An analytic hierarchy process-based self-decision method for welding positions of multi-pass welding of thick plate T-joints

Abstract

The application provides a thick plate T joint multi-pass welding position autonomous decision-making method based on an analytic hierarchy process, which can effectively improve the thick plate welding efficiency. First, a groove contour recognition method is designed based on an improved Gabor filter and a Bayesian classifier; second, a groove contour feature point extraction method is realized based on the slope mutation characteristics; then, a three-layer analytic hierarchy model is established by using the extracted feature points and welding experience, and an automatic acquisition algorithm for the comparison matrix elements is designed; finally, the effective welding position is decided from the extracted feature points based on the maximum posterior weight criterion. The application realizes the autonomous decision-making process of the welding position by taking the groove contour feature points as the candidate welding initial positions and using the machine vision and welding experience, which helps to control the web corner deformation while improving the welding efficiency, and has the advantages of good real-time performance, high precision and strong robustness.

Description

Technical Field

[0001] This invention relates to an autonomous decision-making method for welding position of multi-pass welding of thick plate T-joints based on visual sensing and hierarchical analysis model, belonging to the field of multi-layer multi-pass arc intelligent welding technology. Background Technology

[0002] A major reason for the low efficiency of thick plate arc welding is that the welding position cannot be autonomously determined in multi-pass welding. After one weld is completed, it is often necessary to specify an initial welding position for the next weld based on welding experience and welding conditions, and then move the welding torch or notify the control system to move the welding torch to complete the next weld. Therefore, by utilizing appropriate sensor information and establishing a decision model that incorporates welding experience to implement autonomous welding position decision-making, it is possible to improve welding efficiency and automation levels.

[0003] Currently, numerous studies have focused on weld planning for multi-layer, multi-pass welding. These planning processes often employ auxiliary software or empirical formulas to establish models between weld geometry and welding process parameters. Then, based on the geometry of the filler groove, the required number of weld passes and layers is calculated, resulting in a rough weld plan. A few studies have explored minor adjustments to the weld plan based on welding conditions. However, regardless of the offline weld planning method used, a common challenge during welding is determining a suitable starting and tracking position for the next weld pass. Current solutions involve numbering extracted feature points and then assigning these numbers to determine the welding position, failing to achieve an autonomous decision-making process. This invention first obtains candidate welding positions using visual sensing and machine learning algorithms, then integrates welding experience into a decision model, achieving autonomous welding position decision-making based on a hierarchical analysis process. This method exhibits strong anti-interference capabilities, and the decision results help control web angle deformation, improving both welding efficiency and quality. Summary of the Invention

[0004] This invention addresses the need for autonomous welding position decision-making during multi-layer, multi-pass welding of thick plate T-joints. It provides a welding position autonomous decision-making method based on visual sensing and a hierarchical analysis model. The aim is to significantly improve welding efficiency and quality by making real-time decisions on appropriate initiation and tracking positions during welding, thus providing a technical pathway for the intelligent transformation of welding manufacturing. To achieve the above objectives, the technical solution adopted by this invention is as follows:

[0005] An autonomous decision-making method for multi-pass welding position of thick plate T-joints based on the analytic hierarchy process includes the following steps:

[0006] Step 1: The original weld image is filtered using an improved Gabor filter to obtain the orientation feature map. Then, the orientation feature map is binarized using a local threshold segmentation method. After that, the data is clustered by nearest neighbor and the clustered data clusters are classified using a Naive Bayes classifier to initially remove interference data. Finally, the visual feature competition method is used to further remove interference and obtain the bevel profile.

[0007] Step 2: Calculate the average height of the obtained bevel profile column by column in the vertical direction to obtain a linear profile. Then, calculate the slope at each position of the linear profile and smooth the slope to obtain monotonic intervals. Then, calculate the slope change at each point in the monotonic interval and sort the changes in descending order. Finally, determine the changes that are first in the sequence by specifying the number of bevel profile feature points, thereby determining the first monotonic interval. Determine the position of the weld profile feature points based on the center position of the monotonic interval.

