, resulting in inaccurate deviation correction. Summary of the Invention

[0005] To address the technical problem of inaccurate deviation correction caused by laser interference during the welding process, this invention provides a real-time deviation correction method for GIS shell circumferential seam welding, comprising:

[0006] Within a preset time period, continuous frame images of the GIS shell circumferential weld process are acquired at the same frequency and converted to grayscale. These images are then arranged chronologically to generate a target image sequence containing several target images. Based on the grayscale value ranges of pixels in the target images, the images are binarized and filtered. Laser candidate regions are located using the filtering results. Laser scores are calculated based on the brightness concentration of the candidate regions. All laser scores are calculated, and the region with the highest score is taken as the laser region. All laser regions in the target image sequence are obtained, resulting in a laser region sequence. Laser centroid coordinates are calculated using the exponent of the grayscale value of pixels within the laser region as weights, forming a laser centroid sequence. Based on the geometric characteristics of the triangle formed by three adjacent centroids in the laser centroid sequence, the laser centroid sequence is smoothed to obtain smooth trajectory points, generating the actual laser motion trajectory. The actual laser motion trajectory is then subjected to time delay and spatial smoothing to generate a reference welding trajectory. A real-time correction vector is calculated based on the dynamic correlation between the reference welding trajectory and the actual laser motion trajectory. Based on the real-time correction vector and the welding reference position in the reference welding trajectory, a new target position is calculated, and the welding head position is adjusted to achieve real-time correction of welding deviations.

[0007] This invention addresses various problems in welding by capturing continuous images of the welding process in real time. Through grayscale and binarization filtering, it accurately identifies the laser's location, effectively eliminating interference from factors such as molten pool reflection and metal spatter. The true laser region is selected by calculating the laser fraction, and the coordinates of the particles are determined by weighting the exponential power of the grayscale value. Trajectory smoothing is performed using the triangular features formed by adjacent particles, accurately reflecting the laser's movement. Based on this, a reasonable reference trajectory is generated through time delay and spatial smoothing. A real-time correction vector is calculated based on the dynamic correlation between the reference trajectory and the actual trajectory, adjusting the welding head position promptly to ensure it remains in the correct position. This avoids deviations caused by dynamic interference such as temperature changes and workpiece weight, solving problems like easy deviation and inaccurate correction in traditional methods. This ensures welding accuracy and stability, guaranteeing that the welded product meets design requirements.

[0008] Preferably, locating the laser candidate region includes:

[0009] Extract any target image from the target image sequence, set a threshold of 'a', perform a high threshold filtering, and record the obtained image as a binary image; perform region growing on the white pixels of any binary image, and record the obtained region as the laser candidate region.

[0010] Preferably, the laser fraction satisfies the following expression:

[0011] ;

[0012] In the formula, Indicates the laser score of the laser candidate region; Indicates the pixel coordinates of the laser candidate region; Represents candidate region pixels The grayscale value in the target image; Used to amplify the differences between pixels with different grayscale values; This represents the number of pixels in the candidate region set. This represents the normalization function.

[0013] This scoring method of the present invention, by calculating and normalizing the brightness-related values ​​within the region, can highlight areas with concentrated brightness and high energy, effectively distinguish between real laser areas and interference areas, making the selected laser areas more consistent with the actual situation and improving the accuracy of identification.

[0014] Preferably, obtaining the laser region sequence includes:

[0015] Traverse all candidate laser regions, calculate the laser score for each one, and select the region with the highest score as the real laser region, denoted as the laser region; obtain the laser regions corresponding to all target images in the target image sequence, and arrange the laser regions according to the order of the target images corresponding to the laser regions in the target image sequence to obtain the laser region sequence.

[0016] Preferably, the coordinates of the laser particle satisfy the following expression:

[0017] ;

[0018] In the formula, Indicates the coordinates of the laser particle; Represents the x-coordinate of the laser particle; Represents the ordinate of the laser particle; This represents the coordinate vector of a pixel within the laser area. Indicates the laser area; The coordinates within the laser area are: The grayscale value of the pixel; n represents a constant greater than 1, intended to be used to... replace .

[0019] The method of calculating particle coordinates in this invention assigns greater weight to pixels with high brightness, bringing the particle coordinates closer to the laser energy center, reducing the influence of edges and noise, and achieving more accurate positioning. It can capture minute positional changes and provide more accurate basic data for subsequent trajectory smoothing and correction.

