Pipeline repairing method and system based on defect detection

Abstract

The invention relates to the technical field of defect detection, and discloses a pipeline repairing method and system based on defect detection. The method comprises the following steps: performing acoustic detection on a pipeline based on a multi-frequency sound wave signal to obtain pipeline sound field data; carrying out electromagnetic wave scattering detection on the pipeline to obtain electromagnetic scattering data; constructing a sound field distribution diagram in the pipeline and an electromagnetic property distribution diagram on the surface of the pipeline; performing feature extraction on the sound field distribution diagram to obtain a first detection feature set, and performing feature extraction on the electromagnetic characteristic distribution diagram to obtain a second detection feature set; inputting the first detection feature set and the second detection feature set into a space-time diagram convolutional network model for pipeline defect diagnosis to obtain a defect position and a defect type; and performing causal reasoning and repair scheme matching on the pipeline according to the defect position and the defect type to obtain a target pipeline repair scheme, thereby improving the accuracy of pipeline defect detection, thereby improving the accuracy of pipeline repair.

Description

Technical Field

[0001] The present application relates to the technical field of defect detection, and in particular to a pipeline repair method and system based on defect detection. Background Art

[0002] With the widespread use and long-term operation of pipeline systems, various types of defects are inevitable, such as corrosion, cracks, deformation, etc. These defects may lead to serious accidents such as pipeline leakage and bursting. Traditional pipeline inspection methods often rely on a single inspection method, such as ultrasonic inspection or electromagnetic inspection. These methods perform well in detecting certain specific types of defects, but when faced with complex and diverse pipeline defects, it is often difficult to fully and accurately identify and locate all potential problems.

[0003] The main challenge facing current pipeline defect detection technology is how to effectively integrate multiple detection methods to improve the accuracy and comprehensiveness of defect identification. Existing technologies still have many problems in data fusion, feature extraction, and defect diagnosis. For example, how to deal with heterogeneous data generated by different detection methods, how to extract effective features from massive detection data, and how to establish an accurate defect diagnosis model are all technical problems that need to be solved urgently. In addition, existing pipeline repair solutions often lack pertinence and systematicness, and it is difficult to formulate the optimal repair strategy based on specific defect characteristics and environmental conditions. Summary of the invention

[0004] The present application provides a pipeline repair method and system based on defect detection, which are used to improve the accuracy of pipeline defect detection, thereby improving the accuracy of pipeline repair.

[0005] In a first aspect, the present application provides a pipeline repair method based on defect detection, the pipeline repair method based on defect detection comprising: Acoustically detecting the pipeline based on the multi-frequency sound wave signal and adaptively sampling and phase synchronizing the response signal of the pipeline to obtain pipeline sound field data; Performing electromagnetic wave scattering detection on the pipeline and performing dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; Constructing an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and constructing an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; Performing feature extraction on the acoustic field distribution diagram to obtain a first detection feature set, and performing feature extraction on the electromagnetic characteristic distribution diagram to obtain a second detection feature set; Inputting the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain defect locations and defect types; Causal reasoning and repair scheme matching are performed on the pipeline according to the defect location and the defect type to obtain a target pipeline repair scheme.

[0006] In a second aspect, the present application provides a pipeline repair system based on defect detection, the pipeline repair system based on defect detection comprising: An acoustic detection module, used to perform acoustic detection on the pipeline based on a multi-frequency sound wave signal and to perform adaptive sampling and phase synchronization on the response signal of the pipeline to obtain pipeline sound field data; An electromagnetic detection module, used to perform electromagnetic wave scattering detection on the pipeline and to perform dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; A construction module, used to construct an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and to construct an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; A feature extraction module, used to perform feature extraction on the acoustic field distribution diagram to obtain a first detection feature set, and to perform feature extraction on the electromagnetic characteristic distribution diagram to obtain a second detection feature set; A diagnosis module, used for inputting the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain a defect location and a defect type; A matching module is used to perform causal reasoning and repair scheme matching on the pipeline according to the defect location and the defect type to obtain a repair scheme for the target pipeline.

[0007] The third aspect of the present application provides a pipeline repair device based on defect detection, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the pipeline repair device based on defect detection executes the above-mentioned pipeline repair method based on defect detection.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned pipeline repair method based on defect detection.

