Spatial all-position welding process parameter self-adaptive welding method and equipment

Abstract

The invention provides a welding method and equipment with self-adaptive spatial all-position welding process parameters, and the method comprises the steps: obtaining a plurality of structured light images of a groove of a to-be-welded region through a laser visual sensor, calculating the image feature information corresponding to each image, marking the region between the structured light images of the groove of two adjacent to-be-welded regions as a regulation region, and carrying out the adjustment and control of the region; and calculating each regulation and control area to obtain the volume of a to-be-welded sub-area corresponding to the currently set cladding layer, and inputting the structured light image of the groove of the to-be-welded area, the image feature information, the regulation and control distance and the volume of the to-be-welded sub-area into a trained CNN model to obtain a plurality of groups of welding process parameters, so that in the welding process, a plurality of groups of welding process parameters are obtained. And corresponding welding process parameters are called for each regulation and control area for welding, so that the welding requirements of complex groove working conditions are met, and the welding quality can be improved.

Description

Technical Field

[0001] The present invention relates to the field of welding technology, and in particular to a welding method and equipment with adaptive spatial all-position welding process parameters. Background Art

[0002] Current pipeline welding challenges include uneven roundness, uneven base and filler welds, and irregular groove conditions in the welded area. Existing full-position automatic welding technologies often rely on pre-set automatic welding parameters or manual visual monitoring and control of the welding process. These technologies rely on unchanging, pre-set welding process parameters and oscillation parameters, making them unable to cope with the dynamic demands of groove conditions. This makes pipeline welding prone to defects such as lack of fusion, undercuts, and weld bumps.

[0003] In the prior art, some pipeline all-position welding methods are disclosed, such as the pipeline all-position swing width adaptive welding method, device and equipment disclosed in CN 117066647 A. The above methods have the following disadvantages:

[0004] (1) The calculation formula is highly idealized: Swing width calculation formula L = 2(U e -L S The "-4" at the end of )tanγ+a-4 lacks theoretical basis and is suspected to be an empirical correction value, which may be invalid under complex groove conditions.

[0005] (2) The necessity of the dual-trolley system is questionable: the above solution requires the configuration of two independent welding trolleys on the port and starboard sides, which doubles the equipment cost.

[0006] In existing technology, all-position pipeline welding can be performed using traditional automatic pipeline welding equipment: a welding carriage is mounted on tracks, with a pre-set trajectory and fixed swing width to complete the welding process. However, this fixed swing width cannot adapt to changes in V-groove dimensions, resulting in poor groove adaptability. Furthermore, the welder must monitor the weld pool status throughout the welding process, requiring manual intervention.

[0007] Pipeline all-position welding can also be performed manually using manual arc welding: the welder holds the welding torch and adjusts welding parameters and torch position based on experience, completing the weld along the pipeline. This manual welding process relies on the welder's experience and demands high skill and physical strength. Fatigue and technical variability can easily lead to weld defects such as lack of fusion, undercuts, and weld bumps. Summary of the Invention

[0008] In order to solve the problems existing in the prior art, the present application proposes a welding method and equipment with adaptive spatial all-position welding process parameters, so as to adapt to complex groove working conditions and solve the problem of low welding quality caused by complex groove working conditions in pipeline welding.

[0009] In order to achieve the above objectives, one aspect of the present application proposes a welding method with adaptive spatial all-position welding process parameters, comprising the following steps:

[0010] Step 1: Perform pre-welding cleaning operations;

[0011] Step 2: Perform pipe jointing and manual bottom welding;

[0012] Step 3: Control the welding carriage to move 180 degrees from the initial position in a preset welding direction. During the movement, a sensor system installed on the welding carriage scans and obtains a number of groove structured light images of the area to be welded. The number of groove structured light images of the area to be welded is recorded as F.

[0013] Step 4: The F groove structured light images of the welding area obtained in step 3 are screened according to a preset control distance, and the first and Fth images are retained. After screening, M-2 images are obtained, and a total of M images are obtained. The M images obtained are respectively processed to obtain image feature information corresponding to each image;

[0014] Step 5: The area between two adjacent groove structured light images of the area to be welded is recorded as the control area, and M-1 control areas are obtained. In any control area, the volume of the sub-area to be welded corresponding to the currently set cladding layer in the control area can be calculated based on the image feature information of the two groove structured light images of the area to be welded corresponding to the control area and the control distance;

[0015] Step 6: Input the M images, the image feature information corresponding to each image, the control distance, and the volume of the sub-area to be welded corresponding to the M-1 control areas into the trained CNN model to obtain the corresponding M-1 groups of welding process parameters that need to be controlled. The M-1 control areas correspond one-to-one to the M-1 groups of welding process parameters that need to be controlled.

[0016] Step 7: Return the welding carriage to the initial position, and then control the welding carriage to move 180° along the welding direction preset in step 3. During the movement, when the welding carriage reaches the starting position of each control area, the welding process parameters corresponding to the control area are adopted, including the welding process parameters that need to be regulated obtained in step 6 and the welding process parameters that do not need to be regulated throughout the welding process, until the welding carriage reaches the end position of the control area, and welds the control area using the corresponding welding process parameters until the welding carriage completes the 180° path.

[0017] Step 8: Control the welding carriage to return to the initial position and determine whether the welding operation is completed along the 360° path of the pipeline. If not, jump to step 9; if yes, end the welding process of the current cladding layer.

[0018] Step 9: Reversely install the welding carriage in the initial position, and then repeat steps 3 to 7 to achieve welding of the current cladding layer in the area to be welded on the other side of the pipeline at a 180° path. This completes the full-position welding operation of the current cladding layer of the pipeline.

[0019] Step 10: Repeat steps 3 to 9 to achieve welding of each cladding layer, so as to realize spatial full-position welding operation of the pipeline.

[0020] In some embodiments, in step 3, the sensor system uses a laser vision sensor.