[0020] Preferably, obtaining smooth trajectory points includes:

[0021] Obtain the centroids of all laser regions in the laser region sequence. Take the centroid of the triangle formed by any centroid and its left and right adjacent centroids as the smoothed trajectory point on the laser centroid sequence and record it as the smoothed trajectory point.

[0022] Preferably, generating the actual laser trajectory includes:

[0023] A series of smooth trajectory points are generated for the centroids in the laser centroid sequence. The first and last two centroids in the laser centroid sequence are not included in the calculation but are retained to generate the actual laser motion trajectory.

[0024] Preferably, the reference welding trajectory satisfies the following expression:

[0025] ;

[0026] In the formula, Indicates the first welding trajectory on the reference welding path The coordinates of the trajectory points at time; i represents a variable that defines the range of trajectory points participating in smoothing, ranging from -N to N; D represents a fixed integer representing the time delay; This indicates that the laser's actual trajectory was extracted using... Centered on the axis, a continuous segment of N trajectory points to the left and right; N represents the radius of the trajectory points involved in the smoothing calculation. This indicates the number of points in a continuous segment of trajectory.

[0027] The method of generating reference trajectories in this invention combines time delay and spatial smoothing, taking into account both the trend of historical trajectories and adapting to the slow changes in the workpiece, making the reference trajectory more reasonable and avoiding the problem that fixed references cannot cope with dynamic changes, thus providing a more suitable reference for calibration.

[0028] Preferably, the real-time correction vector satisfies the following expression:

[0029] ;

[0030] In the formula, This represents the real-time correction vector at time t; This indicates the actual welding position at time t; This indicates the actual welding position at time t-1; This indicates the welding reference position at time t.

[0031] The calculation method of the real-time correction vector of the present invention integrates the current reference position and the actual position at the previous moment, making the correction direction more in line with the motion law, avoiding instability caused by excessive correction amplitude, making the welding head move more smoothly, and reducing the adverse effects caused by sudden adjustment.

[0032] Preferably, calculating the new target position and adjusting the weld joint position includes:

[0033] The first The welding reference position at time t is added to the real-time correction vector at time t to obtain the... The corrected welding position at time 1 is recorded as the new target position; the welding position after correction at time 2 is recorded as the new target position. The coordinates of the welding position after the moment correction are immediately sent to the underlying motion controller of the welding robot. The controller sends instructions to drive the robot to smoothly move the welding head to the new target position.

[0034] The beneficial effects of this invention are as follows: Through a series of coherent steps, it achieves precise control over the welding process. It can respond to various changes during welding in real time, automatically adjust the position of the welding head, reduce human intervention, and improve the level of automation in welding. Simultaneously, by optimizing the identification and correction process, it enhances the stability of welding quality, reduces product problems caused by deviations, and saves the cost and time of subsequent repairs. It makes the complex welding process more controllable, providing strong support for efficient and high-quality production and promoting technological progress in related production fields. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a real-time correction method for circumferential weld deviation of a GIS shell according to the present invention;

[0036] Figure 2 This is a schematic diagram illustrating the target image. Detailed Implementation

[0037] This invention discloses a method for real-time correction of circumferential weld deviation in GIS shells, referring to... Figure 1 This includes steps S1-S4:

[0038] S1: Obtain the target image sequence within a preset time period.

[0039] It should be noted that the GIS shell, as a key structure of high-voltage power equipment, has stringent requirements for sealing, strength, and geometric accuracy after welding. During circumferential welding, the deformation at the joint caused by thermal expansion and contraction of the shell due to temperature changes will continuously generate dynamic deviations. This invention tracks the dynamic position of the laser by acquiring images in real time, ensuring the immediacy of deviation monitoring from the source and avoiding the correction failure caused by the lag of traditional offline references. Since interference such as molten pool reflection and metal spatter can obscure the true position of the laser during welding, this invention accurately identifies the laser's action area and filters out instantaneous noise by combining the continuous characteristics of laser movement, stably extracting the true laser trajectory under complex working conditions. During GIS shell welding, the shell deforms slowly due to heat, and the welding head cannot respond to commands instantaneously due to its mechanical characteristics. The dynamic reference trajectory of this invention, based on historical trajectories and incorporating delay and smoothing processing, includes both the actual welding movement trend and adapts to shell deformation. The final smoothing adjustment can avoid equipment vibration or weld defects caused by forced jumps, balancing correction accuracy and process stability.

