CN115041771B

CN115041771B - Automatic advancing pipeline welding and cutting integrated machining equipment and control method thereof

Info

Publication number: CN115041771B
Application number: CN202210787490.9A
Authority: CN
Inventors: 潘小云
Original assignee: Huawei Welding And Cutting Technology Zhejiang Co ltd
Current assignee: Huawei Welding And Cutting Technology Zhejiang Co ltd
Priority date: 2022-07-04
Filing date: 2022-07-04
Publication date: 2023-03-31
Anticipated expiration: 2042-07-04
Also published as: CN115041771A

Abstract

The application relates to the field of intelligent processing of engineering pipelines, and particularly discloses an automatic advancing pipeline welding and cutting integrated processing device and a control method thereof.

Description

Automatic advancing pipeline welding and cutting integrated machining equipment and control method thereof

Technical Field

The present invention relates to the field of intelligent processing of engineering pipes, and more particularly, to an automatic-advancing pipe welding and cutting integrated processing apparatus and a control method thereof.

Background

The pipeline beveling machine is a special tool for chamfering the end face of a pipeline or a flat plate before welding, overcomes the defects of irregular angle, rough slope surface, high working noise and the like of operation processes such as flame cutting, grinding by a polishing machine and the like, and has the advantages of simplicity and convenience in operation, standard angle, smooth surface and the like.

When the pipe cutting machine is used for welding the groove welding all-in-one machine, the welding gun lifting assembly and the wobbler assembly are required to be controlled to control the wobbling angle of the welding gun and the distance between the welding gun and the workpiece. However, in actual conditions, the all-in-one machine may work on pipes with uneven surfaces, and the surfaces of the pipes are poor in state, so that the control of the welding gun lifting assembly and the welding gun swing assembly is delayed, and the welding effect is poor. Therefore, an optimized pipe cutter weld groove all-in-one machine is desired to accurately control the welding gun lifting assembly and the wobbler assembly to improve the welding effect.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. Embodiments of the present application provide an automatic traveling pipe welding and cutting integrated processing apparatus and a control method thereof, which adaptively adjust a swing angle of a welding torch and a distance between the welding torch and a workpiece based on a situation of a current welding region and a situation of an unwelded pipe surface by using a deep neural network model of an artificial intelligence technique so that a controlled parameter can be adapted to a surface characteristic of the pipe surface to improve welding quality.

According to an aspect of the present application, there is provided an automatic traveling pipe welding and cutting integrated processing apparatus, including: the processing state acquisition module is used for acquiring a processing image of the pipeline surface with the welding area, which is acquired by the camera; the machining region coding module is used for enabling the machining image of the pipeline surface to pass through a plurality of convolution layers to obtain a convolution characteristic diagram; the region-of-interest detection module is used for enabling the convolution feature map to pass through a first convolution neural network model serving as a welding region detection network so as to obtain a welding region-of-interest from the convolution feature map; the characteristic distribution correction module is used for carrying out global characteristic value correction based on the convolution characteristic diagram on the welding interesting region to obtain a corrected welding interesting region; the control parameter module is used for acquiring control parameters of a plurality of preset time points including the current time point in a preset time period, and the control parameters comprise the distance between the welding gun and the workpiece and the swing angle of the welding gun; the control parameter structuring module is used for respectively constructing the distances and the swing angles in the control parameters of a plurality of preset time points including the current time point as a first input vector and a second input vector; the control parameter correlation module is used for calculating a control parameter correlation matrix between the first input vector and the second input vector; the control parameter association coding module is used for enabling the control parameter association matrix to pass through a second convolutional neural network to obtain a control parameter characteristic diagram; the characteristic fusion module is used for fusing the control parameter characteristic diagram and the corrected welding interesting area to obtain a decoding characteristic diagram; and the parameter control module is used for decoding and regressing the decoding characteristic diagram through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point.

In the above automatic advancing pipe welding and cutting integrated processing apparatus, the characteristic distribution correcting module includes: a layout feature representation unit, configured to calculate a logarithmic function value of a sum of a feature value and one at each position in the welding interest region as a local feature representation at each position in the welding interest region; a global feature representation unit, configured to calculate a logarithmic function value of a sum of the feature values of all positions in the convolution feature map and a sum of one as a global feature representation of the convolution feature map; and the correction unit is used for dividing the local feature representation of each position in the welding interest region by the global feature representation of the convolution feature map respectively to obtain a corrected welding interest region as a corrected feature value of each position in the welding interest region.

In the above automatic traveling pipe welding and cutting integrated processing apparatus, the control parameter association module includes: a correlation unit, configured to calculate a product between the first input vector and a transposed vector of the second input vector to obtain an initial correlation matrix; the constraint quantity calculating unit is used for calculating the Frobenius norm of the initial incidence matrix; and the constraint unit is used for dividing the characteristic value of each position in the initial correlation matrix by the Frobenius norm of the initial correlation matrix to obtain the control parameter correlation matrix.

In the above automatic advancing pipe welding and cutting integrated processing apparatus, the control parameter association coding module is further configured to perform convolution processing, pooling processing, and nonlinear activation processing on input data in forward transmission of layers using the layers of the second convolutional neural network, respectively, to output the control parameter feature map from the last layer of the second convolutional neural network, where an input of the first layer of the second convolutional neural network is the control parameter association matrix.

In the above-mentioned automatic pipeline welding of advancing and cutting integral type processing equipment, the feature fuses the module, includes: the size adjusting unit is used for adjusting the control parameter characteristic diagram to be the same as the corrected welding interesting area in size through linear transformation; the position-based weighting unit is used for fusing the control parameter characteristic diagram and the corrected welding interesting region according to the following formula to obtain the decoding characteristic diagram; wherein the formula is:

F＝αF ₁ +βF ₂

wherein F is the decoding characteristicFIG. F ₁ For the control parameter profile, F ₂ For the corrected weld region of interest, "+" indicates the addition of elements at the corresponding locations of the control parameter profile and the corrected weld region of interest, α and β are weighting parameters for controlling the balance between the control parameter profile and the corrected weld region of interest.

In the above automatic advancing pipe welding and cutting integrated processing apparatus, the parameter control module is further configured to: decoding the decoded feature map using the decoder to perform a decoding regression to obtain the first decoded value and the second decoded value; wherein the formula is:

wherein X is a regression matrix, Y is a decoded value, W is a weight matrix, and>

representing a matrix multiplication.

In the above automatic traveling pipe welding and cutting integrated processing apparatus, the first convolutional neural network model is centret, extreme net, or RepPoints.

