Carbon steel pipe welding quality evaluation system

Abstract

The invention relates to the technical field of welding quality evaluation, in particular to a carbon steel pipe welding quality evaluation system which comprises a welding area three-dimensional scanning module, a welding seam dynamic grid generation module, a welding form change analysis module, a welding seam structure diagram construction module, a welding seam quality diagram convolutional network module and a welding defect intelligent recognition module. According to the invention, through real-time adjustment of the dynamic grid generation module, tiny changes of the geometric shape of the welding seam can be captured, new dimensions are provided for welding quality evaluation through generation of welding form characteristic data and construction of the welding seam structure diagram, and the introduction of the welding seam quality diagram convolution network module can improve the accuracy of welding quality evaluation. Node relations and structural features in a weld joint diagram structure are analyzed through a deep learning technology, an intelligent means is provided for welding quality evaluation, an intelligent welding defect identification module optimizes an identification process, various welding defects can be accurately classified, and a powerful technical support is provided for guarantee of welding quality.

Description

Technical Field

[0001] The invention relates to the technical field of welding quality assessment, and in particular to a carbon steel pipe welding quality assessment system. Background Art

[0002] The field of welding quality assessment technology focuses on inspecting and verifying the integrity, stability and strength of welded joints or welds to ensure that they meet specific industrial standards and usage requirements. This field combines materials science, mechanical engineering and physical inspection technology, aiming to identify possible defects in welds, such as cracks, holes, inclusions, etc., through a series of inspection methods. These defects affect the safety and reliability of the structure. Assessment technologies include but are not limited to non-destructive testing (NDT), destructive testing, and chemical and physical analysis methods. With the development of technology, the field of welding quality assessment continues to incorporate new technologies and methods, such as digital imaging, automated inspection and machine learning algorithms, to improve the accuracy and efficiency of inspection.

[0003] The carbon steel pipe welding quality assessment system is a system specially designed to assess the quality of carbon steel pipe welds. Its purpose is to ensure that the welds formed during the welding process can meet the predetermined safety and performance standards. This system usually needs to assess the uniformity, continuity and presence of defects in the welds to ensure the safety and reliability of the pipeline when transporting fluids (whether gas or liquid). By using this assessment system, welding problems that lead to pipeline failures, such as cracks, pores or lack of penetration, can be discovered and corrected in a timely manner, thereby reducing operation and maintenance costs, extending the service life of the pipeline and ensuring transportation safety.

[0004] When dealing with welding quality assessment, traditional systems often have difficulty accurately capturing and evaluating subtle defects inside welds due to technical limitations. Especially when the weld morphology is complex or changing, existing methods are difficult to adapt and accurately evaluate in real time, resulting in inaccurate assessment results that cannot fully guarantee the safety and reliability of the structure. The lack of efficient automated and intelligent assessment tools makes the defect identification process time-consuming and experience-dependent, reduces assessment efficiency, and increases operation and maintenance costs. In addition, the lack of ability to identify complex defects affects the comprehensiveness and accuracy of the assessment, limiting the application of the assessment system in scenarios with high standards and strict requirements. The consequences of these deficiencies include increased safety risks, increased operation and maintenance costs, and shortened pipeline service life, which have an adverse impact on production and operations. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a carbon steel pipe welding quality assessment system.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: A carbon steel pipe welding quality assessment system comprises: The welding area 3D scanning module uses 3D scanning technology to scan the welding area of ​​carbon steel pipes, capture the structural characteristics of the weld, record geometric data, and generate 3D welding scanning data; The weld dynamic mesh generation module performs mesh division based on the three-dimensional welding scanning data, adjusts each mesh unit in real time to match the change of the weld geometry, refines the mesh unit and captures detail information, and generates a weld dynamic mesh model; The welding morphology change analysis module monitors the changes in the width, depth and edge profile of the weld based on the weld dynamic grid model, extracts the characteristics of the changes, and generates welding morphology feature data; The weld structure diagram construction module maps the weld and defect area to the diagram nodes through the welding morphological feature data, defines the edges according to the geometric characteristics and positional relationship between the nodes, constructs the diagram representation of the weld, and generates the weld diagram structure of the carbon steel pipe; The weld quality graph convolutional network module adopts the carbon steel pipe weld graph structure to perform feature extraction, analyze the node relationship and structural characteristics in the weld graph representation, extract weld quality correlation information, and generate weld quality feature representation; The welding defect intelligent identification module identifies and classifies the defect types through the weld quality feature representation, distinguishes the differentiated types of welding defects, and generates welding defect identification results.

[0007] As a further solution of the present invention, the three-dimensional welding scanning data includes a position coordinate set, a depth value, a width value, and a surface feature set; the weld dynamic grid model includes a grid unit size, a weld morphology description set, and a unit connection diagram; the welding morphology feature data includes a width change, a depth change, and an edge contour morphology set; the carbon steel pipe weld map structure includes a spatial distribution map, a connection property map, and a regional node classification map; the weld quality feature representation includes an overall evaluation index, a local defect feature vector set, and a quality classification label.

