Pipeline integrity intelligent decision visualization system

Abstract

The application provides a pipeline integrity intelligent decision visual system, comprising a pipeline three-dimensional visualization module and a pipeline intelligent decision module; the pipeline three-dimensional visualization module comprises a three-dimensional model visualization display sub-module, a support environment sub-module, an analysis decision module data interface sub-module, a pipeline information directory tree sub-module and a pipeline engineering data management sub-module; the pipeline intelligent decision module comprises a basic data management sub-module, a detection management sub-module, a defect management sub-module and an integrity intelligent decision sub-module; the integrity intelligent decision sub-module comprises intelligent analysis and decision for pipeline abnormal signals of a single detection method or multiple detection methods which cannot quantize defect size and type. The application realizes intelligent analysis and decision of pipeline integrity through intelligent analysis of detection signals of different detection methods and abnormal signal maintenance decision result data determined by expert experience or according to excavation, and interaction of a buried pipeline three-dimensional visualization platform.

Description

Technical Field

[0001] This invention relates to the field of intelligent evaluation and visualization technology for pipeline integrity, and in particular to an intelligent decision-making visualization system for pipeline integrity. Background Technology

[0002] Integrity assessment is one of the core tasks of pipeline integrity management. Integrity assessment determines the safety status of pipelines and provides guidance for pipeline maintenance. For pipelines in refineries, stations, and oil and gas field gathering and transportation networks where internal inspections or excavation inspections are not feasible, the main problems currently facing pipelines include: 1) Refineries and stations contain a large number of raw material lines, interconnection lines, and power lines. Most power pipelines are underground, with diverse types and complex distribution, making unified management difficult; 2) Oil and gas field gathering and transportation networks suffer from a large number of pipelines, with significant data gaps in older pipelines. Inconsistencies often arise between drawings and actual site conditions, as well as the knowledge of management personnel, causing considerable difficulties for construction, maintenance, and the expansion and upgrading of equipment; 3) Complex underground conditions and widespread pipeline corrosion and thinning are common. Some older pipelines operate beyond their capacity and lifespan, leading to pipe bursts, leaks, and cross-connections due to corrosion and pressure. Leaks occur frequently, posing significant safety hazards; 4) Internal inspections are not possible, making it impossible to promptly grasp the pipeline's safety status. 5) Existing methods of evaluating test data based on human experience cannot guarantee the accuracy of the evaluation results. 6) Using a combination of multiple testing methods, such as magnetic stress and guided waves, for pipeline safety evaluation can improve the accuracy of the test. However, current technologies for combining multiple testing methods mainly rely on human experience to judge the type of maintenance. Human experience judgment is generally subject to strong subjective factors, requires a high level of professional expertise from the evaluators, has high labor costs, and is economical and inefficient. Summary of the Invention

[0003] This invention provides an intelligent decision-making visualization system for pipeline integrity, which addresses the shortcomings of existing pipeline analysis technologies, such as high cost, low efficiency, and low accuracy.

[0004] This invention provides a pipeline integrity intelligent decision-making visualization system, comprising: a pipeline 3D visualization module and a pipeline intelligent decision-making module;

[0005] The pipeline 3D visualization module includes a 3D model visualization display submodule, a support environment submodule, an analysis and decision-making module data interface submodule, a pipeline information directory tree submodule, and a pipeline engineering data management submodule.

[0006] The pipeline intelligent decision-making module includes a basic data management submodule, an inspection management submodule, a defect management submodule, and an integrity intelligent decision-making submodule;

[0007] The data interface submodule of the analysis and decision module is used to provide data interfaces for the basic data management submodule, the detection management submodule, the defect management submodule, and the integrity intelligent decision module.

[0008] The integrity intelligent decision-making submodule includes intelligent analysis and decision-making for pipeline anomaly signals obtained by a single detection method or multiple detection methods that cannot quantify the size and type of defects.

[0009] The 3D model visualization display submodule will display the intelligent evaluation and maintenance decision results in a dynamic 3D interactive visualization.

[0010] According to the present invention, a pipeline integrity intelligent decision-making visualization system is provided. The pipeline 3D visualization module is based on the principle of human-computer interaction. It determines the geographical information or relative position of the modeling object by conducting on-site surveys of refinery equipment and devices, above-ground landmarks and above-ground and underground pipelines connected to the devices. The system then establishes a model of the object in a 3D model and manages the relevant engineering information of the pipeline, devices and equipment through the 3D model or directory tree. The pipeline information serves as the data basis for evaluating the integrity of buried pipelines.

[0011] According to the present invention, a pipeline integrity intelligent decision-making visualization system is provided, wherein the basic data management submodule is used to obtain basic information of the pipeline where the pipeline is located from the pipeline three-dimensional visualization module. The basic information includes pipe diameter, wall thickness, pipe material, conveying medium, design data and connection equipment information, and to add, query, delete and modify the basic information.

[0012] According to the present invention, a pipeline integrity intelligent decision-making visualization system includes a detection management submodule comprising four functions: detection signal sample library management, pipeline under inspection management, detection project management, and detection data management, which enables logical management of detection operations, detection signals, and detection results.

[0013] Among them, the detection signal sample library management is used to manage the historical abnormal detection signals of the pipeline being inspected; the pipeline being inspected management is used to manage the pipeline being inspected and its associated inspection projects; the inspection project management is used to establish an inspection project after an inspection operation is carried out on a section of pipeline, and to manage the inspection project; the detection data management is used to manage the detection data for a specific inspection project, and its functions include supporting the management of detection signal diagrams, detection signal data, and detection result data for different detection methods.

[0014] According to the present invention, a pipeline integrity intelligent decision-making visualization system is provided, wherein the defect management submodule is used to manage the defect data obtained after a pipeline inspection through expert signal analysis, excavation verification, or a built-in intelligent decision-making model or pipeline decision-making model; the defect data can interact with the three-dimensional model to realize the addition, query, deletion, and modification of the detected pipeline defect information; at the same time, the defect information is classified and managed according to different defect types, wherein the corrosion rate of corrosion defects is directly calculated based on the defect depth and commissioning time through a built-in algorithm in the background, or is recorded through experiments or actual measurements.

[0015] According to the present invention, a pipeline integrity intelligent decision-making visualization system is provided, wherein the integrity intelligent decision-making submodule is used to: obtain the detection signal map or detection signal intensity value of the pipe segment from the detection management submodule for pipeline abnormal signals obtained by a single detection method that cannot quantify the size and type of defects, and make intelligent maintenance decisions using an intelligent decision-making model pre-trained based on deep learning.