[0021] In some embodiments, in step 4, the image feature information corresponding to each image includes: groove feature point position, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and cross-sectional area of ​​the to-be-welded sub-region of the cladding layer;

[0022] The process of obtaining the image feature information corresponding to the image is as follows:

[0023] First, the structured light image of the groove in the area to be welded is binarized to extract the point information that constitutes the line structured light. The horizontal coordinate axis is recorded as the X-axis and the vertical coordinate axis is recorded as the Y-axis. The above points are screened and the point at the far left of the image is recorded as point St, whose X coordinate is the smallest; the point at the far right of the image is recorded as point En, whose X coordinate is the largest; point C1 is the point with the smallest X coordinate when the Y coordinate is the smallest among all the line structured light points, that is, the second inflection point on the left of the groove; point C2 is the point with the largest X coordinate when the Y coordinate is the smallest among all the line structured light points, that is, the second inflection point on the right of the groove. The X and Y coordinates of the above points St, En, C1, and C2 are all known;

[0024] Calculate the first turning point on the left of the groove, that is, point A, and the first turning point on the right of the groove, that is, point B. The calculation process is as follows:

[0025] Construct the straight line equation StC1: (YY St )(X c1 -X St )-(XX St )(Y c1 -Y St )=0, where Y St Indicates the Y coordinate of point St, X c1 Indicates the X coordinate of point C1, X St Indicates the X coordinate of point St, Y c1 Indicates the Y coordinate of point C1;

[0026] Construct the straight line equation EnC2: (YY En )(X c2 -X En )-(XX En )(Y c2 -Y En )=0, where Y En Indicates the Y coordinate of point En, X c2 Indicates the X coordinate of point C2, X En Indicates the X coordinate of point En, Y c2 Indicates the Y coordinate of point C2;

[0027] Calculate the straight-line distance from the point of line structured light to the straight lines StC1 and EnC2: Where p represents any point in the line structured light except St, En, C1, and C2, px represents the X coordinate of point p, py represents the Y coordinate of point p, and a, b, and c are the coefficients of the straight line equation;

[0028] Among the points whose X coordinates are less than the X coordinates of point C1, the point farthest from StC1 is defined as point A based on the straight-line distance, i.e., the first left inflection point of the groove. Among the points whose X coordinates are greater than the X coordinates of point C2, the point farthest from EnC2 is defined as point B based on the straight-line distance, i.e., the first right inflection point of the groove.

[0029] So far, the four characteristic points of the groove have been obtained, namely point A, point B, point C1, and point C2, and the position information (X, Y) of the four characteristic points of the groove are also known;

[0030] The calculation process of the groove bottom width is as follows: Among them, X c1 Indicates the X coordinate of point C1, X c2 Indicates the X coordinate of point C2, Y c1 Indicates the Y coordinate of point C1, Y c2 Indicates the Y coordinate of point C2;

[0031] The calculation process of the groove top width is as follows: Among them, X A Indicates the X coordinate of point A, X B Indicates the X coordinate of point B, Y A Indicates the Y coordinate of point A, Y B Indicates the Y coordinate of point B;

[0032] The calculation process of the groove angle is as follows:

[0033] Among them, θ1 is the angle between the straight line AC1 and the Y axis, X A Indicates the X coordinate of point A, X c1 Indicates the X coordinate of point C1, Y A Indicates the Y coordinate of point A, Y c1 Indicates the Y coordinate of point C1;

[0034] Among them, θ2 is the angle between the straight line BC2 and the Y axis, X B Indicates the X coordinate of point B, X c2 Indicates the X coordinate of point C2, Y B Indicates the Y coordinate of point B, Y c2 Indicates the Y coordinate of point C2;

[0035] θ=θ1+θ2, where θ is the groove angle;

[0036] The calculation process of the groove area is as follows: Among them, S is the groove area formed by ABC1C2;

[0037] Cladding layer height: h represents the height of the cladding layer required for a single weld, in mm. The intersection of the cladding layer and the groove boundary AC1 is marked as point A', and the intersection of the cladding layer and the groove boundary BC2 is marked as point B'. The position information (X, Y) of points A' and B' can be known.

[0038] The calculation process of the cross-sectional area of ​​the sub-region to be welded for the cladding layer is as follows: , where S' is the cross-sectional area of ​​the sub-region to be welded composed of A'B'C1C2 for the cladding layer, X A′ Indicates the X coordinate of point A', Y B′ represents the Y coordinate of point B', X B′ Indicates the X coordinate of point B', Y A′ Indicates the Y coordinate of point A'.

[0039] In some embodiments, in step 5, the preset control distance is set to L control , which represents the distance between each two adjacent groove structured light images of the area to be welded, that is, the distance the welding carriage travels through a control area, in mm;

[0040] Among them, V app represents the volume of the sub-area to be welded corresponding to the cladding layer in the control area, S′ start Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the starting image corresponding to the control area, S′ end Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the ending image corresponding to the control area.

[0041] In some embodiments, in step 6, the welding process parameters that need to be regulated include welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed.

[0042] In some embodiments, before step 6, the step of training a CNN model is further included:

[0043] Step a: determining optimal welding process parameters corresponding to the groove of the area to be welded under different preset working conditions, wherein under one preset working condition, each cross-section of the groove of the area to be welded is the same; the welding process parameters are the welding process parameters that need to be adjusted;

[0044] Step b, using a sensor system to capture structured light images of the groove of the area to be welded under different preset working conditions, and only capturing one structured light image of the groove of the area to be welded for each working condition;

[0045] Step c, performing image processing on each structured light image of the groove of the area to be welded captured in step b to obtain image feature information corresponding to each image: groove feature point position, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and cross-sectional area of ​​the sub-area to be welded of the cladding layer;

[0046] Step d: Based on the preset control distance and the image feature information corresponding to each image obtained in step c, calculate the volume of the sub-area to be welded corresponding to the cladding layer under each working condition: V train =S′×L control , where L control represents the control distance, S′ represents the cross-sectional area of ​​the sub-region to be welded of the cladding layer under the corresponding working condition obtained in step c;

[0047] Step e: The optimal welding process parameters corresponding to the groove of the to-be-welded area under different preset working conditions obtained in step a, the structured light images of the groove of the to-be-welded area under different preset working conditions captured in step b, the image feature information corresponding to each image obtained in step c, the control distance, and the volume of the sub-area to be welded corresponding to the cladding layer under each working condition obtained in step d are input into a convolutional neural network (CNN) model for training to obtain the optimal CNN model after training.

[0048] In some embodiments, the step e further comprises the following steps:

[0049] Step e1: The structured light images of the groove of the to-be-welded area under different preset working conditions received by the CNN model input layer are classified into an image branch; the image feature information corresponding to each image obtained in step c, the control distance, and the volume of the to-be-welded sub-area corresponding to the cladding layer under each working condition obtained in step d are classified into a structured feature branch;

[0050] Step e2: performing a convolution operation on the image branch to obtain a local feature vector of the image;

[0051] Step e3: Use a fully connected layer to perform structured learning on the structured feature branch, introduce nonlinearity into the feature vectors in the structured feature branch, capture the high-order relationship between the feature vectors, and then compress the dimension of the feature vector to avoid overfitting;

[0052] Step e4: Using the feature fusion layer in the CNN model, the local feature vector of the image obtained in step e2 and the feature vector output in step e3 are concatenated in the channel dimension, and the concatenated feature vector is then output.