[0009] In the technical solution provided by the present application, by combining multi-frequency acoustic signal detection and electromagnetic wave scattering detection, a comprehensive analysis of the acoustic field and electromagnetic characteristics is realized, and the adaptive sampling and phase synchronization technology is used to process the pipeline response signal, which improves the quality and reliability of the original data. By constructing the acoustic field distribution map and the electromagnetic characteristic distribution map, the visualization expression of the internal and surface characteristics of the pipeline is realized, and the multi-angle feature extraction of the acoustic field and electromagnetic characteristic distribution map is performed. The spatiotemporal graph convolution network model is introduced for defect diagnosis, and the spatiotemporal correlation of pipeline defects is effectively utilized to improve the accuracy and robustness of diagnosis. Through time series correlation analysis and multi-task learning, in-depth mining of the dynamic characteristics of defects is realized, which helps to identify complex and evolving defects. Based on the defect knowledge graph and Bayesian network reasoning, intelligent tracing and analysis of the root causes of defects are realized, and multi-criteria decision analysis and fuzzy comprehensive evaluation methods are adopted to ensure the comprehensiveness and optimality of the repair plan and improve the repair effect. By establishing a mapping relationship between defect characteristics and repair plans, automatic matching of repair plans is realized, the efficiency of pipeline maintenance is improved, and intelligent management of the entire life cycle of pipelines is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 This is a schematic diagram of an embodiment of a pipeline repair method based on defect detection in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a pipeline repair system based on defect detection in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a pipeline repair method and system based on defect detection. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of a pipeline repair method based on defect detection includes: Step S101, acoustically detecting the pipeline based on the multi-frequency sound wave signal and adaptively sampling and phase synchronizing the response signal of the pipeline to obtain pipeline sound field data; It is understandable that the execution subject of the present application may be a pipeline repair system based on defect detection, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0014] Specifically, the acoustic wave signal within the preset frequency range is discretized to obtain multiple discrete frequency points. By selecting an appropriate frequency range, the accuracy and effectiveness of the detection are ensured. Among them, the appropriate frequency range refers to the optimal acoustic wave detection frequency interval determined according to the pipeline material, size and expected defect type to be detected. This frequency range is between 20kHz and 5MHz. For metal pipelines, lower frequencies (20-100kHz) are suitable for detecting large-area corrosion and wall thickness changes, while higher frequencies (1-5MHz) are suitable for detecting tiny cracks. Specifically, for a carbon steel pipeline with a diameter of 100mm and a wall thickness of 5mm, the appropriate frequency range for detecting inner wall corrosion is approximately 500kHz-1MHz. Generate pseudo-random coding sequences corresponding to multiple frequencies based on discrete frequency points. The pseudo-random coding sequence is generated by a specific algorithm, has good autocorrelation and cross-correlation characteristics, and helps to improve the resolution and anti-interference ability of the signal. Among them, the specific algorithm refers to the mathematical method used to generate the pseudo-random coding sequence, including the linear feedback shift register (LFSR) algorithm and the Gold code generation algorithm. The LFSR algorithm uses shift registers and XOR gates to generate a pseudo-random sequence based on a specific feedback polynomial. In this embodiment, a 4-bit LFSR uses the polynomial x 4 +x 3+1 indicates that a pseudo-random sequence is generated through the initial seed and each shift operation. The Gold code algorithm generates a pseudo-random sequence with better cross-correlation characteristics by performing modulo-2 addition operations on two different LFSR sequences. The generated pseudo-random coding sequence is orthogonally modulated to transmit signals of different frequencies independently of each other, thereby improving the utilization efficiency of the signal and the accuracy of detection, and obtaining a multi-frequency orthogonal modulation signal. The piezoelectric transducer array is excited according to the multi-frequency orthogonal modulation signal. The piezoelectric transducer array is a device that converts electrical signals into acoustic signals using the piezoelectric effect. By exciting the piezoelectric transducer array, a multi-frequency acoustic excitation signal is generated. The propagation of the multi-frequency acoustic excitation signal in the pipeline is simulated by time-domain finite difference simulation to establish a prediction model for the acoustic field distribution in the pipeline. The specific steps are as follows: Spatial discretization: Divide the pipeline space into a uniform grid. For three-dimensional space, the grid points are represented as (i, j, k), where i, j, k represent the indexes in the x, y, and z directions, respectively. The grid spacing Δx, Δy, Δz needs to be small enough to ensure the calculation accuracy and should be less than 1 / 10 of the minimum wavelength. Time discretization: Divide the time axis into equally spaced time steps, and use n to represent the time step index. The time step Δt needs to satisfy the Courant-Friedrichs-Lewy (CFL) condition to ensure calculation stability. Discretization of acoustic wave equations: Discretize the partial differential equations of acoustic wave propagation. Boundary condition processing: Process the reflection of acoustic waves at the pipe wall boundary. Boundary conditions include: rigid boundary: p[boundary] = 0, soft boundary: ∂p / ∂n[boundary] = 0, impedance boundary: ∂p / ∂n+(1 / Z)(∂p / ∂t) = 0, where Z is the acoustic impedance. Excitation source modeling: Simulate the acoustic wave excitation source, including point source, line source or surface source. For a point sound source, a time-varying pressure is applied to the grid point: p[is, js, ks, n] = A*sin(2πft), where (is, js, ks) is the sound source position, A is the amplitude, and f is the frequency. Iterative calculation: Starting from the initial conditions, the sound pressure value of each grid point is calculated step by time. Absorbing boundary conditions: Absorbing boundary conditions are used at the boundary of the computational domain to simulate infinite space and avoid boundary reflections. Post-processing: Calculate the energy distribution of the sound field: E=(1 / 2ρc²)*p², calculate the sound intensity: I=p*v, where v is the particle velocity, and perform spectrum analysis. Model verification and optimization: Compare with analytical solutions: For simple geometric shapes, compare the numerical solution with the known analytical solution. Mesh convergence study: Gradually refine the mesh until the result no longer changes significantly. Experimental verification: Compare with the actual pipeline measurement results, adjust the model parameters, and establish a prediction model for the sound field distribution in the pipeline. Time-domain finite difference simulation is a numerical simulation method that predicts the distribution of the sound field by discretizing and numerically solving the propagation process of sound waves in the pipeline. The response signal of the pipeline is collected according to the sound field distribution prediction model to obtain the original response signal data. According to the original response signal data, the sampling rate is dynamically adjusted to obtain an adaptive sampling signal.Adaptive sampling is a method that dynamically adjusts the sampling rate according to the characteristics of the signal, which can improve the sampling efficiency and accuracy of the signal. Wavelet denoising and phase correction are performed on the adaptive sampling signal. Wavelet denoising is a method that uses wavelet transform to decompose and reconstruct signals, which can effectively remove noise components in the signal; phase correction is to adjust the phase information of the signal to eliminate phase deviation and distortion. After wavelet denoising and phase correction, a noise-suppressed synchronization signal is obtained. Time-frequency analysis and spatial interpolation are performed on the noise-suppressed synchronization signal to obtain the pipeline sound field data. Time-frequency analysis is a method that combines the time domain and frequency domain information of the signal, which can provide the time-frequency distribution of the signal; spatial interpolation is to interpolate and reconstruct the spatial distribution of the signal to improve the spatial resolution and accuracy of the data.

[0015] Step S102, performing electromagnetic wave scattering detection on the pipeline and performing dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; Specifically, the preset terahertz frequency range is discretely sampled to ensure that the required frequency range is covered. By phase encoding the frequency scanning sequence, a terahertz pulse sequence with unique phase information is generated to enhance the signal resolution and detection sensitivity. The phase-coded terahertz pulse sequence is used to drive the terahertz wave transmitter to generate an incident electromagnetic wave, which can interact with the pipeline surface with specific phase and frequency characteristics. In the process of interaction between the incident electromagnetic wave and the pipeline surface, a scattering field distribution prediction model is established by a finite element simulation method. Among them, the process of establishing a scattering field distribution prediction model by a finite element simulation method is as follows: Geometric modeling: Create a three-dimensional geometric model of the pipeline and its surrounding space. Define material regions, such as pipeline walls, internal media, and external space. Consider possible defects, such as cracks and corrosion. Meshing: Discretize the geometric model into a finite element mesh using tetrahedral or hexahedral units. Use a finer mesh in areas where the electromagnetic field is expected to change drastically (such as defect edges and material interfaces). Use adaptive mesh technology to dynamically adjust the mesh density based on preliminary calculation results. Material property definition: Specify corresponding electromagnetic properties for each region. For conductors, define the conductivity. Considering the frequency dependence of the material, a dispersion model is required. Governing equation setup: Based on the Maxwell equations, the wave equation can be obtained in the frequency domain. Boundary condition definition: Set absorbing boundary conditions at the boundaries of the computational domain. Apply appropriate continuity conditions on the material interface. For metal surfaces, apply ideal conductor boundary conditions. Excitation source modeling: Establish a terahertz wave transmitter model using Gaussian wave packets. Finite element discretization: Discretize the wave equation using the weighted residual method. Assemble the discretized equations into a global matrix equation. Solve large-scale linear equations: For frequency domain problems, use direct solvers or iterative solvers. For time domain problems, use time integration methods. Post-processing and scattered field calculations: Calculate the scattered field, calculate the scattering cross section, and calculate the radar scattering cross section. Model verification and optimization: Compare with analytical solutions: For simple geometries, compare the numerical solutions with the Mie scattering theory solution. Experimental verification: Compare with actual terahertz scattering measurement results. Error analysis: Calculate the convergence and stability of the numerical solution. Parameter optimization: Use sensitivity analysis and optimization algorithms to adjust model parameters and establish a scattering field distribution prediction model. This model can accurately describe the scattering behavior of electromagnetic waves on the surface and inside of the pipeline. According to the scattering field distribution prediction model, the scattered electromagnetic waves are collected in real time to obtain the original scattering signal. The electromagnetic characteristics and defect information of the pipeline surface are reflected by analyzing the original scattering signal. According to the original scattering signal, the local oscillator frequency is adjusted in real time to generate a dynamic frequency tuning signal to ensure the frequency matching and stability of the signal and enhance the sensitivity of the detection system. The dynamic frequency tuning signal is coherently demodulated with the reference signal to obtain a demodulated complex signal, which can accurately reflect the amplitude and phase information of the electromagnetic wave.The demodulated complex signal is transformed into frequency domain information by spatial Fourier transform, and then combined with back-projection reconstruction technology to obtain electromagnetic scattering data. Spatial Fourier transform can convert the signal from time domain to frequency domain, enhancing the analyzability of the signal, while back-projection reconstruction converts the frequency domain data back to the spatial domain through reconstruction algorithm to obtain electromagnetic scattering images.

