Fracture structure performance evaluation system and method based on geometric graph neural network

Abstract

The invention provides a fracture structure performance evaluation system and method based on a geometric graph neural network. A target fracture structure is expressed as a fracture network knowledge graph composed of nodes, fracture bodies and nodes. And constructing a crack training sample library through the crack network knowledge graph and the damage state, training the geometric graph neural network to form an algorithm capable of performing structure state evaluation, reasoning a to-be-evaluated crack structure, and determining the damage state of the to-be-evaluated crack structure. According to the method, the implicit relationship between the crack and the structure state can be excavated, the method has important significance in realizing efficient and accurate structure performance evaluation, the research and development difficulty is low, the method is easy to popularize and apply, and the market prospect is very good; in addition, richer and structured data can be provided for an artificial intelligence algorithm, and therefore efficient and accurate structural performance evaluation can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of engineering structure cracks and digital technology, and in particular to a crack structure performance evaluation system and method based on a geometric graph neural network. Background Art

[0002] In concrete structures, cracks not only damage the appearance of the building, but also provide channels for moisture and harmful substances to penetrate into the concrete structure, which in turn causes steel corrosion and affects the bearing capacity and durability of the structure. Structural operation and maintenance engineers need to repair the cracks in the concrete structure to ensure the long-term safety and stability of the building. Traditional crack detection technologies such as visual inspection, ultrasonic testing, radiographic testing, and core sampling have limitations such as high reliance on manual operation, low detection quality and efficiency, and greater risks to the safety of operators.

[0003] With the rapid development of artificial intelligence technologies such as deep learning, engineers have begun to apply these advanced technologies to the field of crack detection. At present, crack detection methods based on deep learning technology mainly include convolutional neural networks (CNN), deepLab series, fully convolutional networks (FCN), U-Net networks, and YOLO algorithms. Although artificial intelligence technology has made significant progress in crack identification, intelligent algorithms can only identify whether it is a crack and the location of the crack through features, but cannot obtain the length and width of the crack. The acquisition of these crack parameters still mainly relies on the traditional OpenCV library method, which obtains local discrete crack information by filtering the crack image with candy, Gaussian, etc., but cannot intelligently evaluate the overall state characteristics of the crack structure.

[0004] The deficiency of traditional crack-based structural state assessment methods lies in the lack of crack representation data structure for intelligent algorithms and intelligent algorithms based on crack representation data structure, which limits the efficient application of artificial intelligence algorithms in crack-based structural performance assessment. Summary of the invention

[0005] The purpose of the present invention is to provide a crack structure performance evaluation system and method based on geometric graph neural network, which is used to solve the problems that the prior art cannot intelligently evaluate the overall state characteristics of the crack structure and limits the efficient application of artificial intelligence algorithms in crack-based structural performance evaluation. It can provide more abundant and structured data for artificial intelligence algorithms, which helps to achieve efficient and accurate structural performance evaluation.

[0006] In order to achieve the above-mentioned object, in a first aspect, the present invention provides a fracture structure performance evaluation system based on a geometric graph neural network, comprising an input module, a fracture map module, a database module, a training module and an evaluation module;

[0007] The input module is used to establish a digital model according to the target fracture structure, wherein the digital model includes multiple independent fractures, each fracture includes a node and a fracture body, the node records the position coordinates of the two ends of the fracture, the fracture body records the fracture information, and all the fractures are connected through common nodes to form a fracture network, wherein the fracture information includes fracture position parameters, basic fracture geometric parameters and fracture shape;

[0008] The crack graph module is used to form a crack network knowledge graph of the target crack structure according to the crack network, and store the crack network knowledge graph using an adjacency matrix or a linked list;

[0009] The database module is used to obtain the damage state of the target crack structure, and form a crack training sample library according to the crack network knowledge graph and the damage state of the target crack structure;

[0010] The training module is used to construct a geometric graph neural network, and train the geometric graph neural network using the crack training sample library;

[0011] The evaluation module is used to determine the damage state of the crack structure to be evaluated based on the trained geometric graph neural network.

[0012] According to a fracture structure performance evaluation system based on a geometric graph neural network provided by the present invention, the fracture position parameters include fracture axis intersection endpoint numbers and fracture actual endpoint numbers.

[0013] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the crack shape refers to the curve shapes of the upper side, the lower side and the middle axis of the crack respectively described by three mathematical expressions.

[0014] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the three mathematical expressions specifically include: constructing a global coordinate system based on the crack grid image, and using polynomial fitting to generate mathematical expressions of the upper side, lower side and central axis of the crack.

[0015] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the basic geometric parameters of the crack include an average crack width, a maximum crack width, a minimum crack width, and a crack length.