[0053] Step e5: The feature vector processed by the feature fusion layer is processed by a fully connected layer, and the dimension is compressed using the ReLU activation function before output;

[0054] Step e6: Output the predicted values ​​of the welding process parameters by linear regression through the regression output layer of the CNN model;

[0055] Step e7, construct the loss function: Where N is the total number of samples, and a structured light image of the groove of the area to be welded and its corresponding image feature information is recorded as a sample, where the value of N is consistent with the total number of structured light images in step b, and j represents the jth of the five welding process parameters, namely, welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; y ij represents the true value of the jth welding process parameter of the i-th sample, that is, the value of the corresponding welding process parameter among the optimal welding process parameters under the corresponding working conditions obtained in step a received in the input layer of the CNN model; represents the predicted value of the jth welding process parameter of the i-th sample output in step e6; j Represents the importance weight of the jth welding process parameter;

[0056] In step e8, it is determined whether the loss function is less than the threshold. If so, the training is terminated to obtain the trained CNN model, i.e., the optimal CNN model. If not, the hyperparameters of the CNN model are continuously adjusted until the loss function is less than the threshold.

[0057] In some embodiments, in step e6, the following formula is used in the regression output layer:

[0058] in, The five-dimensional vector representing the predicted welding process parameters output includes welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; W out The matrix dimension is 5xK, where K is the dimension of the input feature vector of the regression output layer, and W out It is the weight matrix of the regression output layer parameters, which can map the high-dimensional feature space to the 5-dimensional welding process parameter space; X in represents the input feature vector, which is also the feature vector output by step e5; b out Represents the regression output layer bias vector, which represents the reference value of welding process parameters.

[0059] Another aspect of the present application provides a welding device with adaptive spatial all-position welding process parameters, comprising a welding carriage, a processing and control system, a sensor system, a welding power source, a wire feeding system, a flexible guide rail, and a welding gun;

[0060] In which, the flexible guide rail is used to be assembled on the pipe to be welded along the circumferential direction; the welding carriage is controlled by the processing and control system to move on the flexible guide rail; the sensor system is mounted on the welding carriage to obtain the structured light image of the groove of the area to be welded; the welding power supply is used to adjust the welding current and arc voltage used in the welding process; the wire feeding system is used to provide welding wire to the welding gun during the welding process, and is controlled by the processing and control system to change the wire feeding speed; the welding gun is installed on the welding carriage to perform welding operations on the area to be welded of the pipeline; the processing and control system is used to perform the corresponding operations of steps 3 to 7 of the welding method with adaptive spatial all-position welding process parameters as described in claim 1.

[0061] The beneficial effect of the scheme of the present application lies in the above-mentioned spatial all-position welding process parameter adaptive welding method and equipment, which obtains several structured light images of the groove of the to-be-welded area through a laser vision sensor, calculates the image feature information corresponding to each image, and records the area between two adjacent structured light images of the groove of the to-be-welded area as a control area, calculates the volume of the sub-area to be welded corresponding to the currently set cladding layer in each control area, and inputs the structured light image of the groove of the to-be-welded area, image feature information, control distance, and volume of the sub-area to be welded into the trained CNN model to obtain several groups of welding process parameters. In this way, during the welding process, the corresponding welding process parameters are called for each control area to perform welding to meet the welding requirements of complex groove working conditions.

[0062] The above-mentioned welding method and equipment with adaptive spatial all-position welding process parameters have the following advantages:

[0063] (1) The detected groove structured light image of the welding area is rich in information, and various groove feature information can be extracted during CNN model training;

[0064] (2) After training, the CNN model involved in this application only requires a laser vision sensor during actual welding, without the need for other sensing equipment;

[0065] (3) The CNN model is used to predict welding process parameters in the area to be welded. It has good adaptability to various complex and changeable pipeline welding groove conditions and can be used for welding pipelines of different specifications and positions. It can effectively avoid common welding defects such as lack of fusion, lack of penetration, undercut and weld bead, thereby improving welding quality.

[0066] (4) It reduces the welding interruption and adjustment time caused by changes in groove working conditions, improves the continuity and stability of the welding process, and thus improves the overall welding efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flow chart of a welding method with adaptive spatial all-position welding process parameters in an embodiment is shown.

[0068] Figure 2 A schematic diagram of characteristic information of the structured light image of the groove of the area to be welded in an embodiment is shown.

[0069] Figure 3 The figure shows the structural diagram of the sub-area to be welded corresponding to two adjacent structured light images of the groove of the area to be welded when the trained CNN model is applied.

[0070] Figure 4 The figure shows the structural diagram of the area to be welded under a single working condition during CNN model training. DETAILED DESCRIPTION

[0071] The specific implementation of this application will be further described below with reference to the accompanying drawings.

[0072] In the description of the present application, it should be understood that the terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a specific order or sequence. The terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present application.

[0073] like Figure 1 As shown, the spatial all-position welding process parameter adaptive welding method involved in this application includes the following steps:

[0074] Step 1: Perform pre-welding cleaning operations.

[0075] Specifically, before the welding trolley performs full-position welding of the pipeline, the area to be welded needs to be cleaned before welding. The cleaning process includes removing oil stains, impurities and oxide layers in the welding area to avoid defects such as slag inclusions in the subsequent welding process, thereby ensuring the stability of the welding process and the welding quality.

[0076] Step 2: Perform pipe joint matching and manual base welding.

[0077] Step 3: Control the welding carriage to move 180° from the initial position in the preset welding direction. During the movement, a sensor system installed on the welding carriage, such as a laser vision sensor, scans and obtains several groove structured light images of the area to be welded. The number of groove structured light images of the area to be welded is recorded as F.

[0078] Step 4: The F structured light images of the groove of the area to be welded obtained in step 3 are screened according to the preset control distance, and the first and Fth images are retained. After screening, M-2 images are obtained, and a total of M images are obtained. The M images obtained are processed separately to obtain the image feature information corresponding to each image: groove feature point position, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and cross-sectional area of ​​the sub-area to be welded for the cladding layer.

[0079] Specifically, the control distance is set in advance, and based on the first groove structured light image of the area to be welded and the control distance, the second groove structured light image of the area to be welded can be selected; based on the second groove structured light image of the area to be welded and the control distance, the third groove structured light image of the area to be welded can be selected, and so on, until M images are obtained.

