A pipeline defect diagnosis method based on guided wave technology

Abstract

The present application relates to the technical field of pipeline nondestructive testing, and discloses a pipeline defect diagnosis method based on guided wave technology. The method obtains original guided wave response signals under multiple excitation frequencies through a guided wave sensor array arranged on the outer wall of the pipeline, performs multi-stage analysis and separation on the signals, extracts a set of independent wave mode components directly related to the defects, including longitudinal wave components, torsional wave components and surface wave components. According to the propagation time delay and energy distribution of each component, a multi-dimensional defect feature tensor is constructed, input into a trained pipeline defect discrimination network for feature mapping and abstraction, and a preliminary diagnosis report of the defect category and geometric profile is output. Combined with the pipeline service environment parameters, the physical field coupling correction is carried out to generate the final diagnosis result. The method solves the problems of serious mode aliasing and poor environmental adaptability in the traditional guided wave technology, and realizes accurate identification and positioning of defects.

Description

Technical Field

[0001] This invention relates to the field of pipeline non-destructive testing technology, specifically a pipeline defect diagnosis method based on guided wave technology. Background Technology

[0002] Guided wave testing technology has become an important tool in the field of pipeline non-destructive testing due to its efficient long-distance pipeline scanning capability. Existing conventional techniques mainly rely on exciting and receiving a single dominant guided wave mode at a specific frequency, and inferring the presence and approximate location of defects by analyzing the reflection, transmission, or attenuation characteristics of this mode signal. This method, in principle, utilizes the mode conversion and energy scattering phenomena that occur when guided wave propagation encounters structural discontinuities.

[0003] Existing technical solutions have shortcomings. The guided wave signals actually propagating in pipelines are essentially complex aliasing of multiple modes, including longitudinal, torsional, and bending modes. Conventional single-mode analysis methods actively discard defect information carried by other modes, resulting in insufficient sensitivity to defects in certain directions or types, and a limited diagnostic dimension. Furthermore, the propagation characteristics of guided waves are strongly dependent on the physical environment of the pipeline. Existing methods lack a mechanism to systematically incorporate these environmental parameters into the diagnostic model, causing diagnostic algorithms based on laboratory calibration data to have poor generalization ability under real-world, variable operating conditions, leading to decreased reliability.

[0004] A diagnostic method is needed to overcome the aforementioned limitations. The core issues to be addressed are: how to effectively analyze and separate different guided wave mode components sensitive to different defects from complex raw signals in order to make full use of multi-mode information; and how to construct an intelligent diagnostic model that can integrate multi-dimensional defect features and adaptively correct environmental interference to achieve high-precision transfer from the laboratory to the engineering field. Summary of the Invention

[0005] The purpose of this invention is to provide a pipeline defect diagnosis method based on guided wave technology to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a pipeline defect diagnosis method based on guided wave technology, the method comprising: By using an array of guided wave sensors installed on the outer wall of the pipe, the original guided wave response signals of the pipe structure at multiple excitation frequencies are obtained; Multi-level signal analysis and separation operations are performed on the original guided wave response signal to extract the set of independent wave mode components directly related to the pipeline defect. The set of independent wave mode components includes longitudinal guided wave components propagating along the pipeline axis, torsional guided wave components distributed along the pipeline circumferentially, and surface guided wave components attenuating along the pipeline radially. Based on the propagation delay and energy distribution of each component in the set of independent wave mode components, a multidimensional defect feature tensor is constructed to characterize the spatial features of the defect. The multi-dimensional defect feature tensor is input into a trained pipeline defect discrimination network for layer-by-layer feature mapping and abstraction, and the output is a preliminary diagnostic report containing defect category labels and defect geometric contours. The preliminary diagnostic report is subjected to physical field coupling correction based on pipeline service environment parameters to generate the final diagnostic result.

[0007] Preferably, the step of performing multi-level signal analysis and separation operations on the original guided wave response signal to extract the set of independent wave mode components directly related to the pipeline defect includes: The original guided wave response signal is decomposed in the time-frequency domain based on wavelet packet transform to obtain a set of sub-signals covering different frequency bands; For each sub-signal, calculate its coherence coefficient with a preset standard guided wave mode template, and identify and separate the noise components and multipath reflection components mixed in the sub-signal based on the strength of the coherence coefficient. The denoised sub-signals are processed using blind source separation technology. Based on the principle of signal statistical independence, the time-domain waveforms of the longitudinal guided wave component, the torsional guided wave component, and the surface guided wave component are iteratively decoupled. Envelope detection and phase synchronization analysis are performed on each decoupled time-domain waveform to extract the complete parameters of arrival time, group velocity, and energy decay curve of each wave mode component, and these parameters are combined to form the set of independent wave mode components.

[0008] Preferably, the step of constructing a multi-dimensional defect feature tensor to characterize the spatial features of defects based on the propagation delay and energy distribution of each component in the set of independent wave mode components includes: Based on the arrival time sequence of the longitudinal guided wave component, the wave velocity disturbance value at different positions along the axial direction of the pipeline is calculated, and an axial wave velocity anomaly distribution map in the first dimension is generated based on the wave velocity disturbance value. Analyze the energy attenuation curve of the torsional guided wave component and, in conjunction with its circumferential propagation path length, calculate the circumferential energy loss rate spectrum of the pipeline, which serves as the second dimension of the circumferential energy loss spectrum. The attenuation characteristics of the surface guided wave components are quantified, and based on their differences in penetration ability in the radial depth, the equivalent scattering intensity profile in the pipe wall thickness direction is derived, thus forming a third-dimensional radial scattering intensity profile. The axial wave velocity anomaly distribution map, the circumferential energy defect map, and the radial scattering intensity profile are aligned and superimposed on spatial coordinates to form a multidimensional data volume with a spatial grid structure, namely the multidimensional defect feature tensor.