[0016] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the average crack width, the maximum crack width, and the minimum crack width are obtained by calculating the shortest distance between the upper side and the lower side of the crack through mathematical expressions, and the crack length is calculated through a mathematical expression of the crack center axis.

[0017] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the construction of the geometric graph neural network includes: adding position coordinate information based on a global coordinate system to the node data of the graph neural network, and the input data of the graph neural network includes an adjacency matrix or a linked list stored based on the crack network knowledge graph.

[0018] According to a crack structure performance evaluation system based on a geometric graph neural network provided by the present invention, the graph neural network includes a graph convolutional neural network, a graph attention network, a graph autoencoder, a graph generation network or a graph heterogeneous network.

[0019] In a second aspect, the present invention also provides a method for evaluating crack structure performance based on a geometric graph neural network, comprising:

[0020] Step 1: Select the target crack structure and use an imaging device to take pictures to obtain a crack grid image;

[0021] Step 2: Identify and digitally locate the crack grid image to determine the position coordinates of each node of the crack network in the crack grid image in the global coordinate system;

[0022] Step 3: Cut the crack network into multiple image units with a single crack, and obtain the crack information in each image unit;

[0023] Step 4: assemble all the cracks through common nodes according to the crack information to form a crack network knowledge graph;

[0024] Step 5: storing the crack network knowledge graph using an adjacency matrix or a linked list;

[0025] Step 6, repeating steps 1 to 5 multiple times to obtain a crack network knowledge graph of multiple target crack structures, and respectively determining the damage states of the multiple target crack structures to obtain a crack training sample library, which includes the crack network knowledge graphs and damage states of the multiple target crack structures;

[0026] Step 7: construct a geometric graph neural network, and train the geometric graph neural network using the crack training sample library;

[0027] Step 8: Based on the trained geometric graph neural network, determine the damage state of the crack structure to be evaluated.

[0028] The present invention provides a crack structure performance evaluation system and method based on a geometric graph neural network. The system includes an input module, a crack map module, a database module, a training module and an evaluation module. The target crack structure is represented as a crack network knowledge map composed of "node-crack body-node". A crack training sample library is constructed through the crack network knowledge map and damage state, and the geometric graph neural network is trained to form an algorithm that can evaluate the structural state. The crack structure to be evaluated is inferred to determine its damage state. The present invention helps to explore the implicit relationship between cracks and structural states, and is of great significance in achieving efficient and accurate structural performance evaluation. It has low research and development difficulty, is easy to promote and apply, and has a good market prospect. In addition, the present invention can provide more abundant and structured data for artificial intelligence algorithms, thereby helping to achieve efficient and accurate structural performance evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] In the attached picture:

[0031] Figure 1 This is a physical diagram of a concrete structure of the present invention;

[0032] Figure 2 A schematic diagram of a fracture network of a target fracture structure of the present invention;

[0033] Figure 3 A schematic diagram of the fracture network knowledge graph of the present invention;

[0034] Figure 4 A schematic diagram of a digital fracture model of the present invention;

[0035] Figure 5 This is a schematic diagram of the fracture geometry fitting of the present invention;

[0036] Figure 6 A schematic diagram of the fracture network knowledge graph of the present invention being stored in an adjacency matrix manner;

[0037] Figure 7 This is a schematic diagram of the information transmission principle of the geometric graph neural network of the present invention;

[0038] Figure 8 It is a structural block diagram of the crack structure performance evaluation system based on geometric graph neural network of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Some embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0041] After in-depth research, the inventors of the present invention found that digital storage of cracks is crucial for structural damage rating based on digital image vision technology. Digital storage can not only improve the accuracy and efficiency of crack detection, but also provide richer and more structured data for artificial intelligence algorithms, thereby helping to explore the implicit relationship between cracks and structural states, and achieving efficient and accurate structural performance assessment.

[0042] Based on this, in order to promote the in-depth application of intelligent technology in the field of concrete structure health monitoring and damage assessment, please refer to Figure 8 , an embodiment of the present invention provides a fracture structure performance evaluation system based on a geometric graph neural network, including an input module, a fracture map module, a database module, a training module and an evaluation module;

[0043] The input module is used to establish a digital model according to the target fracture structure. The digital model includes multiple independent fractures. Each fracture includes a node and a fracture body. The node records the position coordinates of the two ends of the fracture, and the fracture body records the fracture information. All fractures are connected through common nodes to form a fracture network. The fracture information includes fracture position parameters, basic fracture geometric parameters, and fracture shape, etc.

