Boiler pressure vessel inner surface crack detection system and method

Abstract

The invention discloses a boiler pressure vessel inner surface crack detection system and method, and relates to the technical field of equipment anomaly detection.The method comprises the steps that a digital twin body is built according to a boiler geometric model, and a monitoring point set is generated in the digital twin body according to a preset monitoring point generation rule, calling historical operation and maintenance data of each monitoring point and mapping the historical operation and maintenance data to the digital twinborn body; collecting the preprocessed ultrasonic signal data and acoustic emission signal data of each monitoring point to obtain a feature vector set, and associating the feature vector set with the digital twin to form a preliminary anomaly detection graph; and inputting the feature vector set into a crack identification model, outputting a crack position set and a corresponding confidence score set, updating and mapping the crack position set and the corresponding confidence score to the digital twin, and generating a crack distribution diagram, thereby enhancing the visual expression of the crack information, and improving the crack identification accuracy. And comprehensive and accurate identification and detection are provided for the crack state of the boiler pressure vessel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment anomaly detection, and in particular to a boiler pressure vessel inner surface crack detection system and method. BACKGROUND

[0002] As an important component of pressure vessels, boilers are widely used in energy, chemical and other fields, and are long-term operated in high-temperature and high-pressure environments. The inner surface of the shell is prone to cracks due to factors such as thermal fatigue and stress corrosion, which seriously affects the safety and service life of the equipment. In order to ensure the safe and reliable operation of the boiler, crack detection has become a key link in the operation and maintenance management of the boiler.

[0003] The existing crack detection methods mainly rely on non-destructive testing technologies such as ultrasonic detection and acoustic emission detection, and collect detection signals through manual or semi-automatic methods, and combine experience to identify cracks. This kind of method has certain detection accuracy, can realize the identification of the boiler surface crack to a certain extent, has the advantages of relatively low equipment investment and strong detection flexibility, and has been widely used in industrial field. However, with the increase of the complexity of the boiler structure and the diversification of the operating environment, the traditional detection method gradually exposes many shortcomings: Firstly, the existing method usually lacks high-precision modeling of the overall structure of the boiler, resulting in insufficient spatial positioning accuracy in the crack detection process, and it is difficult to realize the global visualization of the crack distribution; Secondly, the existing technology is based on single detection data, lacks deep fusion with historical operation and maintenance data, cannot effectively capture the crack development trend, affects the timeliness and integrity of the detection result, and further, in the crack feature extraction and identification process, limited by data processing means and feature modeling capability, the local accuracy and spatial distribution expression of the detection exist deficiencies, it is difficult to intuitively present the density, confidence and spatial connectivity of the crack, which affects the reliability and decision reference value of the crack detection result. SUMMARY

[0004] In order to solve the above technical problems, the present application provides a boiler pressure vessel inner surface crack detection system, which comprises: A structure modeling module generates a boiler geometric model according to the obtained boiler structure data; A twin body generation module constructs a digital twin body according to the boiler geometric model, generates a set of monitoring points in the digital twin body according to a preset monitoring point generation rule, and maps the historical operation and maintenance data of each monitoring point to the digital twin body; An anomaly detection graph construction module acquires the preprocessed ultrasonic signal data and acoustic emission signal data of each monitoring point to obtain a feature vector set, associates the feature vector set with the digital twin body, and forms a preliminary anomaly detection graph; The crack location module inputs the feature vector set into the crack recognition model, outputs the crack location set and the corresponding confidence score set, updates and maps the crack location set and its corresponding confidence score to the digital twin, and generates a crack distribution map.

[0005] Furthermore, the boiler structural data includes boiler shell length, boiler shell outer diameter, furnace shell outer wall curvature, shell wall thickness, weld location, weld material properties, and welding process parameters.

[0006] Furthermore, the steps for constructing a digital twin based on the boiler's geometric model are as follows: The twin generation module includes a 3D spatial mapping unit and an attribute definition unit; The spatial mapping unit reads the coordinate data of each coordinate point in the boiler geometric model, establishes a spatial coordinate system Σ(x, y, z), and reconstructs the boiler shape point by point in the digital space based on the outer wall curvature field K(x, y, z) and thickness distribution H(x, y, z) defined in the geometric model. A continuous shell surface is generated by a surface reconstruction algorithm, and the weld area and the area with abnormal wall thickness are partitioned to form a multi-resolution spatial mesh. The weld location, weld material properties, and welding process parameters are bound to the corresponding mesh nodes through attribute definition units to form a digital twin with multi-dimensional geometric and material properties, and outputs a digital twin with spatial consistency and attribute mapping integrity corresponding to the physical boiler.