According to another aspect of the present application, a control method of an automatic traveling pipe welding and cutting integrated processing apparatus, includes: acquiring a machining image of a pipeline surface with a welding area, which is acquired by a camera; passing the machined image of the pipeline surface through a multilayer convolution layer to obtain a convolution characteristic diagram; passing the convolved signature through a first convolved neural network model as a weld region detection network to obtain a weld region of interest from the convolved signature; performing global characteristic value correction based on the convolution characteristic diagram on the welding interesting region to obtain a corrected welding interesting region; acquiring control parameters of a plurality of preset time points including a current time point in a preset time period, wherein the control parameters comprise the distance between a welding gun and a workpiece and the swing angle of the welding gun; constructing the distances and the swing angles in the control parameters of the plurality of preset time points including the current time point as a first input vector and a second input vector respectively; calculating a control parameter association matrix between the first input vector and the second input vector; enabling the control parameter incidence matrix to pass through a second convolutional neural network to obtain a control parameter characteristic diagram; fusing the control parameter characteristic diagram and the corrected welding interesting area to obtain a decoding characteristic diagram; and performing decoding regression on the decoding characteristic graph through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point.

In the above method for controlling an automatic traveling pipe welding and cutting integrated machining apparatus, performing global feature value correction based on the convolution feature map on the welding region of interest to obtain a corrected welding region of interest includes: calculating a logarithmic function value of a sum of the eigenvalue and one of each position in the welding interest region as a local characteristic representation of each position in the welding interest region; calculating a logarithmic function value of a summation value of the feature values of all positions in the convolution feature map and a summation value of one as a global feature representation of the convolution feature map; dividing the local feature representation of each position in the welding interest region by the global feature representation of the convolution feature map respectively to obtain a corrected feature value of each position in the welding interest region.

In the above control method of the automatic traveling pipe welding and cutting integrated processing apparatus, calculating a control parameter association matrix between the first input vector and the second input vector includes: calculating a product between the first input vector and a transposed vector of the second input vector to obtain an initial correlation matrix; calculating the Frobenius norm of the initial incidence matrix; and dividing the characteristic value of each position in the initial correlation matrix by the Frobenius norm of the initial correlation matrix to obtain the control parameter correlation matrix.

In the above method for controlling an automatic traveling pipeline welding and cutting integrated machining device, passing the control parameter association matrix through a second convolutional neural network to obtain a control parameter characteristic diagram, the method includes: and respectively performing convolution processing, pooling processing and nonlinear activation processing on input data in forward transmission of layers by using each layer of the second convolutional neural network to output the control parameter characteristic diagram from the last layer of the second convolutional neural network, wherein the input of the first layer of the second convolutional neural network is the control parameter correlation matrix.

In the above method for controlling an automatic advancing pipe welding and cutting integrated processing apparatus, fusing the control parameter feature map and the corrected welding region of interest to obtain a decoded feature map, the method includes: adjusting the control parameter characteristic diagram to have the same size as the corrected welding interesting area through linear transformation; fusing the control parameter characteristic diagram and the corrected welding interesting area according to the following formula to obtain the decoding characteristic diagram; wherein the formula is:

F＝αF ₁ +βF ₂

wherein F is the decoding characteristic diagram, F ₁ For the control parameter profile, F ₂ For the corrected weld region of interest, "+" indicates the addition of elements at the corresponding locations of the control parameter profile and the corrected weld region of interest, α and β are weighting parameters for controlling the balance between the control parameter profile and the corrected weld region of interest.

In the above method for controlling an automatic traveling pipeline welding and cutting integrated processing device, the decoding regression of the decoded feature map by a decoder to obtain a first decoded value and a second decoded value includes: decoding the decoded feature map using the decoder to perform a decoding regression to obtain the first decoded value and the second decoded value; wherein the formula is:

where X is a regression matrix, Y is the decoded value, W is a weight matrix, and>

representing a matrix multiplication.

In the above method for controlling an automatic traveling pipe welding and cutting integrated machining device, the first convolutional neural network model is centret, extremeNet or RepPoints.

Compared with the prior art, the automatic advancing pipeline welding and cutting integrated processing equipment and the control method thereof have the advantages that the rocking angle of the welding gun and the distance between the welding gun and a workpiece are adaptively adjusted based on the condition of the current welding area and the condition of the unwelded pipeline surface by utilizing the deep neural network model of the artificial intelligence technology, so that the controlled parameters can be adapted to the surface characteristics of the pipeline surface, and the welding quality is improved.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.

Fig. 1A is one of schematic structural views of an automated traveling pipeline welding and cutting integrated processing device according to an embodiment of the present application.

Fig. 1B is one of the schematic structural views of an automated pipe welding and cutting integrated processing apparatus according to an embodiment of the present application.

Fig. 1C is one of the schematic structural views of an automated traveling pipeline welding and cutting integrated processing device according to an embodiment of the present application.

Fig. 2 is an application scenario diagram of an automatic traveling pipeline welding and cutting integrated processing device according to an embodiment of the application.

FIG. 3 is a block diagram of an automated traveling pipe welding and cutting integrated processing equipment according to an embodiment of the present application.

FIG. 4 is a block diagram of a feature distribution correction module in an automated pipe welding and cutting integrated processing tool according to an embodiment of the present application.

FIG. 5 is a block diagram of a control parameter association module in an automated guided-tube welding and cutting integrated processing tool in accordance with an embodiment of the present application.

Fig. 6 is a flowchart of a control method of an automated traveling pipe welding and cutting integrated processing device according to an embodiment of the present application.

Fig. 7 is a schematic configuration diagram of a control method of an automatic traveling pipe welding and cutting integrated processing device according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

Overview of scenes

As mentioned above, the pipeline beveling machine is a special tool for chamfering the end face of a pipeline or a flat plate before welding, overcomes the defects of irregular angle, rough slope surface, high working noise and the like of operation processes such as flame cutting, grinding by a polishing machine and the like, and has the advantages of simple and convenient operation, standard angle, smooth surface and the like.

When the pipe cutting machine is used for welding the groove welding all-in-one machine, the welding gun lifting assembly and the wobbler assembly are required to be controlled to control the wobbling angle of the welding gun and the distance between the welding gun and the workpiece. However, in actual working conditions, the all-in-one machine may work on pipes with uneven surfaces, and the surfaces of the pipes are poor, so that the control of the welding gun lifting assembly and the welding gun swing assembly is delayed, and the welding effect is poor. Therefore, an optimized pipe cutter welding groove all-in-one machine is desired to accurately control the welding gun lifting assembly and the wobbler assembly to improve the welding effect.

Correspondingly, in the technical solution of the present application, as shown in three schematic structural views of fig. 1A, 1B and 1C, the automatic advancing pipe welding and cutting integrated processing equipment comprises: the cutting torch comprises a gas circuit assembly 1, a control box assembly 2, a side wall clamping wrench 3, a host machine assembly 4, a steel belt locking assembly 5, a steel belt loosening/tightening handle 6, an angle adjusting screw 7, a roller groove 8, a cutting torch lifting assembly 9, a cutting torch assembly 10 and a steel pipe 11; a traverse assembly 12, a battery assembly 13, a movable side wall 14, a control box assembly 15, a torch lift assembly 16, a wobbler assembly 17, and a torch clamping assembly 18.