[0040] It should be noted that mounting the industrial camera on the side of the laser welding head and moving it synchronously with the welding head can capture detailed information about the circumferential weld at close range, while avoiding direct damage to the camera lens from spatter and strong light during the welding process. At the same time, it can ensure that the shooting angle and the welding area maintain a stable relative position, reducing image deviation caused by positional changes.

[0041] Specifically, an industrial camera is mounted off-axis on the laser welding head to continuously acquire multiple images of the welding process. These images are then converted to grayscale to obtain target images that can be used for subsequent deviation analysis, such as... Figure 2 : Schematic diagram of the target image. The target images are sorted according to the acquisition time to form a target image sequence.

[0042] At this point, the target image sequence has been obtained.

[0043] S2: Based on the differences in grayscale values ​​of pixels in the target image, the target image is binarized and filtered, and the laser candidate region is located based on the filtering results; the laser score is calculated based on the brightness concentration of the laser candidate region; all laser scores are calculated and the region with the highest score is taken as the laser region, and all laser regions in the target image sequence are obtained to obtain the laser region sequence.

[0044] It should be noted that the laser, as the core energy output source in the welding process, directly affects the weld formation. Only by determining its position in each frame of the image can the laser's motion trajectory be analyzed, thereby determining whether welding deviations have occurred and making corrections. Furthermore, during welding, factors such as shell thermal deformation and mechanical vibration can cause welding deviations. Changes in the laser's position directly reflect the dynamic offset of the welding execution end. Determining the laser area fundamentally ensures the accuracy and real-time nature of welding deviation correction, solving the problem of unstable weld quality caused by the lack of precise process monitoring in traditional welding.

[0045] It should be noted that when capturing the target image, there may be complex backgrounds with various high-brightness interferences such as molten pool reflections and spatter. Traditional methods for locating the largest connected component based on a fixed threshold are prone to misjudgment when faced with changes in welding conditions, misidentifying reflective areas as laser areas. This step uses a simple image segmentation algorithm, employing a two-layer filtering approach. First, it determines the candidate laser region based on a threshold, and then performs precise centering of the selected region to obtain the laser area.

[0046] Specifically, extract any target image from the target image sequence, set a threshold of 'a', and perform a high-threshold filtering. The resulting image is recorded as a binary image. Obtain the binary images of all target images and record them as a binary image sequence. Perform region growing on the white pixels of any binary image, and record the resulting region as the laser candidate region. It should be noted that high-threshold filtering of the target image can quickly narrow down the search range, filtering out most of the background and non-critical low-brightness areas. Regarding the value range of the threshold 'a' for high-threshold filtering, a relatively high fixed grayscale threshold can be set here. For example, threshold 'a' can be dynamically adjusted according to the actual welding scenario. An initial value of 200 is recommended, and it can be appropriately increased if there is too much interference. Set any pixel with a grayscale value lower than 'a' to black, and any pixel with a grayscale value higher than or equal to 'a' to white. This operation allows for rapid location of the laser candidate region.

[0047] It should be noted that, as the energy source for welding, the laser spot has the characteristics of concentrated brightness and relatively stable area. In contrast, the bright areas of spatter or reflection are either diffuse in brightness or have a low degree of matching between area and energy. This step, by calculating the laser fraction, can amplify this difference and ensure accurate identification of the real laser area in complex welding scenarios.

[0048] Preferably, all candidate laser regions on the target image are extracted, and the pixels in each candidate laser region are recorded as a set of candidate pixels. The total number of pixels is calculated and recorded as the set pixel count. The gray values ​​of all pixels are obtained and recorded as the pixel gray value set. Based on the set pixel count and pixel gray value set corresponding to each set, the score of the region where this set belongs to the laser region is calculated and recorded as the laser score, including:

[0049] The laser fraction satisfies the following expression:

[0050] ;

[0051] In the formula, Indicates the laser score of the laser candidate region; Indicates the pixel coordinates of the laser candidate region; Represents candidate region pixels The grayscale value in the target image; Used to amplify the differences between pixels with different grayscale values; This represents the number of pixels in the candidate region set. This represents the normalization function.

[0052] In the formula, This means squaring the brightness values ​​of all pixels within the candidate region and then summing them up to obtain a total brightness energy value. This means that when the cumulative luminance energy of a candidate region is high and its area is small, its laser score is high.