[0080] Specifically, the process of performing image processing on each image to obtain the image feature information corresponding to the image is as follows:

[0081] First, the structured light image of the groove in the welding area is binarized to extract the point information that constitutes the line structured light. The horizontal coordinate axis is recorded as the X axis and the vertical coordinate axis is recorded as the Y axis. The above points are screened and the point on the left end of the image is recorded as the St point, which has the smallest X coordinate. Figure 2 As shown in the figure, the rightmost point of the image is recorded as point En, whose X coordinate is the largest; point C1 is the point with the smallest X coordinate when the Y coordinate is the smallest among all the line structure light points (that is, the second turning point on the left of the groove); point C2 is the point with the largest X coordinate when the Y coordinate is the smallest among all the line structure light points (that is, the second turning point on the right of the groove). The X and Y coordinates of the above points St, En, C1, and C2 are all known.

[0082] Calculate the first turning point on the left of the groove, that is, point A, and the first turning point on the right of the groove, that is, point B. The calculation process is as follows:

[0083] Construct the straight line equation StC1: (YY St )(X c1 -X St )-(XX St )(Y c1 -Y St )=0, where Y St Indicates the Y coordinate of point St, X c1 Indicates the X coordinate of point C1, X St Indicates the X coordinate of point St, Y c1 Indicates the Y coordinate of point C1.

[0084] Construct the straight line equation EnC2: (YY En )(X c2 -X En )-(XX En )(Y c2 -Y En )=0, where Y En Indicates the Y coordinate of point En, X c2 Indicates the X coordinate of point C2, X En Indicates the X coordinate of point En, Y c2 Indicates the Y coordinate of point C2.

[0085] Calculate the straight-line distance from the point of line structured light to the straight lines StC1 and EnC2: Wherein, p represents any point in the line structured light except St, En, C1 and C2, px represents the X coordinate of point p, py represents the Y coordinate of point p, and a, b, and c are coefficients of the straight line equation.

[0086] Among the points whose X coordinates are less than the X coordinates of point C1, the point farthest from StC1 is defined as point A (i.e., the first left inflection point of the groove) based on the straight-line distance. Among the points whose X coordinates are greater than the X coordinates of point C2, the point farthest from EnC2 is defined as point B (i.e., the first right inflection point of the groove) based on the straight-line distance.

[0087] Therefore, the four characteristic points of the groove are obtained, namely point A, point B, point C1, and point C2, and the position information (X, Y) of the four characteristic points of the groove is also known.

[0088] The calculation process of the groove bottom width is as follows: Among them, X c1 Indicates the X coordinate of point C1, X c2 Indicates the X coordinate of point C2, Y c1 Indicates the Y coordinate of point C1, Y c2 Indicates the Y coordinate of point C2.

[0089] The calculation process of the groove top width is as follows: Among them, X A Indicates the X coordinate of point A, X B Indicates the X coordinate of point B, Y A Indicates the Y coordinate of point A, Y B Indicates the Y coordinate of point B.

[0090] The calculation process of the groove angle is as follows:

[0091] Among them, θ1 is the angle between the straight line AC1 and the Y axis, X A Indicates the X coordinate of point A, X c1 Indicates the X coordinate of point C1, Y A Indicates the Y coordinate of point A, Y c1 Indicates the Y coordinate of point C1.

[0092] Among them, θ2 is the angle between the straight line BC2 and the Y axis, X B Indicates the X coordinate of point B, X c2 Indicates the X coordinate of point C2, Y B Indicates the Y coordinate of point B, Y c2 Indicates the Y coordinate of point C2.

[0093] θ=θ1+θ2, where θ is the groove angle.

[0094] The calculation process of the groove area is as follows: Among them, S is the groove area formed by ABC1C2.

[0095] Cladding layer height: h represents the height of the cladding layer required for a single weld, measured in mm. The intersection of the cladding layer and the groove boundary AC1 is designated as point A', and the intersection of the cladding layer and the groove boundary BC2 is designated as point B'. The positions (X, Y) of points A' and B' are known.

[0096] The calculation process of the cross-sectional area of ​​the sub-region to be welded for the cladding layer is as follows:

[0097] , where S' is the cross-sectional area of ​​the sub-region to be welded of the cladding layer formed by A'B'C1C2, and Y A′ Indicates the X coordinate of point A', Y B′ represents the Y coordinate of point B', X B′ Indicates the X coordinate of point B', Y A′ Indicates the Y coordinate of point A'.

[0098] In the M images obtained in steps 5 and 4, the distance between each two adjacent groove structured light images of the area to be welded is the control distance. The area between the two adjacent groove structured light images of the area to be welded is recorded as the control area, and M-1 control areas are obtained. In any control area, based on the image feature information of the two groove structured light images of the area to be welded corresponding to the control area and the control distance, the volume of the sub-area to be welded corresponding to the cladding layer in the control area can be calculated.

[0099] The calculation process is as follows:

[0100] Set the preset control distance to L control , which represents the distance between each two adjacent groove structured light images of the area to be welded, that is, the distance that the welding carriage travels through a control area, in mm.

[0101] Among them, V app It represents the volume of the sub-area to be welded corresponding to the cladding layer in the control area when the trained CNN model is applied, such as Figure 3 As shown, S′ start Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the starting image corresponding to the control area, S′ end Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the ending image corresponding to the control area.

[0102] Step 6: Input the M images, the image feature information corresponding to each image, the control distance, and the volume of the sub-area to be welded corresponding to the M-1 control regions into the trained CNN model to obtain the corresponding M-1 sets of welding process parameters that need to be controlled. The M-1 control regions correspond one-to-one to the M-1 sets of welding process parameters that need to be controlled. The welding process parameters that need to be controlled include welding speed, swing amplitude, swing speed, edge dwell time, and wire feed speed. In addition, some welding process parameters do not need to be adjusted during the entire welding process, such as welding voltage.

[0103] Step 7. Return the welding carriage to the initial position, and then control the welding carriage to move 180° along the welding direction preset in step 3. During the movement, when it reaches the starting position of each control area, the welding process parameters corresponding to the control area are adopted: including the welding process parameters that need to be regulated obtained in step 6 and the welding process parameters that do not need to be regulated throughout the welding process, until the welding carriage reaches the end position of the control area, so as to weld the control area using the corresponding welding process parameters, until the welding carriage completes the 180° path.

[0104] Step 8: Control the welding carriage to return to its initial position and determine whether the welding operation is completed along the 360° path of the pipeline. If not, jump to step 9. If yes, end the welding process of the current cladding layer. The specific judgment process can be determined manually.

[0105] Step 9. Reversely install the welding carriage in the initial position, and then repeat steps 3 to 7 to achieve welding of the current cladding layer in the area to be welded on the 180° path on the other side of the pipeline. At this point, the full-position welding operation of the current cladding layer of the pipeline is completed.

[0106] Step 10: Repeat steps 3 to 9 to achieve welding of each cladding layer, so as to realize spatial full-position welding operation of the pipeline.