[0009] Preferably, the step of inputting the multi-dimensional defect feature tensor into a trained pipeline defect discrimination network for layer-by-layer feature mapping and abstraction includes: The first feature extraction layer of the pipeline defect discrimination network receives the multi-dimensional defect feature tensor and extracts the local spatial pattern features of the multi-dimensional data volume by scanning with a three-dimensional convolutional kernel to generate a primary feature map. The second feature abstraction layer of the pipeline defect discrimination network performs pooling and nonlinear transformation on the primary feature map, compressing the data dimension while enhancing the rotation and translation invariance of the features, and generating intermediate abstract features. The third classification decision layer of the pipeline defect discrimination network performs a fully connected operation on the intermediate abstract features, maps the high-dimensional features to a preset defect category space, and outputs the defect category label. In parallel, the fourth contour generation layer of the pipeline defect discrimination network reconstructs the three-dimensional geometric shape of the defect in the pipeline space step by step through deconvolution operation based on the intermediate abstract features, and outputs the geometric contour of the defect.

[0010] Preferably, the physical field coupling correction performed on the preliminary diagnostic report based on pipeline service environment parameters includes: The pressure fluctuation data, temperature gradient distribution data, and external soil stress data of the medium inside the pipeline are obtained, which together constitute the service environment parameters of the pipeline. The defect geometry profile is coupled with the medium pressure fluctuation data for analysis to calculate the stress concentration factor at the defect edge under dynamic pressure load, and the boundary sharpness of the defect geometry profile is adjusted accordingly. The temperature gradient distribution data is correlated with the circumferential energy loss map to correct the influence of material sound velocity changes caused by temperature inhomogeneity on the energy loss rate calculation, and a temperature-compensated circumferential energy loss map is generated. By combining the soil stress data and the axial wave velocity anomaly distribution map, the additional effect of external mechanical constraints on wave velocity disturbance is evaluated, and stress field correction is performed on the wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map.

[0011] Preferably, the calculation of the stress concentration factor at the defect edge under dynamic pressure load, and the adjustment of the boundary sharpness of the defect geometry accordingly, includes: A finite element model of the pipeline containing the geometric contour of the defect is established, and the pressure fluctuation data of the medium is applied as a time-varying boundary condition to the inner wall of the model. The equivalent stress cloud map of the defect region during the complete pressure cycle is calculated by finite element method, and the curve of the maximum principal stress at the defect edge changing with time is extracted from the equivalent stress cloud map. Based on the maximum principal stress variation curve and the fatigue strength limit of the pipeline material, the value of the stress concentration factor is quantified. Based on the value of the stress concentration factor, the boundary nodes with large curvature in the defect geometric profile are subjected to smoothing iteration. The goal of the iteration is to match the local curvature of the boundary nodes with the calculated local stress gradient, thereby outputting the defect geometric profile with adjusted boundary sharpness.

[0012] Preferably, the correction for the influence of material sound velocity variation caused by temperature unevenness on the energy loss rate calculation includes: Based on the temperature gradient distribution data, a lookup table is established to show the correspondence between temperature and material sound velocity in each region of the pipe circumference. The original energy loss rate value is read from the circumferential energy loss map, and the corresponding local temperature value is obtained from the temperature gradient distribution data according to the signal propagation path. Using the aforementioned lookup table, the sound speed correction coefficient corresponding to the local temperature value is found, and the original energy loss rate value is recalibrated using the sound speed correction coefficient. The recalibrated energy loss rate values ​​are filled back into the corresponding positions of the circumferential energy deficit map to generate the temperature-compensated circumferential energy deficit map.

[0013] Preferably, the stress field correction of the wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map includes: The soil stress data is decomposed into components acting on the pipe axis and components perpendicular to the axis. An empirical correlation model between wave velocity disturbance and axial stress component is constructed, and the wave velocity reference offset caused by axial stress component in the soil stress data is calculated using the empirical correlation model. Subtract the wave velocity reference offset from the original wave velocity disturbance value of the axial wave velocity anomaly distribution map to eliminate the systematic deviation caused by external axial stress. Considering the micro-deformation of the pipe cross section caused by the stress component perpendicular to the axial direction, a cross section stiffness factor is introduced to perform a secondary fine-tuning of the wave velocity disturbance value after eliminating the systematic deviation caused by the external axial stress, so as to obtain the final wave velocity disturbance value after stress field correction and update the axial wave velocity anomaly distribution map.

[0014] Preferably, the calculation of the coherence coefficient between it and the preset standard guided wave mode template includes: Each sub-signal is preprocessed with zero mean and then windowed using the Hanning window function to suppress spectral leakage. The windowed sub-signal is cross-correlated with the preset standard guided wave mode templates stored in the database, which correspond to the same excitation frequency and pipe specifications, to obtain a cross-correlation sequence; Calculate the maximum value of the cross-correlation coefficient sequence, normalize it, and use it as the coherence coefficient between the windowed sub-signal and the corresponding standard guided wave mode template; A coherence coefficient threshold is set, and sub-signal components with coherence coefficients lower than the threshold are marked as noise components and multipath reflection components, and are removed from subsequent processing.