[0008] As a further solution of the present invention, the weld dynamic mesh generation module includes: The mesh initialization submodule uses a voxelization method based on the three-dimensional welding scanning data to determine whether the data covers the entire weld area. If so, an initial mesh framework is created, which is automatically divided into multiple basic mesh units according to the geometric shape of the weld, and initial size and shape parameters are set for each unit to generate an initial mesh model. The grid adjustment submodule uses the initial grid model to monitor the real-time changes of the weld shape. If a change is detected, the size and shape of the grid unit are dynamically adjusted to match the current geometric characteristics of the weld. The distribution and density of adjacent grid units are adjusted by calculating the change in the weld geometry to obtain an adjusted grid model. The detail capture submodule uses the adjusted mesh model to iteratively improve the model's ability to capture weld details, and identifies key geometric feature areas by analyzing the geometric characteristics of the weld area. If the model needs to capture geometric features, the density of the mesh is increased in the key area, the model is locally refined, and the size of the mesh unit is fine-tuned to obtain a dynamic mesh model of the weld.

[0009] As a further solution of the present invention, the geometric characteristics of the weld area are analyzed using the formula: ; Calculate the geometric feature deviation value, combine the geometric characteristics of the weld area, calculate the local feature parameters through the key area data, identify the key geometric feature area, and if the model needs to capture the geometric features, increase the density of the grid in the key area, perform local refinement on the model, and fine-tune the size of the grid unit to obtain the dynamic grid model of the weld; in, Represents the geometric feature deviation value, Represents the weld area The local curvature length of the sampling points, represents the average curvature length of all weld sampling points, Representative The area of ​​the mesh unit in the weld area, Representative The density of mesh cells in the weld area, represents the total number of mesh elements in the weld area, represents the summation of all grid cells. Represents the absolute deviation of the weld curvature length from the average curvature length.

[0010] As a further solution of the present invention, the welding morphology change analysis module includes: The width monitoring submodule analyzes the distribution characteristics of the grid units along the length of the weld based on the weld dynamic grid model, and records the width data of each position by comparing the distance changes between adjacent grid units in the lateral direction, thereby generating weld width change data; The depth monitoring submodule uses the weld width variation data to analyze the variation of the grid unit along the weld depth, measures the height difference of the differentiated depth grid unit, records the distance from the weld surface to the deepest point, extracts the depth variation trend along the length of the weld, and obtains the weld depth variation data; The edge profile monitoring submodule identifies the change of the weld edge profile through the weld depth change data, analyzes the peripheral lines of the grid model, compares the edge profile line changes at different time points, records the fluctuation and offset of the edge profile, tracks the trajectory of the edge profile change, and obtains the weld edge profile data; The feature extraction submodule integrates the change information of width, depth and edge profile based on the weld edge profile data, analyzes the key change characteristics of the weld morphology, and generates welding morphology feature data by comparing and analyzing the change laws and patterns of width, depth and edge profile data.

[0011] As a further solution of the present invention, the weld structure diagram building module includes: The welding morphology recognition submodule analyzes the geometric size and morphology of the weld and defect area based on the welding morphology feature data and sets the node attributes, locates the position of the weld and defect, assigns a unique identifier to each graph node, maps the weld and defect features to the graph node attributes, and generates the welding morphology feature analysis results; The graph node mapping submodule adopts the welding morphological feature analysis results and uses a mapping algorithm to convert the description of the weld and defect area into a graph node, assigns a unique identifier to each graph node and assigns a corresponding attribute value, and quantifies the morphological features of the weld and the defect, and connects the nodes to each other according to the spatial relationship between the weld and the defect to generate a graph node mapping result; The edge definition and graph construction submodule uses spatial position information to sort nodes and establish connection paths based on the graph node mapping results, adjusts connection strength parameters according to the characteristics of the connection path, and refines the connection path and strength parameters through iterative optimization to obtain the carbon steel pipe weld graph structure.

[0012] As a further solution of the present invention, the weld quality graph convolutional network module includes: The structural analysis submodule analyzes each node in the carbon steel pipe weld diagram structure and its connection mode, identifies the position and function of the weld or defect area represented by each node in the overall weld structure, analyzes the direct and indirect connection paths between nodes, and generates a deep analysis result of the weld structure; The quality influencing factor identification submodule uses a network analysis algorithm based on the weld structure in-depth analysis results to analyze the interaction between nodes and their potential impact on the overall quality of the weld. If the mining target node connection mode is associated with quality defects, the node connection mode is analyzed in detail to identify the key nodes and connection methods that affect the weld quality and obtain key quality influencing factors. The quality feature synthesis submodule integrates multiple groups of key quality influencing factors, analyzes the joint action mode of multiple groups of key quality influencing factors on weld quality, studies the interaction between key nodes and connection modes, analyzes the influencing factors of the interaction on weld quality, and generates weld quality feature representation.