Before the automatic advancing pipeline welding and cutting integrated processing equipment carries out corresponding cutting and welding operation, corresponding assembly devices need to be installed, specifically, a steel belt loosening/tightening handle is lifted up firstly to quickly install a steel belt assembly so as to improve the working efficiency, and after the steel belt assembly is installed on a proper station, the steel belt is reversely operated and locked. Then, the side arm clamping wrench is lifted up to place the trolley on the steel strip, so that the roller groove I is meshed with the steel strip, and the side arm clamping wrench is operated reversely to fix the trolley on the steel strip. Then, when the trolley is fixed to the steel belt and the roller wheel is found not to be in positive angle engagement with the steel belt, the angle adjusting screw is loosened by using a 5mm hexagon socket wrench, and the side arm is adjusted to a proper angle. And locking the screws reversely.

Wherein, cutting torch lifting unit: used for controlling the distance between the cutting nozzle and the workpiece. A transverse moving component: used for adjusting the transverse distance between the cutting torch and the workpiece. A battery pack: used for providing power for the operation of the trolley. Control box subassembly: used for controlling the working speed and direction of the trolley. Welding gun lifting unit: used for controlling the distance between the weldment and the workpiece. A rocker assembly: used for controlling the swing angle of the welding gun.

Particularly, in the operation process of the automatic traveling pipeline welding and cutting integrated processing equipment, because the integrated machine can work on the pipe with an uneven surface, the control of the welding gun lifting assembly and the rocker assembly is delayed due to poor surface state of the pipe, and the welding effect is poor. Therefore, an optimized pipe cutter welding groove all-in-one machine is desired to accurately control the welding gun lifting assembly and the wobbler assembly to improve the welding effect.

Accordingly, the inventors of the present application desire to adaptively adjust the swing angle of the torch and the distance between the torch and the workpiece based on the condition of the current welding region and the condition of the unwelded pipe face to enable the controlled parameters to be adapted to the surface characteristics of the pipe face to improve the welding quality. Specifically, the current welding area condition and the unwelded pipe surface condition can be represented by using machining image features of the pipe surface of the welding area, deep feature mining is carried out on the pipe surface by using a convolutional neural network model with excellent performance in the aspect of local implicit associated feature extraction, and the parameter value to be adjusted is obtained through decoding regression, so that the welding quality is effectively and accurately improved.

Specifically, in the technical scheme of the application, a machining image of a pipeline surface with a welding area is collected through a camera. Then, the machined image of the pipeline surface is processed by a multilayer convolution layer to extract local high-dimensional feature information in the machined image of the pipeline surface, so that a convolution feature map is obtained.

Further, it should be understood that, since the characteristics of the welding region and the non-welded and welded joint surface region should be more focused in the machining image of the pipeline surface, the convolution feature map is subjected to feature extraction in the first convolution neural network model as the welding region detection network so as to focus more on the hidden characteristics of the welded region in the deep excavation of the features, thereby obtaining the welding region of interest from the convolution feature map.

In particular, here, the first convolutional neural network model is centret, extremeNet or repoints. It should be understood that the target detection method based on deep learning divides the network into two categories, anchor-based (Anchor-based) and Anchor-free (Anchor-free), according to whether an Anchor window is used in the network. Anchor window based methods such as Fast R-CNN, retinaNet, etc., anchor window based methods such as CenterNet, extremeNet, rePoints, etc. The method without the anchor window solves the defects that the target with large scale change is difficult to identify, positive and negative samples are unbalanced in the training process, and the memory is occupied excessively, and the like, which are caused by the anchor window, and is the current mainstream development direction.

The anchor-free window method is subdivided into two categories, namely a center point-based method and a key point-based method. Center point-based methods such as YOLOv1, FCOS, centrnet, etc. directly detect the center point of the target and then regress the boundary information of the target. The bounding box is obtained by predicting the key points of the target based on the key point methods such as CornerNet, extremeNet, rePoints and the like. The method based on the key points is usually slightly higher in detection accuracy than the method based on the center point, but is more expensive in calculation amount.

It should be understood that when welding control is performed by the integrated processing apparatus, it is advantageous to improve the accuracy of welding control if image information of an unwelded area can be combined. That is, feature distribution correction is performed on the welding region-of-interest based on the whole of the convolution feature map (namely, the whole image information of the pipeline surface to be machined, including the welding region and the unwelded region) so as to fuse the information of the unwelded region of the pipeline surface to be machined into the welding region, so that the subsequent welding control has certain foresight and prejudice to improve the welding quality.

Therefore, the welding region of interest is further subjected to global eigenvalue correction based on the convolution signature to obtain a corrected welding region of interest, that is:

wherein, the first and the second end of the pipe are connected with each other,

characteristic values representing various positions of the welding region of interest, based on the characteristic values>

Characteristic values representing respective positions of the corrected welding region of interest, based on the correction value>

And summing the feature values representing the respective positions of the convolved feature maps.

Considering that if one wants to adaptively adjust the swing angle of the welding torch and the distance between the welding torch and the workpiece based on the condition of the current welding area and the condition of the unwelded pipe surface so that the controlled parameters can be adapted to the surface characteristics of the pipe surface to improve the welding quality, it is also necessary to further acquire control parameters at a plurality of predetermined time points including the current time point within a predetermined time period, the control parameters including the distance between the welding torch and the workpiece and the swing angle of the welding torch. Then, the distances and the swing angles in the control parameters of the plurality of preset time points including the current time point are respectively constructed into a first input vector and a second input vector so as to carry out feature relevance mining on the first input vector and the second input vector.

It should be understood that, when calculating the product between the first input vector and the transpose of the second input vector to obtain the control parameter association matrix, since the first input vector and the second input vector correspond to the time-series association features of distance and angle, respectively, when correlating them to obtain the parameter association matrix, there may be a problem that the parameter association matrix is shifted in probability distribution due to the difference in scale of the first input vector and the second input vector.

Thus, for the first input vector V ₁ And a second input vector V ₂ Performing scale migration certainty-based feature vector association, expressed as:

wherein V ₁ Representing said first input vector, V ₂ Representing said second input vector, V ₁ And V ₂ Are all in the form of column vectors, and | · | | | luminance _F The Frobenius norm, exp (·), representing the matrix represents the exponential operation of the matrix, which represents the calculation of a natural exponential function value raised to the eigenvalues of the various positions in the matrix.