[0044] The crack graph module is used to form a crack network knowledge graph of the target crack structure according to the crack network, and store the crack network knowledge graph using an adjacency matrix or a linked list;

[0045] The database module is used to obtain the damage state of the target crack structure and form a crack training sample library according to the crack network knowledge graph and damage state of the target crack structure;

[0046] The training module is used to construct a geometric graph neural network and train the geometric graph neural network using a crack training sample library;

[0047] The evaluation module is used to determine the damage state of the crack structure to be evaluated based on the trained geometric graph neural network.

[0048] Specifically, the crack graph module can represent the crack network as a crack network knowledge graph consisting of "node-crack body-node".

[0049] Specifically, the crack information can be stored through data structures such as JSON and arrays, and parsed through corresponding parsing algorithms.

[0050] Among them, the crack position parameters include the intersection endpoint number of the crack axis and the actual endpoint number of the crack. The basic geometric parameters of the crack include information such as the average width of the crack, the maximum width of the crack, the minimum width of the crack, and the length of the crack, which can be obtained by calculating the mathematical expression of the crack shape. The width of the crack (average width, maximum width, minimum width) is obtained by the shortest distance between the upper side and the lower side of the crack. The length of the crack refers to the length calculated by the mathematical expression of the crack center axis. The crack shape refers to the shape of the crack described by a mathematical expression, including three mathematical expressions for the upper side, lower side and center axis of the crack, specifically including: constructing a global coordinate system based on the image pixel grid map of the crack (crack grid image), using polynomial fitting to generate the mathematical expression of the crack, and storing the coefficients of the mathematical expression in order of the power size of the variable, generally an array or linked list data structure can be selected.

[0051] Specifically, the state of the crack structure mainly refers to the damage state, which can be characterized by the damage level. For example, it can be divided into 5 levels according to "very serious damage, serious damage, moderate damage, slight damage, no damage", corresponding to "level 4, level 3, level 2, level 1, level 0" respectively. In practical applications, the damage level assessment of the crack structure is usually based on relevant industry standards or specifications, such as the "Standard for Inspection and Appraisal of Building Structures" (GB / T50315), "Standard for Technical Condition Assessment of Highway Bridges" (JTG / TH21), etc. In some embodiments, the damage state of the crack structure can be determined by a structural assessment expert through an assessment of the state of the target crack structure to determine the damage level of the crack structure.

[0052] In some embodiments, constructing a geometric graph neural network includes: adding position coordinate information based on a global coordinate system to the node data of the graph neural network, and the input data of the graph neural network includes an adjacency matrix or a linked list stored based on a crack network knowledge graph.

[0053] It is understandable that if Figure 7As shown in the figure, geometric graph neural networks (GGNNs) add location coordinate information based on the Cartesian global coordinate system to the node data of traditional graph neural networks (GNNs). The corresponding input data is the adjacency matrix, linked list and other information stored based on the crack network knowledge graph, and its aggregation function needs to consider the geometric properties of the nodes. Commonly used graph neural networks include graph convolutional neural networks (GCN), graph attention networks (GAT), graph autoencoders (GAE), graph generation networks (GGN), and graph heterogeneous networks (HeteroGNN). Specifically, the comprehensive width, length and other information of the node contains the information of the width, length and other information of all cracks with the node as the endpoint, which is aggregated after considering the influence coefficient. The specific formula is:

[0054] Node information node[i] = [{x i ,y i , z i},{Nwidth i ,Nlength i}]

[0055] Node information node[j] = [{x j ,y j , z j},{Nwjdth j ,Nlength j}]

[0056] Average width of node cracks

[0057] Node crack length

[0058] Crack main body information edge[i, j] = [{node[i], node[j]}, {Ewidth i,j , Elength i,j ,MaxEwidth i,j}]

[0059] Compared with traditional graph neural networks, the geometric graph neural network in the embodiment of the present invention not only focuses more on the structural properties of the graph and the characteristics of nodes / edges, but also needs to consider additional spatial dimensions and geometric constraints when processing data with explicit geometric properties. This is reflected in data preprocessing, network architecture design, and loss function definition. For example, the wider and longer the cracks, the more serious the structural damage caused. The cracks located in high stress areas and key parts of the structure have a more serious impact on structural damage. When the width of the crack exceeds the threshold, the damage to the structure will have a sudden effect. The formula considering the above-mentioned influencing factors is as follows, and the expression of the influence coefficient a[i,j] is:

[0060]

[0061] Where Ewidth i,j is the width of the crack located by the two nodes node[i] and node[j],

[0062] Elength i,j is the length of the crack located by the two nodes node[i] and node[j],

[0063] e i,j is the attention value between node[i] and node[j],

[0064] Limwidth is the critical crack width set by the code that seriously affects the durability of the structure.