[0016] According to the intelligent decision-making visualization system for pipeline integrity provided by the present invention, the training process of the intelligent decision-making model includes:

[0017] Obtain sample magnetic stress detection signal images or detection signal intensity values ​​from expert analysis data in the detection management submodule; wherein, the sample magnetic stress detection signal images or intensity values ​​need to cover various defect pixel locations and various repair levels;

[0018] The pipeline anomaly area mileage location and maintenance level are obtained from the sample magnetic stress detection signal image or intensity value from the expert analysis data in the detection management submodule; wherein, the sample data is divided into training set and test set;

[0019] The model takes a sample magnetic stress detection signal image or intensity value with a pre-determined defect pixel location and maintenance level as input, and takes the pipeline abnormal area mileage location and maintenance level in the sample magnetic stress detection signal image or intensity value as output. The neural network model is trained and tested based on the intelligent learning algorithm to obtain the intelligent decision model.

[0020] According to the present invention, a pipeline integrity intelligent decision-making visualization system is provided, wherein the integrity intelligent decision-making submodule is further configured to: for pipeline abnormal signals obtained by various detection methods that cannot quantify defect size and type, first obtain the detection signal map and detection signal intensity value of the pipe segment under each detection method from the detection management submodule, extract the feature vectors of the detection signal map and detection signal intensity value under each detection method, and input each feature vector into the weight prediction model to obtain the weight coefficients corresponding to each feature vector output by the weight prediction model;

[0021] After updating the weight coefficients of the pipeline decision model with the weight coefficients corresponding to each feature vector, the feature vectors are input into the pipeline decision model to obtain the pipeline analysis results output by the pipeline decision model.

[0022] Each feature vector is input into the maintenance level prediction model, which then fuses the feature vectors to obtain a fused vector. The maintenance level prediction result is then obtained based on the fused vector.

[0023] According to the intelligent decision-making visualization system for pipeline integrity provided by the present invention, when excavation data is not available in the early stage, the intelligent decision-making model is trained and tested by expert analysis data in the defect management submodule;

[0024] Once the accumulated excavation data meets the preset conditions, excavation defect data is obtained from the detection management submodule, and the intelligent decision-making model is optimized through data migration.

[0025] According to the intelligent decision-making visualization system for pipeline integrity provided by the present invention, the weight prediction model is trained based on the sample feature vector and the sample weight coefficients corresponding to the sample feature vector;

[0026] The sample weighting coefficients are determined based on the following steps:

[0027] Determine the initial sample weight coefficients corresponding to the feature vectors of each sample, and iterate through the initial sample weight coefficients with the target step size to obtain the iterative weight coefficients; wherein, the sum of the iterative weight coefficients obtained in each iteration is 1;

[0028] Based on the weight coefficients of each iteration, the feature vectors of each sample are weighted and fused to obtain each fused feature vector;

[0029] Based on each fused feature vector, the pipeline defect prediction result corresponding to each fused feature vector is determined, and the iterative weight coefficient corresponding to the pipeline defect prediction result with the same pipeline defect label is used as the sample weight coefficient corresponding to the sample feature vector.

[0030] As can be seen from the above technical solution, the intelligent decision-making and visualization system for pipeline integrity provided by the present invention realizes intelligent analysis and decision-making on pipeline integrity, especially pipelines in refineries, stations, and gathering and transportation networks, through intelligent analysis of detection signals from different detection methods and maintenance decision-making results data of abnormal signals determined by expert experience or based on excavation, and interaction with a three-dimensional visualization platform for buried pipelines. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the structure of the intelligent decision-making visualization system for pipeline integrity provided by the present invention;

[0033] Figure 2 This is a flowchart illustrating the weight prediction model training method provided by the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0035] The intelligent decision-making and visualization system for pipeline integrity provided in this invention realizes intelligent analysis and decision-making for pipeline integrity, especially pipelines within refineries, stations, and gathering and transportation networks, through intelligent analysis of detection signals from different detection methods and maintenance decision-making results data of abnormal signals determined by expert experience or based on excavation, and interaction with a three-dimensional visualization platform for buried pipelines.

[0036] Figure 1 This is a schematic diagram of the intelligent decision-making and visualization system for pipeline integrity provided by the present invention, as shown below. Figure 1 As shown, the system includes: a pipeline 3D visualization module and a pipeline intelligent decision-making module;

[0037] The pipeline 3D visualization module includes a 3D model visualization display submodule, a support environment submodule, an analysis and decision-making module data interface submodule, a pipeline information directory tree submodule, and a pipeline engineering data management submodule.

[0038] The pipeline intelligent decision-making module includes a basic data management submodule, an inspection management submodule, a defect management submodule, and an integrity intelligent decision-making submodule;

[0039] Among them, the data interface submodule of the analysis and decision module is used to provide data interfaces for the basic data management submodule, the detection management submodule, the defect management submodule, and the integrity intelligent decision module;

[0040] Among them, the integrity intelligent decision-making submodule includes intelligent analysis and decision-making for pipeline abnormal signals obtained by a single detection method or multiple detection methods that cannot quantify the size and type of defects. For pipeline abnormal signals obtained by a single detection method that cannot quantify the size and type of defects, the detection signal map and detection signal intensity value of the pipe section are first obtained from the detection management submodule, and intelligent maintenance decision-making is carried out by using an intelligent decision-making model based on pre-trained signal map or signal data.

[0041] For pipeline anomaly signals from multiple detection methods where defect size and type cannot be quantified, the detection signal map and detection signal intensity value of the pipe segment under each detection method are first obtained from the detection management submodule. A pre-trained weight prediction model based on deep learning is used to predict the weight coefficients of each detection method to obtain the corresponding weight coefficients. The pre-trained pipeline decision model is then updated with the corresponding weight coefficients of each detection method to obtain the pipeline evaluation result. Finally, a pre-trained maintenance level prediction model is used to fuse the feature vectors corresponding to the detection signal map and detection signal intensity value under each detection method to obtain the maintenance decision result.

[0042] The 3D model visualization submodule displays the intelligent evaluation and maintenance decision results in a dynamic, interactive 3D visualization.

[0043] As can be seen from the above technical solution, the intelligent decision-making visualization system for pipeline integrity provided in this embodiment achieves intelligent analysis and decision-making on pipeline integrity, especially pipelines within refineries, stations, and gathering and transportation networks, through intelligent analysis of detection signals obtained by different detection methods and maintenance decision-making results data of abnormal signals determined by expert experience or excavation, and interaction with the three-dimensional visualization platform for buried pipelines.