[0008] Step 3: Establish a three-layer hierarchical analysis model based on the obtained bevel contour feature points, then calculate the elements of each comparison matrix based on the coordinate information of the feature points to obtain the weights from the target layer to the scheme layer, and finally select the corresponding feature points as the welding positions at the current sampling time according to the maximum a posteriori weight principle.

[0009] Furthermore, the specific steps of step one are as follows:

[0010] Step 1: Design an improved Gabor filter and filter the original image:

[0011]

[0012] Where: x′=x cosθ+y sinθ, y′=-x sinθ+y cosθ, f is the frequency, σ is the template size, and θ is the filter angle.

[0013] Step 2: Apply Gabor filtering to the original image using different filtering angles to obtain feature maps in each direction. Then, linearly combine the obtained feature maps to obtain a comprehensive feature map.

[0014] Step 3: Obtain the average gray value at each location in the comprehensive directional feature image by sliding an m×m window in a column manner.

[0015]

[0016] Where i represents the row of the image, j represents the column of the image, i:i+m-1 represents the row to be covered, j:j+m-1 represents the column to be covered, and f(x(), y()) represents the gray value at the specified row and column positions.

[0017] Step 4: Perform linear filtering on the obtained average gray value, and then obtain the monotonic interval of the filtered average gray value;

[0018] Step 5: Calculate the deviation of the average gray value at the end point of each monotonic interval before and after filtering, and obtain the positions p where the deviation is greater than T. k Average gray value before filtering If k = 0, the threshold for the current loop coverage area is set to 255; if k = 1, the threshold for the current loop coverage area is set to... If k > 1, then the region covered by this loop is determined by p. k It is divided into two parts; the threshold for the upper part is set to 255, and the threshold for the lower part is set to...

[0019] Step 6: Set the Euclidean distance threshold E and perform nearest neighbor clustering on the data;

[0020] Step 7: Define the prior probabilities of weld profile thickness L, profile thickness uniformity J, and profile thickness compactness Z as follows:

[0021]

[0022] in: N j w is the number of members in the j-th data cluster. ij Let μ1 represent the thickness of the i-th member of the j-th data cluster, μ1 represent the average value of the weld profile thickness monitoring, and δ1 represent the corresponding variance.

[0023]

[0024] in: α≥3, μ2 is the average value of weld profile thickness uniformity monitoring, and δ2 is the corresponding variance;

[0025]

[0026] in: t ij denoted by , μ3 represents the number of missing data in the i-th column of the j-th data cluster, μ3 is the average value of weld profile thickness compactness monitoring, δ3 is the corresponding variance, and β≥3.

[0027] Step 8: Calculate the following posterior probabilities:

[0028] p(C i |L j J j Z j )=p(L j J j Z j |Ci )p(C i ) / p(L j J j Z j (i = 1, 2)

[0029] Where C1 and C2 represent weld contour data class and interference data class, respectively; p(C1) = p(C2).

[0030] Step 9: Classify the data using the following criteria: If

[0031] p(L j |C1)p(J j |C1)p(Z j |C1)>p(L j |C2)p(J j |C2)p(Z j If |C2), then the j-th data cluster belongs to category C1; otherwise, it belongs to category C2.

[0032] Step 10: For data clusters belonging to C2, their grayscale values ​​are assigned to 0, thereby achieving the initial removal of interfering data;

[0033] Step 11: Analyze the remaining data clusters from Step 4. If any two clusters overlap in the horizontal direction, calculate the Euclidean distance between each cluster. Clusters with smaller Euclidean distances are further classified as interference clusters, and their grayscale values ​​are assigned to 0 to further remove interference and obtain the bevel profile.

[0034] Furthermore, the specific steps for identifying the bevel contour feature points are as follows:

[0035] Step 1: Scan the obtained bevel profile column by column, count the ordinate of the data points with a gray value of 255 in each column, and calculate the average value. Use this average value as the position of all data points in the column to obtain the linear bevel profile.