[0065] MaxEwidth i,j is the maximum width of the crack located by the two nodes node[i] and node[j].

[0066] It should be noted that Figure 1 This is a physical picture of a concrete structure (i.e., the target crack structure). Figure 2 is a schematic diagram of the crack network of the target crack structure, Figure 2 There are three cracks in the figure, marked as crack 1, crack 2, and crack 3. The intersection of the axes of crack 1 and crack 2 is point A. Crack 1 and crack 3 do not intersect, but the intersection of their axes is point B, that is, one of the nodes of crack 3 is at a distance from the starting point of crack 1. Accordingly, Figure 3 In the figure, the two nodes of crack 1 are numbered as node[1] and node[5]; the two nodes of crack 2 are numbered as node[2] and node[4]; the two nodes of crack 3 are actually numbered as node[3,1] and node[6,1], and the intersection endpoints of the axes are numbered as node[3] and node[6]; there are five crack bodies formed between the three cracks: edge[1,2], edge[2,3], edge[3,5], edge[2,4], and edge[3,6]. Figure 4 This is a schematic diagram of the digital crack model, showing the crack body edge[i,j] between nodes node[i] and node[j]. The mathematical expression is: Y = a0+a1x+a2x 2 +a3x 3 +……, the corresponding node coordinates, the variable coefficients of the mathematical expression constitute the geometric expression, geometric characteristics and the crack body. Figure 5 The three mathematical expressions of the upper side, lower side and middle axis of the crack are Y u =f(x1),Y c=a0+a1x+a2x 2 +a3x 3 +……、Y b =f(x2), and the maximum width, average width and length of the crack are shown. Figure 6 Schematic diagram of the storage of the crack network knowledge graph using the adjacency matrix method.

[0067] Based on the same inventive concept, another embodiment of the present invention provides a method for evaluating crack structure performance based on a geometric graph neural network, which corresponds to the system of the aforementioned embodiment, and includes:

[0068] Step 1: Select the target crack structure and use an imaging device to take pictures to obtain a crack grid image;

[0069] Specifically, in step 1, a digital camera is used manually or by an unmanned aerial vehicle to take photos of the cracks or record videos to obtain crack grid images, or convert video frames into crack grid images, and form a crack image library.

[0070] Step 2: Identify and digitally locate the crack grid image, and determine the position coordinates of each node of the crack network in the crack grid image in the global coordinate system;

[0071] Specifically, pre-trained deep learning algorithms such as YOLO and UNET with crack recognition function can be used to identify cracks in the crack grid image and perform digital positioning to determine the position coordinates of each node in the crack network in the global coordinate system.

[0072] Step 3: Cut the crack network into multiple image units with a single crack, and obtain the crack information in each image unit;

[0073] Specifically, the crack network can be positioned and cropped through the OpenCV library to form an image unit with a single crack, and then the background noise of the single crack can be removed through filtering algorithms such as candy and Gaussian, and the skeleton of the single crack can be stripped out. Based on each image unit with a single crack, the geometric shapes of the upper side, lower side and central axis of the crack are polynomially fitted to form a mathematical expression. According to the proportional relationship between the crack length and the pixel point, mathematical calculations are performed based on the above three mathematical expressions to obtain the crack length, average crack width, maximum crack width, minimum crack width and crack position parameters. The traversal loop performs the above processing on all image units with a single crack to obtain the crack information of all cracks.

[0074] Step 4: Assemble all cracks through common nodes according to the crack information to form a crack network knowledge graph;

[0075] Step 5: Store the crack network knowledge graph in the form of an adjacency matrix or a linked list.

[0076] Specifically, the crack network knowledge graph can be stored in a physical hard disk in the form of an adjacency matrix or a linked list.

[0077] Step 6, repeating steps 1 to 5 multiple times to obtain a crack network knowledge graph of multiple target crack structures, and respectively determining the damage states of the multiple target crack structures to obtain a crack training sample library, which includes the crack network knowledge graphs and damage states of the multiple target crack structures;

[0078] Step 7: construct a geometric graph neural network, and train the geometric graph neural network using a crack training sample library;

[0079] Step 8: Based on the trained geometric graph neural network, determine the damage state of the crack structure to be evaluated. This makes it easier to make scientific decisions on policies such as reinforcement, repair and maintenance of the structure based on the damage state of the crack structure to be evaluated.