[0107] Before step 6, the following steps are also included to train the CNN model:

[0108] Step a: Determine the optimal welding process parameters corresponding to the groove of the area to be welded under different preset working conditions, wherein, under one preset working condition, the cross sections of the groove of the area to be welded are the same, such as Figure 4 shown.

[0109] Specifically, the optimal welding process parameters corresponding to the groove of the area to be welded under different pre-set working conditions can be determined through experiments or database retrieval.

[0110] The different working conditions of the groove in the area to be welded mainly include the working conditions when the groove width, groove angle and groove symmetry change; the above-mentioned welding process parameters refer to the welding process parameters that need to be regulated, including welding speed, swing amplitude, swing speed, edge residence time, and wire feeding speed. In addition, there are some welding process parameters that do not need to be adjusted during the entire welding process, such as welding voltage, etc. These welding process parameters that do not change are no longer mentioned. This application is designed for the welding process parameters that need to be regulated, so that the optimal welding process parameters that are compatible with the corresponding working conditions can be output for different working conditions of the groove in the area to be welded. The standard for determining the optimal welding process parameters is that after welding using the welding process parameters, the weld is well formed and there are no obvious welding defects on the surface.

[0111] Step b: using a sensor system, such as a laser vision sensor, to capture structured light images of the groove of the area to be welded under different preset working conditions, and only capturing one structured light image of the groove of the area to be welded for each working condition.

[0112] Step c: Process each structured light image of the groove to be welded captured in step b to obtain the corresponding image feature information: groove feature point location, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and the cross-sectional area of ​​the sub-area to be welded for that cladding layer. Refer to step 4 for the image processing process in this step.

[0113] Step d: Based on the preset control distance and the image feature information corresponding to each image obtained in step c, the volume of the to-be-welded area corresponding to the cladding layer under each working condition is calculated.

[0114] The specific calculation process is as follows: V train =S′×L control , where V train Indicates the volume of the sub-area to be welded corresponding to the cladding layer under one working condition (each cross section of the groove in the area to be welded is the same) when training the CNN model, L control Represents the control distance, and S′ represents the cross-sectional area of ​​the sub-region to be welded of the cladding layer under the corresponding working condition obtained in step c.

[0115] Step e: The optimal welding process parameters corresponding to the groove of the to-be-welded area under different preset working conditions obtained in step a, the structured light images of the groove of the to-be-welded area under different preset working conditions captured in step b, the image feature information corresponding to each image obtained in step c, the control distance, and the volume of the sub-area to be welded corresponding to the cladding layer under each working condition obtained in step d are input into a convolutional neural network (CNN) model for training to obtain the optimal CNN model after training.

[0116] The specific step e also includes the following steps:

[0117] Step e1: The structured light images of the groove of the area to be welded under different pre-set working conditions received by the CNN model input layer are classified into the image branch; the image feature information corresponding to each image obtained in step c and the control distance received by the input layer, as well as the volume of the sub-area to be welded corresponding to the cladding layer under each working condition obtained in step d are classified into the structured feature branch.

[0118] Specifically, the image branch is responsible for directly learning local image features that cannot be described using the image feature information calculated based on the structured light image of the groove in the weld area. The structured feature branch directly inputs known precise image feature information, the control distance, and the volume of the weld sub-area. The CNN model can use these two different types of information to make predictions.

[0119] Step e2: Perform a convolution operation on the image branch to obtain a local feature vector of the image.

[0120] In this embodiment, the convolution operation is performed on the image branch, and the process is as follows:

[0121] (1) Convolutional layer of the CNN model: 32 3x3 convolution kernels (stride 2, no padding) are used to perform convolution operations on the image. A 224x224x1 (1-channel grayscale image, pixel size 224x224) image is input. Each convolution kernel slides across the image, calculating the weighted sum of the local area to obtain a 222x222x32 linear feature map after convolution. After the convolution, the ReLU activation function is used to introduce nonlinearity into the result of the convolution operation, allowing the CNN network to learn more complex feature representations. This convolution layer initially extracts structured light edge features.

[0122] (2) Max pooling layer of the CNN model: Each feature map obtained by the convolutional layer is downsampled using a 2x2 window (with a stride of 2). The input is a 222x222x32 feature map, and the output is a 111x111x32 feature map. The purpose of the max pooling layer is to reduce the spatial size of the feature map and reduce the complexity of subsequent calculations.

[0123] (3) Convolutional layer of the CNN model: 64 3x3 convolution kernels (stride 2, no padding) are used, with a 111x111x32 feature map as input. Each convolution kernel slides over the feature map, calculating the weighted sum of the local region to obtain a 109x109x64 feature map after convolution. After the convolution, the ReLU activation function is used to introduce nonlinearity into the result of the convolution operation, allowing the CNN network to learn more complex feature representations. Increasing the number of convolutions can increase the depth of the feature hierarchy extracted by the CNN model.

[0124] (4) Max pooling layer of the CNN model: Each feature map obtained by the convolutional layer is downsampled using a 2x2 window (with a stride of 2), inputting a 109x109x64 feature map and outputting a 54x54x64 feature map. The purpose of the max pooling layer is to reduce the spatial size of the feature map and reduce the complexity of subsequent calculations.

[0125] (5) Convolutional layer of the CNN model: 128 3x3 convolution kernels (stride 2, padding 1) are used, with a 54x54x64 feature map as input. Each convolution kernel slides over the feature map, calculating the weighted sum of the local area to obtain a 54x54x128 feature map after convolution. After the convolution, the ReLU activation function is used to introduce nonlinearity into the result of the convolution operation, allowing the CNN network to learn more complex feature representations. Increasing the number of convolutions can increase the depth of the feature hierarchy extracted by the model.

[0126] (6) Max pooling layer of the CNN model: Each feature map obtained by the convolutional layer is downsampled using a 2x2 window (with a stride of 2), inputting a 54x54x128 feature map and outputting a 27x27x128 feature map. The purpose of the max pooling layer is to reduce the spatial size of the feature map and reduce the complexity of subsequent calculations.

[0127] (7) Flattening layer of CNN model: convert the multi-dimensional feature map into a one-dimensional feature vector, input 27x27x128 feature vectors, and output 27x27x128=93312 one-dimensional feature vectors.

[0128] (8) Fully connected layer of CNN model: Since the number of 1-dimensional feature vectors obtained in (7) is too large, a fully connected layer is used for dimensionality reduction compression. 93,312 feature vectors are input and compressed using the ReLU activation function to output a 256-dimensional feature vector, allowing the features to be compressed while retaining key information.