In this way, the feature vector association based on the scale migration certainty performs low rank constraint on the embedding of the feature vector based on the relative position of the associated feature, so that the long-range dependency relationship of the high-dimensional feature expressed by the feature vector on the associated feature under the scale migration during association is retained, and thus, the consistency of the probability distribution of the parameter association matrix and the first input vector and the second input vector can be ensured to some extent.

Therefore, the control parameter incidence matrix can be obtained, and then the control parameter incidence matrix is processed through a second convolutional neural network to extract high-dimensional incidence implicit characteristic information of the control parameters, so that a control parameter characteristic diagram is obtained.

Further, after the control parameter feature map is adjusted to have the same size as the corrected welding region of interest through linear transformation, the control parameter feature map and the corrected welding region of interest may be fused to obtain a decoding feature map. Then, the decoding regression of the decoding characteristic diagram is carried out through a decoder to obtain a first decoding value used for representing the distance between the welding gun and the workpiece at the next time point and a second decoding value used for representing the swing angle of the welding gun at the next time point.

Based on this, the application provides an automatic pipeline welding of advancing and cutting integral type processing equipment, it includes: the processing state acquisition module is used for acquiring a processing image of the pipeline surface with the welding area, which is acquired by the camera; the machining region coding module is used for enabling the machining image of the pipeline surface to pass through a plurality of convolution layers to obtain a convolution characteristic diagram; the region-of-interest detection module is used for enabling the convolution feature map to pass through a first convolution neural network model serving as a welding region detection network so as to obtain a welding region-of-interest from the convolution feature map; the characteristic distribution correction module is used for carrying out global characteristic value correction based on the convolution characteristic diagram on the welding interesting region to obtain a corrected welding interesting region; the control parameter module is used for acquiring control parameters of a plurality of preset time points including the current time point in a preset time period, wherein the control parameters comprise the distance between a welding gun and a workpiece and the swing angle of the welding gun; the control parameter structuring module is used for respectively constructing the distances and the swing angles in the control parameters of a plurality of preset time points including the current time point as a first input vector and a second input vector; the control parameter correlation module is used for calculating a control parameter correlation matrix between the first input vector and the second input vector; the control parameter association coding module is used for enabling the control parameter association matrix to pass through a second convolutional neural network to obtain a control parameter characteristic diagram; the characteristic fusion module is used for fusing the control parameter characteristic diagram and the corrected welding interesting area to obtain a decoding characteristic diagram; and the parameter control module is used for decoding and regressing the decoding characteristic diagram through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point.

Fig. 2 illustrates an application scenario of an automated traveling pipeline welding and cutting integrated processing device according to an embodiment of the present application. As shown in fig. 2, in this application scenario, first, a machining image of a face of a pipe (e.g., N as illustrated in fig. 2) having a welding area is acquired by a camera (e.g., C as illustrated in fig. 2) disposed at an automatic travel pipe welding and cutting integrated machining apparatus (e.g., P as illustrated in fig. 2), and control parameters including a distance between a welding gun and a workpiece and a swing angle of the welding gun at a plurality of predetermined time points including a current time point within a predetermined time period are acquired by sensors (e.g., a distance sensor T1 and an angle sensor T2 as illustrated in fig. 2). Then, the obtained machined image of the pipe surface having the welding region and the control parameters of the plurality of predetermined time points are input into a server (for example, a cloud server S as illustrated in fig. 2) in which an automatic traveling pipe welding and cutting integrated machining apparatus algorithm is deployed, wherein the server can process the machined image of the pipe surface having the welding region and the control parameters of the plurality of predetermined time points with the automatic traveling pipe welding and cutting integrated machining apparatus algorithm to generate a first decoded value representing a distance between the welding gun and the workpiece at a next time point and a second decoded value representing a swing angle of the welding gun at the next time point.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary System

FIG. 3 illustrates a block diagram of an automated traveling pipe welding and cutting integrated processing equipment according to an embodiment of the present application. As shown in fig. 3, the automatic traveling pipe welding and cutting integrated machining apparatus 200 according to the embodiment of the present application includes: a machining state acquisition module 210 for acquiring a machining image of a pipe surface having a welding region acquired by a camera; a processing region encoding module 220, configured to pass the processed image of the pipeline surface through a plurality of convolutional layers to obtain a convolutional feature map; a region-of-interest detection module 230, configured to pass the convolution feature map through a first convolution neural network model as a welding region detection network to obtain a welding region-of-interest from the convolution feature map; a feature distribution correction module 240, configured to perform global feature value correction based on the convolution feature map on the welding region of interest to obtain a corrected welding region of interest; a control parameter module 250, configured to obtain control parameters of multiple predetermined time points within a predetermined time period, where the multiple predetermined time points include a current time point, where the control parameters include a distance between a welding gun and a workpiece and a swing angle of the welding gun; a control parameter structuring module 260, configured to construct distances and swing angles in the control parameters of the plurality of predetermined time points including the current time point as a first input vector and a second input vector, respectively; a control parameter association module 270, configured to calculate a control parameter association matrix between the first input vector and the second input vector; the control parameter association coding module 280 is configured to pass the control parameter association matrix through a second convolutional neural network to obtain a control parameter feature map; a feature fusion module 290, configured to fuse the control parameter feature map and the corrected welding region of interest to obtain a decoded feature map; and the parameter control module 300 is configured to perform decoding regression on the decoded feature map through a decoder to obtain a first decoded value and a second decoded value, where the first decoded value is a distance between the welding gun and the workpiece at a next time point, and the second decoded value is a swing angle of the welding gun at the next time point.

Specifically, in this embodiment of the present application, the machining state acquisition module 210 and the machining region encoding module 220 are configured to acquire a machining image of a pipe surface with a welding region acquired by a camera, and pass the machining image of the pipe surface through multiple layers of convolution layers to obtain a convolution signature. As described above, in the technical solution of the present application, it is desirable to adaptively adjust the rocking angle of the torch and the distance between the torch and the workpiece based on the condition of the current welding region and the condition of the unwelded pipe face so that the controlled parameters can be adapted to the surface characteristics of the pipe face to improve the welding quality. Accordingly, in the technical scheme of the application, the situation of the current welding area and the situation of the unwelded pipeline surface can be represented by using the machining image features of the pipeline surface of the welding area, deep-level feature mining is performed on the machining image features by using a convolutional neural network model which has excellent performance in the aspect of local implicit associated feature extraction, and the parameter values which are to be adjusted are obtained through decoding regression, so that the welding quality is effectively and accurately improved.

That is, specifically, in the technical solution of the present application, a processed image of a pipe surface having a welding region is first acquired by a camera disposed at an automatic traveling pipe welding and cutting integrated processing apparatus. Then, the machined image of the pipeline surface passes through a plurality of convolution layers to extract local high-dimensional feature information in the machined image of the pipeline surface, and therefore a convolution feature map is obtained.