[0080] In summary, the present invention provides a crack structure performance evaluation system and method based on a geometric graph neural network, which represents the target crack structure as a crack network knowledge graph composed of "node-crack body-node". A crack training sample library is constructed through the crack network knowledge graph and damage state, and the geometric graph neural network is trained to form an algorithm that can evaluate the structural state, and the crack structure to be evaluated is inferred to determine its damage state. The relevant knowledge graph technology, crack feature recognition technology, and geometric graph neural network technology are mature existing technologies, and the present invention helps to explore the implicit relationship between cracks and structural states, which is of great significance for achieving efficient and accurate structural performance evaluation. It has low research and development difficulty, is easy to promote and apply, and has a good market prospect. In addition, the present invention can realize the digital storage of cracks, which is crucial for structural damage rating based on digital image vision technology. Digital storage can not only improve the accuracy and efficiency of crack detection, but also provide more abundant and structured data for artificial intelligence algorithms, thereby helping to achieve efficient and accurate structural performance evaluation.

[0081] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the embodiments disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A crack structure performance evaluation system based on geometric graph neural network, characterized in that: It includes input module, crack map module, database module, training module and evaluation module; The input module is used to establish a digital model according to the target fracture structure, wherein the digital model includes multiple independent fractures, each fracture includes a node and a fracture body, the node records the position coordinates of the two ends of the fracture, the fracture body records the fracture information, and all the fractures are connected through common nodes to form a fracture network, wherein the fracture information includes fracture position parameters, basic fracture geometric parameters and fracture shape; The crack graph module is used to form a crack network knowledge graph of the target crack structure according to the crack network, and store the crack network knowledge graph using an adjacency matrix or a linked list; The database module is used to obtain the damage state of the target crack structure, and form a crack training sample library according to the crack network knowledge graph and the damage state of the target crack structure; The training module is used to construct a geometric graph neural network, and train the geometric graph neural network using the crack training sample library; The evaluation module is used to determine the damage state of the crack structure to be evaluated based on the trained geometric graph neural network.

2. The crack structure performance evaluation system based on geometric graph neural network according to claim 1 is characterized in that: The crack position parameters include the crack axis intersection endpoint number and the crack actual endpoint number.

3. The crack structure performance evaluation system based on geometric graph neural network according to claim 1 is characterized in that: The crack shape refers to the curve shapes of the upper side, lower side and middle axis of the crack described by three mathematical expressions respectively.

4. The crack structure performance evaluation system based on geometric graph neural network according to claim 3 is characterized in that: The three mathematical expressions specifically include: constructing a global coordinate system based on the crack grid image, and generating mathematical expressions of the upper side, lower side and middle axis of the crack by using a polynomial fitting method.

5. The crack structure performance evaluation system based on geometric graph neural network according to claim 3 is characterized in that: The basic geometric parameters of the crack include the average crack width, the maximum crack width, the minimum crack width, and the crack length.

6. The crack structure performance evaluation system based on geometric graph neural network according to claim 5 is characterized in that: The average width of the crack, the maximum width of the crack, and the minimum width of the crack are obtained by calculating the shortest distance between the upper side and the lower side of the crack through a mathematical expression, and the length of the crack is calculated through a mathematical expression of the center axis of the crack.

7. The crack structure performance evaluation system based on geometric graph neural network according to claim 6 is characterized in that: The construction of the geometric graph neural network includes: adding position coordinate information based on a global coordinate system to the node data of the graph neural network, and the input data of the graph neural network includes an adjacency matrix or a linked list stored based on the crack network knowledge graph.

8. The crack structure performance evaluation system based on geometric graph neural network according to claim 7 is characterized in that: The graph neural network includes a graph convolutional neural network, a graph attention network, a graph autoencoder, a graph generation network or a graph heterogeneous network.

9. A method for evaluating crack structure performance based on geometric graph neural network, characterized in that: include: Step 1: Select the target crack structure and use an imaging device to take pictures to obtain a crack grid image; Step 2: Identify and digitally locate the crack grid image to determine the position coordinates of each node of the crack network in the crack grid image in the global coordinate system; Step 3: Cut the crack network into multiple image units with a single crack, and obtain the crack information in each image unit; Step 4: assemble all the cracks through common nodes according to the crack information to form a crack network knowledge graph; Step 5: storing the crack network knowledge graph using an adjacency matrix or a linked list; Step 6, repeating steps 1 to 5 multiple times to obtain a crack network knowledge graph of multiple target crack structures, and respectively determining the damage states of the multiple target crack structures to obtain a crack training sample library, which includes the crack network knowledge graphs and damage states of the multiple target crack structures; Step 7: construct a geometric graph neural network, and train the geometric graph neural network using the crack training sample library; Step 8: Based on the trained geometric graph neural network, determine the damage state of the crack structure to be evaluated.

Images