[0015] Preferably, the step of calculating wave velocity disturbance values ​​at different positions along the pipe axis based on the arrival time series of the longitudinal guided wave components, and generating an axial wave velocity anomaly distribution map in the first dimension based on the wave velocity disturbance values, includes: The arrival time sequence is formed by extracting the arrival time of a series of known discrete detection points along the pipe axis from the complete parameters of the longitudinal guided wave component. Based on the known axial distance between adjacent detection points and the corresponding arrival time difference, the measured longitudinal guided wave propagation group velocity of the section between the adjacent detection points is calculated. Obtain the longitudinal guided wave reference group velocity of the corresponding pipe segment under defect-free conditions; Calculate the percentage of the relative deviation between the measured group velocity and the reference group velocity, and define the percentage value as the wave velocity disturbance value at the corresponding axial position; Using the axial position of the pipeline as the abscissa and the calculated wave velocity disturbance value as the ordinate, data interpolation and gridding are performed to generate a continuous axial wave velocity anomaly distribution map that reflects the spatial distribution.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By performing multi-level signal analysis and separation operations on the original guided wave response signal, a set of independent wave mode components directly related to pipeline defects is extracted. This set includes longitudinal guided wave components, torsional guided wave components, and surface guided wave components. This technique overcomes the limitations of traditional methods that rely on single-mode signals. When guided waves of different modes interact with pipeline defects during propagation, their scattering characteristics, mode conversions, and energy attenuation patterns differ. The separated independent mode components can reveal the characteristic information of the defect from the axial, circumferential, and radial dimensions, making the defect characterization information obtained from the aliased signal more comprehensive and refined. Joint analysis using multi-mode information enhances the physical interpretability of signal characteristics, laying a more reliable signal foundation for subsequent defect quantification and classification.

[0017] Based on the propagation delay and energy distribution of each independent wave mode component, a multi-dimensional defect feature tensor is constructed. This tensor organizes and fuses the time-domain and energy-domain features of different modes within a unified high-dimensional data structure, thus forming a composite description of the defect's spatial morphology. This feature tensor is input into a trained pipeline defect discrimination network, and the network's nonlinear mapping capability learns the complex correlation between defect features, types, and contours. Physical field coupling correction is performed on the network's preliminary diagnostic results based on pipeline service environment parameters. The correction process uses environmental variables such as temperature and pressure as inputs to compensate and adjust the feature tensor or network output, establishing a dynamic mapping relationship between the diagnostic model and the actual physical environment. This reduces the interference of environmental factors on guided wave propagation characteristics, ensuring the robustness and accuracy of the diagnostic system's output under actual operating conditions. The feature tensor provides a structured high-dimensional input, while physical field coupling correction guarantees the model's stable performance transfer from ideal conditions to complex field environments. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the pipeline defect diagnosis method based on guided wave technology described in this invention. Figure 2 A flowchart for constructing a multi-dimensional defect feature tensor; Figure 3 A flowchart for feature mapping and abstraction of pipeline defect discrimination network; Figure 4 A heat map showing the spatial distribution of the multidimensional defect feature tensor (axial × circumferential) of the pipeline; Figure 5 This is a diagram showing the dynamic response of medium pressure load and principal stress at the defect edge. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1This invention provides a pipeline defect diagnosis method based on guided wave technology. The method includes: using a guided wave sensor array arranged circumferentially on the outer wall of the pipeline to acquire raw guided wave response signals at multiple excitation frequencies; performing multi-level analysis and separation on the signals to extract independent wave mode sets of longitudinal, torsional, and surface guided wave components; constructing a multi-dimensional defect feature tensor based on the propagation delay and energy distribution of each component, inputting it into a pre-trained pipeline defect discrimination network, and outputting a preliminary diagnostic report of defect category and geometric contour; finally, combining pipeline service environment parameters for physical field coupling correction to generate the final diagnostic result. The sensor array needs to cover the circumferential range of the pipeline, the excitation frequency range is 10-100kHz, and the signal sampling rate is not less than 1MHz to ensure mode separation accuracy.

[0021] In one embodiment of the present invention, the original guided wave response signal is decomposed in the time-frequency domain using wavelet packet transform to obtain a set of sub-signals covering different frequency bands. For each sub-signal, after windowing and zero-mean normalization using a Hanning window function, cross-correlation is performed with a preset standard guided wave mode template, and the maximum value of the normalized cross-correlation coefficient sequence is calculated as the coherence coefficient. A threshold of 0.85 is set, and components with coherence coefficients below the threshold are marked as noise or multipath reflection components and removed. Blind source separation technology is used to iteratively decouple the denoised sub-signals, and the time-domain waveforms of the longitudinal, torsional, and surface guided wave components are separated based on the principle of statistical independence. Envelope detection and phase synchronization analysis are performed on the decoupled waveforms to extract the arrival time, group velocity, and energy attenuation curve parameters of each wave mode, which are then combined to form an independent wave mode component set.