Specifically, in this embodiment, the region-of-interest detecting module 230 is configured to pass the convolution feature map through a first convolution neural network model as a welding region detection network to obtain a welding region-of-interest from the convolution feature map. It should be understood that, since the machined image of the pipe surface should focus more on the characteristics of the welded region and the region of the non-welded and welded joint surface, in the technical solution of the present application, the convolution feature map is further subjected to feature extraction in the first convolution neural network model as the welding region detection network, so as to focus more on the implicit characteristics of the welded region in the deep excavation of the features, thereby obtaining the welding region of interest from the convolution feature map.

In particular, here, the first convolutional neural network model is centret, extremeNet or repoints. It should be understood that the target detection method based on deep learning divides the network into two categories, anchor-based (Anchor-based) and Anchor-free (Anchor-free) according to whether an Anchor window is used in the network. Anchor window based methods such as Fast R-CNN, faster R-CNN, retinaNet, etc., and anchorless window based methods such as CenterNet, extremNet, rePoints, etc. The method without the anchor window solves the defects that the target with large scale change is difficult to identify, positive and negative samples are unbalanced in the training process, and the memory is occupied excessively, and the like, which are caused by the anchor window, and is the current mainstream development direction.

The anchor-free window method is subdivided into two categories, namely a center point-based method and a key point-based method. Center point-based methods such as YOLOv1, FCOS, centrnet, etc. directly detect the center point of the target and then regress the boundary information of the target. The bounding box is obtained by predicting the key points of the target based on the key point methods such as CornerNet, extremeNet, rePoints and the like. The keypoint-based method is generally slightly higher in detection accuracy than the center-point-based method, but is more computationally expensive.

Specifically, in this embodiment of the present application, the feature distribution correction module 240 is configured to perform global feature value correction on the welding region of interest based on the convolution feature map to obtain a corrected welding region of interest. It should be understood that when welding control is performed by the integrated processing apparatus, it is advantageous to improve the accuracy of the welding control if the image information of the non-welding region can be combined. That is, feature distribution correction is carried out on the welding region-of-interest based on the whole situation of the convolution feature map (namely, the whole situation image information of the pipeline surface to be machined, including a welding region and a non-welding region) so as to fuse the information of the non-welding region of the pipeline surface to be machined into the welding region, so that the subsequent welding control has certain foresight and predictability, and the welding quality is improved. Therefore, in the technical solution of the present application, a global feature value correction based on the convolution feature map is further performed on the welding region of interest to obtain a corrected welding region of interest.

More specifically, in this embodiment of the present application, the feature distribution correction module includes: first, a logarithmic function value of a sum of the eigenvalue and one of each position in the welding interest region is calculated as a local eigen representation of each position in the welding interest region. Then, a logarithmic function value of the sum of the feature values of all positions in the convolution feature map and the sum of one is calculated as a global feature representation of the convolution feature map. And finally, dividing the local feature representation of each position in the welding interest region by the global feature representation of the convolution feature map respectively to obtain a corrected feature value of each position in the welding interest region. That is, the formula for global eigenvalue correction based on the convolved signature for the weld region of interest is:

characteristic values representing the respective position of the welding region of interest, <' > or>

Characteristic values representing the respective position of the corrected welding region of interest->

To representAnd adding the characteristic values of all the positions of the convolution characteristic diagram.

FIG. 4 illustrates a block diagram of a feature distribution correction module in an automated traveling pipeline welding and cutting integrated processing tool in accordance with an embodiment of the present application. As shown in fig. 4, the feature distribution correction module 240 includes: a layout feature representation unit 241, configured to calculate a logarithmic function value of a sum of a feature value and one at each position in the welding interest region as a local feature representation at each position in the welding interest region; a global feature representation unit 242, configured to calculate a logarithmic function value of a sum of feature values of all positions in the convolution feature map and a sum of one as a global feature representation of the convolution feature map; a correction unit 243, configured to divide the local feature representation of each position in the welding region of interest by the global feature representation of the convolution feature map to obtain a corrected welding region of interest.

Specifically, in the embodiment of the present application, the control parameter module 250 and the control parameter structuring module 260 are configured to obtain control parameters of a plurality of predetermined time points including a current time point within a predetermined time period, where the control parameters include a distance between a welding gun and a workpiece and a swing angle of the welding gun, and configure the distance and the swing angle in the control parameters of the plurality of predetermined time points including the current time point as a first input vector and a second input vector, respectively. It should be understood that, in view of the idea of adaptively adjusting the swing angle of the welding torch and the distance between the welding torch and the workpiece based on the condition of the current welding area and the condition of the unwelded pipe surface so that the controlled parameters can be adapted to the surface characteristics of the pipe surface to improve the welding quality, in the technical solution of the present application, it is further required to acquire the control parameters of a plurality of predetermined time points including the current time point within the predetermined time period, wherein the control parameters include the distance between the welding torch and the workpiece and the swing angle of the welding torch. Then, the distances and the swing angles in the control parameters of the plurality of preset time points including the current time point are respectively constructed into a first input vector and a second input vector so as to carry out feature relevance mining on the first input vector and the second input vector.

Specifically, in this embodiment, the control parameter association module 270 is configured to calculate a control parameter association matrix between the first input vector and the second input vector. It should be understood that when calculating the product between the first input vector and the transpose of the second input vector to obtain the control parameter correlation moment, since the first input vector and the second input vector correspond to the time-series correlation characteristics of distance and angle, respectively, there may be a problem that the parameter correlation matrix shifts due to the difference in scale of the first input vector and the second input vector when correlating them to obtain the parameter correlation matrix. Therefore, in the technical solution of the present application, the first input vector V is further subjected to ₁ And said second input vector V ₂ And performing characteristic vector association based on scale migration certainty to obtain the control parameter association matrix.

More specifically, in this embodiment of the present application, the control parameter association module includes: first, a product between the first input vector and the transposed vector of the second input vector is calculated to obtain an initial correlation matrix. Then, the Frobenius norm of the initial correlation matrix is calculated. And finally, dividing the characteristic value of each position in the initial incidence matrix by the Frobenius norm of the initial incidence matrix to obtain the control parameter incidence matrix. That is, in one specific example, the formula for calculating the control parameter correlation matrix between the first input vector and the second input vector is:

wherein V ₁ Representing said first input vector, V ₂ Representing said second input vector, V ₁ And V ₂ Are in the form of column vectors, and | · | | non-calculation _F Frobenius norm representing matrix, exp (-) representing exponential operation of matrixThe exponential operation of the matrix means that a natural exponent function value raised to the eigenvalue of each position in the matrix is calculated. It should be understood that, in this way, the feature vector association based on the scale migration certainty performs low rank constraint on the relative position embedding of the feature vector based on the associated feature, so as to retain the long-range dependency relationship of the high-dimensional feature expressed by the feature vector on the associated feature at the scale migration time of association, and thus, the probability distribution consistency of the parameter association matrix and the first input vector and the second input vector can be achieved to some extent.