[0129] After the above process, the local feature vector of the image (such as the boundary edge feature between the line structured light and the background, the position information of the inflection point of the line structured light, etc.) is finally obtained.

[0130] Step e3: Use a fully connected layer to perform structured learning on the structured feature branch, introduce nonlinearity into the feature vectors in the structured feature branch, capture the high-order relationship between the feature vectors, and then compress the dimension of the feature vector to avoid overfitting.

[0131] Specifically, under each working condition received by the CNN model input layer, the image feature information corresponding to the structured light image of the groove of the area to be welded, the control distance, and the volume of the sub-area to be welded corresponding to the cladding layer under the corresponding working condition are 16-dimensional feature vectors, including the X-coordinate, Y-coordinate, groove bottom width, groove top width, groove angle, groove area, cladding layer height, cross-sectional area of ​​the sub-area to be welded for the cladding layer, control distance, and volume of the sub-area to be welded corresponding to the cladding layer. The ReLU activation function is used to introduce nonlinearity into the input feature vector, input a 16-dimensional feature vector, and output a 32-dimensional feature vector to capture high-order relationships between feature vectors (such as the relationship between the groove bottom width and the groove area). The ReLU activation function is used again to compress the feature vector dimension, input a 32-dimensional feature vector, and output a 16-dimensional feature vector to avoid overfitting.

[0132] Step e4: Through the feature fusion layer in the CNN model, the local feature vector of the image obtained in step e2 and the feature vector output in step e3 are vector-concatenated in the channel dimension, and then the concatenated feature vector is output.

[0133] For example, the 256-dimensional feature vector output by step e2 and the 16-dimensional feature vector output by step e3 are concatenated in the channel dimension to output a 272-dimensional feature vector.

[0134] Step e5: The feature vector processed by the feature fusion layer is processed by the fully connected layer and the dimension is compressed using the ReLU activation function before output. For example, if a 272-dimensional feature vector is input, a 128-dimensional feature vector X is output. in .

[0135] Step e6: Output the predicted values ​​of the welding process parameters by linear regression through the regression output layer of the CNN model.

[0136] In this embodiment, the following formula is used in the regression output layer:

[0137] in, The five-dimensional vector representing the predicted welding process parameters output includes welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; W out The matrix dimension is 5xK, where K is the dimension of the input feature vector of the regression output layer, for example, 128, and W out It is the weight matrix of the regression output layer parameters, which can map the high-dimensional feature space to the 5-dimensional welding process parameter space; X in represents the input feature vector, which is also the feature vector output by step e5; b outIt represents the regression output layer bias vector, which represents the reference value of welding process parameters (for example, when the groove bottom width = 0, there must still be a minimum welding speed, minimum wire feeding speed, etc.).

[0138] Specifically, the welding speed dependent characteristics are the cross-sectional area of ​​the sub-area to be welded for the cladding layer and the width of the top of the groove; the swing amplitude dependent characteristics are the width of the bottom of the groove; the swing speed dependent characteristics are the groove angle; the edge residence time dependent characteristics are the height of the cladding layer; and the wire feeding speed dependent characteristics are the volume of the sub-area to be welded corresponding to the cladding layer.

[0139] The advantages of using the above linear activation function regression are: the output range is all real numbers, the fitting result can approximate any real number target; and there is no saturation zone in terms of training stability, and the training will not produce stagnation due to small gradients.

[0140] Step e7, construct the loss function: Where N is the total number of samples, and a structured light image of the groove of the area to be welded and its corresponding image feature information is recorded as a sample, where the value of N is consistent with the total number of structured light images in step b, and j represents the jth of the five welding process parameters, namely, welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; y ij represents the true value of the jth welding process parameter of the i-th sample, that is, the value of the corresponding welding process parameter among the optimal welding process parameters under the corresponding working conditions obtained in step a received in the input layer of the CNN model; represents the predicted value of the jth welding process parameter of the i-th sample output in step e6; j It represents the importance weight of the j-th welding process parameter. For example, the importance weight of welding speed is 0.15, the importance weight of swing amplitude is 0.20, the importance weight of swing speed is 0.25, the importance weight of edge dwell time is 0.30, and the importance weight of wire feeding speed is 0.10.

[0141] In step e8, it is determined whether the loss function is less than a threshold value, for example, 0.05. If so, the training is terminated to obtain a trained CNN model, i.e., the optimal CNN model. If not, the hyperparameters of the CNN model are continuously adjusted until the loss function is less than the threshold value.

[0142] The spatial all-position welding process parameter adaptive welding equipment involved in this application includes a welding carriage, a processing and control system, a sensor system, a welding power supply, a wire feeding system, a flexible guide rail, and a welding gun, which is used to achieve all-position adaptive welding for complex groove working conditions.

[0143] Specifically, the flexible guide rail is used to be assembled on the pipe to be welded along the circumferential direction; the welding carriage is controlled by the processing and control system to move on the flexible guide rail; the sensor system is mounted on the welding carriage, for example, a laser vision sensor is used to obtain a structured light image of the groove of the area to be welded; the welding power supply is used to adjust the welding current and arc voltage used in the welding process; the wire feeding system is used to provide welding wire to the welding gun during the welding process, and is controlled by the processing and control system to change the wire feeding speed; the welding gun is installed on the welding carriage for welding operations on the area to be welded of the pipeline; the processing and control system is used to perform the corresponding operations of steps 3 to 7 of the spatial all-position welding process parameter adaptive welding method involved in this application.

[0144] The present application relates to a welding method and apparatus with adaptive spatial all-position welding process parameters, which obtains several structured light images of the groove of the area to be welded through a laser vision sensor, calculates the image feature information corresponding to each image, records the area between two adjacent structured light images of the groove of the area to be welded as a control area, calculates the volume of the sub-area to be welded corresponding to the currently set cladding layer in each control area, and inputs the structured light image of the groove of the area to be welded, the image feature information, the control distance, and the volume of the sub-area to be welded into a trained CNN model to obtain several sets of welding process parameters. In this way, during the welding process, the corresponding welding process parameters are called for each control area to perform welding to meet the welding requirements of complex groove working conditions.

[0145] The above-mentioned welding method and equipment with adaptive spatial all-position welding process parameters have the following advantages:

[0146] (1) The detected groove structured light image of the welding area is rich in information, and various groove feature information can be extracted during CNN model training;

[0147] (2) After training, the CNN model involved in this application only requires a laser vision sensor during actual welding, without the need for other sensing equipment;

[0148] (3) The CNN model is used to predict welding process parameters in the area to be welded. It has good adaptability to various complex and changeable pipeline welding groove conditions and can be used for welding pipelines of different specifications and positions. It can effectively avoid common welding defects such as lack of fusion, lack of penetration, undercut and weld bead, thereby improving welding quality.