[0016] Step S103, constructing an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and constructing an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; Specifically, the pipeline sound field data is subjected to a time domain Fourier transform to obtain frequency domain sound field data. By processing these frequency domain sound field data, the propagation characteristics of sound waves inside the pipeline are analyzed. The sound wave propagation equation is solved according to the frequency domain sound field data to obtain the sound pressure distribution inside the pipeline. The sound wave propagation equation in the medium is numerically solved to determine the distribution of sound pressure at various positions inside the pipeline. The sound pressure distribution is spatially interpolated and smoothed to obtain the initial sound field distribution map. The purpose of spatial interpolation is to fill the data gaps in the sound pressure distribution so that the sound field distribution is more continuous and smooth. Smoothing is to reduce noise and errors in the data, thereby improving the accuracy of the distribution map. According to the initial sound field distribution map, the boundary conditions are iteratively optimized to obtain the sound field distribution map inside the pipeline. Iterative optimization is a process of repeatedly adjusting the boundary conditions to make the sound field distribution more consistent with the actual situation and improve its accuracy and reliability. At the same time, the electromagnetic scattering data is polarization decomposed to obtain a scattering matrix. Polarization decomposition is to decompose the scattered signal into components of different polarization states to extract more physical information. According to the scattering matrix, the inverse scattering problem is solved to obtain the dielectric constant distribution of the pipeline surface. The solution of the inverse scattering problem is a complex process. The physical properties of the pipeline surface are inferred from the scattered signal through numerical algorithms. The dielectric constant is an important parameter of the material. Its distribution can be used to understand the material properties of the pipeline surface. The dielectric constant distribution is spatially filtered and denoised to obtain the initial electromagnetic distribution map. The purpose of spatial filtering is to remove high-frequency noise in the data to make the distribution map smoother and clearer. Denoising is to reduce the random noise generated during the measurement process, thereby improving the accuracy of the data. According to the initial electromagnetic distribution map, the material parameters are constrained optimized and registered mapping is performed to obtain the electromagnetic characteristic distribution map of the pipeline surface. Constrained optimization is to optimize the accuracy of the electromagnetic distribution map by introducing some prior knowledge or physical constraints. Registration mapping is to match the optimized electromagnetic distribution map with the actual pipeline surface to obtain a highly accurate electromagnetic characteristic distribution map.

[0017] Step S104, extracting features from the sound field distribution map to obtain a first detection feature set, and extracting features from the electromagnetic characteristic distribution map to obtain a second detection feature set; Specifically, the sound field distribution map is subjected to wavelet transformation to obtain a multi-scale sound field coefficient matrix, and the sound field data is decomposed into components of different scales and frequencies, so as to better analyze its local features and overall features. The local energy distribution is calculated according to the multi-scale sound field coefficient matrix to obtain a sound field energy feature map. The sound field energy feature map reflects the energy distribution of the sound field at different positions and can effectively characterize the local characteristics of the sound field. The sound field energy feature map is subjected to local binary pattern transformation to extract the texture features of the sound field. Local binary pattern transformation is an effective texture analysis method, which extracts the texture information of the image by comparing the pixel value with its neighboring pixel value. According to the sound field texture features, the gray level co-occurrence matrix is ​​calculated to obtain the sound field statistical features. The gray level co-occurrence matrix is ​​a statistical tool that describes the spatial distribution features of the gray value of an image. The statistical features of the image are extracted by analyzing the spatial relationship between the gray value pairs. The principal component analysis is performed on the sound field statistical features to reduce the dimension and extract the main features to obtain the first detection feature set. At the same time, the electromagnetic characteristic distribution map is subjected to Gabor filtering to obtain multi-directional electromagnetic characteristic responses. Gabor filtering is a filtering method that simulates the human visual system. It extracts the directional features of the image by filtering the image in different directions and scales. According to the multi-directional electromagnetic characteristic response, the directional gradient histogram is calculated to obtain the directional features of the electromagnetic characteristics. The directional gradient histogram is a statistical tool that describes the local directional distribution characteristics of the image. By analyzing the distribution of the image gradient direction, the directional features of the image are extracted. The directional features of the electromagnetic characteristics are transformed by Fourier descriptor to transform the spatial domain features into frequency domain features to obtain the frequency domain features of the electromagnetic characteristics. Fourier descriptor transform is a method that transforms the edge features of the image into frequency domain representation. By analyzing the frequency domain characteristics of the image, the frequency components of the image are extracted. According to the frequency domain characteristics of the electromagnetic characteristics, the statistical moment is calculated to extract the statistical features of the image. The statistical moment is a statistical tool that describes the grayscale distribution characteristics of the image. By calculating the statistical quantities such as the mean, variance, skewness and kurtosis of the image, the statistical features of the image are extracted. Independent component analysis is performed on the statistical features of the electromagnetic characteristics to extract the main features that are independent of each other and obtain the second detection feature set. Independent component analysis is a method that decomposes a signal into independent components and extracts the essential characteristics of the signal by analyzing the independence of the signal.

[0018] Step S105: input the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain defect locations and defect types; Specifically, the first detection feature set and the second detection feature set are subjected to feature fusion to obtain a multimodal feature matrix. Feature fusion can comprehensively utilize information of two different modes by combining feature sets of acoustic field and electromagnetic characteristics, thereby improving the accuracy and robustness of defect detection. The multimodal feature matrix is ​​input into a preset spatiotemporal graph convolutional network model to construct a spatiotemporal graph structure to obtain a spatiotemporal graph of pipeline defects. The spatiotemporal graph convolutional network model can effectively capture the dependency of features in time and space, and reflect the distribution and evolution of pipeline defects by constructing a spatiotemporal graph structure. A graph convolution operation is performed on the spatiotemporal graph of pipeline defects to extract graph feature representation. The graph convolution operation is a convolution calculation performed on a graph structure, which can capture the relationship and feature transfer between graph nodes. Through the graph convolution operation, a high-order feature representation of pipeline defects can be extracted. Based on the graph feature representation, multi-scale feature extraction is performed to obtain a multi-scale feature vector. Multi-scale feature extraction can capture different levels of information of defects by performing feature analysis at different scales, thereby improving the richness and expression ability of feature representation. The multi-scale feature vector is weighted by an attention mechanism to obtain a weighted feature vector. The attention mechanism can enhance the model's attention to key features and improve the effectiveness of feature representation by giving higher weights to important features. Nonlinear transformation is performed on the weighted feature vector to obtain a defect representation vector. Nonlinear transformation can enhance the complexity and discrimination ability of feature representation by introducing nonlinear functions, and better characterize the characteristics of defects. Time series correlation analysis is performed on the defect representation vector to obtain a time series feature sequence. Time series correlation analysis can capture the dynamic characteristics of defects by analyzing the temporal change law of features. Based on the time series feature sequence, defect dynamic features are generated. The defect dynamic features reflect the evolution of defects over time. Multi-task learning is performed on the defect dynamic features to simultaneously predict the probability distribution of defect locations and defect type. Multi-task learning can improve the generalization ability and prediction accuracy of the model by optimizing the objective functions of multiple related tasks at the same time. Threshold judgment is performed based on the probability distribution of defect locations and defect type to obtain the defect location and defect type of the pipeline. Threshold judgment can effectively filter noise and false predictions by setting a reasonable threshold, thereby improving the accuracy and reliability of defect diagnosis.

[0019] Step S106: Perform causal reasoning and repair scheme matching on the pipeline according to the defect location and defect type to obtain a repair scheme for the target pipeline.