[0053] It should be noted that the welding process is continuous, and the laser's position on the circumferential seam of the shell changes constantly as the welding progresses. Selecting the region with the highest laser score from the candidate regions ensures that each target image has a unique and accurate laser position record. Arranging these laser regions according to the order of the target images in the sequence creates a laser region sequence that fully reflects the dynamic movement of the laser during the welding process.

[0054] Preferably, all candidate laser regions are traversed, and laser scores are calculated one by one. The region with the highest score is selected as the real laser region and denoted as the laser region. The laser regions corresponding to all target images in the target image sequence are obtained, and the laser regions are arranged according to the order of the target images corresponding to the laser regions in the target image sequence to form a laser region sequence.

[0055] At this point, the laser region sequence was obtained.

[0056] S3: Calculate the coordinates of the laser particles using the exponent of the gray values ​​of the pixels within the laser region as weights, and form a laser centroid sequence; based on the geometric features of the triangle formed by three adjacent centroids in the laser centroid sequence, smooth the laser centroid sequence to obtain smooth trajectory points and generate the actual laser motion trajectory.

[0057] It should be noted that although the traditional centroid method is simple, its core flaw is that its response to brightness is linear and it cannot effectively distinguish between core brightness and edge brightness. Therefore, it is easily affected by irregular spot shapes or edge noise. To avoid this problem, this step exponentially weights the brightness values, disproportionately amplifying the weight of the core high-brightness pixels in the centroid calculation. This automatically suppresses the interference of edges and noise in a single calculation, allowing the centroid to naturally converge to the true energy center.

[0058] Specifically, the coordinate vectors of all pixels in any laser region within the laser region sequence are obtained and denoted as laser pixel coordinates; the grayscale values ​​of all pixels in any laser region are obtained and denoted as laser pixel grayscale values; based on the laser pixel coordinates and laser pixel grayscale values, the coordinates of the mass points in the laser region are calculated and denoted as laser mass point coordinates, including:

[0059] The coordinates of the laser particle satisfy the following expression:

[0060] ;

[0061] In the formula, Indicates the coordinates of the laser particle; Represents the x-coordinate of the laser particle; Represents the ordinate of the laser particle; This represents the coordinate vector of a pixel within the laser area. Indicates the laser area; The coordinates within the laser area are: The grayscale value of the pixel; n represents a constant greater than 1, intended to be used to... replace .

[0062] In the formula, or This means that for each pixel within the laser area, the nth power of the gray value is used as the weight, multiplied by the corresponding horizontal and vertical coordinates, and then summed to obtain the corresponding coordinate value. This represents the sum of weights, where the nth power of the gray value of each pixel within the laser region is used as the weight, and is used to normalize the coordinate vector of the molecule. or This means that after dividing the numerator by the denominator, the result is... It is a two-dimensional coordinate system. It should be noted that when n=1, this formula is the traditional method for calculating the grayscale centroid. When n>1, this formula greatly and non-linearly amplifies the influence of the brightest pixel group. The value range of n is [2,4]. If n is less than 2, its focusing enhancement effect on the core region is not significant; if n is greater than 4, it may cause the calculation to be overly sensitive to a single brightest pixel. In this invention, setting n=3 achieves a better balance between enhancing the core weight and maintaining the stability of the algorithm.

[0063] It should be noted that the coordinates of the center point calculated by the formula are... For example, a two-dimensional coordinate vector containing floating-point numbers. =123.45, =245.67. This invention retains all decimal places of this coordinate without any rounding or approximation, achieving sub-pixel level positioning accuracy. In all subsequent steps, including but not limited to the construction of the motion trajectory, the calculation of the deviation vector, and the generation of the correction signal, operations are performed in the floating-point domain, thereby ensuring high-sensitivity capture of minute displacements of the laser spot.

[0064] It should be noted that directly generating the trajectory using the original centroid coordinates will result in noise points in the centroid coordinates due to interference during the welding process, such as spatter and arc flash. This noise will cause burrs and jumps in the trajectory, failing to accurately reflect the stable motion trend of the laser. If left untreated, this noise will be misinterpreted as actual offset when judging welding offset based on the trajectory, affecting the correction accuracy. This step processes the centroid sequence to generate the actual laser motion trajectory.

[0065] Preferably, the centroids of all laser regions in the laser region sequence are obtained, and the centroid of the triangle formed by any centroid and the left and right adjacent centroids is taken as the smoothed trajectory point on the laser centroid sequence and recorded as the smoothed trajectory point.