[0044] In this embodiment, the intelligent pipeline integrity evaluation and visualization system comprises two parts: 3D visualization functionality and intelligent pipeline integrity analysis and decision-making. The 3D visualization functionality mainly includes the visualization and management of simplified pipeline connection equipment, devices, ground markers, and all pipelines connected to the equipment using 3D models. The intelligent pipeline analysis and decision-making functionality mainly includes pipeline basic data management, inspection / maintenance decision data management, defect management, and intelligent pipeline integrity analysis and decision-making through dynamic data interaction with the 3D model.

[0045] The pipeline 3D visualization module is described below:

[0046] The pipeline 3D visualization module includes 3D model visualization, a secure and reliable support environment, and application interfaces for the pipeline integrity intelligent decision-making module (pipeline basic data management, inspection management, defect management, and intelligent analysis and decision-making). Following the principle of human-computer interaction, it determines the geographical information or relative location of the modeling objects by conducting on-site surveys of pipeline-connected equipment and devices, ground landmarks (such as roads and manhole covers), and all above-ground and underground pipelines connected to the devices. Models of these objects are then created in the 3D model. Simultaneously, the established pipeline, device, and equipment-related engineering information can be added, searched, deleted, and modified through the 3D model or directory tree. The pipeline information serves as the data foundation for the integrity evaluation of buried pipelines.

[0047] The intelligent decision-making module for pipelines is described below:

[0048] The pipeline intelligent decision-making module mainly includes: a basic data management submodule, an inspection management submodule, a defect management submodule, and an integrity intelligent decision-making submodule that enables dynamic data interaction with the 3D model.

[0049] The basic data management submodule includes basic information about the pipeline obtained from the 3D visualization platform, such as pipe diameter, wall thickness, pipe material, transport medium, design data, and connecting equipment, as well as functions for adding, querying, deleting, and modifying this type of information.

[0050] The inspection management submodule mainly includes four functions: inspection signal sample library management, inspected pipeline management, inspection project management, and inspection data management. It primarily focuses on the logical management of inspection data such as inspection operations, inspection signals, and inspection results. ① Inspection signal sample library management: This is used to manage historical abnormal inspection signals of the inspected pipelines. This includes adding, querying, deleting, and modifying information such as pipeline information, inspection equipment name, inspection method, inspection equipment manufacturer, evaluation expert level, abnormal signals, maintenance decision type, and maintenance decision results. For historical abnormal inspection signals from various non-contact pipeline inspection methods to be added to the library, the information includes the inspected pipeline information (pipe diameter, pipe material, interference scenario and factors), inspection equipment name, inspection method, inspection equipment manufacturer, evaluation expert level, abnormal signals, maintenance decision type (excavation verification, expert analysis), and maintenance decision results (including but not limited to abnormal location, defect type, maintenance level, etc.). All of this data is stored in the intelligent decision model sample library. ① As a sample set for the model, it enables training, testing, and continuous optimization of the model; ② Management of inspected pipelines: This mainly involves the management of the inspected pipelines and their associated inspection projects. The management of inspected pipelines primarily focuses on the inspection range of the inspected pipelines. Based on the built-in pipeline numbering and naming rules, it automatically generates pipeline numbers for newly added inspected pipelines according to the pipeline where the inspected pipeline is located. This allows for the maintenance, modification, and viewing of information such as inspection length, inspection location (or inspection start and end markers), and remarks for newly added inspected pipelines. Simultaneously, after creating a new inspected pipeline, it highlights and positions the inspected pipeline on the 3D model using its pipeline number. Furthermore, under the added inspected pipelines, the inspection projects and related information for the inspected pipelines can be added, searched, deleted, and modified. After a pipeline inspection, on the pipeline 3D model, the inspection range of the inspected pipeline is determined manually by selecting points on the pipeline to be inspected using information such as inspection markers. The system automatically generates an inspection pipeline number based on the pipeline where the inspected pipeline is located, and simultaneously, based on the inspection pipeline number, it achieves highlighting and positioning of the inspected pipeline in the 3D model. The module manages and maintains information about the inspected pipelines, including inputting and modifying information such as inspection length, inspection location points, and design coefficients, as well as deleting the inspected pipelines. When an inspected pipeline is deleted, its associated inspection projects, defect locations, and defect data are also deleted. ③ Inspection Project Management: After an inspection is conducted on a pipeline section, an inspection project is created, including creating a new inspection project under a newly added inspected pipeline. The module manages and maintains the inspection project and its related information (inspection project name, inspection method, inspection equipment, inspection time, inspection unit, etc.) through adding, querying, deleting, and modifying data. When an inspection project is deleted, its associated defect locations and defect data are also deleted. This module also allows viewing a list of all inspection projects under a corresponding inspected pipeline by its inspection pipeline number. ④ Inspection Data Management Function: This function manages the addition, querying, deletion, and modification of inspection data for a specific inspection project.The functionality includes management of detection signal diagrams, detection signal data, and detection result data for different detection methods. Detection signal diagram management primarily involves adding, querying, deleting, and modifying detection signal diagrams for pipelines using different detection methods. Detection signal data management imports the signal strength values ​​along the detection mileage into the system in form, based on the detector sampling results. Detection result data is also managed by importing data into the system in form. Furthermore, it includes downloading detection signal and result templates, data import, and the ability to add, query, delete, and modify data.

[0051] The defect management submodule primarily manages defect data (including defect number, defect type, defect size, defect location, corrosion rate, defect location, and decision results) obtained after a pipeline inspection through expert signal analysis, excavation verification, or a built-in intelligent decision-making model. This defect data can interact with the 3D model, enabling the addition, querying, deletion, and modification of pipeline defect information. Furthermore, it can classify and manage defect information according to different defect types, such as corrosion defects, manufacturing defects, deformation defects, and weld anomalies. The corrosion rate for corrosion defects can be directly calculated using a built-in algorithm based on defect depth and commissioning time, or it can be directly entered from experimental or measured values. Adding new defects can be done in two ways: ① For a single defect obtained through excavation verification and expert analysis, given the known defect location, the defect point location can be directly added on the 3D pipeline model by measuring markers (pipeline start point, road, manhole cover, etc.) on the inspected pipeline. Then, the defect data is entered, and the excavated and verified pipeline information, inspection information, and maintenance decision results are stored in the historical inspection and maintenance decision sample library for optimization of the intelligent decision model; ② Through the intelligent decision model, intelligent analysis and decision-making are performed on the entered inspection signal images or data, outputting the defect location, defect type, and maintenance level. The intelligent analysis and decision results (including defect number, defect type, defect size, defect location, defect corrosion rate, defect location, decision result, etc.) are editable. The intelligent decision can also be further verified by expert analysis or excavation to check for errors. For the verified defect information, especially after excavation verification, the intelligent decision results are further corrected. At the same time, the defect-related data is stored in the historical inspection and maintenance decision sample library in the same way as in ① to optimize the intelligent decision model. The decision results can automatically mark abnormal sections on the corresponding pipe segments on the 3D platform and automatically synchronize the maintenance level to the corresponding abnormal sections of the pipeline. For abnormal sections that need immediate repair, they are displayed with a highlighted or red second-level warning sign and are automatically saved for display when re-entering the intelligent analysis and decision interface; ③ For batch detection signals, batch intelligent addition and positioning can be performed in the same way as step ② through intelligent analysis and decision. Defect revision can be achieved by batch exporting the intelligent analysis and decision results, modifying the defect data for excavation verification after export, and then importing the modified data back into the system. The modified data is also synchronized to the sample library in the same way as in ①. On the defect management page, defects can be queried through pipeline information, inspection projects, etc., and defects can also be deleted. When a defect is deleted, both the defect data and the defect on the 3D model are deleted simultaneously.