[0020] Specifically, the defect location and defect type are encoded, and the specific location and type of the defect are converted into a standardized data format to obtain defect feature information. According to the defect feature information, the preset defect knowledge graph is retrieved to obtain the corresponding defect node set. The defect knowledge graph is a knowledge base that systematically organizes and represents different types of defects and their relationships. By retrieving the knowledge graph, the node set related to the current defect can be quickly found. According to the defect node set, a defect causal relationship graph is constructed. The defect causal relationship graph is a relationship graph established by analyzing the causal relationship between different defects. According to the defect causal relationship graph, path importance analysis is performed to determine the key causal chain. By evaluating the importance of different causal paths, path importance analysis can identify the most critical causal chain and help determine the main cause and impact of the defect. The key causal chain is traced back to obtain a candidate set of defect root causes. Reverse tracing is to find the root cause that may cause the current defect by reversely analyzing the causal chain. According to the candidate set of defect root causes, Bayesian network reasoning is performed to generate an initial set of repair solutions. Bayesian network is a model based on probabilistic reasoning that generates a series of possible repair solutions by combining defect root causes and repair measures. Based on the initial set of repair solutions, multi-criteria decision analysis is performed to obtain a repair solution scoring table. Multi-criteria decision analysis is to evaluate and score different repair solutions by comprehensively considering multiple evaluation criteria to select the best repair solution. Fuzzy comprehensive evaluation is performed on the repair solution scoring table to obtain the target pipeline repair solution. Fuzzy comprehensive evaluation is a method that applies fuzzy logic to decision analysis. By dealing with the uncertainty and ambiguity in the scoring, the pros and cons of the repair solution can be more accurately evaluated.

[0021] In the embodiment of the present application, by combining multi-frequency acoustic signal detection and electromagnetic wave scattering detection, a comprehensive analysis of the acoustic field and electromagnetic characteristics is achieved, and the pipeline response signal is processed by adaptive sampling and phase synchronization technology, which improves the quality and reliability of the original data. By constructing the acoustic field distribution map and the electromagnetic characteristic distribution map, the visual expression of the internal and surface characteristics of the pipeline is achieved, and the multi-angle feature extraction of the acoustic field and electromagnetic characteristic distribution map is performed. The spatiotemporal graph convolution network model is introduced for defect diagnosis, and the spatiotemporal correlation of pipeline defects is effectively utilized to improve the accuracy and robustness of diagnosis. Through time series correlation analysis and multi-task learning, in-depth mining of the dynamic characteristics of defects is achieved, which helps to identify complex and evolving defects. Based on the defect knowledge graph and Bayesian network reasoning, intelligent tracing and analysis of the root causes of defects are achieved, and multi-criteria decision analysis and fuzzy comprehensive evaluation methods are adopted to ensure the comprehensiveness and optimality of the repair plan and improve the repair effect. By establishing a mapping relationship between defect characteristics and repair plans, automatic matching of repair plans is achieved, the efficiency of pipeline maintenance is improved, and intelligent management of the entire life cycle of pipelines is achieved.

[0022] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Discretize the acoustic wave signal within a preset frequency range to obtain multiple discrete frequency points; (2) generating a plurality of pseudo-random coding sequences corresponding to the frequencies according to the discrete frequency points, and performing orthogonal modulation on the pseudo-random coding sequences corresponding to the plurality of frequencies to obtain a multi-frequency orthogonal modulation signal; (3) The piezoelectric transducer array is excited according to the multi-frequency orthogonal modulation signal to obtain a multi-frequency acoustic wave excitation signal, and the propagation of the multi-frequency acoustic wave excitation signal in the pipeline is simulated by time-domain finite difference to obtain a prediction model of the acoustic field distribution in the pipeline; (4) Collect the response signal of the pipeline according to the prediction model of the sound field distribution in the pipeline to obtain the original response signal data; (5) Dynamically adjust the sampling rate according to the original response signal data to obtain an adaptive sampling signal, and perform wavelet denoising and phase correction on the adaptive sampling signal to obtain a synchronization signal after noise suppression; (6) Perform time-frequency analysis and spatial interpolation on the noise-suppressed synchronization signal to obtain the pipeline sound field data.

[0023] Specifically, the sound wave signal within the preset frequency range is discretized to obtain multiple discrete frequency points. Assume that the preset frequency range is arrive , dividing it into equal intervals discrete frequency points, that is, the discrete frequency points are ,in, For the frequency points, is the frequency interval. For example, assuming kHz, kHz, and 10 discrete frequency points are required, then the interval between each frequency point is kHz, the discrete frequency points obtained are 1kHz, 2kHz, 3kHz, etc. Generate pseudo-random code sequences corresponding to multiple frequencies based on discrete frequency points. Pseudo-random code sequences are a type of pseudo-random signal with good autocorrelation and cross-correlation characteristics, which can distinguish and identify different frequencies. When generating pseudo-random code sequences, algorithms such as linear feedback shift registers (LFSRs) or Gold codes are usually used. For example, for each discrete frequency point , generates a pseudo-random sequence . Orthogonal modulation is performed on the pseudo-random coding sequences corresponding to multiple frequencies to obtain a multi-frequency orthogonal modulation signal. Pseudo-random sequences of different frequencies are modulated to make them orthogonal to each other in the frequency domain to avoid interference. The orthogonal modulation signal can be expressed as: ; in, It is The pseudo-random code sequence corresponding to the frequency, It is discrete frequency points, is the initial phase. The piezoelectric transducer array is excited according to the multi-frequency orthogonal modulation signal to obtain a multi-frequency acoustic wave excitation signal. The piezoelectric transducer array is a device that converts electrical signals into mechanical waves. By exciting the piezoelectric transducer array, a multi-frequency acoustic wave excitation signal is generated. When these acoustic wave excitation signals propagate in the pipeline, their propagation characteristics can be simulated by finite difference time domain simulation. In the finite difference time domain simulation, the wave equation of acoustic wave propagation can be expressed as: ; in, is the sound pressure, is the speed of sound, It's time. is the Laplace operator. By discretizing the wave equation, the propagation of sound waves in the pipeline is numerically simulated to obtain the prediction model of the sound field distribution in the pipeline. According to the prediction model of the sound field distribution in the pipeline, the response signal of the pipeline is collected to obtain the original response signal data. The collected original response signal data contains the sound pressure information at various positions inside the pipeline, which is used for subsequent signal processing and analysis. According to the original response signal data, the sampling rate is dynamically adjusted to obtain an adaptive sampling signal. Adaptive sampling is a method of dynamically adjusting the sampling rate according to the signal characteristics, which can improve the sampling efficiency and accuracy of the signal. By analyzing the original response signal data, the optimal sampling rate is determined and adaptively adjusted to obtain a higher quality sampling signal. Wavelet denoising and phase correction are performed on the adaptive sampling signal to obtain a noise-suppressed synchronization signal. Wavelet denoising is a method of decomposing and reconstructing signals using wavelet transform, which can effectively remove noise components in the signal. Phase correction is to adjust the phase information of the signal to eliminate phase deviation and distortion. Through wavelet denoising and phase correction processing, a clearer and more accurate synchronization signal is obtained. Time-frequency analysis and spatial interpolation are performed on the noise-suppressed synchronous signal to obtain the pipeline sound field data. Time-frequency analysis is a method that combines the time domain and frequency domain information of the signal to provide the time-frequency distribution of the signal. Spatial interpolation interpolates and reconstructs the spatial distribution of the signal to improve the spatial resolution and accuracy of the data. Through time-frequency analysis and spatial interpolation, the sound field data inside the pipeline is obtained to provide sound field distribution information.