[0066] It should be noted that the centroid of the triangle is geometrically the most stable and central equilibrium point among the three points. Using this centroid as a point on the final smooth trajectory can eliminate random noise at individual points.

[0067] Preferably, a series of new, smooth trajectory points are generated for each point in the laser centroid sequence. The first and last two centroids in the laser centroid sequence are not included in the calculation and are retained to obtain the actual laser motion trajectory.

[0068] At this point, the actual trajectory of the laser was obtained.

[0069] S4: Perform time delay and spatial smoothing on the actual laser motion trajectory to generate a reference welding trajectory; calculate the real-time correction vector based on the dynamic correlation between the reference welding trajectory and the actual laser motion trajectory; calculate the new target position based on the real-time correction vector and the welding reference position in the reference welding trajectory, and adjust the welding head position to achieve real-time correction of welding deviation.

[0070] It should be noted that traditional welding correction systems heavily rely on a pre-generated, fixed reference trajectory offline. This method has two major drawbacks: it requires time-consuming and laborious offline programming, and it cannot adapt to overall workpiece offsets or systemic deformations, potentially causing a complete shift in the welding path. To address this issue, this step, while acknowledging that the actual welding trajectory is generally correct on a macroscopic scale, identifies microscopic features such as jitter and deviations that require correction. It uses a past, deeply smoothed actual trajectory as the current evaluation benchmark.

[0071] Specifically, the actual laser trajectory is acquired, and then subjected to a time delay and spatial smoothing to obtain a reference welding trajectory, including:

[0072] The reference welding trajectory satisfies the following expression:

[0073] ;

[0074] In the formula, Indicates the first welding trajectory on the reference welding path The coordinates of the trajectory points at time; i represents a variable that defines the range of trajectory points participating in smoothing, ranging from -N to N; D represents a fixed integer representing the time delay; This indicates that the laser's actual trajectory was extracted using... Centered on the axis, a continuous segment of N trajectory points to the left and right; N represents the radius of the trajectory points involved in the smoothing calculation. This represents the number of points in a continuous trajectory segment. To ensure the trajectory capture window is effective, D > N must be satisfied to avoid the smoothing range exceeding the actual trajectory sequence due to insufficient time delay.

[0075] In the formula, This indicates that the laser's actual trajectory was extracted using... Centered on a point, take N consecutive trajectory points on either side of this point as a reference point. Calculate the arithmetic mean of these consecutive trajectory points. This average value is the coordinate of the reference welding trajectory at time t. .

[0076] It should be noted that in welding scenarios, due to the inertia of the welding head movement, workpiece thermal deformation, and arc spatter, the actual welding position is difficult to perfectly coincide with the reference trajectory. This step utilizes the actual position from the previous moment, combined with the current reference position, and uses the midpoint between the two as the dynamic target position. This ensures that the correction action conforms to the physical laws of the continuous movement of the welding head, while gradually closing the distance to the reference trajectory, avoiding equipment vibration and weld defects caused by forced jumps.

[0077] Preferably, the real-time correction vector is calculated based on the reference welding trajectory and the actual laser motion trajectory, including:

[0078] The real-time correction vector satisfies the following expression:

[0079] ;

[0080] In the formula, This represents the real-time correction vector at time t; This indicates the actual welding position at time t; This indicates the actual welding position at time t-1; This indicates the welding reference position at time t.

[0081] In the formula, Indicates the use of the first The midpoint between the actual welding position and the reference position is used to represent the dynamic target position. This means that by fusing the actual position and the reference position, a dynamic target position is constructed, the deviation between the current actual position and the dynamic target is calculated, and the output correction vector allows the welding head to smoothly and continuously correct its position.

[0082] Preferably, the first The welding reference position at time t is added to the real-time correction vector at time t to obtain the... The corrected welding position at time 1 is recorded as the new target position; the welding position after correction at time 2 is recorded as the new target position. After the correction at time t, the coordinates of the welding position are immediately sent to the underlying motion controller of the welding robot. The controller then sends instructions to drive the robot to smoothly move the welding head to the new target position. It should be noted that due to the characteristics of workpiece deformation and slow change of laser trajectory during the welding process, the difference between the real-time correction vector at time t and the actual deviation at time t+1 can be controlled within a very small range within an extremely short time interval. Therefore, the real-time correction vector at time t can be adapted to the reference position at time t+1.