[0022] In the specific implementation, wavelet packet transform is used to perform time-frequency domain decomposition on the original guided wave response signal. Specifically, the DB8 wavelet basis function is selected to perform 5-level complete wavelet packet decomposition on the original guided wave response signal acquired by each channel in the sensor array, thereby obtaining a set of 32 time-frequency sub-signals covering the low-frequency sub-band to the high-frequency sub-band.

[0023] In some embodiments, for each time-frequency sub-signal obtained after decomposition, its coherence coefficient with a preset standard guided wave mode template needs to be calculated. The preset standard guided wave mode templates are stored in a database, and each template corresponds to a standard guided wave mode response for an excitation frequency, pipe specification, and defect-free pipe section. In a specific implementation, each time-frequency sub-signal to be processed first undergoes zero-mean preprocessing to eliminate the DC bias component in the signal. Subsequently, a Hanning window function is used to window the zero-mean sub-signal, with the window function length being the length of the sub-signal. The windowing operation aims to suppress spectral leakage in subsequent spectral analysis. The specific form of the Hanning window function is given by the following formula:

[0024] in: This indicates the window function at discrete time points. The value at that location, This represents the total length of the window function. The value range is from 0 to The windowed time-frequency sub-signal is cross-correlated with a preset standard guided wave mode template retrieved from the database, matching its excitation frequency and pipe specifications, to obtain a cross-correlation coefficient sequence. The absolute maximum value of this cross-correlation coefficient sequence is calculated and normalized. The normalization factor is the square root of the product of the zero-hysteresis values ​​of the autocorrelation functions of the two signals. The normalized value is defined as the coherence coefficient between the time-frequency sub-signal and the corresponding preset standard guided wave mode template. A coherence coefficient threshold of 0.85 is set. Time-frequency sub-signal components with coherence coefficients below 0.85 are marked as noise components or multipath reflection components, and these marked components are removed from subsequent signal processing.

[0025] It is understandable that after identifying and removing noise and multipath reflection components, blind source separation technology is used to process the remaining denoised time-frequency sub-signal set. Blind source separation technology is based on the principle of signal statistical independence, assuming that the source signals mixed in the observed signal are statistically independent. In specific implementation, a fixed-point iterative algorithm based on maximizing negative entropy is used to iteratively decouple the denoised multi-channel time-frequency sub-signal data. Specifically, the remaining denoised time-frequency sub-signal set is first constructed into a multi-channel observation data matrix. The matrix has rows equal to the number of sub-signals and columns equal to the number of time sampling points for each sub-signal. The input observation signal is used for blind source separation. Then, the... Centering is performed by subtracting the mean of each row, making the mean of the observed signal for each channel zero. Then, the centered data is whitened, and the eigenvalue decomposition of the covariance matrix is ​​calculated using principal component analysis to obtain the whitening matrix. This makes the whitened data The components are uncorrelated and have unit variance. Subsequently, a fixed-point iterative algorithm is used to estimate the unmixing matrix one by one. Each row vector Each row vector corresponds to the extraction direction of an independent source component. Random initialization is performed at the start of the iteration. It is a unit vector, and then updated according to the following formula:

[0026] in: This represents the updated row vector; It represents the expected value of a mathematical operation, which is approximated by the sample mean in actual calculations; Represents the whitened multi-channel data matrix A column vector in the vector, representing the observations of all channels at a given moment; This represents the transpose of the current row vector; It is a nonlinear function; the hyperbolic tangent function is chosen here. ; It is the derivative of the nonlinear function, that is... After each iteration, for Perform orthogonalization and normalization to ensure that different row vectors are mutually orthogonal and have a magnitude of 1. Repeat the above iterative process until... If the change in the value is less than the preset convergence tolerance, the row vector is considered to have converged. Repeat the above process for each independent source component to be extracted to obtain the complete unmixing matrix. Ultimately, through Calculate the separated source signal matrix ,matrix The rows correspond to the time-domain waveform estimates of the longitudinal guided wave component, torsional guided wave component, and surface guided wave component, respectively. These separated time-domain waveforms are the input data for subsequent envelope detection and phase synchronization analysis.

[0027] Each iteration adjusts the unmixing matrix to maximize the sum of the negative entropy of the output components. Through multiple iterations, the unmixing matrix converges to a stable state, thereby separating statistically independent source signal estimates from the mixed observation signals. These source signal estimates are the approximate waveforms of the longitudinal guided wave component, the torsional guided wave component, and the surface guided wave component in the time domain.

[0028] Optionally, after decoupling the time-domain waveforms of each wave mode component using blind source separation technology, these time-domain waveforms need to be further analyzed to extract complete parameters. Envelope detection is performed on the time-domain waveform of each decoupled wave mode component, and the analytic signal of the waveform is calculated using the Hilbert transform method. The modulus of the analytic signal is taken as the envelope of the waveform. Simultaneously, phase synchronization analysis is performed, and the arrival time of the waveform is determined by calculating the instantaneous phase of the analytic signal. Combining the known spatial location of the sensor array and the signal propagation path, the group velocity of each wave mode is calculated based on the arrival time difference and propagation distance. Furthermore, the attenuation trend of the wave mode component envelope is analyzed, and an energy attenuation curve is fitted. The arrival time, group velocity, and energy attenuation curve of each extracted wave mode component together constitute the complete parameters of that component. The complete parameters of all components are combined to ultimately form a set of independent wave mode components, including longitudinal guided wave components, torsional guided wave components, and surface guided wave components.