[0036] Step 2: Calculate the slope at each point on the linear bevel profile using the following formula:

[0037]

[0038] Where x() and y() represent the coordinates of the data points, M represents the number of data points used in each calculation, i represents the index of the data point, and j represents the index of the data point adjacent to i.

[0039] Step 3: Use a 1×Q one-dimensional linear filter to filter the slope and obtain the monotonic interval of the filtered slope;

[0040] Step 4: Calculate the change in slope for each monotonic interval, sort the changes in slope in descending order, and sort the corresponding monotonic intervals as well.

[0041] Step 5: Set the number P of feature points for the bevel profile, extract the first P monotonic intervals that have been sorted, and obtain the middle position of each corresponding monotonic interval. The position of the feature point is the middle position of these monotonic intervals.

[0042] Furthermore, the specific steps of step three are as follows:

[0043] Step 1: Calculate the horizontal and vertical distances between the other weld contour feature points and the rightmost feature point, and use these two distances as evaluation indicators:

[0044]

[0045] Where X i and Y i X and Y represent the x and y coordinates of the i-th feature point from left to right, respectively. n and Y n These represent the x and y coordinates of the rightmost feature point, respectively.

[0046] Step 2: Apply formula W i H i Two distances are mapped to scales 1-9:

[0047]

[0048] Step 3: Define two distance thresholds, W0 and H0, in the horizontal and vertical directions respectively, and map them to scales 1-9.

[0049]

[0050] Step 4: Calculate the elements of the judgment matrix from the target layer to the indicator layer using the following formulas:

[0051] a ij =γw i / h j i = 2k + 1, j = 2k

[0052] a ij =w i / γh j i = 2k, j = 2k+1

[0053]

[0054] a ij =h i / h j i ≠ j, i, j = 2k

[0055] a ji =1 / a ij

[0056] a ij =1, i=j

[0057] Where γ is the weighting coefficient, and γ≥0.2, k∈z + .

[0058] Step 5: Calculate W for the index layer using the following formulas. i H i Calculation coefficients at the scheme layer:

[0059]

[0060]

[0061] Where n represents the number of feature points, and s = 1, 2, ..., (n-1).

[0062] Step 6: Calculate the elements of the judgment matrix from the indicator layer to the solution layer using the following formulas:

[0063]

[0064] d ji =1 / d ij

[0065] Step 7: Calculate the weights of the hierarchical single-ranking results of the judgment matrices of the target layer and the indicator layer using the following formula:

[0066]

[0067] Step 8: Calculate the weights of the hierarchical single-ranking results of the judgment matrices for the indicator layer and the scheme layer using the following formula:

[0068]

[0069] Step 9: Calculate the importance weight of the scheme layer to the target layer using the following formula:

[0070]

[0071] Step 10: Based on the principle of maximizing weight, select the option that best matches the given value. The feature point with the largest weight is used as the current welding position, thus completing the autonomous decision-making of the welding position.

[0072] Beneficial effects of the present invention: The present invention proposes an autonomous decision-making method for the welding position of multi-pass welding of thick plate T-joints based on visual methods and tomographic analysis models. It is an economical, reliable and universally applicable method. This method can not only greatly improve the welding efficiency and automation level of thick plate T-joints, but also realize the control of web angle deformation, and provide a reference for online autonomous decision-making of the welding position of multi-pass welding of thick plate butt joints. Attached Figure Description

[0073] Figure 1 A flowchart for deciding the welding position of a multi-pass weld on a thick plate T-joint, provided by the present invention.

[0074] Figure 2 This is a flowchart of the bevel contour extraction process described in step one of this invention;

[0075] Figure 3 This is a flowchart of the bevel contour feature point extraction process described in step two of this invention;

[0076] Figure 4 This is an example diagram of the hierarchical analysis model described in step three of this invention;

[0077] Figure 5 This is an example image of the weld seam with bevel profile in multi-pass welding of T-joints in this invention;

[0078] Figure 6 This is an example of the welding position decision described in step three of this invention. Figure 1 (Specify 3 feature points, with feature point 3 being the decision result);

[0079] Figure 7 This is an example of the welding position decision described in step three of this invention. Figure 2 (Specify 4 feature points, with feature point 4 representing the decision result);

[0080] Figure 8 This is an example diagram of the welding result implemented using the welding position decision result described in step three of this invention. Detailed Implementation

[0081] The present invention will be further described below with reference to specific embodiments, but the present invention is not limited to the embodiments.