FIG. 5 illustrates a block diagram of a control parameter association module in an automated traveling pipeline welding and cutting integrated processing tool in accordance with an embodiment of the present application. As shown in fig. 5, the control parameter association module 270 includes: a correlation unit 271, configured to calculate a product between the first input vector and a transposed vector of the second input vector to obtain an initial correlation matrix; a constraint quantity calculating unit 272 configured to calculate a Frobenius norm of the initial correlation matrix; and a constraining unit 273, configured to divide the eigenvalue of each position in the initial correlation matrix by the Frobenius norm of the initial correlation matrix to obtain the control parameter correlation matrix.

Specifically, in this embodiment of the present application, the control parameter association coding module 280 is configured to pass the control parameter association matrix through a second convolutional neural network to obtain a control parameter feature map. That is, in the technical solution of the present application, after the control parameter association matrix is obtained, the control parameter association matrix is further processed through a second convolutional neural network to extract high-dimensional association implicit feature information of the control parameter, so as to obtain a control parameter feature map.

Specifically, in the embodiment of the present application, the feature fusion module 290 and the parameter control module 300 are configured to fuse the control parameter feature map and the corrected welding region of interest to obtain a decoded feature map, and perform decoding regression on the decoded feature map through a decoder to obtain a first decoded value and a second decoded value, where the first decoded value and the second decoded value are obtainedThe first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point. That is, after the control parameter feature map is adjusted to have the same size as the corrected welding region of interest by linear transformation, the control parameter feature map and the corrected welding region of interest may be fused to obtain a decoded feature map. Then, the decoding characteristic map is decoded and regressed by a decoder to obtain a first decoded value used for representing the distance between the welding gun and the workpiece at the next time point and a second decoded value used for representing the swing angle of the welding gun at the next time point. Accordingly, in one specific example, the decoder is used to perform decoding regression on the decoded feature map in the following formula to obtain the first decoded value and the second decoded value; wherein the formula is:

representing a matrix multiplication.

More specifically, in this embodiment, the feature fusion module includes: the size adjusting unit is used for adjusting the control parameter characteristic diagram to be the same as the corrected welding interesting area in size through linear transformation; the position-based weighting unit is used for fusing the control parameter characteristic diagram and the corrected welding interesting region according to the following formula to obtain the decoding characteristic diagram;

wherein the formula is:

F＝αF ₁ +βF ₂

wherein F is the decoding feature map, F ₁ For the control parameter profile, F ₂ For the corrected weld region of interest, "+" indicates the control parameter profileAnd the elements at the corresponding positions of the corrected welding interesting area are added, and alpha and beta are weighting parameters used for controlling the balance between the control parameter feature map and the corrected welding interesting area.

In conclusion, the automatic traveling pipe welding and cutting integrated machining apparatus 200 based on the embodiment of the present application is illustrated, which adaptively adjusts a swing angle of a welding torch and a distance between the welding torch and a workpiece based on a situation of a current welding region and a situation of an unwelded pipe surface by using a deep neural network model of an artificial intelligence technique, so that controlled parameters can be adapted to surface characteristics of a pipe surface to improve welding quality.

As described above, the automatic traveling pipe welding and cutting integrated machining device 200 according to the embodiment of the present application may be implemented in various terminal devices, such as a server of an automatic traveling pipe welding and cutting integrated machining device algorithm, and the like. In one example, the automated traveling pipe welding and cutting integrated processing device 200 according to embodiments of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the automated pipe welding and cutting integrated processing device 200 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the self-propelled pipe welding and cutting integrated processing device 200 could equally be one of many hardware modules of the terminal device.

Alternatively, in another example, the automated pipe welding and cutting integrated machining device 200 and the terminal device may be separate devices, and the automated pipe welding and cutting integrated machining device 200 may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.

Exemplary method

Fig. 6 illustrates a flowchart of a control method of the automatic traveling pipe welding and cutting integrated machining apparatus. As shown in fig. 6, the control method of the automatic traveling pipe welding and cutting integrated processing apparatus according to the embodiment of the present application includes the steps of: s110, acquiring a machining image of a pipeline surface with a welding area, which is acquired by a camera; s120, passing the machined image of the pipeline surface through a multilayer convolution layer to obtain a convolution characteristic diagram; s130, enabling the convolution characteristic diagram to pass through a first convolution neural network model serving as a welding area detection network so as to obtain a welding interesting area from the convolution characteristic diagram; s140, performing global characteristic value correction based on the convolution characteristic diagram on the welding interesting region to obtain a corrected welding interesting region; s150, acquiring control parameters of a plurality of preset time points including the current time point in a preset time period, wherein the control parameters comprise the distance between a welding gun and a workpiece and the swing angle of the welding gun; s160, constructing the distances and the swing angles in the control parameters of the plurality of preset time points including the current time point as a first input vector and a second input vector respectively; s170, calculating a control parameter incidence matrix between the first input vector and the second input vector; s180, passing the control parameter incidence matrix through a second convolutional neural network to obtain a control parameter characteristic diagram; s190, fusing the control parameter characteristic diagram and the corrected welding interesting area to obtain a decoding characteristic diagram; and S200, performing decoding regression on the decoding characteristic diagram through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point.

Fig. 7 illustrates an architectural schematic of a control method of an automated traveling pipe welding and cutting integrated processing tool according to an embodiment of the application. As shown in fig. 7, in the network architecture of the control method of the automatic traveling pipe welding and cutting integrated machining device, first, the obtained machined image (for example, P1 as illustrated in fig. 7) of the pipe surface is passed through a plurality of convolution layers (for example, CL as illustrated in fig. 7) to obtain a convolution signature (for example, F1 as illustrated in fig. 7); next, passing the convolution signature through a first convolution neural network model (e.g., CNN1 as illustrated in fig. 7) as a welding region detection network to obtain a welding region of interest (e.g., RI1 as illustrated in fig. 7) from the convolution signature; then, performing global eigenvalue correction based on the convolution feature map on the welding region of interest to obtain a corrected welding region of interest (for example, RI2 as illustrated in fig. 7); then, constructing the obtained distance (e.g., Q1 as illustrated in fig. 7) and swing angle (e.g., Q2 as illustrated in fig. 7) in the control parameters of the plurality of predetermined time points including the current time point as a first input vector (e.g., V1 as illustrated in fig. 7) and a second input vector (e.g., V2 as illustrated in fig. 7), respectively; then, a control parameter association matrix (e.g., M1 as illustrated in fig. 7) between the first input vector and the second input vector is calculated; then, passing the control parameter association matrix through a second convolutional neural network (e.g., CNN2 as illustrated in fig. 7) to obtain a control parameter feature map (e.g., F2 as illustrated in fig. 7); then, fusing the control parameter profile and the corrected weld region of interest to obtain a decoded profile (e.g., FC as illustrated in fig. 7); and finally, performing decoding regression on the decoding characteristic map through a decoder (for example, as shown in D in FIG. 7) to obtain a first decoded value (for example, as shown in T1 in FIG. 7) and a second decoded value (for example, as shown in T2 in FIG. 7), wherein the first decoded value is the distance between the welding gun and the workpiece at the next time point, and the second decoded value is the swing angle of the welding gun at the next time point.