[0029] In one embodiment of the present invention, see [reference] Figure 2 Based on the arrival time series of the longitudinal guided wave component, the measured group velocity between discrete detection points along the pipeline axis is calculated. Using the reference group velocity of a defect-free pipe section as a benchmark, the percentage deviation of wave velocity disturbance values ​​at each location is calculated. An axial wave velocity anomaly distribution map is generated using interpolation. The energy attenuation curve of the torsional guided wave component is analyzed, and the energy loss rate spectrum is calculated in conjunction with the circumferential propagation path length. The radial attenuation characteristics of the surface guided wave component are quantified, and the equivalent scattering intensity profile in the wall thickness direction is derived. The above three sets of data are aligned and superimposed on a spatial grid to form a defect feature tensor containing axial, circumferential, and radial dimensions.

[0030] In practical implementation, the execution process starts with the complete parameters of the longitudinal guided wave component, torsional guided wave component, and surface guided wave component contained in the independent wave mode component set. Specifically, the complete parameters of the longitudinal guided wave component include its arrival time at a series of known discrete detection points along the pipe axis. These discrete detection points correspond to the installation positions of sensors in the guided wave sensor array, and an arrival time sequence of the longitudinal guided wave component is formed based on these arrival times. The measured longitudinal guided wave propagation group velocity of the section between adjacent discrete detection points is calculated based on the known axial distance between adjacent discrete detection points and the corresponding arrival time difference extracted from the arrival time sequence. The longitudinal guided wave reference group velocity of the corresponding pipe section of the same specification pipe under defect-free conditions is obtained from the database. The percentage deviation of the measured group velocity relative to the reference group velocity is calculated, and this percentage value is defined as the wave velocity disturbance value at the corresponding axial position. Using the pipe axial position as the abscissa and the calculated wave velocity disturbance value as the ordinate, cubic spline interpolation is used to interpolate and mesh the discrete wave velocity disturbance values, generating a continuous axial wave velocity anomaly distribution map reflecting the spatial distribution in the first dimension.

[0031] In some embodiments, processing is performed based on the complete parameters of the torsional guided wave component, which include its energy attenuation curve. The trend of the energy attenuation curve is analyzed, and combined with the known path length of the torsional guided wave propagating circumferentially along the pipe, the energy loss rate at different angular positions along the circumferential direction of the pipe is calculated. The formula for calculating the energy loss rate R is:

[0032] in: Indicates the position of the circumferential angle. Energy loss rate at the location, Indicates the position of the circumferential angle. The initial energy of the incident torsional guided wave component. Indicates the position of the circumferential angle. The reflected energy of the torsional guided wave component received at the location, Indicates the position of the circumferential angle. The effective length of the guided wave propagation path.

[0033] It is understandable that the processing of surface guided wave components focuses on their radial attenuation characteristics, and the complete parameters of the surface guided wave components include information on their penetration capability in the radial depth. In specific implementations, the attenuation characteristics of the surface guided wave components are quantified, and based on the differences in energy attenuation rates of surface guided waves of different frequency components in the pipe wall thickness direction, the equivalent scattering intensity at a series of discrete depth points in the pipe wall thickness direction is derived. The equivalent scattering intensity profile constitutes a third-dimensional radial scattering intensity profile. Optionally, after generating the axial wave velocity anomaly distribution map, the circumferential energy deficiency map, and the radial scattering intensity profile, these three sets of data need to be aligned and superimposed in spatial coordinates to construct a multi-dimensional defect feature tensor. In specific implementations, the pipe axial coordinates, circumferential angular coordinates, and radial depth coordinates are unified into a three-dimensional spatial grid system. The axial wave velocity anomaly distribution map provides data on the axial-circumferential plane, the circumferential energy deficiency map is mapped to data on the circumferential-radial plane, and the radial scattering intensity profile provides data on the axial-radial plane. Using a spatial interpolation algorithm, these three sets of two-dimensional distributed data are filled into a common three-dimensional discrete grid data volume, which is the final multi-dimensional defect feature tensor.

[0034] In one embodiment of the present invention, see [reference] Figure 3The pipeline defect discrimination network's first feature extraction layer uses a 3D convolutional kernel to scan the defect feature tensor, extracting local spatial patterns to generate a primary feature map. The second feature abstraction layer compresses the data dimensionality through pooling and nonlinear transformations, enhancing feature rotation invariance to generate intermediate abstract features. The third classification decision layer performs fully connected operations on the intermediate features, mapping them to the defect category space to output labels. In parallel, the fourth contour generation layer reconstructs the 3D geometric contour of the defect through deconvolution. The network is trained using a labeled pipeline defect dataset, with the loss function combining cross-entropy and contour reconstruction error. The third classification decision layer of the pipeline defect discrimination network receives the intermediate abstract features output from the second feature abstraction layer. This intermediate abstract feature is a 3D feature map after pooling and nonlinear transformation; its spatial dimension is much smaller than the original input multi-dimensional defect feature tensor, but the number of channels is preserved or increased to encode richer semantic information. The third classification decision layer first performs global average pooling on this intermediate abstract feature map, calculating the average pixel value of all spatial locations for each channel's feature map, thereby compressing the entire 3D feature map into a one-dimensional feature vector whose length is equal to the number of channels in the feature map. Subsequently, the one-dimensional feature vector is input into a classification sub-network consisting of two fully connected layers. The first fully connected layer maps the dimension of the input vector to an intermediate hidden layer dimension, which is set to 128, and uses a modified linear unit as the activation function. The second fully connected layer further maps the 128-dimensional hidden feature vector to a vector space with a dimension equal to the preset total number of defect categories, including but not limited to corrosion thinning, cracks, dents, and weld anomalies. Finally, the Softmax activation function is applied to the output vector, transforming it into a probability distribution vector. The value of each element in the vector represents the probability that the input defect feature tensor belongs to the corresponding defect category. The category with the highest probability value is used as the defect category label output by the pipeline defect discrimination network.