[0082] Example 1: An autonomous decision-making method for multi-pass welding position of thick plate T-joints based on the analytic hierarchy process, such as... Figure 1 As shown, the main steps include:

[0083] Step 1: The original weld image is filtered using an improved Gabor filter to obtain the orientation feature map. Then, the orientation feature map is binarized using a local threshold segmentation method. After that, the data is clustered by nearest neighbor and the clustered data clusters are classified using a Naive Bayes classifier to initially remove interference data. Finally, the visual feature competition method is used to further remove interference and obtain the bevel profile.

[0084] Step 2: Calculate the average height of the obtained bevel profile column by column in the vertical direction to obtain a linear profile. Then, calculate the slope at each position of the linear profile and smooth the slope to obtain monotonic intervals. Then, calculate the slope change of the monotonic intervals and sort the changes in descending order. Finally, determine the changes that are first in the sequence by specifying the number of bevel profile feature points, thereby determining the first monotonic intervals. Determine the position of the weld profile feature points based on the center position of the monotonic intervals.

[0085] Step 3: Establish a three-layer hierarchical analysis model based on the obtained bevel contour feature points, then calculate the elements of each comparison matrix based on the coordinate information of the feature points to obtain the weights from the target layer to the scheme layer, and finally select the corresponding feature points as the welding positions at the current sampling time according to the maximum a posteriori weight principle.

[0086] In the bevel contour extraction method described in step one of this example, a Naive Bayes classification and visual feature competition method are used to gradually remove interference, such as... Figure 2 As shown, the calculations were performed using MATLAB software, and the data was visualized for verification. The specific steps are as follows:

[0087] Step 1: Design an improved Gabor filter:

[0088]

[0089] Where: x′=x cosθ+y sinθ, y′=-x sinθ+y cosθ, in the study we take f=1 / 7.8, ∑=4, and the filtering angle θ takes values ​​of 10°, 90°, and -15° respectively;

[0090] Step 2: Apply Gabor filtering to the original image using three filtering angles to obtain three directional feature maps. Then, linearly combine the obtained directional feature maps to obtain a comprehensive directional feature map.

[0091] Step 3: Obtain the average gray value at each location in the comprehensive directional feature image by sliding a 5×5 window in a column manner.

[0092]

[0093] Where i represents the row of the image, j represents the column of the image, i:i+4 represents the row to be covered, j:j+4 represents the column to be covered, and f(x(), y()) represents the gray value at the specified row and column positions.

[0094] Step 4: Use a linear filter with a window size of 1×9 to filter the average gray value, and then obtain the monotonic interval of the filtered average gray value.

[0095] Step 5: Calculate the deviation of the average gray value at the end of each monotonic interval before and after filtering, and obtain the positions p where the deviation is greater than 15. k Average gray value before filtering If k = 0, the threshold for the current loop coverage area is set to 255; if k = 1, the threshold for the current loop coverage area is set to... If k > 1, then the region covered by this loop is determined by p. k It is divided into two parts; the threshold for the upper part is set to 255, and the threshold for the lower part is set to...

[0096] Step 6: Set the Euclidean distance threshold to 2 pixels and perform nearest neighbor clustering on the data;

[0097] Step 7: Define the prior probabilities of weld profile thickness L, profile thickness uniformity J, and profile thickness density Z respectively:

[0098]

[0099] in: N j w is the number of members in the j-th data cluster. ij Let μ1 represent the thickness of the i-th member of the j-th data cluster, where μ1 = 5 and δ1 = 1.2.