[0007] Furthermore, the logic for generating a set of monitoring points in the digital twin according to preset monitoring point generation rules, and retrieving historical operation and maintenance data of each monitoring point and mapping it to the digital twin is as follows: Based on the established rules for generating monitoring points, the surface of the digital twin is traversed, and spatial coordinate points that conform to the rules are selected to generate a set of monitoring points. 'm' is an identifier symbol, representing each monitoring point. The three-dimensional coordinates are ( , , ), ∈m; For each monitoring point, the data interface module is invoked to retrieve the historical operation and maintenance dataset for the corresponding area from the operation and maintenance database. The historical operation and maintenance dataset includes temperature, pressure, and runtime. Historical operation and maintenance datasets are bound to monitoring points in chronological order, generating operation and maintenance mapping data in the digital twin.

[0008] Furthermore, the preset rules for generating monitoring points are as follows: Based on the principle of spatial uniformity, with a fixed spacing Distribute dots evenly along the shell surface; distribute dots more densely within an area of ​​A mm around the weld; distribute dots more densely in areas where the wall thickness gradient exceeds a preset gradient threshold.

[0009] Furthermore, the logic for obtaining the anomaly detection map is as follows: Based on each monitoring point, quantitative features are extracted from the preprocessed ultrasonic signal data and acoustic emission signal data. The quantitative features include ultrasonic signal features and acoustic emission signal features. The ultrasonic signal features include dominant frequency, signal energy, and peak amplitude. The acoustic emission signal features include cumulative energy, signal count, and signal duration. Each monitoring point is constructed based on ultrasonic signal characteristics and acoustic emission signal characteristics. Corresponding feature vector Construct a set of feature vectors ; Set of feature vectors of all monitoring points The data is sequentially mapped to the corresponding monitoring points of the digital twin to update the monitoring point attributes of the digital twin and form a binding relationship. " In the spatial coordinate system Σ(x, y, z) of the digital twin, with the feature vector As node attributes, a feature field for monitoring points is constructed; based on the updated digital twin, an anomaly detection map is generated according to the following logic: Using the feature vector set V as node attributes, the node connection relationship is established based on the spatial proximity relationship of the monitoring points. The anomaly score of each node is calculated, and nodes with anomaly scores higher than the preset score threshold are marked as preliminary anomaly points. Finally, an anomaly detection map is output.

[0010] Furthermore, the logic for outputting the set of crack locations and the corresponding set of confidence scores is as follows: Set of feature vectors of monitoring points The data is input into a preset crack recognition model to obtain recognition results, which include a judgment result and a confidence score. The output logic for the recognition result is as follows: When the judgment result output is 0, the boiler has no cracks; when the judgment result output for crack generation is 1, the boiler has cracks. The confidence score ranges from 0 to 1. The higher the confidence score, the greater the confidence that the crack exists. Iterate through the judgment results of all monitoring points, extract the monitoring points whose judgment result is 1, and construct a set of crack location points. , For the Kth crack location, construct a confidence set based on the confidence scores of each location. , This represents the confidence score corresponding to the Kth crack location.

[0011] Furthermore, the generation logic of the crack recognition model is as follows: Acquire historical monitoring point identification data, and divide the historical monitoring point identification data into an identification training set and an identification test set. The historical monitoring point identification data includes monitoring point feature vectors and corresponding judgment results and confidence scores. Configure the initial classifier by taking the feature vectors of the monitoring points of the sub-components in the identification training set as the input data of the initial classifier, and taking the corresponding judgment results and confidence scores in the identification training set as the output data of the initial classifier. Train the initial classifier to obtain the initial classification network. The anomaly detection graph construction module S13 verifies the initial classification network by recognizing the test set, and outputs an initial classification network with an accuracy greater than or equal to the preset test accuracy as a pre-built crack recognition model.