[0035] In practical implementation, the first feature extraction layer of the pipeline defect discrimination network receives a multi-dimensional defect feature tensor as network input. This layer contains multiple three-dimensional convolutional kernels, which scan the multi-dimensional data volume of the defect feature tensor in spatial dimensions. The size of each three-dimensional convolutional kernel is set to 3x3x3. Through convolution operations, it extracts local spatial pattern features contained in the multi-dimensional data volume, generating a primary feature map containing multiple channels. During the convolution operation, the parameters of each three-dimensional convolutional kernel are learned during the training phase using a backpropagation algorithm, used to capture spatial correlation patterns between different combinations of axial wave velocity anomalies, circumferential energy deficiencies, and radial scattering intensity.

[0036] In some embodiments, the second feature abstraction layer of the pipeline defect discrimination network processes the primary feature map by performing pooling and nonlinear transformations. The pooling operation uses max pooling with a 2x2x2 pooling window and a stride of 2. This operation compresses the data dimension of the primary feature map and achieves feature downsampling. The nonlinear transformation uses a modified linear unit activation function, performing the operation on each pooled data point. The output is the original value when the input value is greater than 0, and 0 when the input value is less than or equal to 0. After pooling and nonlinear transformation, the primary feature map is transformed into a more robust intermediate abstract feature, which exhibits higher invariance to spatial translation and rotation changes of the input multi-dimensional defect feature tensor. The third classification decision layer of the pipeline defect discrimination network performs a classification task on the intermediate abstract feature. This third classification decision layer consists of a series of fully connected layers. The intermediate abstract feature is first flattened into a one-dimensional feature vector, which serves as the input to the fully connected layers. Fully connected layers map high-dimensional intermediate-level abstract features to a vector space with a dimension equal to the preset number of defect categories through matrix multiplication and bias addition. This operation can be represented as follows:

[0037] in: This represents the intermediate-level abstract feature vector of the input. This represents the weight matrix of the fully connected layer. This represents the bias vector. This represents the Softmax activation function. This represents the output class probability vector.

[0038] Optionally, the fourth contour generation layer of the pipeline defect discrimination network operates in parallel with the feature abstraction process. The fourth contour generation layer reconstructs the geometric contour of the defect based on intermediate-level abstract features. This layer consists of multiple deconvolutional layers, which upsample the intermediate-level abstract features using learned deconvolutional kernels, gradually increasing the spatial resolution of the feature maps. The upsampling process and the weighted combination of the convolutional kernels work together to decode and reconstruct the abstract defect information contained in the intermediate-level abstract features into a geometric representation of the defect in the three-dimensional space of the pipeline, ultimately outputting the defect's geometric contour.

[0039] In one embodiment of the present invention, pressure fluctuations, temperature gradient distributions, and external soil stress data of the pipeline's internal medium are acquired as service environment parameters. The defect geometry is coupled with the pressure fluctuation data, and the stress concentration factor at the defect edge is calculated using finite element analysis, thereby adjusting the sharpness of the profile boundary. Temperature gradient data is correlated with the circumferential energy loss map, and the energy loss rate is corrected using a sound velocity-temperature relationship lookup table. By combining soil stress data with the axial wave velocity anomaly distribution map, stress components are decomposed, and an empirical model is introduced to correct wave velocity disturbance values, eliminating systematic deviations caused by external stress.

[0040] In practice, pipeline service environment parameters include internal medium pressure fluctuation data, internal temperature gradient distribution data, and external soil stress data. These parameters are simultaneously measured and recorded using pressure sensor arrays installed on the pipeline system, distributed temperature sensing optical fibers, and earth pressure cells buried around the pipeline. Refer to Table 1 for the measurement methods, typical data ranges, and units of pipeline service environment parameters.