[0024] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) discretely sampling a preset terahertz frequency range to generate a frequency scanning sequence, and phase-encoding the frequency scanning sequence to obtain a phase-encoded terahertz pulse sequence; (2) driving a terahertz wave transmitter according to a phase-encoded terahertz pulse sequence to obtain an incident electromagnetic wave; (3) Perform finite element simulation on the interaction between the incident electromagnetic wave and the pipeline surface to obtain a scattering field distribution prediction model, and collect the scattered electromagnetic wave based on the scattering field distribution prediction model to obtain the original scattering signal; (4) adjusting the local oscillator frequency in real time according to the original scattered signal to obtain a dynamic frequency tuning signal, and coherently demodulating the dynamic frequency tuning signal and the reference signal to obtain a demodulated complex signal; (5) The electromagnetic scattering data is obtained by performing spatial Fourier transform and back-projection reconstruction on the demodulated complex signal.

[0025] Specifically, the preset terahertz frequency range is discretely sampled to generate a frequency scanning sequence. A pseudo-random coding sequence corresponding to multiple frequencies is generated according to the discrete frequency points. The pseudo-random coding sequence has good autocorrelation and cross-correlation characteristics, and can distinguish and identify different frequencies. For example, for each discrete frequency point , generates a pseudo-random sequence . These pseudo-random sequences can be generated by linear feedback shift registers or other common algorithms. The generated pseudo-random coding sequence is phase-encoded to obtain a phase-encoded terahertz pulse sequence. The phase-encoded terahertz pulse sequence is used to drive a terahertz wave transmitter to generate incident electromagnetic waves. The terahertz wave transmitter can convert the encoded signal into electromagnetic waves in the corresponding terahertz frequency range based on technologies such as photoconductive antennas or optical rectification. These incident electromagnetic waves will interact with the pipeline surface, and through this interaction, the scattering characteristics of the pipeline surface are obtained. The finite element method is used to accurately simulate the interaction between the incident electromagnetic wave and the pipeline surface. Finite element simulation is a numerical calculation method used to solve the electromagnetic field distribution of complex geometric structures and material properties. By performing finite element simulation on the propagation of the incident electromagnetic wave on the surface and inside of the pipeline, a scattering field distribution prediction model is obtained. This model can accurately describe the scattering behavior of electromagnetic waves in the pipeline. According to the scattering field distribution prediction model, the scattered electromagnetic waves are collected to obtain the original scattering signal. According to the change of the signal, the local oscillator frequency is adjusted in real time to obtain a dynamic frequency tuning signal. Dynamic frequency tuning signals can ensure that the reception and processing of signals at different frequencies are highly stable and accurate. The dynamic frequency tuning signal is coherently demodulated with the reference signal to obtain a demodulated complex signal. Coherent demodulation is a method that can effectively extract useful information from the signal by matching the phase and frequency with the reference signal. The demodulated complex signal It can be expressed as: ; in, is the amplitude of the signal, is the phase of the signal. Through coherent demodulation, the received signal is decomposed into two components: amplitude and phase. The demodulated complex signal is subjected to spatial Fourier transform and back-projection reconstruction to obtain electromagnetic scattering data. Spatial Fourier transform is a method of converting time domain signals into frequency domain signals. The distribution of the signal in the frequency domain is obtained through Fourier transform. Back-projection reconstruction uses the frequency domain data after Fourier transform to reconstruct the original spatial distribution image. Assume that the demodulated complex signal is represented in the frequency domain as , through spatial Fourier transform we can get: ; in, is the frequency domain signal, is the time domain signal, is the time variable, is the frequency variable. Through the back-projection algorithm, the frequency domain signal The inverse transformation is then performed back to the spatial domain to obtain the electromagnetic scattering data of the pipeline surface.

[0026] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Perform a time-domain Fourier transform on the pipeline sound field data to obtain the frequency-domain sound field data, and solve the sound wave propagation equation based on the frequency-domain sound field data to obtain the sound pressure distribution inside the pipeline; (2) Perform spatial interpolation and smoothing on the sound pressure distribution to obtain an initial sound field distribution map, and iteratively optimize the boundary conditions based on the initial sound field distribution map to obtain the sound field distribution map inside the pipeline; (3) Perform polarization decomposition on the electromagnetic scattering data to obtain a scattering matrix, and solve the inverse scattering problem based on the scattering matrix to obtain the dielectric constant distribution on the pipeline surface; (4) The dielectric constant distribution is spatially filtered and denoised to obtain an initial electromagnetic distribution map. Based on the initial electromagnetic distribution map, the material parameters are constrained and optimized and mapped to obtain an electromagnetic characteristic distribution map of the pipeline surface.

[0027] Specifically, the pipeline sound field data is subjected to a time domain Fourier transform to obtain the frequency domain sound field data. The time domain Fourier transform is a method of converting a signal from the time domain to the frequency domain. Through the Fourier transform, the distribution of the sound field data in the frequency domain is analyzed. Assume that the sound field data is , whose Fourier transform It is expressed as: ; in, is the frequency domain signal, is the time domain signal, is the frequency, is the time variable, is an imaginary unit. The sound wave propagation equation is solved according to the frequency domain sound field data to obtain the sound pressure distribution inside the pipe. The propagation of sound waves in the pipe can be described by the sound wave propagation equation, which is: ; in, is the sound pressure, is the speed of sound, is the Laplace operator, which represents the spatial derivative. According to the frequency domain sound field data, the time derivative is converted into the frequency derivative, thereby simplifying the process of solving the equation. The sound wave propagation equation is solved by numerical methods (such as the finite element method or the boundary element method) to obtain the sound pressure distribution at various positions inside the pipeline. The sound pressure distribution is spatially interpolated and smoothed to obtain the initial sound field distribution map. The purpose of spatial interpolation is to fill the data gaps in the sound pressure distribution so that the sound field distribution is more continuous and smooth. The smoothing process is to reduce the noise and error in the data, thereby improving the accuracy of the distribution map. For example, using bilinear interpolation or spline interpolation for spatial interpolation can effectively improve the resolution and accuracy of the data. According to the initial sound field distribution map, the boundary conditions are iteratively optimized to finally obtain the sound field distribution map inside the pipeline. Iterative optimization is a process of repeatedly adjusting the boundary conditions to make the sound field distribution more in line with the actual situation, thereby improving its accuracy and reliability. By continuously adjusting and optimizing the boundary conditions, the calculation error can be reduced and a more accurate sound field distribution map can be obtained. At the same time, the electromagnetic scattering data is polarized to obtain the scattering matrix. Polarization decomposition is a method of decomposing scattered signals into components of different polarization states, which can extract more physical information. According to the scattering matrix, the inverse scattering problem is solved to obtain the dielectric constant distribution of the pipeline surface. The solution of the inverse scattering problem is a process of inferring the surface or internal characteristics of an object through known scattering signals. Commonly used solution methods include iterative reconstruction algorithm and Bayesian inversion method. By solving the inverse scattering problem, the dielectric constant distribution of the pipeline surface is obtained. The dielectric constant distribution is spatially filtered and denoised to obtain the initial electromagnetic distribution map. The purpose of spatial filtering is to remove high-frequency noise in the data to make the distribution map smoother and clearer. Denoising is to reduce the random noise generated during the measurement process, thereby improving the accuracy of the data. According to the initial electromagnetic distribution map, the material parameters are constrained optimized and registered mapping is performed to obtain the electromagnetic characteristic distribution map of the pipeline surface. Constrained optimization is to optimize the accuracy of the electromagnetic distribution map by introducing some prior knowledge or physical constraints. Registration mapping is to match the optimized electromagnetic distribution map with the actual pipeline surface to obtain a highly accurate electromagnetic characteristic distribution map.