[0083] This completes the real-time correction of the circumferential weld deviation of the GIS shell.

[0084] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for real-time correction of circumferential weld deviation in GIS shells, characterized in that, include: Within a preset time period, continuous frame images of the GIS shell circumferential weld process are acquired at the same frequency and converted to grayscale. These images are then arranged in chronological order to generate a target image sequence containing several target images. Based on the grayscale value range of pixels in the target images, the target images are binarized and filtered. The laser candidate region is then located based on the filtering results. The laser score is calculated based on the brightness concentration of the laser candidate region, satisfying the expression: ； In the formula, Indicates the laser score of the laser candidate region; Indicates the pixel coordinates of the laser candidate region; Represents candidate region pixels The grayscale value in the target image; Used to amplify the differences between pixels with different grayscale values; This represents the number of pixels in the candidate region set. Represents the normalization function; Calculate all laser scores and take the region with the highest score as the laser region to obtain all laser regions in the target image sequence, thus obtaining the laser region sequence; calculate the coordinates of the laser particles using the exponent of the gray value of the pixels in the laser region as the weight, and form the laser centroid sequence; based on the geometric features of the triangle formed by three adjacent centroids in the laser centroid sequence, smooth the laser centroid sequence to obtain smooth trajectory points and generate the actual laser motion trajectory. The actual laser trajectory is time-delayed and spatially smoothed to generate a reference welding trajectory that satisfies the expression: ； In the formula, Indicates the first welding trajectory on the reference welding path The coordinates of the trajectory points at time; i represents a variable that defines the range of trajectory points participating in smoothing, ranging from -N to N; D represents a fixed integer representing the time delay; This indicates that the laser's actual trajectory was extracted using... Centered on the axis, a continuous segment of N trajectory points to the left and right; N represents the radius of the trajectory points involved in the smoothing calculation. This indicates the number of points in a continuous segment of trajectory; Based on the dynamic correlation between the reference welding trajectory and the actual laser motion trajectory, the real-time correction vector is calculated, satisfying the expression: ； In the formula, This represents the real-time correction vector at time t; This indicates the actual welding position at time t; This indicates the actual welding position at time t-1; Based on the real-time correction vector and the welding reference position in the reference welding trajectory, the new target position is calculated, and the welding head position is adjusted to achieve real-time correction of welding deviation.

2. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The positioning laser candidate region includes: Extract any target image from the target image sequence, set a threshold of 'a', perform a high threshold filtering, and record the obtained image as a binary image; perform region growing on the white pixels of any binary image, and record the obtained region as the laser candidate region.

3. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The process of obtaining the laser region sequence includes: Traverse all candidate laser regions, calculate the laser score for each one, and select the region with the highest score as the real laser region, denoted as the laser region; obtain the laser regions corresponding to all target images in the target image sequence, and arrange the laser regions according to the order of the target images corresponding to the laser regions in the target image sequence to obtain the laser region sequence.

4. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The coordinates of the laser particle satisfy the following expression: ； In the formula, Indicates the coordinates of the laser particle; Represents the x-coordinate of the laser particle; Represents the ordinate of the laser particle; This represents the coordinate vector of a pixel within the laser area. Indicates the laser area; The coordinates within the laser area are: The grayscale value of the pixel; n represents a constant greater than 1, intended to be used to... replace .

5. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The process of obtaining smooth trajectory points includes: Obtain the centroids of all laser regions in the laser region sequence. Take the centroid of the triangle formed by any centroid and its left and right adjacent centroids as the smoothed trajectory point on the laser centroid sequence and record it as the smoothed trajectory point.

6. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The generation of the actual laser trajectory includes: A series of smooth trajectory points are generated for the centroids in the laser centroid sequence. The first and last two centroids in the laser centroid sequence are not included in the calculation but are retained to generate the actual laser motion trajectory.

7. The method for real-time correction of circumferential weld deviation of GIS shell according to claim 1, characterized in that, The calculation of the new target position and adjustment of the weld joint position includes: The first The welding reference position at time t is added to the real-time correction vector at time t to obtain the... The corrected welding position at time 1 is recorded as the new target position; the welding position after correction at time 2 is recorded as the new target position. The coordinates of the welding position after the moment correction are immediately sent to the underlying motion controller of the welding robot. The controller sends instructions to drive the robot to smoothly move the welding head to the new target position.

Images