[0013] As a further solution of the present invention, the network analysis algorithm adopts the formula: ; Compute the improved betweenness centrality of the nodes, where For Node The improved betweenness centrality of For Node The weight of For Node and The weight of the path between For Node The position parameters, For path arrive The length parameter, Passing Node The number of shortest paths, For Node To Node The total number of shortest paths.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the real-time adjustment of the dynamic grid generation module can capture the slight changes in the geometry of the weld in detail, thereby making the analysis of welding morphology changes more accurate. The generation of welding morphological feature data and the construction of the weld structure diagram provide a new dimension for welding quality assessment, so that defect identification is not limited to the surface, but combined with deep structural correlation analysis, greatly improving the comprehensiveness and depth of defect identification. The introduction of the weld quality graph convolutional network module analyzes the node relationships and structural features in the weld graph structure through deep learning technology, providing an intelligent means for welding quality assessment, and effectively improving the assessment speed and accuracy. The welding defect intelligent identification module further optimizes the identification process, can quickly and accurately classify various welding defects, and provides strong technical support for the guarantee of welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 It is a schematic diagram of the system framework of the present invention; Figure 3 This is a flow chart of the welding area three-dimensional scanning module of the present invention; Figure 4 It is a flow chart of the weld dynamic mesh generation module of the present invention; Figure 5 It is a flow chart of the welding morphology change analysis module of the present invention; Figure 6 A module flow chart for constructing a weld structure diagram of the present invention; Figure 7 It is a flow chart of the weld quality graph convolutional network module of the present invention; Figure 8This is a flow chart of the welding defect intelligent identification module of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1

[0019] See also Figure 1 to Figure 2 , a carbon steel pipe welding quality assessment system comprises: The welding area 3D scanning module uses 3D scanning technology to scan the welding area of ​​carbon steel pipes, capture the structural characteristics of the weld, record geometric data, and generate 3D welding scanning data; The weld dynamic mesh generation module uses 3D welding scanning data to perform mesh division, adjust each mesh unit in real time to match the changes in the weld geometry, refine the mesh unit and capture detail information to generate a weld dynamic mesh model; The welding morphology change analysis module is based on the dynamic mesh model of the weld, monitors the changes in weld width, depth and edge contour, extracts the characteristics of the changes, and generates welding morphology feature data; The weld structure diagram construction module maps welds and defect areas to graph nodes through welding morphological feature data, defines edges based on the geometric characteristics and positional relationships between nodes, constructs a graph representation of welds, and generates a weld graph structure for carbon steel pipes. The weld quality graph convolutional network module uses the carbon steel pipe weld graph structure to extract features, analyze the node relationship and structural features in the weld graph representation, extract weld quality correlation information, and generate weld quality feature representation; The intelligent welding defect recognition module identifies and classifies the defect types through the weld quality characteristics, distinguishes the differentiated types of welding defects, and generates welding defect recognition results.

[0020] The three-dimensional welding scanning data includes a position coordinate set, a depth value, a width value, and a surface feature set. The dynamic mesh model of the weld includes a mesh unit size, a weld morphology description set, and a unit connection diagram. The welding morphology feature data includes a width change, a depth change, and an edge contour morphology set. The carbon steel pipe weld map structure includes a spatial distribution map, a connection property map, and a regional node classification map. The weld quality feature representation includes an overall evaluation index, a local defect feature vector set, and a quality classification label. The welding defect recognition results include a type recognition set, position coordinates, a size estimate, and a severity level.

[0021] See also Figure 3 , the welding area 3D scanning module includes: The weld positioning submodule is based on 3D scanning technology to preliminarily locate the welding area of ​​the carbon steel pipe, adjust the scanning equipment to align the welding line, record the coordinates of the welding area, measure the length and width of the welding area, control the scanning path to cover the entire welding area, and generate weld positioning information; The weld positioning submodule is based on 3D scanning technology to preliminarily locate the welding area of ​​the carbon steel pipe. It uses an optical 3D scanner to perform scanning operations, sets scanner parameters including a scanning speed of 5mm / s and a scanning resolution of 0.01mm, locates the center point coordinates of the welding area through a 3D coordinate measurement system, uses QR code positioning technology to automatically align the welding line, collects the coordinate data of the welding area, calculates the length and width of the welding area using a linear interpolation algorithm, controls the scanning path to cover the entire welding area in a spiral form, and generates weld positioning information.

[0022] The structural feature capture submodule uses the weld positioning information, adjusts the scanning resolution to refine and capture the weld geometry, synchronously adjusts the scanning speed to match the weld size, records the weld surface texture information, and obtains the weld geometry data; The structural feature capture submodule uses the weld positioning information, adjusts the scanning resolution to refine and capture the weld geometry, and simultaneously adjusts the scanning speed to match the weld size, records the texture information on the weld surface, and obtains the weld geometry data. In this process, the fine scanning mode is used, the scanning resolution is set to 0.005mm, and the scanning speed is adjusted to 2mm / s to match the subtle size changes of the weld. High dynamic range imaging technology (HDR) is used to capture the weld surface texture, and the edge detection algorithm (Canny algorithm) is applied to refine the weld edge. The surface fitting algorithm (least squares method) is simultaneously applied to obtain the accurate geometric structure data of the weld.