[0028] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Performing wavelet transform on the sound field distribution map to obtain a multi-scale sound field coefficient matrix, and calculating the local energy distribution based on the multi-scale sound field coefficient matrix to obtain a sound field energy characteristic map; (2) Performing local binary pattern transformation on the sound field energy feature map to obtain the sound field texture feature, and calculating the gray level co-occurrence matrix based on the sound field texture feature to obtain the sound field statistical feature; (3) performing principal component analysis on the sound field statistical features to obtain a first detection feature set; (4) Performing Gabor filtering on the electromagnetic characteristic distribution map to obtain multi-directional electromagnetic characteristic responses, and calculating the directional gradient histogram based on the multi-directional electromagnetic characteristic responses to obtain the electromagnetic characteristic directional features; (5) Performing Fourier descriptor transformation on the directional characteristics of the electromagnetic characteristics to obtain the frequency domain characteristics of the electromagnetic characteristics, and calculating the statistical moment based on the frequency domain characteristics of the electromagnetic characteristics to obtain the statistical characteristics of the electromagnetic characteristics; (6) Perform independent component analysis on the statistical features of the electromagnetic characteristics to obtain a second detection feature set.

[0029] Specifically, the sound field distribution map is subjected to wavelet transform. Wavelet transform is a mathematical tool that decomposes a signal into components of different scales and frequencies, and can effectively extract the local features of the signal. Suppose the sound field distribution map is , the multi-scale sound field coefficient matrix can be obtained through wavelet transform , the formula is: ; in, The scale is and the location is The wavelet basis function is and is the spatial coordinate. By performing wavelet transform on the sound field distribution map, the detailed characteristics of the sound field are obtained at different scales and positions. The local energy distribution is calculated according to the multi-scale sound field coefficient matrix to obtain the sound field energy characteristic map. The local energy distribution can be expressed by calculating the square sum of the coefficients at each scale and position, that is: ; in, Indicated in scale and location The local energy distribution reveals the energy concentration of the sound field at different locations and scales, which helps to identify abnormal areas. Perform local binary pattern transformation. Local binary pattern encodes the result into a binary pattern by comparing the grayscale value of each pixel with its neighboring pixels. For example, the value of the central pixel is compared with the values ​​of the surrounding pixels. If the central pixel value is greater than or equal to the surrounding pixel value, it is recorded as 1, otherwise it is recorded as 0, and the binary values ​​are combined into a binary number. The result of the local binary pattern transformation is a binary image containing texture information. The grayscale co-occurrence matrix is ​​calculated based on the sound field texture features. The grayscale co-occurrence matrix describes the texture by calculating the joint probability distribution of grayscale value pairs in the image. For each pair of grayscale values ​​in the image and , the gray-level co-occurrence matrix calculates the frequency of simultaneous occurrence of these gray-level value pairs at a specific distance and direction. The gray-level co-occurrence matrix obtained Used to extract texture features, such as contrast, energy, homogeneity, etc. Perform principal component analysis on the sound field statistical features. Principal component analysis reduces the dimension of the data by projecting the data onto the principal component direction while retaining most of the variance information. Suppose the sound field statistical features are , whose covariance matrix is , the principal component is decomposed by eigenvalue to obtain the eigenvector matrix and the corresponding eigenvalue matrix By selecting the largest The eigenvalues ​​corresponding to the eigenvectors are converted into the original data Project to low-dimensional space to get the first detection feature set Similarly, the electromagnetic characteristic distribution map is subjected to Gabor filtering to obtain multi-directional electromagnetic characteristic responses. The Gabor filter is a filter with directional selectivity that can effectively extract the directional information of the image. The response of the Gabor filter can be expressed as: ; in, , , is the direction angle, is the wavelength, is the spatial aspect ratio, is the standard deviation of the Gaussian envelope, is the phase shift. Through multi-directional Gabor filtering, the electromagnetic characteristic responses in different directions are obtained. The directional gradient histogram is calculated based on the multi-directional electromagnetic characteristic responses. The directional gradient histogram is a feature that describes the distribution of local gradient directions. The image is divided into small units, the gradient direction histogram in each unit is calculated, and these histograms are combined into a feature vector. The directional gradient histogram feature can effectively capture the edge and direction information of the image. The electromagnetic characteristic directional features are transformed by Fourier descriptor. Fourier descriptor is a shape description method based on Fourier transform. The frequency domain features are obtained by Fourier transforming the shape contour. Let the shape contour be , whose Fourier transform is: ; in, is the frequency domain feature, is the frequency. Through the Fourier descriptor, the characteristics of electromagnetic characteristics in the frequency domain are obtained. The statistical moment is calculated based on the frequency domain characteristics of electromagnetic characteristics. The statistical moment is a characteristic quantity that describes the shape of data distribution, including mean, variance, skewness and kurtosis. Suppose the frequency domain characteristics of electromagnetic characteristics are , its mean ,variance , skewness and Kurtosis They can be expressed as: ; By calculating these statistical moments, the statistical characteristics of electromagnetic characteristics are obtained. Independent component analysis is performed on the statistical characteristics of electromagnetic characteristics. Independent component analysis is a method to separate mixed signals into independent components and find independent signal sources by maximizing the non-Gaussianity of the signal.

[0030] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Fusing the first detection feature set and the second detection feature set to obtain a multimodal feature matrix; (2) Input the multimodal feature matrix into the preset spatiotemporal graph convolutional network model to construct the spatiotemporal graph structure and obtain the spatiotemporal graph of pipeline defects; (3) Perform graph convolution on the spatiotemporal graph of pipeline defects to obtain graph feature representation, and perform multi-scale feature extraction based on the graph feature representation to obtain a multi-scale feature vector; (4) Perform an attention mechanism weighting on the multi-scale feature vector to obtain a weighted feature vector, and perform a nonlinear transformation on the weighted feature vector to obtain a defect representation vector; (5) Perform time series correlation analysis on the defect characterization vector to obtain a time series feature sequence, and generate defect dynamic features based on the time series feature sequence; (6) Perform multi-task learning on the dynamic characteristics of defects to obtain the probability distribution of defect locations and defect types; (7) Based on the probability distribution of defect location and defect type, threshold judgment is performed to obtain the defect location and defect type of the pipeline.

[0031] Specifically, feature fusion is performed on the first detection feature set and the second detection feature set, and features of different modalities are combined to form a feature matrix containing multimodal information. The multimodal feature matrix is ​​input into the preset spatiotemporal graph convolutional network model to construct the spatiotemporal graph structure to obtain the pipeline defect spatiotemporal graph. The spatiotemporal graph convolutional network model can effectively capture the dependencies of features in time and space. A graph convolution operation is performed on the pipeline defect spatiotemporal graph to obtain a graph feature representation. The graph convolution operation is to capture the relationship between a node and its neighboring nodes by performing a convolution calculation on the graph structure. Assume that the node The characteristic is expressed as , the process of graph convolution can be expressed as: ; in, represents the updated node features, Representation Node The set of neighbor nodes of is the weight matrix, is the bias term, is an activation function. Through multi-layer graph convolution operations, high-level feature representations of nodes are extracted. Multi-scale feature extraction is performed based on the graph feature representation to obtain a multi-scale feature vector. Multi-scale feature extraction is achieved by aggregating features at different scales to capture different levels of feature information. Multi-scale feature extraction is achieved through different convolution kernel sizes or pooling operations. The multi-scale feature vector is weighted by the attention mechanism to obtain a weighted feature vector. The attention mechanism aggregates features in a weighted manner by calculating the importance weights of different features. A nonlinear transformation is performed based on the weighted feature vector to obtain a defect representation vector. Nonlinear transformation can be achieved through a multi-layer feedforward neural network to increase the expressive power of the model. A temporal correlation analysis is performed on the defect representation vector to obtain a temporal feature sequence. Temporal correlation analysis can be achieved through temporal models such as long short-term memory networks or gated recurrent units. Defect dynamic features are generated based on the temporal feature sequence. Defect dynamic features can be further processed to extract dynamic change information. Multi-task learning is performed on the defect dynamic features to obtain the probability distribution of defect locations and defect type. Multi-task learning improves the generalization ability of the model by optimizing multiple objective functions at the same time. According to the probability distribution of defect location and defect type, the threshold is judged to obtain the defect location and defect type of the pipeline. A threshold is set. When the probability of a certain category in the probability distribution exceeds the threshold, it is considered that there is a defect at that location and the defect type is determined.