[0052] The integrity intelligent decision-making submodule mainly includes intelligent analysis and decision-making for pipeline abnormality signals obtained by different detection methods that cannot quantify the size and type of defects, and automatic evaluation and maintenance decisions for quantified defects.

[0053] For pipeline anomaly signals where the size and type of defects cannot be quantified, for pipeline signals obtained using a single detection method, the detection signal diagram and signal strength values ​​for that pipe segment are first retrieved from the "Detection Data Management" system. Then, using a pre-trained intelligent maintenance decision-making model for the integrity of the signal diagram or signal data, a standard length of signal diagram or signal data value is determined based on the pipeline's signal diagram and signal strength data. If the signal diagram or signal value length Then scale the signal graph and values ​​to the standard size; if the signal length Starting from the left endpoint of the signal array, establish a length of... The sliding window, each The signal is trimmed length once until the right side of the sliding window reaches the boundary of the signal's maximum value, resulting in... (Round down) signal diagrams or signal arrays. These signal diagrams or signal arrays are input into the intelligent maintenance decision model for prediction. Based on the network prediction results, the mileage interval where the abnormal signal is located and the maintenance level of the abnormal signal (immediate repair, planned repair, and monitoring) are output respectively. For detection signals obtained by multiple detection methods, such as ultrasonic guided waves and magnetic stress signals, based on the pre-established multi-detection method fusion intelligent analysis and decision, the detection data extracted in the unknown environment is first preprocessed to extract feature vectors. The feature vectors are input into the weight prediction model, which outputs the weight prediction results, i.e., weight coefficients. The weights predicted by the weight prediction model are used to optimize the weight settings of the pipeline decision model (such as a multi-flow hidden Markov model), realizing the decision fusion model of multiple detection methods to output the probability of each defect type. When predicting the maintenance level, the maintenance level prediction model uses the feature matrix processing method to preprocess the feature signals to be identified to obtain the identification feature vectors. The identification feature vectors are input into the maintenance level prediction model, which predicts the maintenance level probability (immediate repair, planned repair, and monitoring recommendation).

[0054] In this embodiment, it should be noted that after obtaining the defect identification results and maintenance level prediction results, management is carried out in accordance with ② and ③ in the defect management submodule.

[0055] For automated evaluation and maintenance decisions regarding quantitative defects, the system assesses both batch and single defects by obtaining information such as size and type through inspection. Different defect types require different evaluation methods. Users can select an evaluation method for a specific type of defect, and the system can then call upon its built-in evaluation algorithm to perform batch or single evaluations and maintenance decisions on defects of the same type or a single defect using the same evaluation method. The system saves the evaluation results upon successful evaluation and displays them on subsequent visits to the evaluation page. Automatic evaluation of single defects is for defects with known type and size. In the single evaluation module, users select the defect to be evaluated, choose the evaluation method, and the system automatically evaluates the defect. Evaluation results are saved and displayed upon re-entry, with the option to re-evaluate using a different method. Automatic evaluation of batch defects occurs on the batch evaluation page. Users select the device, pipeline number, inspection project, and defect type, then select the evaluation method, input the required evaluation information, and the system's built-in evaluation algorithm performs batch evaluations of a specific type of defect. Users can modify the evaluation method and re-evaluate already evaluated defects.

[0056] When excavation data is unavailable in the early stages, the intelligent decision-making model can be trained and tested using expert-analyzed data from the detection and management submodule.

[0057] In this embodiment, the training process of the intelligent decision-making model includes:

[0058] Obtain sample magnetic stress images from expert analysis data in the detection management submodule; wherein, the sample magnetic stress images need to cover various defect pixel locations and various repair levels;

[0059] The mileage location and maintenance level of the pipeline anomaly area are obtained from the expert analysis data in the detection management submodule;

[0060] The model takes a sample magnetic stress image with pre-determined defect pixel location and maintenance level as input, and the pipeline abnormal area mileage location and maintenance level in the sample magnetic stress image as output. The neural network model is trained and tested based on the intelligent learning algorithm to obtain the intelligent decision model.

[0061] Therefore, this embodiment, based on magnetic induction signal image data, makes an intelligent maintenance decision that is highly adaptable and unaffected by subjective human factors. This can effectively improve the accuracy of the detection result analysis, avoid blind maintenance, greatly improve economy and reduce labor, and at the same time improve pipeline safety.

[0062] It should be noted that pipeline inspection signal data is generally in the form of signal strength corresponding to pipeline mileage. This data is usually displayed and analyzed in the form of signal graphs. For example, based on the intensity value of the obtained magnetic stress signal, a corresponding magnetic stress signal image can be drawn, and then an intelligent algorithm can be used to predict the severity of pipeline defects based on the signal image. However, since digital signals lose some precision when converted into images, the accuracy of the algorithm is limited. Furthermore, processing and calculating images requires a large amount of computing resources, is slow, and places very high demands on computer hardware. Therefore, this embodiment proposes a pipeline intelligent maintenance decision-making method that directly targets signal data. This method is not only applicable to the intelligent evaluation of pipeline integrity based on signal data from non-contact pipeline inspections such as magnetic stress detection, but can also be used for the direct intelligent evaluation of other types of inspection data. On the one hand, it has a wide range of applicability; on the other hand, it improves the accuracy and efficiency of pipeline integrity evaluation while reducing labor costs, improving economic benefits and pipeline safety, thus filling the technological gap in existing intelligent pipeline integrity evaluation.