[0100]

[0101] in: α=3, μ2=0.9, δ2=0.3;

[0102]

[0103] in: t ij Let μ3 represent the number of missing data in the i-th column of the j-th data cluster, where μ3 = 0, δ3 = 0.2, and β = 3.

[0104] Step 8: Calculate the following posterior probabilities:

[0105] p(C i |L j J j Zj )=p(L j J j Z j |C i )p(C i ) / p(L j J j Z j (i = 1, 2)

[0106] Where C1 and C2 represent weld contour data class and interference data class, respectively; p(C1) = p(C2).

[0107] Step 9: Classify the data using the following criteria: If

[0108] p(L j |C1)p(J j |C1)p(Z j |C1)>p(L j |C2)p(J j |C2)p(Z j If |C2), then the j-th data cluster belongs to category C1; otherwise, it belongs to category C2.

[0109] Step 10: For data clusters belonging to C2, their grayscale values ​​are assigned to 0, thereby achieving the initial removal of interfering data;

[0110] Step 11: Count the remaining data clusters. If two clusters overlap in the horizontal direction, calculate the Euclidean distance of all data members in each cluster. Take the largest length as the Euclidean distance of the cluster. The cluster with the smaller Euclidean distance is further divided into interference clusters, and its gray value is assigned to 0 to further remove interference and obtain the bevel profile.

[0111] Step two is based on the detection of the change in the slope of the bevel profile. Figure 3 The specific steps for obtaining the feature points of the bevel profile are as follows:

[0112] Step 1: Scan the obtained bevel profile column by column, count the ordinate of the data points with a gray value of 255 in each column, and calculate the average value. Use this average value as the position of all data points in the column to obtain the linear bevel profile.

[0113] Step 2: Calculate the slope at each point on the linear bevel profile using the following formula:

[0114]

[0115] Where x() and y() represent the coordinates of the data points, i represents the index of the data point, and j represents the index of the data point adjacent to i.

[0116] Step 3: Use a 1×9 one-dimensional linear filter to filter the slope and obtain the monotonic interval of the filtered slope;

[0117] Step 4: Calculate the change in slope for each monotonic interval, sort the changes in slope in descending order, and sort the corresponding monotonic intervals as well.

[0118] Step 5: Set the number of feature points for the bevel contour to 3, and extract the first 3 sorted monotonic intervals. Obtain the midpoint of each corresponding monotonic interval. The position of the feature point is the midpoint of these monotonic intervals. Figure 6 Set the number of feature points for the bevel contour to 4, and extract the first 4 sorted monotonic intervals. Obtain the midpoint of each corresponding monotonic interval. The position of the feature point is the midpoint of these monotonic intervals. Figure 7 );

[0119] Step three involves autonomous decision-making regarding welding positions based on a hierarchical analysis model. Figure 4 The specific steps include:

[0120] Step 1: Calculate the horizontal and vertical distances between the other weld contour feature points and the rightmost feature point, and use these two distances as evaluation indicators:

[0121]

[0122] Where X i and Y i X and Y represent the x and y coordinates of the i-th feature point from left to right, respectively. n and Y n Let x and y represent the x and y coordinates of the rightmost feature point, respectively, where n = 3 and n = 4.

[0123] Step 2: Apply formula W i H i Two distances are mapped to scales 1-9:

[0124]

[0125] Step 3: Define two distance thresholds, W0 = 150 and H0 = 60, in the horizontal and vertical directions respectively, and map them to scales 1-9.

[0126]

[0127] Step 4: Calculate the elements of the judgment matrix from the target layer to the indicator layer using the following formulas:

[0128] a ij =γw i / h ji = 2k + 1, j = 2k

[0129] a ij =w i / γh j i = 2k, j = 2k+1

[0130]

[0131] a ij =h i / h j i ≠ j, i, j = 2k

[0132] d ji =1 / a ij

[0133] a ij =1, i=j

[0134] Where γ = 0.2, k ∈ z + .

[0135] Step 5: Calculate W for the index layer using the following formulas. i H i Calculation coefficients at the scheme layer:

[0136]

[0137]

[0138] Where n = 3 and n = 4.