More specifically, in steps S110 and S120, a machined image of a pipe face having a weld region acquired by a camera is acquired, and the machined image of the pipe face is passed through a plurality of convolution layers to obtain a convolution signature. That is, in the technical solution of the present application, it is desirable to adaptively adjust the rocking angle of the torch and the distance between the torch and the workpiece based on the condition of the current welding region and the condition of the unwelded pipe face so that the controlled parameters can be adapted to the surface characteristics of the pipe face to improve the welding quality. Accordingly, in the technical scheme of the application, the condition of the current welding area and the condition of the unwelded pipeline surface can be represented by using the machining image characteristics of the pipeline surface of the welding area, deep level characteristic mining is carried out on the machining image characteristics by using a convolutional neural network model with excellent performance in the aspect of local implicit association characteristic extraction, and the parameter value which needs to be adjusted is obtained through decoding regression, so that the welding quality is effectively and accurately improved.

That is, specifically, in the technical solution of the present application, a machining image of a pipe surface having a welding region is first acquired by a camera disposed at an automatic traveling pipe welding and cutting integrated machining apparatus. Then, the machined image of the pipeline surface passes through a plurality of convolution layers to extract local high-dimensional feature information in the machined image of the pipeline surface, and therefore a convolution feature map is obtained.

More specifically, in step S130, the convolution signature is passed through a first convolution neural network model as a welding region detection network to obtain a welding region of interest from the convolution signature. It should be understood that, since the machined image of the pipe surface should focus more on the features of the welded region and the region of the non-welded and welded joint surface, in the technical solution of the present application, the convolution feature map is further subjected to feature extraction in the first convolution neural network model as the welding region detection network, so as to focus more on the hidden features of the welded region in the deep mining of the features, thereby obtaining the welding region of interest from the convolution feature map. In particular, here, the first convolutional neural network model is centret, extreme net, or RepPoints.

More specifically, in step S140, global feature value correction based on the convolution feature map is performed on the welding region of interest to obtain a corrected welding region of interest. It should be understood that when welding control is performed by the integrated processing apparatus, it is advantageous to improve the accuracy of the welding control if the image information of the non-welding region can be combined. That is, feature distribution correction is carried out on the welding region-of-interest based on the whole situation of the convolution feature map (namely, the whole situation image information of the pipeline surface to be machined, including a welding region and a non-welding region) so as to fuse the information of the non-welding region of the pipeline surface to be machined into the welding region, so that the subsequent welding control has certain foresight and predictability, and the welding quality is improved. Therefore, in the technical solution of the present application, a global feature value correction based on the convolution feature map is further performed on the welding region of interest to obtain a corrected welding region of interest.

More specifically, in steps S150 and S160, control parameters including a distance between the welding torch and the workpiece and a swing angle of the welding torch at a plurality of predetermined time points including a current time point within a predetermined time period are acquired, and the distance and the swing angle of the control parameters at the plurality of predetermined time points including the current time point are respectively configured as a first input vector and a second input vector. It should be understood that, in consideration of adaptively adjusting the swing angle of the welding torch and the distance between the welding torch and the workpiece based on the condition of the current welding region and the condition of the unwelded pipe surface so that the controlled parameters can be adapted to the surface characteristics of the pipe surface to improve the welding quality, in the technical solution of the present application, it is further required to acquire the control parameters of a plurality of predetermined time points including the current time point within the predetermined time period, wherein the control parameters include the distance between the welding torch and the workpiece and the swing angle of the welding torch. Then, the distances and the swing angles in the control parameters of the plurality of preset time points including the current time point are respectively constructed into a first input vector and a second input vector so as to carry out feature relevance mining on the first input vector and the second input vector.

More specifically, in step S170, a control parameter association matrix between the first input vector and the second input vector is calculated. It will be appreciated that in calculating the product between the first input vector and the transpose of the second input vector to obtain the control parameter correlation matrix, since the first input vector and the second input vector correspond to time-series correlation characteristics of distance and angle, respectively, in correlating them to obtain the parameter correlation matrix, there may be a possibility that the parameter correlation matrix is obtained from the first input vector and the second input vectorThe scale difference of the input vector causes the problem of probability distribution shift. Therefore, in the technical solution of the present application, the first input vector V is further subjected to ₁ And said second input vector V ₂ And performing characteristic vector association based on the scale migration certainty to obtain the control parameter association matrix.

More specifically, in step S180, the control parameter correlation matrix is passed through a second convolutional neural network to obtain a control parameter feature map. That is, in the technical solution of the present application, after the control parameter incidence matrix is obtained, the control parameter incidence matrix is further processed through a second convolutional neural network to extract high-dimensional correlation implicit characteristic information of the control parameter, so as to obtain a control parameter characteristic diagram.

More specifically, in step S190 and step S200, the control parameter feature map and the corrected welding region of interest are fused to obtain a decoded feature map, and the decoded feature map is decoded and regressed by a decoder to obtain a first decoded value and a second decoded value, where the first decoded value is a distance between the welding gun and the workpiece at the next time point, and the second decoded value is a swing angle of the welding gun at the next time point. That is, after the control parameter feature map is adjusted to have the same size as the corrected welding region of interest by linear transformation, the control parameter feature map and the corrected welding region of interest may be fused to obtain a decoded feature map. Then, the decoding regression of the decoding characteristic diagram is carried out through a decoder to obtain a first decoding value used for representing the distance between the welding gun and the workpiece at the next time point and a second decoding value used for representing the swing angle of the welding gun at the next time point. Accordingly, in one specific example, the decoder is used to perform decoding regression on the decoded feature map in the following formula to obtain the first decoded value and the second decoded value; wherein the formula is:

representing a matrix multiplication.

In conclusion, the control method of the automatic traveling pipe welding and cutting integrated machining apparatus based on the embodiment of the present application is elucidated, which adaptively adjusts the swing angle of the welding torch and the distance between the welding torch and the workpiece based on the situation of the current welding region and the situation of the unwelded pipe surface by using the deep neural network model of the artificial intelligence technology, so that the controlled parameters can be adapted to the surface characteristics of the pipe surface to improve the welding quality.