[0063] Based on the above embodiments, in this embodiment, after obtaining sample magnetic stress signal data from the expert analysis data in the detection management submodule, manual and intelligent sample magnetic stress signal data processing is also included. The manual sample magnetic stress signal processing procedure is as follows:

[0064] The sample magnetic stress signal data is subjected to uniform scaling adjustment, specifically including:

[0065] First, the detection mileage X of the signal data for each inspected pipe section is calculated. max Average detection mileage Pixel length L src and average original pixel length Average width Round to the nearest integer;

[0066] To ensure that the detection mileage ratio of all signal data is uniform, the pixel length L of all signal data is first determined. src Adjustments were made to make the length of the adjusted sample signal data... Statistical signal data length L0 pixel average Round the data to the nearest integer, and then perform the next scaling adjustment on the signal data.

[0067] It is understandable that by uniformly adjusting the scale of the magnetic induction detection signal map, the training process of the aforementioned intelligent maintenance decision model based on the detection signal map becomes more accurate and effective. This can solve the problem that the abscissa scale of the magnetic stress signal data obtained each time is different due to the different mileage measured each time, which leads to the inability to directly use the obtained magnetic stress signal data for intelligent identification of pipeline maintenance level based on neural network.

[0068] In this embodiment, by first determining the relationship between the detection mileage of the magnetic stress signal data and the average detection mileage of the magnetic stress signal data of the inspected pipe section, and then adjusting the scale of the magnetic stress signal data according to the relationship between the detection mileage of the magnetic stress signal data and the average detection mileage of the magnetic stress signal data of the inspected pipe section, the problem of the abscissa scale of the magnetic stress signal data obtained each time being different due to the different mileage measured each time can be solved, which leads to the inability to directly use the obtained magnetic stress signal data for intelligent identification of pipeline maintenance level based on neural network.

[0069] After a certain amount of excavation data has been accumulated, excavation defect data can be obtained from the detection management submodule, and the intelligent decision-making model can be optimized through data migration.

[0070] The training and optimization process of the above model will be illustrated below with specific examples:

[0071] First, the image is normalized to vectors of the same size based on pixel values. These vectors are then processed by a grayscale VGG convolutional neural network to extract features, resulting in feature map vectors. After being extracted by a convolutional head, these feature maps are fed into an RPN (Region Proposal Network) to obtain the target region, which is then used for prediction.

[0072] Based on manually labeled image anomaly locations (which can be understood as defect pixel locations) and pipeline signal anomaly types (which can be understood as maintenance levels), the ratio of abnormal to normal signal quantities is statistically analyzed. The existing data is then randomly divided into training and testing sets based on this ratio. The training set is used to train the network model. The model input consists of normalized images, and the model output is the defect pixel location and maintenance level. The maintenance levels for magnetic stress anomaly signals are categorized into three cases: immediate repair, planned repair, and monitoring use. The loss function for the samples in the network is calculated.

[0073] Based on the above embodiments, in this embodiment, after obtaining the sample magnetic stress detection signal intensity value from the expert analysis data in the detection management submodule, the following steps are also included:

[0074] The sample magnetic stress detection signal intensity values ​​are processed to achieve length uniformity, specifically including:

[0075] The magnetic induction detection signal for each section of the pipeline is typically a two-dimensional array, in the form of:

[0076]

[0077] Where m is the number of samplings, and the magnetic induction detection signal obtained at each sampling point is:

[0078]

[0079] Where n is the dimension of the magnetic induction detection signal, h ij This represents the numerical value of the j-th signal feature in the magnetic induction detection signal;

[0080] The sampling frequency of the magnetic induction detection signal data is uniform, and the length of H represents the length of the detected pipe segment. The magnetic induction detection signal data is standardized so that the length of each detection signal vector is the same, so as to facilitate network training.

[0081] It is understandable that by uniformly adjusting the scale of the magnetic induction detection signal data, the training process of the aforementioned intelligent maintenance decision-making model based on the detection signal intensity value becomes more accurate and effective. This solves the problem that the horizontal axis scale of the magnetic induction detection signal data varies with each measurement mileage, making it impossible to directly use the obtained magnetic induction detection signal data for intelligent identification of pipeline maintenance levels based on neural networks. In this embodiment, by first uniformly adjusting the scale of the magnetic induction detection signal data according to the relationship between the array length of the magnetic induction detection signal data and the average array length of the inspected pipe section, the problem of the horizontal axis scale of the magnetic induction detection signal data varying with each measurement mileage, thus preventing the direct use of the obtained magnetic induction detection signal data for intelligent identification of pipeline maintenance levels based on neural networks, can be solved.

[0082] In this embodiment, for pipeline anomaly signals from various detection methods where defect size and type cannot be quantified, the process of defect analysis based on the pipeline decision model and maintenance decision-making based on the maintenance level prediction model is as follows:

[0083] Determine the detection signal data for buried pipelines corresponding to at least two detection methods. Here, "detection method" refers to the method used to detect defects in buried pipelines, such as magnetic stress detection and guided wave detection. The two detection methods are complementary. For example, magnetic stress detection is sensitive to the surrounding environment and is easily affected by factors such as trees, metal, and power lines, while guided wave detection is mainly affected by pipe bends and is less affected by environmental factors. In other words, guided wave detection can compensate for the shortcomings of magnetic stress detection, meaning that magnetic stress detection and guided wave detection are complementary.

[0084] After obtaining the detection signal data of buried pipelines corresponding to at least two detection methods, principal component analysis, discrete cosine transform, and other methods can be used to extract the feature vectors of each detection signal data. Since each detection signal data is determined based on different detection methods, the weight of each feature vector in pipeline analysis varies under different detection environments.

[0085] For example, if there are trees, metal, or power lines near the buried pipeline, the magnetic stress detection method is easily affected by interference factors such as trees, metal, and power lines, which will affect the accuracy of the pipeline analysis results. In contrast, the guided wave detection method is not easily affected by environmental factors. In this case, the weight coefficient of the corresponding eigenvector of the guided wave detection method can be increased, while the weight coefficient of the corresponding eigenvector of the magnetic stress detection method can be decreased. This will avoid the problem of reduced accuracy of pipeline analysis results due to interference factors affecting the magnetic stress detection method.

[0086] Similarly, if there are no interfering factors such as trees, metal, or power lines near the buried pipeline, but the buried pipeline has a large number of bends, the guided wave detection method is easily affected by the bends in the pipeline, which will affect the accuracy of the pipeline analysis results. On the other hand, the magnetic stress detection method is not easily affected by the bends in the pipeline. In this case, the weight coefficient of the corresponding eigenvector of the guided wave detection method can be reduced, while the weight coefficient of the corresponding eigenvector of the magnetic stress detection method can be increased. This will avoid the problem of reduced accuracy of pipeline analysis results caused by the bends in the pipeline in the guided wave detection method.