[0041] Table 1: Measurement Table of Pipeline Service Environment Parameters

[0042] In practice, the defect geometric contours contained in the preliminary diagnostic report are coupled with the medium pressure fluctuation data for analysis. The analysis process is completed by establishing a finite element model of the pipeline containing the defect geometric contours. The finite element model accurately reflects the shape, size, and material properties of the defect and the pipeline. In the specific implementation, when establishing the finite element model of the pipeline containing the defect geometric contours, firstly, the three-dimensional spatial data of the defect geometric contours are extracted from the preliminary diagnostic report, including the specific dimensions and shape parameters of the defect in the axial, circumferential, and radial directions of the pipeline. Based on the basic geometric specifications and material properties of the pipeline, a three-dimensional solid model of the pipeline is constructed using finite element analysis software. By embedding the discrete node data of the defect geometric contours, the geometric representation of the defect is accurately generated at the corresponding locations in the model. Adaptive mesh generation technology is used to discretize the pipeline model, and the mesh elements are densified in the defect region to ensure the accuracy of stress and deformation calculations. At the same time, the boundary conditions of the model are set to the actual support state of the pipeline to simulate real constraints. Finally, the medium pressure fluctuation data is used as a time-varying load and applied point by point to the inner wall surface of the model according to the time series, completing the initialization of the finite element model coupled with the physical field. The pressure fluctuation data of a complete cycle of medium obtained by real-time measurement or simulation is used as time-varying boundary conditions and sequentially applied to the inner wall surface of the pipeline finite element model. The equivalent stress distribution of the defect region during the entire pressure fluctuation cycle is calculated by transient finite element solution. The curve of the maximum principal stress at the defect edge changing with time is extracted, and the stress concentration factor is calculated in combination with the fatigue strength limit of the pipeline material.

[0043] Stress Concentration Factor The calculation formula is:

[0044] in: Indicates the stress concentration factor. This represents the peak value of the maximum principal stress at the defect edge, calculated from the finite element model. This represents the nominal stress in a defect-free region, calculated according to the standard pipeline formula, under the same medium pressure load. It is based on the calculated stress concentration factor. The value is used to adjust the boundary sharpness of the defect geometric profile. The adjustment method adopts an iterative smoothing algorithm. The algorithm aims to smooth the local curvature of the boundary nodes with large curvature in the defect geometric profile to a level that matches the calculated local stress gradient, thereby outputting the defect geometric profile with adjusted boundary sharpness.

[0045] In some embodiments, the temperature gradient distribution data inside the pipeline is correlated with the circumferential energy loss map in the preliminary diagnostic report to correct the impact of material sound velocity variations caused by uneven temperature distribution on the energy loss rate calculation. The correction process is based on a pre-calibrated pipeline material sound velocity-temperature correspondence lookup table, which stores calibration coefficients for material sound velocity variations with temperature. The original energy loss rate value is read from the circumferential energy loss map, and simultaneously, based on the propagation path of the guided wave signal that generated the energy loss rate value in the actual pipeline, the local temperature value of the area traversed by the signal propagation path is obtained from the temperature gradient distribution data. Using the sound velocity-temperature correspondence lookup table, the sound velocity correction coefficient corresponding to the local temperature value is found. The original energy loss rate value is recalibrated using the sound velocity correction coefficient, and the recalibrated energy loss rate value is filled back into the corresponding spatial position of the circumferential energy loss map, generating a temperature-compensated circumferential energy loss map.

[0046] It is understandable that by combining the external soil stress data of the pipeline with the axial wave velocity anomaly distribution map in the preliminary diagnostic report, the additional effect of external mechanical constraints on guided wave velocity disturbance can be assessed, thus achieving stress field correction. The soil stress data contains magnitude and direction information, which is decomposed into components acting on the pipeline axis and components perpendicular to the pipeline axis. An empirical correlation model between wave velocity disturbance and axial stress components is constructed. This model is based on a large amount of calibration experimental data under known stress states and describes the linear variation of wave velocity disturbance with axial stress. The empirical correlation model is used to calculate the wave velocity reference offset caused by the axial stress component in the soil stress data. This offset is subtracted from the original wave velocity disturbance value in the axial wave velocity anomaly distribution map to eliminate the systematic deviation caused by external axial stress. Optionally, considering the elliptic micro-deformation of the pipeline cross-section caused by the stress component perpendicular to the pipeline axis in the soil stress data, a stiffness factor of the pipeline cross-section is introduced to perform a secondary fine-tuning of the wave velocity disturbance value after eliminating the systematic deviation caused by external axial stress. The pipe section stiffness factor is calculated using the elastic modulus, Poisson's ratio, and geometric dimensions of the pipe material. The fine-tuning process is based on the deformation theory model of the pipe section under transverse stress. Finally, the final wave velocity disturbance value after stress field correction is obtained, and the axial wave velocity anomaly distribution map is updated accordingly.

[0047] See Figure 4 In the spatial distribution representation of multi-dimensional defect feature tensors, the heatmap visually presents the normalized distribution of defect feature tensors in the axial position (0-10m) and circumferential angle (0-360°) dimensions of the pipeline. Specifically, the values ​​of the feature tensors are obtained through guided wave signal analysis, wave mode component separation, and multi-dimensional feature fusion. Their normalized values ​​(1.0-3.0) correspond to the defect feature intensities in different regions. The color gradient of the heatmap is positively correlated with the feature values: the yellow region (feature value ≈ 3.0) corresponds to areas with high defect feature intensity, concentrated in the axial 2-6m and circumferential 60-240° range, reflecting the significant feature response of defects within this spatial range; the purple / dark blue region (feature value ≈ 1.0) corresponds to areas with weaker defect features, mainly distributed in the axial 8-10m and circumferential edge regions. In the parameter configuration of this heatmap, the sampling interval for the axial position is equidistant (0-10m), the discrete precision of the circumferential angle is 1°, and the normalization processing of the feature values ​​is based on the extreme value scaling of the full space features to ensure the comparability of cross-regional features.