[0023] The 3D data generation submodule refines the recorded data through the weld geometry data, adjusts the data model to match the actual weld geometry features, reconstructs the 3D model of the weld, and obtains 3D welding scanning data.

[0024] The 3D data generation submodule refines the recorded data through the weld geometry data, adjusts the data model to match the actual weld geometry, reconstructs the weld 3D model, and obtains 3D welding scanning data. In this step, a 3D reconstruction algorithm (voxel-based Marching Cubes algorithm) is used, the voxel size is set to 0.001 mm³, the spatial distribution of 3D data points is adjusted using the acquired geometry data, the point cloud is filtered and denoised using point cloud data processing software (such as PointCloudLibrary, PCL), and surface reconstruction technology (such as Poisson reconstruction) is used to convert the processed point cloud into a continuous 3D surface, and finally 3D model data matching the actual weld is generated.

[0025] See also Figure 4 , the weld dynamic mesh generation module includes: The mesh initialization submodule uses voxelization method based on 3D welding scanning data to determine whether the data covers the entire weld area. If so, it creates an initial mesh framework, automatically divides it into multiple basic mesh units according to the geometric shape of the weld, sets the initial size and shape parameters for each unit, and generates an initial mesh model. The mesh initialization submodule is based on the 3D welding scan data and adopts the voxelization method. The voxel meshing technology is used to convert the 3D scan data into voxel meshes. The voxel size is set to 1mm³ to determine whether the data covers the entire weld area. If so, an initial mesh framework is created and an automatic meshing algorithm (such as Octree partitioning) is used to automatically divide the weld into multiple basic mesh units according to the geometric shape of the weld. The initial size and shape parameters are set for each unit, including a unit side length of 2mm. The shape parameters are automatically adjusted according to the local geometric characteristics of the weld to generate an initial mesh model.

[0026] The grid adjustment submodule uses the initial grid model to monitor the real-time changes of the weld shape. If changes are detected, the size and shape of the grid cells are dynamically adjusted to match the current geometric characteristics of the weld. The distribution and density of adjacent grid cells are adjusted by calculating the change in the weld geometry to obtain the adjusted grid model. The mesh adjustment submodule adopts the initial mesh model, monitors the real-time changes of the weld shape, and uses real-time 3D scanning feedback in combination with dynamic mesh adjustment algorithms, such as the elastic mesh method. If changes are detected, the size and shape of the mesh unit are dynamically adjusted to match the current geometric characteristics of the weld. By calculating the change in the weld geometry, the distribution and density of adjacent mesh units are adjusted using dynamic interpolation technology, including adjusting the mesh unit spacing to 1.5 mm. The density is dynamically increased or decreased according to the changes in the weld to obtain the adjusted mesh model.

[0027] The detail capture submodule iteratively improves the model's ability to capture weld details through the adjusted mesh model. It identifies key geometric feature areas by analyzing the geometric characteristics of the weld area. If the model needs to capture geometric features, the density of the mesh is increased in the key area, the model is locally refined, and the size of the mesh unit is fine-tuned to obtain a dynamic mesh model of the weld.

[0028] The detail capture submodule uses the adjusted mesh model to iteratively improve the model's ability to capture weld details. It uses detail enhancement algorithms, such as mesh refinement technology, to analyze the geometric characteristics of the weld area and combine local refinement strategies to identify key geometric feature areas. If the model needs to capture geometric features, the density of the mesh is increased in the key area, and the density is adjusted to have more mesh units within 2mm². The model is locally refined and the size of the mesh units is fine-tuned, such as adjusting the unit size to 1mm in the detail area to obtain a dynamic mesh model of the weld.

[0029] The geometric characteristics of the weld area are analyzed using the formula: ; Calculate the geometric feature deviation value, combine the geometric characteristics of the weld area, calculate the local feature parameters through the key area data, identify the key geometric feature area, and if the model needs to capture the geometric features, increase the density of the grid in the key area, perform local refinement on the model, and fine-tune the size of the grid unit to obtain the dynamic grid model of the weld; in, Represents the geometric feature deviation value, Represents the weld area The local curvature length of the sampling points, represents the average curvature length of all weld sampling points, Representative The area of ​​the mesh unit in the weld area, Representative The density of mesh cells in the weld area, represents the total number of mesh elements in the weld area, represents the summation of all grid cells. Represents the absolute deviation of the weld curvature length from the average curvature length.