[0087] Therefore, after extracting the feature vectors of each detection signal data, each feature vector is input into the weight prediction model, so that the weight coefficients corresponding to each feature vector can be determined according to different detection environments, thereby ensuring the accuracy of subsequent pipeline analysis results.

[0088] The weight prediction model is trained based on the sample feature vector and the sample weight coefficients corresponding to the sample feature vector. The sample feature vector is determined based on the sample detection signal data obtained by at least two detection methods. Here, the at least two detection methods are complementary detection methods. For example, the detection methods can be complementary magnetic stress detection methods and guided wave detection methods.

[0089] Before inputting each feature vector into the weight prediction model, the weight prediction model can be pre-trained. This can be achieved by performing the following steps: First, collect a large number of sample feature vectors and determine their corresponding sample weight coefficients through manual annotation. Then, train the initial model based on the sample feature vectors and their corresponding sample weight coefficients to obtain the weight prediction model.

[0090] After determining the weight coefficients corresponding to each feature vector, the corresponding weight coefficients in the pipeline decision model are updated. This allows the pipeline decision model to fuse the pipeline analysis results obtained from each feature vector based on the weight coefficients, yielding the final pipeline analysis result. The pipeline decision model is trained based on sample feature vectors and the corresponding sample analysis results. The sample feature vectors are determined based on sample detection signal data obtained from at least two detection methods.

[0091] Understandably, the pipeline decision model can be trained based on a multi-flow Hidden Markov Model (HMM). The HMM uses a weighted product approach to simultaneously fuse the state observation probabilities corresponding to different detection methods and calculate the maximum likelihood probability lgP(O|C).

[0092] lgP(O|C)=λ1log(pr(O1|C1))+…+λ i log(pr(O i |C i ))+…+λ n log(pr(O n |C n ))

[0093] λ1+λ2+…+λ n =1

[0094] 0<λ1,λ2,…,λ n <1

[0095] Among them, C i These are the state parameters of the Hidden Markov Model.

[0096] Therefore, the embodiments of the present invention update the pipeline decision model with the weight coefficients corresponding to each feature vector, so that the weight coefficients corresponding to the detection methods that are more affected by interference factors are smaller, and the weight coefficients corresponding to the detection methods that are less affected by interference factors are larger, thereby avoiding interference factors from affecting the accuracy of pipeline analysis results.

[0097] After obtaining the feature vectors of each detection signal data, these feature vectors are input into the maintenance level prediction model to obtain maintenance decision results. These maintenance decision results can include options such as "immediate repair," "planned repair," and "recommendation for monitoring."

[0098] Before inputting the feature vectors into the maintenance level prediction model, the model can be pre-trained. This can be achieved by performing the following steps: First, collect a large number of sample feature vectors and manually label them to determine their corresponding maintenance level labels. Then, train the initial model based on the sample feature vectors and their corresponding maintenance level labels to obtain the maintenance level prediction model. The maintenance level prediction model can include a fusion layer and a prediction layer. The fusion layer is used to cross-combine and reduce the dimensionality of the feature vectors to create a new feature vector, i.e., a fusion vector. After obtaining the fusion vector, input it into the prediction layer to obtain the maintenance decision result.

[0099] When training the maintenance level prediction model, maintenance level labels can be divided into three types: "Immediate Repair," "Planned Repair," and "Recommended Monitoring." Taking the fusion of magnetic induction force detection and guided wave detection methods as an example, the feature vector data of the "Immediate Repair" label using the magnetic induction force detection method is m. i (i is the number of sample data), the feature vector data of the guided wave detection method labeled "immediate repair" is n. j (j is the number of sample data). For m i With n j Linear interpolation transformation and principal component analysis are reduced to the same dimension l, and the feature vectors are fused by cross-combining feature vectors to obtain the sample fusion vector. Sample fusion vector It can be represented as:

[0100] Therefore, the sample fusion vector The matrix has a dimension of 2l, and the number of samples in the sample fusion vector data is k = i * j. For the sample fusion vector... After reducing the dimensionality to the same dimension l as the original feature matrix, the number of feature vector matrices becomes i*j+i+j. Applying the same processing method to the "Planned Repair" and "Suggested Detection" label matrices, the feature vector M of the "Immediate Repair" dataset in the maintenance level prediction model can be expressed as:

[0101] Based on the above embodiments, the sample feature vector can be obtained through the following steps:

[0102] It includes two parts: interference data A1 affected by typical factors and non-interference data A2. Taking the fusion of magnetic stress detection and guided wave detection methods as an example, the magnetic stress detection method is easily affected by environmental factors such as trees, metals, and wires. The data collected and sorted out that the magnetic stress method is affected by such factors is data set A1. Data set A1 contains magnetic stress detection data affected by interference factors, and uses real anomaly points based on excavation, the maintenance levels of anomaly points (immediate repair, planned repair, monitoring) and defect type data (depression, metal loss, weld abnormality) as label data. Specific label information includes the signal length, the start point and end point of the defect position, and the corresponding maintenance level and defect type. The detection signal data and label data that are not interfered by the magnetic stress interference factors are A2. Data sets A1 and A2 together constitute the data basis of the multi-detection method fusion detection model with adaptive weights. The guided wave detection method is not easily interfered by the above interference factors, and the collected guided wave detection data is used as data set A3. The above collected data together constitute a sample data set that integrates magnetic stress detection and guided wave detection methods.

[0103] Data normalization preprocessing is performed on the established sample data set. The collected detection data is usually a two-dimensional array. Taking the fusion of magnetic stress and guided wave as an example, the magnetic stress detection signal data is The guided wave detection signal data is where m is the number of samplings, the detection signal can be expressed as:

[0104]

[0105]

[0106] The magnetic induction signals and guided wave signals at each sampling point are respectively:

[0107]

[0108]

[0109] where m is the dimension of the magnetic induction signal, and n is the dimension of the guided wave signal. In order to achieve effective data information fusion for the data of the two detection methods, it is necessary to perform data normalization processing. Calculate the average length of each detection signal in the data set, calculate the average length L of the detection signal through the linear interpolation transformation method. For signal data with data length l<L, the interpolation transformation method is used to interpolate the signal data, and for the signal data after interpolation transformation and data with data length l>L, the principal component analysis dimension reduction method is used to reduce the data length to L. So far, The data dimensions are the same. Since the calculated signal feature vectors differ significantly in value, the feature vectors after linear interpolation are normalized to reduce the impact of these numerical differences on modeling.

[0110]

[0111]

[0112] The normalized eigenvectors are

[0113]

[0114]

[0115] With the normalized eigenvectors The sample feature vectors are used to train the weight prediction model and the pipeline decision model.