[0012] Furthermore, the logic for updating and mapping the set of crack locations and their corresponding confidence scores to the digital twin to generate a crack distribution map is as follows: In the spatial coordinate system Σ(x, y, z) of the digital twin, the crack location points are marked as crack nodes; Confidence score As attribute values ​​of crack nodes, to form attribute mapping relationships. ”; In the digital twin, with the crack node as the center, a spatial connection threshold Q is set. If the Euclidean distance between two crack nodes is less than or equal to Q, an undirected connection edge is established between the two crack nodes. An edge set E is constructed based on multiple sets of undirected connection edges. Construct a crack distribution map G using a set of nodes L and a set of edges E, where G = (L, E, P), and output the crack distribution map G.

[0013] A method for detecting cracks on the inner surface of a boiler or pressure vessel, applicable to any boiler or pressure vessel inner surface crack detection system, the method comprising: Boiler structure modeling: Generate a boiler geometric model based on the acquired boiler structure data; Digital twin construction: A digital twin is constructed based on the boiler geometric model. A set of monitoring points is generated in the digital twin according to the preset monitoring point generation rules. Historical operation and maintenance data of each monitoring point are retrieved and mapped into the digital twin. Anomaly detection map construction: Collect preprocessed ultrasonic signal data and acoustic emission signal data from each monitoring point to obtain a set of feature vectors, associate the set of feature vectors with the digital twin to form a preliminary anomaly detection map; Crack identification and localization: The feature vector set is input into the crack identification model, which outputs the crack location set and the corresponding confidence score set. The crack location set and its corresponding confidence score are updated and mapped to the digital twin to generate a crack distribution map.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention significantly improves data spatial consistency by performing high-precision three-dimensional modeling of key geometric characteristics and material parameters of boiler shell, combined with the construction of spatial coordinate system and attribute binding, and enhances the deep correlation of structural information based on the mapping of monitoring point set and multi-dimensional operation and maintenance data, so as to achieve the synchronous unification of spatial positioning and state perception. This invention also improves the detection accuracy and distribution clarity of local abnormal states through multimodal feature extraction and spatial feature field construction. Combined with crack identification and node confidence mapping, the generated crack distribution map can intuitively present the specific location, quantity density and spatial connectivity of cracks in three-dimensional space, which greatly enhances the visualization of crack information and provides comprehensive and accurate identification and detection of crack states in boiler pressure vessels. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a block diagram of a boiler pressure vessel internal surface crack detection system provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a method for detecting cracks on the inner surface of a boiler pressure vessel, provided in Embodiment 2 of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0018] Please see Figure 1 As shown in the figure, this embodiment discloses a system for detecting cracks on the inner surface of a boiler pressure vessel. The system includes: The structural modeling module S11 generates a boiler geometric model based on the acquired boiler structural data. It should be noted that the boiler geometric model is a three-dimensional model; Specifically, the boiler structural data includes boiler shell length, boiler shell outer diameter, furnace shell outer wall curvature, shell wall thickness, weld location, weld material properties, and welding process parameters; It should be noted that: the boiler shell length and outer diameter are obtained using laser ranging; the outer wall curvature is obtained using 3D laser scanning; the shell wall thickness is collected using an ultrasonic thickness gauge; weld information is collected; weld location is obtained using 3D visual scanning; and weld material properties and welding process parameters are obtained using historical welding data. Based on the above data, a 3D modeling software was used to generate a boiler geometric model, and the curvature K of the outer wall of the furnace shell was defined. , , With shell wall thickness H ( , , ); Wherein, K( , , ) indicates in spatial coordinates ( , , The curvature of the outer wall of the shell; H( , , ) indicates in spatial coordinates ( , , The shell wall thickness.