[0030] Analyze the geometric characteristics of the weld area. First, obtain the initial mesh model of the weld area, select the coordinate data of the discrete points of the weld edge points and the internal area, and calculate the curvature change rate of adjacent mesh units based on the coordinate data. For the curvature change rate of the weld area, calculate the local curvature mean of each mesh unit, and obtain the weld geometric feature area by comparing the curvature mean change of adjacent units. Further, calculate the area of ​​each mesh unit. and mesh density , these data are used to evaluate the geometric change characteristics of the weld area. For example, in the transition area at the edge of the weld, there is usually a large curvature change rate, while in the center area of ​​the weld, the curvature change rate is relatively small. In order to determine whether a certain area is a key feature area, the geometric feature deviation value of the mesh unit in the weld area is calculated. , among which, if Above the set threshold , it is considered that the area contains key geometric features and needs to increase the mesh density for local refinement. According to the complexity of the weld shape, a higher value is taken when the weld shape is relatively simple, for example ; When the weld shape is complex, take a lower value, such as For example, if the mean curvatures of the five mesh elements in a weld area are , , , , , then the mean curvature , and then calculate the absolute deviation value of each grid cell , and find their sum, and then divide it by the sum of the area densities of all grid cells to get ,If the value is higher than the set threshold, the mesh density needs to be increased in the local area, otherwise it remains as it is.,Finally, according to the above calculation results, the mesh density is increased in the key geometric feature area, and local refinement is performed, and the size of the mesh unit is finally,fine-tuned to obtain the dynamic mesh model of the weld.

[0031] It is known that the mesh unit areas of a certain weld area are mm2, mm2, mm2, mm2, mm2, and the grid densities are , , , , , calculate the geometric feature deviation value: ; ; ; Compare With threshold ,like Less than , then the current area does not belong to the key geometric feature area, and there is no need to increase the grid density. If the calculation is , then the mesh density in this area needs to be increased and local refinement processing needs to be performed. The result shows that the geometric features in the weld area change little, and the existing mesh division accuracy can meet the modeling requirements without refinement processing.

[0032] Innovations in the formula: The benefit of the formula is that it quantifies the local changes in the weld geometry by calculating the local curvature changes and calculating the area density ratio, and determines the mesh refinement requirements by setting thresholds, thereby achieving accurate capture of key areas.

[0033] See also Figure 5 , welding morphology change analysis module includes: The width monitoring submodule is based on the dynamic grid model of the weld, analyzes the distribution characteristics of the grid units along the length of the weld, and records the width data of each position by comparing the distance changes between adjacent grid units in the lateral direction, thus generating the weld width change data; The width monitoring submodule is based on the dynamic grid model of the weld and uses the grid analysis algorithm. It calls the NumPy library through the Python programming language to process the grid data, traverses the grid cells along the length of the weld, calculates the Euclidean distance between adjacent grid cells in the horizontal direction, uses a for loop to traverse each position, and records the width data of each position as a NumPy array to generate weld width change data.

[0034] The depth monitoring submodule uses the weld width change data to analyze the change of the grid unit along the weld depth. By measuring the height difference of the differentiated depth grid unit, the distance from the weld surface to the deepest point is recorded, the depth change trend along the length of the weld is extracted, and the weld depth change data is obtained. The depth monitoring submodule uses the weld width change data and applies the depth measurement algorithm to draw the depth-width graph through the Matplotlib library, analyzes the change trend of the grid unit along the weld depth, calculates the height difference of adjacent depth grid units through the difference method, and uses the map function to map each height difference to depth data. The distance from the weld surface to the deepest point is recorded as a list to generate the weld depth change data.

[0035] The edge profile monitoring submodule identifies changes in the weld edge profile through weld depth change data, analyzes the outer lines of the grid model, compares changes in edge profile lines at different time points, records the fluctuations and offsets of the edge profile, tracks the trajectory of edge profile changes, and obtains weld edge profile data; The edge contour monitoring submodule uses the edge detection algorithm through the weld depth change data, and identifies the weld edge contour through the Canny edge detection function in the OpenCV library. It sets the thresholds to 100 and 200, applies edge detection to the weld model, analyzes the obtained edge image, and records the ups and downs and offsets of the edge contour as a two-dimensional array by traversing the edge pixels. It tracks the trajectory of the edge contour changes and generates the weld edge contour data.

[0036] The feature extraction submodule is based on the weld edge contour data, integrates the change information of width, depth and edge contour, analyzes the key change characteristics of the weld morphology, and generates welding morphology feature data by comparing and analyzing the change laws and patterns of width, depth and edge contour data.

[0037] The feature extraction submodule is based on the weld edge profile data, integrates the change information of width, depth and edge profile, uses the data clustering algorithm, executes the K-means algorithm through the Scikit-learn library, sets the number of clusters to 3, and performs cluster analysis on the key change characteristics of the weld morphology. The fit method is applied to the collection of width, depth and edge profile data, and the change rules and patterns of width, depth and edge profile data are compared and analyzed to generate welding morphology feature data.