[0116] Based on any of the above embodiments, in this embodiment, the weight prediction model is trained based on the sample feature vector and the sample weight coefficients corresponding to the sample feature vector;

[0117] The sample weight coefficients are determined based on the following steps:

[0118] Determine the initial sample weight coefficients corresponding to the feature vectors of each sample, and iterate through the initial sample weight coefficients with the target step size to obtain the iterative weight coefficients; wherein, the sum of the iterative weight coefficients obtained in each iteration is 1;

[0119] Based on the weight coefficients of each iteration, the feature vectors of each sample are weighted and fused to obtain each fused feature vector;

[0120] Based on each fusion feature vector, the pipeline defect prediction result corresponding to each fusion feature vector is determined, and the iterative weight coefficient corresponding to the pipeline defect prediction result with the same pipeline defect label is used as the sample weight coefficient corresponding to the sample feature vector.

[0121] Specifically, after obtaining the feature vectors of each sample, initial sample weight coefficients can be assigned to each sample feature vector, where the sum of all initial sample weight coefficients is 1. Then, the initial sample weight coefficients are iterated over with a target step size to obtain the iterative weight coefficients corresponding to each sample feature vector. The target step size can be set according to the actual situation, and this embodiment of the invention does not impose a specific limitation on it.

[0122] In the case of pipeline defect prediction based on fused feature vectors, a pre-trained pipeline defect prediction model can be used to make predictions and obtain pipeline defect prediction results.

[0123] For example, when there are two detection methods, i.e., n=2, the initial sample weight coefficients corresponding to the two detection methods are set as λ1=0 and λ2=1, respectively, and λ1+λ2=1. The weight values ​​are traversed with the target step size (l=0.1, 0<λ1, λ2<1), and the feature vectors in the dataset are modeled and analyzed respectively. Based on the output pipeline defect prediction result and the pipeline defect label, if it is the same as the label result, the corresponding weight value is recorded as the sample weight coefficient.

[0124] For example, when there are two or more detection methods, i.e., n > 2, the same method as above is used to traverse the weights. Taking n = 3 as an example, the initial sample weight coefficients are set as λ1 = 1, λ2 = 0, λ3 = 1 - λ1 - λ2. With the target step size (l = 0.1, 0 < λ1, λ2, λ3 < 1), λ1 is fixed, and λ2 and λ3 are traversed exhaustively. Table 1 is the weight traversal table. As shown in Table 1, when λ1 = 1, λ2 = 0, λ3 = 1 - λ1 - λ2 = 0. The value of λ1 is reduced by the target step size l = 0.1, so λ1 = 0.9. When λ1 = 0.9, λ2 and λ3 are traversed exhaustively. When λ2 = 0, then λ3 = 0.1. When λ3 = 0, then λ2 = 0.1, and so on, traversing all values ​​exhaustively.

[0125] Table 1

[0126]

[0127] The values ​​of λ1, λ2, and λ3 are compared with the pipeline defect label results. If they match the label results, the weight is recorded as the sample weight coefficient. When multiple detection methods are integrated, specific environmental impact factors can be analyzed. For example, when detection method 3 is complementary to detection methods 1 and 2, λ1 and λ2 can be bound as optimal weights and fitted to the extracted feature vector for prediction, then λ3 = 1 - λ1 - λ2. When detection methods 1, 2, and 3 are complementary, an adaptive weight model can be established to predict λ1 and λ2, then λ3 = 1 - λ1 - λ2. λ1, λ2, and λ3 always satisfy 0 < λ1, λ2, λ3 < 1 and λ1 + λ2 + λ3 = 1.

[0128] At this point, the sample weight coefficients corresponding to different sample feature vectors have been calculated using the pre-trained model. Based on the feature vectors O1, O2, ..., O in the dataset... n The sample weight coefficients λ1,λ2,…,λ that are matched with it n Fitting can be achieved through methods such as multilayer linear perceptrons, decision trees, and random forests.

[0129] λ i =f(O i )

[0130] Taking the fusion of magnetic stress detection and guided wave detection methods as an example, feature vectors O are extracted from the normalized dataset using the discrete cosine transform method. c O d Then, the likelihood probability lgP(O|C) is calculated by simultaneously fusing the state observation probabilities of the two detection methods in the form of a weighted product decision fusion using a multi-stream hidden Markov model, where C c C d Here, λ1 and λ2 are the state parameters of the Hidden Markov Model, calculated during training, and the weight coefficients for the two methods are: lgP(O|C)=λ1log(pr ... c |C c ))+λ2log(pr(O d |C c )), λ1+λ2=1, 0<λ1, λ2<1.

[0131] like Figure 2 As shown, when there are n detection methods, the initial sample weight coefficients corresponding to each detection method are set as λ1=1, λ2=0, ..., λ n =0, and λ1+λ2+...+λ n =1, traverse the initial sample weight coefficients with the target step size, model and analyze the feature vectors in the dataset respectively, such as using a multi-detection method to analyze the prediction model, and obtain the pipeline defect prediction result. Based on the output pipeline defect prediction result and the pipeline defect label, if it is the same as the pipeline defect label, record the corresponding weight as the sample weight coefficient, and train the weight prediction model based on the sample feature vector and the sample weight coefficient.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A pipeline integrity intelligent decision-making visualization system, characterized in that, include: Pipeline 3D visualization module and pipeline intelligent decision-making module; The pipeline 3D visualization module includes a 3D model visualization display submodule, a support environment submodule, an analysis and decision-making module data interface submodule, a pipeline information directory tree submodule, and a pipeline engineering data management submodule. The pipeline intelligent decision-making module includes a basic data management submodule, an inspection management submodule, a defect management submodule, and an integrity intelligent decision-making submodule; The data interface submodule of the analysis and decision module is used to provide data interfaces for the basic data management submodule, the detection management submodule, the defect management submodule, and the integrity intelligent decision module. The integrity intelligent decision-making submodule includes intelligent analysis and decision-making for pipeline anomaly signals obtained by a single detection method or multiple detection methods that cannot quantify defect size and type. The process of defect analysis based on the pipeline decision model and maintenance decision-making based on the maintenance level prediction model for pipeline anomaly signals from multiple detection methods is as follows: Determine the detection signal data of buried pipelines corresponding to at least two detection methods; Principal component analysis or discrete cosine transform is used to extract the feature vectors of each detected signal data; Each feature vector is input into a weight prediction model to determine the weight coefficients corresponding to each feature vector according to different detection environments; wherein, the weight prediction model is trained based on the sample feature vector and the sample weight coefficients corresponding to the sample feature vector, and the sample feature vector is determined based on sample detection signal data obtained by at least two detection methods, wherein the at least two detection methods are complementary detection methods. After determining the weight coefficients corresponding to each feature vector, the corresponding weight coefficients in the pipeline decision model are updated so that the pipeline decision model can fuse the pipeline analysis results obtained from each feature vector based on the weight coefficients to obtain the final pipeline analysis result; wherein, the pipeline decision model is trained based on the sample feature vectors and the sample analysis results corresponding to the sample feature vectors; the sample feature vectors are determined based on the sample detection signal data obtained by at least two detection methods; Each feature vector is input into the maintenance level prediction model to obtain the maintenance decision result. The maintenance level prediction model includes a fusion layer and a prediction layer. The fusion layer is used to cross-combine and reduce the dimensionality of each feature vector to form a fusion vector. After obtaining the fusion vector, the fusion vector is input into the prediction layer to obtain the maintenance decision result. When fusing the feature vectors corresponding to each detection method to predict the maintenance level, the feature vectors corresponding to each detection method are subjected to linear interpolation transformation and principal component analysis to reduce them to the same dimension. The fusion vector is obtained by cross-combining feature vectors, and the fusion vector is reduced to the same dimension as the original feature matrix. Before extracting the feature vectors of the detection signal data under each detection method, the detection signal data is preprocessed by normalization: the average length of each detection signal data in the dataset is calculated, and for signal data whose data length is less than the average length, the data is interpolated by the interpolation transformation method. For the interpolated signal data and the signal data whose data length is greater than the average length, the data length is reduced to the average length by the principal component analysis dimensionality reduction method. The 3D model visualization display submodule will display the intelligent evaluation and maintenance decision results in a dynamic 3D interactive visualization.

2. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The pipeline 3D visualization module is based on the principle of human-computer interaction. It conducts on-site surveys of refinery equipment and devices, above-ground landmarks, and above-ground and underground pipelines connected to the devices to determine the geographical information or relative location of the modeling objects. It then establishes the object model in the 3D model and manages the relevant engineering information of the pipelines, devices, and equipment through the 3D model or directory tree. Among these, the pipeline information serves as the data basis for evaluating the integrity of buried pipelines.

3. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The basic data management submodule is used to obtain basic information about the pipeline from the pipeline 3D visualization module. The basic information includes pipe diameter, wall thickness, pipe material, conveying medium, design data, and connection equipment information, and can add, query, delete, and modify the basic information.

4. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The detection management submodule includes four functions: detection signal sample library management, pipeline under test management, detection project management, and detection data management, which are used for the logical management of detection operations, detection signals, and detection results. Among them, the detection signal sample library management is used to manage the historical abnormal detection signals of the pipeline being inspected; the pipeline being inspected management is used to manage the pipeline being inspected and its associated inspection projects; the inspection project management is used to establish an inspection project after an inspection operation is carried out on a section of pipeline, and to manage the inspection project; the detection data management is used to manage the detection data for a specific inspection project, and its functions include supporting the management of detection signal diagrams, detection signal data, and detection result data for different detection methods.

5. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The defect management submodule is used to manage defect data obtained after a pipeline inspection, through expert signal analysis, excavation verification, or a built-in intelligent decision-making model. The defect data can interact with the 3D model to add, query, delete, and modify the defect information of the inspected pipeline. At the same time, defect information is managed according to different defect types. For example, the corrosion rate of corrosion defects is directly calculated based on the defect depth and commissioning time through a built-in algorithm in the background, or it can be recorded through experiments or actual measurements.

6. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The integrity intelligent decision-making submodule is used to: obtain the detection signal map or detection signal intensity value of the pipe section from the detection management submodule for pipeline abnormal signals obtained by a single detection method that cannot quantify the size and type of defects, and make intelligent maintenance decisions using an intelligent decision-making model pre-trained based on deep learning.

7. The intelligent decision-making and visualization system for pipeline integrity according to claim 6, characterized in that, When there is no excavation data available in the early stages, the intelligent decision-making model is trained and tested by using expert-analyzed data in the detection and management submodule; Once the accumulated excavation data meets the preset conditions, excavation defect data is obtained from the detection management submodule, and the intelligent decision-making model is optimized through data migration.

8. The intelligent decision-making and visualization system for pipeline integrity according to claim 6, characterized in that, The training process of the intelligent decision-making model includes: Obtain sample magnetic stress detection signal images or detection signal intensity values ​​from expert analysis data in the detection management submodule; wherein, the sample magnetic stress detection signal images or intensity values ​​need to cover various defect pixel locations and various repair levels; The pipeline anomaly area mileage location and maintenance level are obtained from the sample magnetic stress detection signal image or intensity value from the expert analysis data in the detection management submodule; wherein, the sample data is divided into training set and test set; The model takes a sample magnetic stress detection signal image or intensity value with pre-determined defect pixel location and maintenance level as input, and takes the pipeline abnormal area mileage location and maintenance level in the sample magnetic stress detection signal image or intensity value as output. The neural network model is trained and tested based on the intelligent learning algorithm to obtain the intelligent decision model.

9. The intelligent decision-making and visualization system for pipeline integrity according to claim 1, characterized in that, The integrity intelligent decision-making submodule is also used for: obtaining the detection signal map and detection signal intensity value of the pipe segment under each detection method from the detection management submodule for the pipeline abnormal signal obtained by various detection methods that cannot quantify the defect size and type, extracting the feature vector of the detection signal map and detection signal intensity value under each detection method, and inputting each feature vector into the weight prediction model to obtain the weight coefficient corresponding to each feature vector output by the weight prediction model; After updating the weight coefficients of the pipeline decision model with the weight coefficients corresponding to each feature vector, the feature vectors are input into the pipeline decision model to obtain the pipeline analysis results output by the pipeline decision model. Each feature vector is input into the maintenance level prediction model, which then fuses the feature vectors to obtain a fused vector. The maintenance level prediction result is then obtained based on the fused vector.

10. The intelligent decision-making and visualization system for pipeline integrity according to claim 9, characterized in that, The weight prediction model is trained based on the sample feature vector and the sample weight coefficients corresponding to the sample feature vector. The sample weighting coefficients are determined based on the following steps: Determine the initial sample weight coefficients corresponding to the feature vectors of each sample, and iterate through the initial sample weight coefficients with the target step size to obtain the iterative weight coefficients; wherein, the sum of the iterative weight coefficients obtained in each iteration is 1; Based on the weight coefficients of each iteration, the feature vectors of each sample are weighted and fused to obtain each fused feature vector; Based on each fused feature vector, the pipeline defect prediction result corresponding to each fused feature vector is determined, and the iterative weight coefficient corresponding to the pipeline defect prediction result with the same pipeline defect label is used as the sample weight coefficient corresponding to the sample feature vector.

Images