[0019] It should be noted that the 3D modeling software mentioned includes, but is not limited to, CAD, CATIA, and UG; The structural modeling module S11 performs high-precision 3D modeling of the spatial morphology and wall thickness variation of boiler pressure vessels, effectively ensuring the accurate reproduction of physical structural characteristics in digital space, improving the geometric consistency and spatial accuracy of subsequent inspection objects, and providing reliable data support for crack detection and analysis under complex structural conditions. The twin generation module S12 constructs a digital twin based on the boiler geometric model, generates a set of monitoring points in the digital twin according to the preset monitoring point generation rules, and retrieves the historical operation and maintenance data of each monitoring point and maps it into the digital twin. Specifically, the steps for constructing a digital twin based on the boiler's geometric model are as follows: The twin generation module includes a 3D spatial mapping unit and an attribute definition unit; The spatial mapping unit reads the coordinate data of each coordinate point in the boiler geometric model, establishes a spatial coordinate system Σ(x, y, z), and reconstructs the boiler shape point by point in the digital space based on the outer wall curvature field K(x, y, z) and thickness distribution H(x, y, z) defined in the geometric model. A continuous shell surface is generated by a surface reconstruction algorithm, and the weld area and the area with abnormal wall thickness are partitioned to form a multi-resolution spatial mesh. It should be noted that the surface reconstruction algorithm includes, but is not limited to, Bezier surface fitting or B-spline surface reconstruction. The weld location, weld material properties, and welding process parameters are bound to the corresponding mesh nodes through attribute definition units to form a digital twin with multi-dimensional geometric and material properties, and outputs a digital twin with spatial consistency and attribute mapping integrity corresponding to the physical boiler.

[0020] Specifically, the logic for generating a set of monitoring points in the digital twin according to preset monitoring point generation rules, and retrieving historical operation and maintenance data of each monitoring point and mapping it to the digital twin is as follows: S121, Based on the established monitoring point generation rules, traverse the surface of the digital twin, select spatial coordinate points that conform to the rules, and generate a set of monitoring points. m is an identifier, m≥1, for each monitoring point The three-dimensional coordinates are ( , , ), ∈m; Specifically, the preset rules for generating monitoring points are as follows: Based on the principle of spatial uniformity, with a fixed spacing Distribute dots evenly along the shell surface; increase the density of dots within an A mm radius around the weld; increase the density of dots in areas where the wall thickness gradient exceeds a preset gradient threshold. S122, for each monitoring point The system calls the data interface module to retrieve the historical operation and maintenance dataset for the corresponding region from the operation and maintenance database. The historical operation and maintenance dataset includes temperature, pressure, and runtime. S123, bind historical operation and maintenance datasets to monitoring points in chronological order, and generate operation and maintenance mapping data in the digital twin; The twin generation module S12 establishes a deep correlation between structural information and operation and maintenance data based on three-dimensional spatial mapping and multi-dimensional attribute fusion, laying a data foundation for the accurate identification of crack anomaly features; The anomaly detection map construction module S13 collects preprocessed ultrasonic signal data and acoustic emission signal data from each monitoring point to obtain a set of feature vectors, and associates the set of feature vectors with the digital twin to form a preliminary anomaly detection map. It should be noted that the preprocessing of the ultrasound signal data is as follows: noise suppression is performed using a Butterworth bandpass filter with a filter frequency range of 0.5MHz to 10MHz, and then the envelope extraction algorithm is used to extract the signal energy profile. Acoustic emission signal data: Wavelet denoising technology was used, and four-level wavelet decomposition was performed based on Daubechies wavelet (db4) to retain the principal component signal and normalize the amplitude.