[0032] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Encode the defect location and defect type to obtain defect feature information, and retrieve the preset defect knowledge graph based on the defect feature information to obtain the corresponding defect node set; (2) Construct a defect causal relationship graph based on the defect node set, and perform path importance analysis based on the defect causal relationship graph to obtain the key causal chain; (3) Reversely trace the key causal chain to obtain a candidate set of defect root causes, and perform Bayesian network reasoning based on the candidate set of defect root causes to generate an initial set of repair solutions; (4) Perform multi-criteria decision analysis based on the initial repair plan set to obtain a repair plan scoring table, and perform fuzzy comprehensive evaluation on the repair plan scoring table to obtain the target pipeline repair plan.

[0033] Specifically, the defect location and defect type are converted into a processable encoding format. For example, one-hot encoding or embedded encoding is used for encoding. According to the defect feature information, the preset defect knowledge graph is retrieved to obtain the corresponding defect node set. The defect knowledge graph is a structured knowledge representation that contains various defect types and their attributes, relationships and other information. Through retrieval, a defect node set matching the encoded defect feature information can be found. A defect causal relationship graph is constructed according to the defect node set, and path importance analysis is performed on the defect causal relationship graph to obtain a key causal chain. The defect causal relationship graph represents the causal relationship between defects, and the cause and impact of the defect can be analyzed through the graph structure. Path importance analysis can find the most critical causal chain by calculating the weight of the path. The key causal chain is traced back to obtain a defect root cause candidate set, and Bayesian network reasoning is performed based on the defect root cause candidate set to generate an initial repair solution set. Reverse tracing is to find the possible defect root cause by tracing back from the defect result node to the cause node. Let the key causal chain be k, and the reverse tracing process can be expressed as starting from the end node of k and gradually tracing back to the starting node to obtain the defect root cause candidate set R. A Bayesian network is constructed based on the defect root cause candidate set, and an initial set of repair solutions is generated through Bayesian reasoning. A Bayesian network is a directed acyclic graph, where nodes represent random variables and edges represent conditional dependencies between variables. Let the Bayesian network be B ,in Represents conditional dependencies, initial repair solution set Through Bayesian reasoning, we can get: ; in, Indicates defect feature information Next repair plan The probability of It is a repair plan Defect feature information is observed below The probability of It is a repair solution The prior probability of Defect feature information According to the initial repair solution set Perform multi-criteria decision analysis to obtain a repair plan scoring table, and perform fuzzy comprehensive evaluation on the repair plan scoring table to obtain the target pipeline repair plan. Multi-criteria decision analysis is to comprehensively evaluate and rank the repair plans by considering multiple evaluation criteria. Suppose the repair plan set is , the evaluation criteria set is , each repair solution In the guidelines The ratings below are The repair scheme scoring table can be expressed as a scoring matrix , whose elements Indicates the repair plan In the guidelines Fuzzy comprehensive evaluation is a method based on fuzzy mathematics, which comprehensively evaluates the score through fuzzy transformation and weighted summation. Let the fuzzy weight vector be , represents the weight of each criterion, and the fuzzy comprehensive evaluation result It can be expressed as: .in, is the comprehensive score vector of the repair scheme. Finally, the scheme with the highest comprehensive score is selected as the target pipeline repair scheme.

[0034] The above describes the pipeline repair method based on defect detection in the embodiment of the present application. The following describes the pipeline repair system based on defect detection in the embodiment of the present application. Figure 2 In the embodiments of the present application, an embodiment of a pipeline repair system based on defect detection includes: The acoustic detection module 201 is used to perform acoustic detection on the pipeline based on the multi-frequency sound wave signal and adaptively sample and phase synchronize the response signal of the pipeline to obtain pipeline sound field data; The electromagnetic detection module 202 is used to perform electromagnetic wave scattering detection on the pipeline and perform dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; A construction module 203 is used to construct an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and to construct an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; A feature extraction module 204 is used to extract features from the sound field distribution map to obtain a first detection feature set, and to extract features from the electromagnetic characteristic distribution map to obtain a second detection feature set; The diagnosis module 205 is used to input the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain a defect location and a defect type; The matching module 206 is used to perform causal reasoning and repair scheme matching on the pipeline according to the defect location and defect type to obtain a repair scheme for the target pipeline.

[0035] Through the synergy of the above components, by combining multi-frequency acoustic signal detection and electromagnetic wave scattering detection, the comprehensive analysis of acoustic field and electromagnetic characteristics is realized. Adaptive sampling and phase synchronization technology are used to process pipeline response signals, which improves the quality and reliability of original data. Through the construction of acoustic field distribution map and electromagnetic characteristic distribution map, the visualization of pipeline internal and surface characteristics is realized. Multi-angle feature extraction is performed on the acoustic field and electromagnetic characteristic distribution map. The spatiotemporal graph convolution network model is introduced for defect diagnosis, which effectively utilizes the spatiotemporal correlation of pipeline defects and improves the accuracy and robustness of diagnosis. Through time series correlation analysis and multi-task learning, in-depth mining of dynamic characteristics of defects is realized, which helps to identify complex and evolving defects. Based on defect knowledge graph and Bayesian network reasoning, intelligent tracing and analysis of defect root causes are realized. Multi-criteria decision analysis and fuzzy comprehensive evaluation methods are adopted to ensure the comprehensiveness and optimality of repair solutions and improve the repair effect. By establishing a mapping relationship between defect characteristics and repair solutions, automatic matching of repair solutions is realized, the efficiency of pipeline maintenance is improved, and intelligent management of the entire life cycle of pipelines is realized.

[0036] The present application also provides a pipeline repair device based on defect detection, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the pipeline repair method based on defect detection in the above-mentioned embodiments.

[0037] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the pipeline repair method based on defect detection.

[0038] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0039] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0040] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A pipeline repair method based on defect detection, characterized in that: The method comprises: Acoustically detecting the pipeline based on the multi-frequency sound wave signal and adaptively sampling and phase synchronizing the response signal of the pipeline to obtain pipeline sound field data; Performing electromagnetic wave scattering detection on the pipeline and performing dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; Constructing an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and constructing an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; Performing feature extraction on the acoustic field distribution diagram to obtain a first detection feature set, and performing feature extraction on the electromagnetic characteristic distribution diagram to obtain a second detection feature set; Inputting the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain defect locations and defect types; Causal reasoning and repair scheme matching are performed on the pipeline according to the defect location and the defect type to obtain a target pipeline repair scheme.

2. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The method of performing acoustic detection on a pipeline based on a multi-frequency sound wave signal and adaptively sampling and phase synchronizing a response signal of the pipeline to obtain pipeline sound field data includes: Discretize the sound wave signal within a preset frequency range to obtain multiple discrete frequency points; Generate a plurality of pseudo-random code sequences corresponding to the frequencies according to the discrete frequency points, and perform orthogonal modulation on the pseudo-random code sequences corresponding to the plurality of frequencies to obtain a multi-frequency orthogonal modulation signal; The piezoelectric transducer array is excited according to the multi-frequency orthogonal modulation signal to obtain a multi-frequency acoustic wave excitation signal, and a time-domain finite difference simulation is performed on the propagation of the multi-frequency acoustic wave excitation signal in the pipeline to obtain a prediction model of the acoustic field distribution in the pipeline; The response signal of the pipeline is collected according to the prediction model of the sound field distribution in the pipeline to obtain original response signal data; Dynamically adjusting the sampling rate according to the original response signal data to obtain an adaptive sampling signal, and performing wavelet denoising and phase correction on the adaptive sampling signal to obtain a noise-suppressed synchronization signal; The noise-suppressed synchronization signal is subjected to time-frequency analysis and spatial interpolation to obtain pipeline sound field data.

3. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The electromagnetic wave scattering detection is performed on the pipeline and the electromagnetic waves of the pipeline are dynamically frequency tuned and coherently detected to obtain electromagnetic scattering data, including: Discretely sampling a preset terahertz frequency range to generate a frequency scanning sequence, and phase-encoding the frequency scanning sequence to obtain a phase-encoded terahertz pulse sequence; driving the terahertz wave transmitter according to the phase-encoded terahertz pulse sequence to obtain an incident electromagnetic wave; Performing finite element simulation on the interaction between the incident electromagnetic wave and the pipeline surface to obtain a scattering field distribution prediction model, and collecting the scattered electromagnetic wave according to the scattering field distribution prediction model to obtain an original scattering signal; Adjusting the frequency of a local oscillator in real time according to the original scattered signal to obtain a dynamic frequency tuning signal, and coherently demodulating the dynamic frequency tuning signal and a reference signal to obtain a demodulated complex signal; The electromagnetic scattering data is obtained by performing spatial Fourier transformation and back-projection reconstruction on the demodulated complex signal.

4. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The step of constructing a sound field distribution map inside the pipeline according to the pipeline sound field data, and constructing an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data, comprises: Performing a time-domain Fourier transform on the pipeline sound field data to obtain frequency-domain sound field data, and solving the sound wave propagation equation according to the frequency-domain sound field data to obtain the sound pressure distribution inside the pipeline; Performing spatial interpolation and smoothing processing on the sound pressure distribution to obtain an initial sound field distribution map, and iteratively optimizing boundary conditions according to the initial sound field distribution map to obtain a sound field distribution map inside the pipeline; Performing polarization decomposition on the electromagnetic scattering data to obtain a scattering matrix, and solving an inverse scattering problem based on the scattering matrix to obtain a dielectric constant distribution on the pipeline surface; The dielectric constant distribution is spatially filtered and denoised to obtain an initial electromagnetic distribution map, and the material parameters are constrained optimized and registered according to the initial electromagnetic distribution map to obtain an electromagnetic characteristic distribution map of the pipeline surface.

5. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The extracting features from the acoustic field distribution diagram to obtain a first detection feature set, and extracting features from the electromagnetic characteristic distribution diagram to obtain a second detection feature set, comprises: Performing wavelet transform on the sound field distribution map to obtain a multi-scale sound field coefficient matrix, and calculating local energy distribution according to the multi-scale sound field coefficient matrix to obtain a sound field energy characteristic map; Performing a local binary pattern transform on the sound field energy feature map to obtain a sound field texture feature, and calculating a gray level co-occurrence matrix according to the sound field texture feature to obtain a sound field statistical feature; Performing principal component analysis on the sound field statistical features to obtain a first detection feature set; Performing Gabor filtering on the electromagnetic characteristic distribution map to obtain multi-directional electromagnetic characteristic responses, and calculating a directional gradient histogram based on the multi-directional electromagnetic characteristic responses to obtain electromagnetic characteristic directional features; Performing Fourier descriptor transformation on the electromagnetic characteristic directional feature to obtain the electromagnetic characteristic frequency domain feature, and calculating the statistical moment according to the electromagnetic characteristic frequency domain feature to obtain the electromagnetic characteristic statistical feature; Independent component analysis is performed on the statistical features of the electromagnetic characteristics to obtain a second detection feature set.

6. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The step of inputting the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain a defect location and a defect type includes: Performing feature fusion on the first detection feature set and the second detection feature set to obtain a multimodal feature matrix; Inputting the multimodal feature matrix into a preset spatiotemporal graph convolutional network model to construct a spatiotemporal graph structure, thereby obtaining a spatiotemporal graph of pipeline defects; Performing a graph convolution operation on the pipeline defect spatiotemporal graph to obtain a graph feature representation, and performing multi-scale feature extraction based on the graph feature representation to obtain a multi-scale feature vector; Performing attention mechanism weighting on the multi-scale feature vector to obtain a weighted feature vector, and performing nonlinear transformation on the weighted feature vector to obtain a defect representation vector; Performing a time series correlation analysis on the defect characterization vector to obtain a time series feature sequence, and generating a defect dynamic feature according to the time series feature sequence; Performing multi-task learning on the dynamic features of the defects to obtain a probability distribution of defect locations and a probability distribution of defect types; A threshold value is judged according to the defect position probability distribution and the defect type probability distribution to obtain the defect position and defect type of the pipeline.

7. The pipeline repair method based on defect detection according to claim 1 is characterized in that: The performing causal reasoning and repair scheme matching on the pipeline according to the defect location and the defect type to obtain a target pipeline repair scheme includes: Encoding the defect location and the defect type to obtain defect feature information, and searching a preset defect knowledge graph according to the defect feature information to obtain a corresponding defect node set; Constructing a defect causal relationship graph according to the defect node set, and performing path importance analysis according to the defect causal relationship graph to obtain a key causal chain; The key causal chain is traced back to obtain a candidate set of defect root causes, and Bayesian network reasoning is performed based on the candidate set of defect root causes to generate an initial repair solution set; A multi-criteria decision analysis is performed on the initial repair solution set to obtain a repair solution scoring table, and a fuzzy comprehensive evaluation is performed on the repair solution scoring table to obtain a target pipeline repair solution.

8. A pipeline repair system based on defect detection, characterized in that: For executing the pipeline repair method based on defect detection as described in any one of claims 1 to 7, the system comprises: An acoustic detection module, used to perform acoustic detection on the pipeline based on a multi-frequency sound wave signal and to perform adaptive sampling and phase synchronization on the response signal of the pipeline to obtain pipeline sound field data; An electromagnetic detection module, used to perform electromagnetic wave scattering detection on the pipeline and to perform dynamic frequency tuning and coherent detection on the electromagnetic waves of the pipeline to obtain electromagnetic scattering data; A construction module, used to construct an acoustic field distribution map inside the pipeline according to the pipeline acoustic field data, and to construct an electromagnetic characteristic distribution map on the pipeline surface according to the electromagnetic scattering data; A feature extraction module, used to perform feature extraction on the acoustic field distribution diagram to obtain a first detection feature set, and to perform feature extraction on the electromagnetic characteristic distribution diagram to obtain a second detection feature set; A diagnosis module, used for inputting the first detection feature set and the second detection feature set into a preset spatiotemporal graph convolutional network model to perform pipeline defect diagnosis to obtain a defect location and a defect type; A matching module is used to perform causal reasoning and repair scheme matching on the pipeline according to the defect location and the defect type to obtain a repair scheme for the target pipeline.

9. A pipeline repair device based on defect detection, characterized in that: The pipeline repair device based on defect detection includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the pipeline repair device based on defect detection to execute the pipeline repair method based on defect detection according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, a pipeline repair method based on defect detection as described in any one of claims 1 to 7 is implemented.

Images