[0149] (4) It reduces the welding interruption and adjustment time caused by changes in groove working conditions, improves the continuity and stability of the welding process, and thus improves the overall welding efficiency.

[0150] The above is only a preferred specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solution and concept of the present application within the technical scope disclosed in the present application, and they should be covered by the scope of protection of the present application.

Claims

1. A welding method with adaptive welding process parameters in all spatial positions, characterized by: The following steps are involved: Step 1: Perform pre-welding cleaning operations; Step 2: Perform pipe jointing and manual bottom welding; Step 3: Control the welding carriage to move 180 degrees from the initial position in a preset welding direction. During the movement, a sensor system installed on the welding carriage scans and obtains a number of groove structured light images of the area to be welded. The number of groove structured light images of the area to be welded is recorded as F. Step 4: The F groove structured light images of the welding area obtained in step 3 are screened according to a preset control distance, and the first and Fth images are retained. After screening, M-2 images are obtained, and a total of M images are obtained. The M images obtained are respectively processed to obtain image feature information corresponding to each image; Step 5: The area between two adjacent groove structured light images of the area to be welded is recorded as the control area, and M-1 control areas are obtained. In any control area, the volume of the sub-area to be welded corresponding to the currently set cladding layer in the control area can be calculated based on the image feature information of the two groove structured light images of the area to be welded corresponding to the control area and the control distance; Step 6: Input the M images, the image feature information corresponding to each image, the control distance, and the volume of the sub-area to be welded corresponding to the M-1 control areas into the trained CNN model to obtain the corresponding M-1 groups of welding process parameters that need to be controlled. The M-1 control areas correspond one-to-one to the M-1 groups of welding process parameters that need to be controlled. Step 7: Return the welding carriage to the initial position, and then control the welding carriage to move 180° along the welding direction preset in step 3. During the movement, when the welding carriage reaches the starting position of each control area, the welding process parameters corresponding to the control area are adopted, including the welding process parameters that need to be regulated obtained in step 6 and the welding process parameters that do not need to be regulated throughout the welding process, until the welding carriage reaches the end position of the control area, and welds the control area using the corresponding welding process parameters until the welding carriage completes the 180° path. Step 8: Control the welding carriage to return to the initial position and determine whether the welding operation is completed along the 360° path of the pipeline. If not, jump to step 9; if yes, end the welding process of the current cladding layer. Step 9: Reversely install the welding carriage in the initial position, and then repeat steps 3 to 7 to achieve welding of the current cladding layer in the area to be welded on the other side of the pipeline at a 180° path. This completes the full-position welding operation of the current cladding layer of the pipeline. Step 10: Repeat steps 3 to 9 to achieve welding of each cladding layer, so as to realize spatial full-position welding operation of the pipeline.

2. The welding method with adaptive spatial all-position welding process parameters according to claim 1, characterized in that: In step 3, the sensor system uses a laser vision sensor.

3. The welding method with adaptive spatial all-position welding process parameters according to claim 1, characterized in that: In step 4, the image feature information corresponding to each image includes: groove feature point position, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and cross-sectional area of ​​the sub-area to be welded for the cladding layer; The process of obtaining the image feature information corresponding to the image is as follows: First, the structured light image of the groove in the area to be welded is binarized to extract the point information that constitutes the line structured light. The horizontal coordinate axis is recorded as the X-axis and the vertical coordinate axis is recorded as the Y-axis. The above points are screened and the point at the far left of the image is recorded as point St, whose X coordinate is the smallest; the point at the far right of the image is recorded as point En, whose X coordinate is the largest; point C1 is the point with the smallest X coordinate when the Y coordinate is the smallest among all the line structured light points, that is, the second inflection point on the left of the groove; point C2 is the point with the largest X coordinate when the Y coordinate is the smallest among all the line structured light points, that is, the second inflection point on the right of the groove. The X and Y coordinates of the above points St, En, C1, and C2 are all known; Calculate the first turning point on the left of the groove, that is, point A, and the first turning point on the right of the groove, that is, point B. The calculation process is as follows: Construct the straight line equation StC1: (YY St )(X c1 -X St )-(XX St )(Y c1 -Y St )=0, where Y St Indicates the Y coordinate of point St, X c1 Indicates the X coordinate of point C1, X St Indicates the X coordinate of point St, Y c1 Indicates the Y coordinate of point C1; Construct the straight line equation EnC2: (YY En )(X c2 -X En )-(XX En )(Y c2 -Y En )=0, where Y En Indicates the Y coordinate of point En, X c2 Indicates the X coordinate of point C2, X En Indicates the X coordinate of point En, Y c2 Indicates the Y coordinate of point C2; Calculate the straight-line distance from the point of line structured light to the straight lines StC1 and EnC2: Where p represents any point in the line structured light except St, En, C1 and C2. x represents the X coordinate of point p, p y represents the Y coordinate of point p, a, b, c are the coefficients of the straight line equation; Among the points whose X coordinates are less than the X coordinates of point C1, the point farthest from StC1 is defined as point A based on the straight-line distance, i.e., the first left inflection point of the groove. Among the points whose X coordinates are greater than the X coordinates of point C2, the point farthest from EnC2 is defined as point B based on the straight-line distance, i.e., the first right inflection point of the groove. So far, the four characteristic points of the groove have been obtained, namely point A, point B, point C1, and point C2, and the position information (X, Y) of the four characteristic points of the groove are also known; The calculation process of the groove bottom width is as follows: Among them, X c1 Indicates the X coordinate of point C1, X c2 Indicates the X coordinate of point C2, Y c1 Indicates the Y coordinate of point C1, Y c2 Indicates the Y coordinate of point C2; The calculation process of the groove top width is as follows: Among them, X A Indicates the X coordinate of point A, X B Indicates the X coordinate of point B, Y A Indicates the Y coordinate of point A, Y B Indicates the Y coordinate of point B; The calculation process of the groove angle is as follows: Among them, θ1 is the angle between the straight line AC1 and the Y axis, X A Indicates the X coordinate of point A, X c1 Indicates the X coordinate of point C1, Y A Indicates the Y coordinate of point A, Y c1 Indicates the Y coordinate of point C1; Among them, θ2 is the angle between the straight line BC2 and the Y axis, X B Indicates the X coordinate of point B, X c2 Indicates the X coordinate of point C2, Y B Indicates the Y coordinate of point B, Y c2 Indicates the Y coordinate of point C2; θ=θ1+θ2, where θ is the groove angle; The calculation process of the groove area is as follows: Among them, S is the groove area formed by ABC1C2; Cladding layer height: h represents the height of the cladding layer required for a single weld, in mm. The intersection of the cladding layer and the groove boundary AC1 is marked as point A', and the intersection of the cladding layer and the groove boundary BC2 is marked as point B'. The position information (X, Y) of points A' and B' can be known. The calculation process of the cross-sectional area of ​​the sub-region to be welded for the cladding layer is as follows: , where S' is the cross-sectional area of ​​the sub-region to be welded of the cladding layer formed by A'B'C1C2, and Y A′ Indicates the X coordinate of point A', Y B′ represents the Y coordinate of point B', X B′ Indicates the X coordinate of point B', Y A′ Indicates the Y coordinate of point A'.