[0021] Specifically, the logic for obtaining the anomaly detection map is as follows: S131, based on each monitoring point Quantitative features are extracted from the preprocessed ultrasonic signal data and acoustic emission signal data, respectively. The quantitative features include ultrasonic signal features and acoustic emission signal features. The ultrasonic signal characteristics include dominant frequency, signal energy, and peak amplitude; It should be noted that: the dominant frequency is the maximum frequency component of the signal energy extracted through Short Time Fourier Transform (STFT), the signal energy envelope is the integral area of ​​the signal, and the peak amplitude is the maximum amplitude of the signal in the time domain. The acoustic emission signal is characterized by cumulative energy, signal count, and signal duration; It should be noted that: the cumulative energy of the feature is the cumulative value of the signal energy on the time axis; the signal count is the number of signal peaks that exceed the set threshold; and the signal duration is the time span from the first trigger to the end of the signal. S132, each monitoring point is constructed based on ultrasonic signal characteristics and acoustic emission signal characteristics. Corresponding feature vector Construct a set of feature vectors , ∈m; Feature vectors of all monitoring points The sets are sequentially mapped to the monitoring points corresponding to the digital twins. To update the monitoring point attributes of the digital twin and establish a binding relationship. ”; S133, in the spatial coordinate system Σ(x, y, z) of the digital twin, with the feature vector As node attributes, construct the feature field of monitoring points; It should be noted that the execution of step S33 is to ensure that each spatial location not only contains coordinate information, but also carries feature information related to crack detection; S134, Based on the updated digital twin, generate an anomaly detection graph according to the following logic: Using the feature vector set V as node attributes, node connections are established based on the spatial proximity of monitoring points. Anomaly scores are calculated for each node, and nodes with anomaly scores higher than a preset score threshold are marked as preliminary anomalies. The final output is an anomaly detection map. It should be noted that node connections can be calculated using Delaunay triangulation or the K-nearest neighbor algorithm. The formula for calculating the anomaly score is as follows: ; In the formula, For abnormal scoring, For monitoring points In the Node attributes on each node This represents the average value of the attributes of each node. The standard deviation of the data at each node. Let α be a constant greater than 0, and let α be the local deviation amplification exponent. Preferably, α > 1. For the first The weight parameters corresponding to each node The preferred amplification index for global anomalies is... >1; It should be noted that: The settings are based on historical experimental data; By combining anomaly detection with spatial correlation analysis based on feature vectors, the local accuracy and spatial clarity of anomaly detection results are significantly improved, enabling sensitive capture of potential anomalous states and precise characterization of spatial distribution characteristics. The crack location module S4 inputs the feature vector set into the crack recognition model, outputs the crack location set and the corresponding confidence score set, updates and maps the crack location set and its corresponding confidence score to the digital twin, and generates a crack distribution map. Specifically, the logic for outputting the set of crack locations and the corresponding set of confidence scores is as follows: Set of feature vectors of monitoring points The data is input into a preset crack recognition model to obtain recognition results, which include a judgment result and a confidence score. Specifically, the generation logic of the crack recognition model is as follows: Acquire historical monitoring point identification data, and divide the historical monitoring point identification data into an identification training set and an identification test set. The historical monitoring point identification data includes monitoring point feature vectors and corresponding judgment results and confidence scores. Configure the initial classifier by taking the feature vectors of the monitoring points of the sub-components in the identification training set as the input data of the initial classifier, and taking the corresponding judgment results and confidence scores in the identification training set as the output data of the initial classifier. Train the initial classifier to obtain the initial classification network. The anomaly detection graph construction module S13 verifies the initial classification network by recognizing the test set, and outputs an initial classification network with an accuracy greater than or equal to the preset test accuracy as a pre-built crack recognition model.

[0022] It should be noted that the crack recognition model includes, but is not limited to, convolutional neural networks, graph convolutional networks, and transformer networks. The crack recognition model includes at least one feature extraction layer, multiple nonlinear mapping layers, and a classification and discrimination layer. The output logic for the recognition result is as follows: When the judgment result output is 0, the boiler has no cracks; when the judgment result output for crack generation is 1, the boiler has cracks. The confidence score ranges from 0 to 1. The higher the confidence score, the greater the confidence that the crack exists.

[0023] Iterate through the judgment results of all monitoring points, extract the monitoring points whose judgment result is 1, and construct a set of crack location points. , For the Kth crack location, construct a confidence set based on the confidence scores of each location. , The confidence score is the score corresponding to the Kth crack location. Specifically, the logic for updating and mapping the set of crack locations and their corresponding confidence scores to a digital twin to generate a crack distribution map is as follows: In the spatial coordinate system Σ(x, y, z) of the digital twin, the crack location points are marked as crack nodes; Confidence score As attribute values ​​of crack nodes, to form attribute mapping relationships. ”; In the digital twin, with the crack node as the center, a spatial connection threshold Q is set. If the Euclidean distance between two crack nodes is less than or equal to Q, an undirected connection edge is established between the two crack nodes. An edge set E is constructed based on multiple sets of undirected connection edges. Construct a crack distribution map G using the node set L and the edge set E, where G = (L, E, P), and output the crack distribution map G. The crack location module S4 can intuitively identify the location distribution, quantity density, detection confidence level, and spatial connectivity between cracks in three-dimensional space through the crack distribution map, making it easy to clearly understand the overall spatial distribution and local concentrated areas of cracks. Example