[0038] See also Figure 6 , the weld structure diagram building modules include: The welding morphology recognition submodule analyzes the geometric size and morphology of the weld and defect area based on the welding morphology feature data and sets the node attributes. It also locates the positions of the weld and defects, assigns a unique identifier to each graph node, maps the weld and defect features to the graph node attributes, and generates the welding morphology feature analysis results. The welding morphology recognition submodule is based on welding morphology feature data and adopts graph theory analysis algorithm. It calls the NetworkX library through the Python programming language to create a graph model, quantitatively analyzes the geometric size and morphology of the weld and defect area, sets a unique identifier for each graph node, and uses the set_node_attributes method to map the weld and defect characteristics to the graph node attributes. The node attributes including geometric size and morphology description are recorded to generate the welding morphology feature analysis results.

[0039] The graph node mapping submodule adopts the welding morphological feature analysis results and uses the mapping algorithm to convert the description of the weld and defect area into graph nodes, assigns a unique identifier to each graph node and gives it a corresponding attribute value, and quantifies it according to the morphological characteristics of the weld and defect. At the same time, the nodes are connected to each other according to the spatial relationship between the weld and the defect to generate the graph node mapping result; The graph node mapping submodule adopts the analysis results of welding morphological characteristics, utilizes the data mapping algorithm, and continues to use the NetworkX library to process graph data through the Python programming language. It converts the description of welds and defect areas into graph nodes, assigns a unique identifier to each node and gives it the corresponding attribute value. The nodes are added to the graph through the add_node method, and are quantified according to the morphological characteristics of welds and defects. The add_edge method is used to connect the nodes according to the spatial relationship between welds and defects to generate the graph node mapping results.

[0040] The edge definition and graph construction submodule uses the spatial position information to sort the nodes and establish the connection path based on the graph node mapping results, adjusts the connection strength parameters according to the characteristics of the connection path, and refines the connection path and strength parameters through iterative optimization to obtain the carbon steel pipe weld graph structure.

[0041] The edge definition and graph construction submodule is based on the graph node mapping results and adopts the spatial relationship analysis algorithm. It continues to use the NetworkX library through the Python programming language to operate the graph, sort the nodes and establish connection paths, and use the sorted function to sort according to the spatial position information of the nodes. The connection strength parameters are adjusted through the set_edge_attributes method according to the characteristics of the connection path, and iterative optimization is performed to refine the connection path and strength parameters to generate the carbon steel pipe weld graph structure.

[0042] See also Figure 7 ,The weld quality graph convolutional network module includes: The structural analysis submodule is based on the carbon steel pipe weld diagram structure, analyzes each node in the diagram and its connection mode, identifies the position and role of the weld or defect area represented by each node in the overall weld structure, analyzes the direct and indirect connection paths between nodes, and generates the weld structure in-depth analysis results; The structural analysis submodule is based on the carbon steel pipe weld graph structure and adopts graph theory analysis methods. It uses the Python programming language to analyze the graph structure using the NetworkX library, conducts in-depth research on each node and its connection method, and uses the graph.nodes(data=True) method to extract each node and its attributes. The graph.edges method is used to analyze the direct and indirect connection paths between nodes, identify the position and role of the weld or defect area represented by each node in the overall weld structure, record the connection information and path characteristics between nodes, and generate deep analysis results of the weld structure.

[0043] The quality influencing factor identification submodule uses the network analysis algorithm based on the deep analysis results of the weld structure to analyze the interaction between nodes and their potential impact on the overall quality of the weld. If the mining target node connection mode is associated with quality defects, the node connection mode is analyzed in detail to identify the key nodes and connection methods that affect the weld quality and obtain the key quality influencing factors. The quality influencing factor identification submodule is based on the deep analysis results of the weld structure, applies the network analysis algorithm, continues to process data through the Python programming language and the NetworkX library, analyzes the interactions between nodes and their potential impact on the overall quality of the weld, and uses the nx.betweenness_centrality(graph) method to calculate the betweenness centrality of the nodes to identify the key nodes and connection methods that have a significant impact on the weld quality. It focuses on analyzing the node connection patterns related to quality defects and identifying key quality influencing factors.

[0044] The network analysis algorithm uses the formula: ; Compute the improved betweenness centrality of the nodes, where For Node The improved betweenness centrality of For Node The weight of For Node and The weight of the path between For Node The position parameters, For path arrive The length parameter, Passing Node The number of shortest paths, For Node To Node The total number of shortest paths.

[0045] First, the betweenness centrality of each node is calculated, and then new weights and parameters are introduced according to the location of the node and the characteristics of the path, including the node Weight , which reflects the importance or functionality of the node. and The weight of the path between , considering the importance or capacity of the path, nodes Positional parameters , indicating the geographical or logical location of a node in the network, path arrive The length parameter , taking into account the actual length or cost of the path, all of these parameters are combined to adjust the calculation of betweenness centrality to make it more accurately reflect the influence of the node in the network, the weight coefficient and The specific value of can be obtained by analyzing the properties of nodes and paths. For example, the weight value can be determined based on the node connectivity, the path congestion or the transmission rate. It can be determined by analyzing the position of the node in the network layout. For example, edge nodes and core nodes have different position parameters and length parameters. It can be determined based on the actual distance or cost of the path. These improvements not only enhance the calculation accuracy of betweenness centrality, but also provide more dimensions of consideration for network analysis.