[0139] Step 6: Calculate the elements of the judgment matrix from the indicator layer to the solution layer using the following formulas:

[0140]

[0141] d ji =1 / d ij

[0142] Step 7: Calculate the weights of the hierarchical single-ranking results of the judgment matrices of the target layer and the indicator layer using the following formula:

[0143]

[0144] Where n = 3 and n = 4.

[0145] Step 8: Calculate the weights of the hierarchical single-ranking results of the judgment matrices for the indicator layer and the scheme layer using the following formula:

[0146]

[0147] Where n = 3 and n = 4.

[0148] Step 9: Calculate the importance weight of the scheme layer to the target layer using the following formula:

[0149]

[0150] Where n = 3 and n = 4.

[0151] Step 10: Based on the principle of maximizing weight, select the option that best matches the given value. The feature point with the largest weight is taken as the current welding position, completing the autonomous decision-making of the welding position: when n=3, the feature point marked as 3 is the result of the decision. Figure 6 When n=4, the feature point marked as 4 is the result of the decision. Based on the decision result, a welding test is conducted. Figure 8 The experimental results demonstrate the effectiveness of the decision-making method proposed in this invention.

[0152] The above description merely illustrates preferred embodiments of the present invention, and while the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for autonomous decision-making on welding position of multi-pass welding of thick plate T-joints based on the analytic hierarchy process, characterized in that, Includes the following steps: Step 1: The original weld image is filtered using an improved Gabor filter to obtain the orientation feature map. Then, the orientation feature map is binarized using a local threshold segmentation method. After that, the data is clustered by nearest neighbor and the clustered data clusters are classified using a Naive Bayes classifier to initially remove interference data. Finally, the visual feature competition method is used to further remove interference and obtain the bevel profile. Step Two: Calculate the average height of the obtained bevel profile column by column in the vertical direction to obtain a linear profile. Then, calculate the slope at each position of the linear profile and smooth the slope to obtain monotonic intervals. Calculate the slope change of each monotonic interval and sort the changes in descending order. Finally, determine the changes that precede the specified number of bevel profile feature points to determine the preceding monotonic intervals. Determine the position of the weld profile feature points based on the center position of each monotonic interval. The weld profile feature point identification method in Step Two includes the following steps: Step 1: Scan the bevel profile obtained in Step 1 column by column, count the ordinate of the data points with a gray value of 255 in each column, and calculate the average value. Use this average value as the position of all data points in this column to obtain the linear bevel profile. Step 2: Calculate the slope at each point on the linear bevel profile using the following formula: （7） in , Represents the coordinates of the data points. This indicates the number of data points used in each calculation. Indicates data point subscripts, Indicates and Subscripts of adjacent data points; Step 3: Use a 1×Q one-dimensional linear filter to filter the slope and obtain the monotonic interval of the filtered slope; Step 4: Calculate the change in slope for each monotonic interval, sort the changes in slope in descending order, and sort the corresponding monotonic intervals as well. Step 5: Set the number P of feature points for the bevel contour, extract the first P monotonic intervals that have been sorted, and obtain the middle position of each corresponding monotonic interval. The position of the feature point is the middle position of these monotonic intervals. Step 3: Establish a three-layer hierarchical analysis model based on the obtained bevel contour feature points, then calculate the elements of each comparison matrix based on the coordinate information of the feature points to obtain the weights from the target layer to the scheme layer, and finally select the corresponding feature points as the welding positions at the current sampling time according to the maximum a posteriori weight principle.