[0024] Please see Figure 2 As shown, based on a unified inventive concept, this embodiment discloses a method for detecting cracks on the inner surface of a boiler pressure vessel, the method comprising: S21, Boiler structure modeling: Generate a boiler geometric model based on the acquired boiler structure data; S22, Digital Twin Construction: Construct a digital twin based on the boiler geometric model, generate a set of monitoring points in the digital twin according to the preset monitoring point generation rules, and retrieve the historical operation and maintenance data of each monitoring point and map it into the digital twin; S23, Anomaly Detection Map Construction: Collect preprocessed ultrasonic signal data and acoustic emission signal data from each monitoring point to obtain a set of feature vectors, associate the set of feature vectors with the digital twin, and form a preliminary anomaly detection map; S24, Crack Identification and Localization: Input the feature vector set into the crack identification model, output the crack location set and the corresponding confidence score set, update the crack location set and its corresponding confidence score to the digital twin, and generate a crack distribution map; The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A boiler pressure vessel internal surface crack detection system, characterized by, The system comprises: a structure modeling module, which generates a boiler geometric model according to acquired boiler structure data; a twin generation module, which constructs a digital twin according to the boiler geometric model, generates a set of monitoring points in the digital twin according to a preset monitoring point generation rule, and calls historical operation and maintenance data of each monitoring point and maps the data into the digital twin; an anomaly detection graph construction module, which collects preprocessed ultrasonic signal data and acoustic emission signal data of each monitoring point to obtain a set of feature vectors, associates the set of feature vectors with the digital twin, and forms a preliminary anomaly detection graph; a crack positioning module, which inputs the set of feature vectors into a crack recognition model, outputs a set of crack locations and a corresponding set of confidence scores, and updates and maps the set of crack locations and the corresponding set of confidence scores to the digital twin to generate a crack distribution graph.

2. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 1, wherein The boiler structure data comprises a boiler shell length, a boiler shell outer diameter, a shell outer wall curvature, a shell wall thickness, a weld position, a weld material attribute, and a welding process parameter.

3. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 2, characterized by, The step of constructing a digital twin according to a boiler geometric model comprises: The twin generation module comprises a three-dimensional space mapping unit and an attribute definition unit; The space mapping unit reads coordinate point data in the boiler geometric model, establishes a space coordinate system Σ(x, y, z), and reconstructs a boiler shape in a digital space point by point according to a shell outer wall curvature field K(x, y, z) and a thickness distribution H(x, y, z) defined in the geometric model; A continuous shell surface is generated through a surface reconstruction algorithm, and weld regions and wall thickness abnormal regions are partitioned to form a multi-resolution space grid; The weld position, the weld material attribute, and the welding process parameter are bound to corresponding grid nodes through the attribute definition unit to form a digital twin with geometric and material multi-dimensional attributes, and a digital twin with spatial consistency and attribute mapping integrity corresponding to a physical boiler is output.

4. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 3, characterized by The logic of generating a set of monitoring points in the digital twin according to a preset monitoring point generation rule and calling historical operation and maintenance data of each monitoring point and mapping the data into the digital twin comprises: According to the monitoring point generation rule, the digital twin surface is traversed, and a spatial coordinate point meeting the rule is selected to generate a monitoring point set , m is an identifier, and each monitoring point Three-dimensional coordinates are (x, y, z) , , ), ∈m; For each monitoring point, a data interface module is called to retrieve a historical operation and maintenance data set of a corresponding region from an operation and maintenance database, and the historical operation and maintenance data set comprises temperature, pressure, and running time length; The historical operation and maintenance data set is bound to the monitoring point in chronological order to generate operation and maintenance mapping data in the digital twin.

5. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 4, wherein The preset monitoring point generation rule comprises: Based on the principle of spatial uniformity, with fixed intervals Uniformly distribute points along the surface of the shell; encrypt the distribution of points within A millimeters of the weld perimeter; encrypt the distribution of points for areas where the wall thickness gradient exceeds a preset gradient threshold.

6. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 5, wherein The anomaly detection graph acquisition logic comprises: Based on each monitoring point, quantitative features are extracted from preprocessed ultrasonic signal data and acoustic emission signal data, and the quantitative features comprise ultrasonic signal features and acoustic emission signal features; the ultrasonic signal features comprise a main frequency, signal energy, and a peak amplitude; the acoustic emission signal features comprise cumulative energy, signal count, and signal duration; constructing a feature vector set based on the ultrasonic signal features and the acoustic emission signal features corresponding feature vectors , constructing a feature vector set ; mapping the feature vector set of all monitoring points to the corresponding monitoring points of the digital twin in sequence to update the monitoring point attributes of the digital twin and form a binding relationship ​ In the spatial coordinate system of the digital twin Σ(x, y, z), the feature vector As a node attribute, a monitoring point feature field is constructed; based on the updated digital twin, an anomaly detection graph is generated according to the following logic: A feature vector set V is taken as a node attribute, a node connection relationship is established based on a spatial proximity relationship of the monitoring points, an anomaly score of each node is calculated, nodes with anomaly scores higher than a preset score threshold are marked as preliminary anomaly points, and finally an anomaly detection graph is output.

7. A system for detecting cracks in the internal surface of a boiler pressure vessel according to claim 6, characterized in that, The logic of outputting a set of crack locations and a corresponding set of confidence scores comprises: The monitoring point feature vector set The input is input into a preset crack identification model to obtain an identification result, the identification result including a determination result and a confidence score; The output logic of the identification result is: When the determination result output is 0, the boiler has no crack, and when the determination result output is 1, the boiler has a crack; The numerical range of the confidence score is greater than 0 and less than 1, and the greater the value of the confidence score, the greater the credibility of the crack; Traverse the determination results of all monitoring points, extract the monitoring points with determination result output of 1, and construct a crack position point set , For the Kth crack position point, a confidence set is constructed based on the confidence of each position point , The Kth crack position point corresponds to the confidence score.

8. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 7, wherein The generation logic of the crack identification model is: Obtain historical monitoring point identification data, divide the historical monitoring point identification data into an identification training set and an identification test set, the historical monitoring point identification data includes monitoring point feature vectors and corresponding determination results and confidence scores; Configure an initial classifier, use the monitoring point feature vectors of the sub-components in the identification training set as the input data of the initial classifier, use the corresponding determination results and confidence scores in the identification training set as the output data of the initial classifier, train the initial classifier to obtain an initial classification network; The anomaly detection graph construction module S13 verifies the initial classification network through the identification test set, outputs the initial classification network greater than or equal to the preset test accuracy, and takes it as a pre-constructed crack identification model.

9. The system for detecting cracks in the inner surface of a boiler pressure vessel according to claim 8, wherein The logic for updating and mapping the crack position set and its corresponding corresponding confidence score to the digital twin to generate a crack distribution graph is: In the spatial coordinate system Σ(x, y, z) of the digital twin, mark the crack position points as crack nodes; The confidence score is calculated as follows: The attribute value of the crack node is taken to form an attribute mapping relationship "; In the digital twin, set a spatial connection threshold Q with the crack nodes as the center, if the Euclidean distance between two crack nodes is less than or equal to Q, then establish an undirected connection edge between the two crack nodes, and construct an edge set E based on multiple groups of undirected connection edges; Construct a crack distribution graph G with the node set L and the edge set E, where G=(L, E, P), and output the crack distribution graph G.

10. A method for detecting cracks on the inner surface of a boiler pressure vessel, applied to the crack detection system for the inner surface of a boiler pressure vessel according to any one of claims 1 to 9, characterized in that, The method comprises: Boiler structure modeling: generating a boiler geometric model according to the obtained boiler structure data; Digital twin construction: constructing a digital twin according to the boiler geometric model, generating a monitoring point set in the digital twin according to a preset monitoring point generation rule, and calling historical operation and maintenance data of each monitoring point and mapping it to the digital twin; Anomaly detection graph construction: collecting preprocessed ultrasonic signal data and acoustic emission signal data of each monitoring point to obtain a feature vector set, associating the feature vector set with the digital twin to form a preliminary anomaly detection graph; Crack identification and positioning: input the feature vector set into the crack identification model to output a crack position set and a corresponding confidence score set, update and map the crack position set and its corresponding corresponding confidence score to the digital twin to generate a crack distribution graph.

Images