[0046] The quality feature synthesis submodule integrates multiple groups of key quality influencing factors, analyzes the joint effect of multiple groups of key quality influencing factors on weld quality, studies the interaction between key nodes and connection modes, analyzes the influencing factors of the interaction on weld quality, and generates weld quality feature representation.

[0047] The quality feature synthesis submodule integrates multiple groups of key quality influencing factors, applies statistical analysis methods, aggregates data on key quality influencing factors through Python programming language and Pandas library, analyzes the joint effect of multiple groups of key quality influencing factors on weld quality, uses df.groupby(['factor'])['influence'].aggregate('mean').reset_index() method to study the interaction between key nodes and connection modes, analyzes the influencing factors of these interactions on weld quality, and generates weld quality feature representation.

[0048] See also Figure 8 , the welding defect intelligent identification module includes: The feature analysis submodule analyzes the dimensions in the feature vector based on the weld quality feature representation, identifies key features and analyzes the impact on weld quality. Through the feature analysis process, it analyzes the reflection of each dimension in the feature vector on the weld quality status, determines the decisive factors affecting weld quality, and generates weld key quality factor data; The feature analysis submodule is based on the weld quality feature representation and applies the principal component analysis (PCA) algorithm. PCA is performed using the Python programming language and the Scikit-learn library. The n_components parameter is set to automatic selection to analyze the dimensions in the feature vector. The PCA.fit_transform method is used to reduce the dimension of the feature data, identify and analyze the impact of key features on weld quality, analyze the way each dimension in the feature vector reflects the weld quality status, determine the decisive factors affecting weld quality, and generate weld key quality factor data.

[0049] The defect classification submodule uses logic analysis to set the correspondence between multiple weld quality features and differentiated defect types based on the weld key quality factor data, subdivides the welds, classifies the welds into defect categories according to the quality features, and outputs detailed defect classification information; The defect classification submodule uses a decision tree classification algorithm based on the key quality factor data of the weld. It continues to use the Scikit-learn library to execute DecisionTreeClassifier through the Python programming language, sets the criterion parameter to "gini" to optimize the classification accuracy, and uses the fit method to apply to the data between the weld quality characteristics and the differentiated defect types. The welds are classified into defect categories according to the quality characteristics, and the detailed defect classification information is output.

[0050] The result output submodule conducts in-depth analysis of defects based on the refined defect classification information, identifies the causes of defects and analyzes the formation mechanism. For each defect type, it provides a category report, builds a welding defect database, and generates welding defect identification results.

[0051] Based on the refined defect classification information, the result output submodule applies the natural language processing (NLP) algorithm, uses the Natural Language Toolkit (NLTK) library in Python programming language to analyze the defect description text, sets appropriate text processing parameters for word segmentation and annotation, identifies the cause of the defect and analyzes the formation mechanism, provides a detailed category report based on the description text of each defect type, builds a welding defect database, and generates welding defect recognition results.

[0052] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A carbon steel pipe welding quality assessment system, characterized in that: The system comprises: The welding area 3D scanning module uses 3D scanning technology to scan the welding area of ​​carbon steel pipes, capture the structural characteristics of the weld, record geometric data, and generate 3D welding scanning data; The weld dynamic mesh generation module performs mesh division based on the three-dimensional welding scanning data, adjusts each mesh unit in real time to match the change of the weld geometry, refines the mesh unit and captures detail information, and generates a weld dynamic mesh model; The welding morphology change analysis module monitors the changes in the width, depth and edge profile of the weld based on the weld dynamic grid model, extracts the characteristics of the changes, and generates welding morphology feature data; The weld structure diagram construction module maps the weld and defect area to the diagram nodes through the welding morphological feature data, defines the edges according to the geometric characteristics and positional relationship between the nodes, constructs the diagram representation of the weld, and generates the weld diagram structure of the carbon steel pipe; The weld quality graph convolutional network module adopts the carbon steel pipe weld graph structure to perform feature extraction, analyze the node relationship and structural characteristics in the weld graph representation, extract weld quality correlation information, and generate weld quality feature representation; The welding defect intelligent identification module identifies and classifies the defect types through the weld quality feature representation, distinguishes the differentiated types of welding defects, and generates welding defect identification results.

2. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The three-dimensional welding scanning data includes a position coordinate set, a depth value, a width value, and a surface feature set. The weld dynamic grid model includes a grid unit size, a weld morphology description set, and a unit connection diagram. The welding morphology feature data includes a width change, a depth change, and an edge contour morphology set. The carbon steel pipe weld map structure includes a spatial distribution map, a connection property map, and a regional node classification map. The weld quality feature representation includes an overall evaluation index, a local defect feature vector set, and a quality classification label. The welding defect identification result includes a type identification set, position coordinates, a size estimate, and a severity level.

3. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The weld dynamic mesh generation module comprises: The mesh initialization submodule uses a voxelization method based on the three-dimensional welding scanning data to determine whether the data covers the entire weld area. If so, an initial mesh framework is created, which is automatically divided into multiple basic mesh units according to the geometric shape of the weld, and initial size and shape parameters are set for each unit to generate an initial mesh model. The grid adjustment submodule uses the initial grid model to monitor the real-time changes of the weld shape. If a change is detected, the size and shape of the grid unit are dynamically adjusted to match the current geometric characteristics of the weld. The distribution and density of adjacent grid units are adjusted by calculating the change in the weld geometry to obtain an adjusted grid model. The detail capture submodule uses the adjusted mesh model to iteratively improve the model's ability to capture weld details, and identifies key geometric feature areas by analyzing the geometric characteristics of the weld area. If the model needs to capture geometric features, the density of the mesh is increased in the key area, the model is locally refined, and the size of the mesh unit is fine-tuned to obtain a dynamic mesh model of the weld.

4. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The geometric characteristics of the weld area are analyzed using the formula: ； Calculate the geometric feature deviation value, combine the geometric characteristics of the weld area, calculate the local feature parameters through the key area data, identify the key geometric feature area, and if the model needs to capture the geometric features, increase the density of the grid in the key area, perform local refinement on the model, and fine-tune the size of the grid unit to obtain the dynamic grid model of the weld; in, Represents the geometric feature deviation value, Represents the weld area The local curvature length of the sampling points, Represents the average curvature length of all weld sampling points, Representative The area of ​​the mesh unit in the weld area, Representative The density of mesh cells in the weld area, represents the total number of mesh elements in the weld area, represents the summation of all grid cells. Represents the absolute deviation of the weld curvature length from the average curvature length.

5. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The welding morphology change analysis module comprises: The width monitoring submodule analyzes the distribution characteristics of the grid units along the length of the weld based on the weld dynamic grid model, and records the width data of each position by comparing the distance changes between adjacent grid units in the lateral direction, thereby generating weld width change data; The depth monitoring submodule uses the weld width variation data to analyze the variation of the grid unit along the weld depth, measures the height difference of the differentiated depth grid unit, records the distance from the weld surface to the deepest point, extracts the depth variation trend along the length of the weld, and obtains the weld depth variation data; The edge profile monitoring submodule identifies the change of the weld edge profile through the weld depth change data, analyzes the peripheral lines of the grid model, compares the edge profile line changes at different time points, records the fluctuation and offset of the edge profile, tracks the trajectory of the edge profile change, and obtains the weld edge profile data; The feature extraction submodule integrates the change information of width, depth and edge profile based on the weld edge profile data, analyzes the key change characteristics of the weld morphology, and generates welding morphology feature data by comparing and analyzing the change laws and patterns of width, depth and edge profile data.

6. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The weld structure diagram building module includes: The welding morphology recognition submodule analyzes the geometric size and morphology of the weld and defect area based on the welding morphology feature data and sets the node attributes, locates the position of the weld and defect, assigns a unique identifier to each graph node, maps the weld and defect features to the graph node attributes, and generates the welding morphology feature analysis results; The graph node mapping submodule adopts the welding morphological feature analysis results and uses a mapping algorithm to convert the description of the weld and defect area into a graph node, assigns a unique identifier to each graph node and assigns a corresponding attribute value, and quantifies the morphological features of the weld and the defect, and connects the nodes to each other according to the spatial relationship between the weld and the defect to generate a graph node mapping result; The edge definition and graph construction submodule uses spatial position information to sort nodes and establish connection paths based on the graph node mapping results, adjusts connection strength parameters according to the characteristics of the connection path, and refines the connection path and strength parameters through iterative optimization to obtain the carbon steel pipe weld graph structure.

7. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The weld quality graph convolutional network module includes: The structural analysis submodule analyzes each node in the carbon steel pipe weld diagram structure and its connection mode, identifies the position and function of the weld or defect area represented by each node in the overall weld structure, analyzes the direct and indirect connection paths between nodes, and generates a deep analysis result of the weld structure; The quality influencing factor identification submodule uses a network analysis algorithm based on the weld structure in-depth analysis results to analyze the interaction between nodes and their potential impact on the overall quality of the weld. If the mining target node connection mode is associated with quality defects, the node connection mode is analyzed in detail to identify the key nodes and connection methods that affect the weld quality and obtain key quality influencing factors. The quality feature synthesis submodule integrates multiple groups of key quality influencing factors, analyzes the joint action mode of multiple groups of key quality influencing factors on weld quality, studies the interaction between key nodes and connection modes, analyzes the influencing factors of the interaction on weld quality, and generates weld quality feature representation.

8. A carbon steel pipe welding quality assessment system according to claim 1, characterized in that: The network analysis algorithm uses the formula: ； Compute the improved betweenness centrality of the nodes, where For Node The improved betweenness centrality of For Node The weight of For Node and The weight of the path between For Node The position parameters, For path arrive The length parameter, Passing Node The number of shortest paths, For Node To Node The total number of shortest paths.

Images