4. The welding method with adaptive spatial all-position welding process parameters according to claim 3, characterized in that: In step 5, the preset control distance is set to L control , which represents the distance between each two adjacent groove structured light images of the area to be welded, that is, the distance the welding carriage travels through a control area, in mm; Among them, V app represents the volume of the sub-area to be welded corresponding to the cladding layer in the control area, S′ start Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the starting image corresponding to the control area, S′ end Indicates the cross-sectional area of ​​the sub-region to be welded of the cladding layer in the ending image corresponding to the control area.

5. The welding method with adaptive spatial all-position welding process parameters according to claim 4, characterized in that: In step 6, the welding process parameters that need to be adjusted include welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed.

6. The welding method with adaptive spatial all-position welding process parameters according to claim 5, characterized in that: Before step 6, the following steps are also included to train the CNN model: Step a: determining optimal welding process parameters corresponding to the groove of the area to be welded under different preset working conditions, wherein under one preset working condition, each cross-section of the groove of the area to be welded is the same; the welding process parameters are the welding process parameters that need to be adjusted; Step b, using a sensor system to capture structured light images of the groove of the area to be welded under different preset working conditions, and only capturing one structured light image of the groove of the area to be welded for each working condition; Step c, performing image processing on each structured light image of the groove of the area to be welded captured in step b to obtain image feature information corresponding to each image: groove feature point position, groove bottom width, groove top width, groove angle, groove area, cladding layer height, and cross-sectional area of ​​the sub-area to be welded of the cladding layer; Step d: Based on the preset control distance and the image feature information corresponding to each image obtained in step c, calculate the volume of the sub-area to be welded corresponding to the cladding layer under each working condition: V train =S′×L control , where L control represents the control distance, S′ represents the cross-sectional area of ​​the sub-region to be welded of the cladding layer under the corresponding working condition obtained in step c; Step e: The optimal welding process parameters corresponding to the groove of the to-be-welded area under different preset working conditions obtained in step a, the structured light images of the groove of the to-be-welded area under different preset working conditions captured in step b, the image feature information corresponding to each image obtained in step c, the control distance, and the volume of the sub-area to be welded corresponding to the cladding layer under each working condition obtained in step d are input into a convolutional neural network (CNN) model for training to obtain the optimal CNN model after training.

7. The welding method with adaptive spatial all-position welding process parameters according to claim 6, characterized in that: The step e further comprises the following steps: Step e1: The structured light images of the groove of the to-be-welded area under different preset working conditions received by the CNN model input layer are classified into an image branch; the image feature information corresponding to each image obtained in step c and the control distance received by the input layer, as well as the volume of the to-be-welded sub-area corresponding to the cladding layer under each working condition obtained in step d, are classified into a structured feature branch; Step e2: performing a convolution operation on the image branch to obtain a local feature vector of the image; Step e3: Use a fully connected layer to perform structured learning on the structured feature branch, introduce nonlinearity into the feature vectors in the structured feature branch, capture the high-order relationship between the feature vectors, and then compress the dimension of the feature vector to avoid overfitting; Step e4: Using the feature fusion layer in the CNN model, the local feature vector of the image obtained in step e2 and the feature vector output in step e3 are concatenated in the channel dimension, and the concatenated feature vector is then output. Step e5: The feature vector processed by the feature fusion layer is processed by a fully connected layer, and the dimension is compressed using the ReLU activation function before output; Step e6: Output the predicted values ​​of the welding process parameters by linear regression through the regression output layer of the CNN model; Step e7, construct the loss function: Where N is the total number of samples, and a structured light image of the groove of the area to be welded and its corresponding image feature information is recorded as a sample, where the value of N is consistent with the total number of structured light images in step b, and j represents the jth of the five welding process parameters, namely, welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; y ij represents the true value of the jth welding process parameter of the i-th sample, that is, the value of the corresponding welding process parameter among the optimal welding process parameters under the corresponding working conditions obtained in step a received in the input layer of the CNN model; represents the predicted value of the jth welding process parameter of the i-th sample output in step e6; j Represents the importance weight of the jth welding process parameter; In step e8, it is determined whether the loss function is less than the threshold. If so, the training is terminated to obtain the trained CNN model, i.e., the optimal CNN model. If not, the hyperparameters of the CNN model are continuously adjusted until the loss function is less than the threshold.

8. The welding method with adaptive spatial all-position welding process parameters according to claim 7, characterized in that: In step e6, in the regression output layer, the following formula is used: in, The five-dimensional vector representing the predicted welding process parameters output includes welding speed, swing amplitude, swing speed, edge dwell time, and wire feeding speed; W out The matrix dimension is 5xK, where K is the dimension of the input feature vector of the regression output layer, and W out It is the weight matrix of the regression output layer parameters, which can map the high-dimensional feature space to the 5-dimensional welding process parameter space; X in represents the input feature vector, which is also the feature vector output by step e5; b out Represents the regression output layer bias vector, which represents the reference value of welding process parameters.

9. A welding device with adaptive welding process parameters in all spatial positions, characterized by: Includes welding carriage, processing and control system, sensor system, welding power source, wire feeding system, flexible guide rail, and welding gun; In which, the flexible guide rail is used to be assembled on the pipe to be welded along the circumferential direction; the welding carriage is controlled by the processing and control system to move on the flexible guide rail; the sensor system is mounted on the welding carriage to obtain the structured light image of the groove of the area to be welded; the welding power supply is used to adjust the welding current and arc voltage used in the welding process; the wire feeding system is used to provide welding wire to the welding gun during the welding process, and is controlled by the processing and control system to change the wire feeding speed; the welding gun is installed on the welding carriage to perform welding operations on the area to be welded of the pipeline; the processing and control system is used to perform the corresponding operations of steps 3 to 7 of the welding method with adaptive spatial all-position welding process parameters as described in claim 1.

10. The welding equipment with adaptive spatial all-position welding process parameters according to claim 9, characterized in that: The sensor system adopts a laser vision sensor.

Images