2. The method for autonomous decision-making of welding position in multi-pass welding of thick plate T-joints based on the analytic hierarchy process according to claim 1, characterized in that: The specific steps of step one are as follows: Step 1: Design an improved Gabor filter and filter the original image: （1） in: , , It's frequency. It is the template size. It is the filtering angle; Step 2: Apply Gabor filtering to the original image using different filtering angles to obtain feature maps in each direction. Then, linearly combine the obtained feature maps to obtain a comprehensive feature map. Step 3: Binarize the comprehensive directional feature map using a local threshold segmentation method. The specific steps are as follows: 1) Adopt The window is slidable in a column to obtain the average gray value at each location in the comprehensive directional feature image: （2） in Represents the rows of an image. Columns representing images, Indicates the rows to be covered. Indicates the column to be covered. Represents the grayscale value at the specified row and column position; 2) Perform linear filtering on the obtained average gray value; 3) Obtain the monotonic interval of the filtered average gray value; 4) Calculate the deviation of the average gray value at the end point of each monotonic interval before and after filtering, and obtain the positions where the deviation is greater than T. Average gray value before filtering ; 5) If k=0, the threshold for the current loop coverage area is set to 255; if k=1, the threshold for the current loop coverage area is set to... If k > 1, then the area covered by this loop is... It is divided into two parts; the threshold for the upper part is set to 255, and the threshold for the lower part is set to... ; Step 4: Set the Euclidean distance threshold E and perform nearest neighbor clustering on the data; Step 5: Use a Naive Bayes classifier to classify the clustered data clusters and initially remove interfering data. The specific steps are as follows: 1) Define the prior probabilities of weld profile thickness L, profile thickness uniformity J, and profile thickness compactness Z as follows: （3） in: , It is the number of members in the j-th data cluster. This represents the thickness of the i-th member of the j-th data cluster. This represents the average value of the weld contour thickness monitoring. Indicates the corresponding variance; （4） in: , , It is the average value for monitoring the uniformity of weld contour thickness. That is the corresponding variance; （5） in: , This represents the number of missing data items in the i-th column of the j-th data cluster. It is the average value of weld contour thickness density monitoring. That is the corresponding variance. ; 2) Calculate the following posterior probability: （6） in and These represent weld contour data and interference data, respectively. ; 3) Data classification is implemented using the following criteria: If Then the j-th data cluster belongs to category Otherwise it belongs to ; 4) For those belonging to The data clusters are assigned a grayscale value of 0, thereby achieving the initial removal of interfering data; Step 6: Analyze the remaining data clusters. If two clusters overlap in the horizontal direction, calculate the Euclidean distance between each cluster. Clusters with smaller Euclidean distances are further classified as interference clusters, and their grayscale values ​​are assigned to 0 to further remove interference and obtain the bevel profile.

3. The method for autonomous decision-making of welding position in multi-pass welding of thick plate T-joints based on the analytic hierarchy process according to claim 1, characterized in that, The specific steps of step three are as follows: Step 1: Calculate the horizontal and vertical distances between the other weld contour feature points and the rightmost feature point, and use these two distances as evaluation indicators: （8） in and These represent the x and y coordinates of the i-th feature point from left to right, respectively. and These represent the x and y coordinates of the rightmost feature point, respectively. Step 2: Map the two distances in formula (8) to scale 1-9: （9） Step 3: Define two distance thresholds in the horizontal and vertical directions respectively. , And map it to scale 1-9 using formula (8): （10） Step 4: Calculate the elements of the judgment matrix from the target layer to the indicator layer using formulas (11)-(16): （11） （12） （13） （14） （15） （16） in It is a weighting coefficient, and , ; Step 5: Calculate the index layer using formulas (17) and (18) respectively. , Calculation coefficients at the scheme layer: （17） （18） in Indicates the number of feature points. ; Step 6: Calculate the elements of the judgment matrix from the indicator layer to the scheme layer using formulas (19) and (20) respectively: （19） （20） Step 7: Calculate the weights of the hierarchical single-ranking results of the judgment matrices of the target layer and the indicator layer using formula (21): （21） Step 8: Calculate the weights of the hierarchical single-ranking results of the judgment matrices for the indicator layer and the scheme layer using formula (22): （22） Step 9: Calculate the importance weight of the scheme layer to the target layer using formula (23): （23） Step 10: Based on the principle of maximizing weight, select the option that best matches the given value. The feature point with the largest weight is used as the current welding position, thus completing the autonomous decision-making of the welding position.

Images