The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by one skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably herein. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations should be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. The utility model provides an automatic pipeline welding of marcing and cutting integral type processing equipment which characterized in that includes:

the processing state acquisition module is used for acquiring a processing image of the pipeline surface with the welding area, which is acquired by the camera;

the machining region coding module is used for enabling the machining image of the pipeline surface to pass through a plurality of convolution layers to obtain a convolution characteristic diagram;

the region-of-interest detection module is used for enabling the convolution feature map to pass through a first convolution neural network model serving as a welding region detection network so as to obtain a welding region-of-interest from the convolution feature map;

the characteristic distribution correction module is used for carrying out global characteristic value correction based on the convolution characteristic diagram on the welding interesting region to obtain a corrected welding interesting region;

the control parameter module is used for acquiring control parameters of a plurality of preset time points including the current time point in a preset time period, wherein the control parameters comprise the distance between a welding gun and a workpiece and the swing angle of the welding gun;

the control parameter structuring module is used for respectively constructing the distances and the swing angles in the control parameters of a plurality of preset time points including the current time point into a first input vector and a second input vector;

the control parameter correlation module is used for calculating a control parameter correlation matrix between the first input vector and the second input vector;

the control parameter association coding module is used for enabling the control parameter association matrix to pass through a second convolutional neural network to obtain a control parameter characteristic diagram;

the characteristic fusion module is used for fusing the control parameter characteristic diagram and the corrected welding interesting region to obtain a decoding characteristic diagram; and

the parameter control module is used for performing decoding regression on the decoding characteristic graph through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point;

wherein the parameter control module is further configured to: decoding the decoded feature map using the decoder to perform a decoding regression to obtain the first decoded value and the second decoded value; wherein the formula is:

representing a matrix multiplication.

2. The automated guided tube welding and cutting integrated processing apparatus of claim 1, wherein said feature profile correction module comprises:

a layout feature representation unit, configured to calculate a logarithmic function value of a sum of a feature value and one at each position in the welding interest region as a local feature representation at each position in the welding interest region;

a global feature representation unit, configured to calculate a logarithmic function value of a sum of the feature values of all the positions in the convolution feature map and a sum of one as a global feature representation of the convolution feature map;

and the correction unit is used for dividing the local feature representation of each position in the welding interest region by the global feature representation of the convolution feature map respectively to obtain a corrected welding interest region as a corrected feature value of each position in the welding interest region.

3. The automated guided tube welding and cutting integrated processing tool of claim 2, wherein the control parameter association module comprises:

a correlation unit, configured to calculate a product between the first input vector and a transposed vector of the second input vector to obtain an initial correlation matrix;

the constraint quantity calculating unit is used for calculating the Frobenius norm of the initial incidence matrix; and

and the constraint unit is used for dividing the characteristic value of each position in the initial incidence matrix by the Frobenius norm of the initial incidence matrix to obtain the control parameter incidence matrix.

4. The automated traveling pipe welding and cutting integrated processing apparatus according to claim 3, wherein the control parameter correlation encoding module is further configured to perform convolution processing, pooling processing and nonlinear activation processing on input data in forward pass of layers, respectively, using layers of the second convolutional neural network to output the control parameter profile from the last layer of the second convolutional neural network, wherein an input of the first layer of the second convolutional neural network is the control parameter correlation matrix.

5. The automated guided pipe welding and cutting integrated processing apparatus of claim 4, wherein said feature fusion module comprises:

the size adjusting unit is used for adjusting the control parameter characteristic diagram to be the same as the corrected welding interesting area in size through linear transformation; and

a position-based weighting unit for fusing the control parameter feature map and the corrected welding interesting region by the following formula to obtain the decoding feature map;

wherein the formula is:

F＝αF ₁ +βF ₂

6. The automated guided tube welding and cutting integrated processing device of claim 5, wherein the first convolutional neural network model is CenterNet, extrmeNet, or RePoints.

7. A control method of an automatic advancing pipeline welding and cutting integrated processing device is characterized by comprising the following steps:

acquiring a machining image of a pipeline surface with a welding area, which is acquired by a camera;

passing the machining image of the pipeline surface through a plurality of convolution layers to obtain a convolution characteristic diagram;

passing the convolved signature through a first convolved neural network model as a weld region detection network to obtain a weld region of interest from the convolved signature;

performing global eigenvalue correction based on the convolution feature map on the welding region of interest to obtain a corrected welding region of interest;

acquiring control parameters of a plurality of preset time points including a current time point in a preset time period, wherein the control parameters comprise the distance between a welding gun and a workpiece and the swing angle of the welding gun;

constructing the distances and the swing angles in the control parameters of a plurality of preset time points including the current time point as a first input vector and a second input vector respectively;

calculating a control parameter association matrix between the first input vector and the second input vector;

enabling the control parameter incidence matrix to pass through a second convolutional neural network to obtain a control parameter characteristic diagram;

fusing the control parameter characteristic diagram and the corrected welding interesting area to obtain a decoding characteristic diagram; and

performing decoding regression on the decoding characteristic graph through a decoder to obtain a first decoding value and a second decoding value, wherein the first decoding value is the distance between the welding gun and the workpiece at the next time point, and the second decoding value is the swing angle of the welding gun at the next time point;

performing decoding regression on the decoded feature map through a decoder to obtain a first decoded value and a second decoded value, wherein the decoding regression includes: decoding the decoded feature map using the decoder to perform a decoding regression to obtain the first decoded value and the second decoded value; wherein the formula is:

x, where X is a regression matrix, Y is the decoded value, W is a weight matrix, and>

representing a matrix multiplication.

8. The control method of an automated guided tube welding and cutting integrated processing tool of claim 7, wherein said weld region of interest is corrected for global eigenvalue corrections based on said convolved signature to yield a corrected weld region of interest, comprising:

calculating a logarithmic function value of a sum of the feature value and one of each position in the welding interest area as a local feature representation of each position in the welding interest area;

calculating a logarithm function value of a summation value of the feature values of all positions in the convolution feature map and a summation value of one as a global feature representation of the convolution feature map;

dividing the local feature representation of each position in the welding interest region by the global feature representation of the convolution feature map respectively to obtain a corrected welding interest region.

9. The method of controlling an automated guided tube welding and cutting integrated processing tool of claim 8, wherein calculating a control parameter correlation matrix between the first input vector and the second input vector comprises:

calculating a product between the first input vector and a transposed vector of the second input vector to obtain an initial correlation matrix;

calculating the Frobenius norm of the initial incidence matrix; and

and dividing the characteristic value of each position in the initial incidence matrix by the Frobenius norm of the initial incidence matrix to obtain the control parameter incidence matrix.