[0048] In one embodiment of the present invention, a finite element model of a pipeline with a defective geometric profile is established, time-varying pressure boundary conditions are applied, the maximum principal stress curve at the defect edge is extracted, and the stress concentration factor is calculated. Iterative smoothing is used to match the curvature of the profile boundary with the stress gradient. Based on temperature gradient data, a local sound velocity correction coefficient is queried, and the circumferential energy loss rate is recalibrated to generate a temperature compensation map. Soil stress is decomposed into axial and vertical components, the wave velocity reference offset is calculated and fine-tuned, and the axial wave velocity anomaly distribution map is updated.

[0049] In practical implementation, establishing a finite element model of the pipeline that includes the geometric contour of the defects is the foundation for subsequent corrections. The pipeline finite element model needs to accurately reflect the pipeline dimensions, material properties, and the spatial morphology of the defect geometric contour obtained from the preliminary diagnostic report. The measured medium pressure fluctuation data is used as a time-varying load condition, applied completely to the inner wall surface of the pipeline finite element model. The equivalent stress distribution of the defect region within a complete pressure fluctuation cycle is calculated using a transient dynamic finite element solver. The curve of the maximum principal stress at the defect edge location changing with time is extracted from the equivalent stress cloud map generated by the solution results. Based on the maximum principal stress change curve and combined with the fatigue strength limit of the pipeline material, the specific value of the stress concentration factor is quantified by comparing the peak stress with the allowable stress of the material.

[0050] In some embodiments, based on the calculated stress concentration factor, a smoothing iteration process is performed on boundary nodes with large curvature in the defect geometry. The goal of the iteration process is to match the local curvature of the boundary nodes with the local stress gradient distribution calculated from the finite element model. The smoothing iteration process follows a curvature adjustment algorithm, which calculates the curvature adjustment amount of the boundary nodes based on the local stress gradient value. The calculation formula is:

[0051] in: This indicates the amount of adjustment needed to the local curvature in the current iteration step. Represents the convergence coefficient. This represents the local stress gradient value at the current node of the defect edge, extracted from the finite element model. This represents the reference stress gradient value calculated based on material properties and average load. Through multiple iterative calculations, the coordinates of the boundary nodes of the defect geometry are gradually updated until the local curvature of all boundary nodes meets the matching condition with the calculated stress gradient, thus outputting the defect geometry with adjusted boundary sharpness.

[0052] It is understandable that correcting the impact of material sound velocity variations caused by temperature inhomogeneity on the calculation of energy loss rate requires a pre-established sound velocity-temperature correspondence lookup table. In practice, for a specific grade of pipe steel, complete data on the variation of guided wave sound velocity with temperature at different frequencies within a temperature range of -20°C to 150°C is obtained through laboratory calibration, and a correspondence lookup table is constructed based on this data. While reading the original energy loss rate value from the circumferential energy loss map, the local average temperature value of the area covered by the propagation path of the guided wave that generated the signal is obtained from the temperature gradient distribution data. Using the sound velocity-temperature correspondence lookup table, a precise sound velocity correction coefficient corresponding to the local average temperature value is found through linear interpolation. This sound velocity correction coefficient is used to recalibrate the original energy loss rate value, and the recalibrated energy loss rate value is filled back into the corresponding position in the circumferential energy loss map, ultimately generating a temperature-compensated circumferential energy loss map.

[0053] Optionally, stress field correction is performed on the wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map, involving the decomposition and application of external soil stress data. The soil stress data is decomposed into stress components parallel to and perpendicular to the pipe axis. An empirical correlation model between the wave velocity disturbance and the axial stress components is constructed. This model takes the axial stress value as input and outputs a wave velocity reference offset. For example, for an X52 grade steel pipeline with an outer diameter of 323.9 mm and a wall thickness of 7.1 mm, the empirical correlation model shows that every 1 MPa increase in axial compressive stress leads to a negative shift of approximately 0.05% in the longitudinal guided wave group velocity. This model is used to calculate the wave velocity reference offset caused by the axial stress component in the soil stress data, and this offset is subtracted from the original wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map to eliminate systematic measurement bias caused by external axial stress. To further consider the slight elliptic deformation of the pipe section induced by the stress component perpendicular to the axial direction, a section stiffness factor calculated based on the pipe section moment of inertia and material elastic modulus is introduced. The wave velocity disturbance value after the axial stress effect has been eliminated is then finely adjusted to obtain the final wave velocity disturbance value after stress field correction, which is then used to update the axial wave velocity anomaly distribution map.

[0054] See Figure 5The figure, with time as the horizontal axis, synchronously presents the dynamic changes of the normalized medium pressure load (blue curve) and the maximum principal stress at the defect edge (orange curve). Specifically, the medium pressure load exhibits periodic fluctuations, while the maximum principal stress at the defect edge shows a strong correlation with the change in medium pressure load: when the medium pressure load increases, the maximum principal stress at the defect edge increases synchronously; when the medium pressure load decreases, the maximum principal stress at the defect edge also decreases accordingly, and the fluctuation amplitude of the principal stress is positively correlated with the fluctuation amplitude of the pressure load. This figure intuitively reflects the response law of the defect edge stress under dynamic pressure load, providing data support for subsequent calculation of stress concentration factors and correction of defect geometry.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pipeline defect diagnosis method based on guided wave technology, characterized in that, include: By using an array of guided wave sensors installed on the outer wall of the pipe, the original guided wave response signals of the pipe structure at multiple excitation frequencies are obtained; Multi-level signal analysis and separation operations are performed on the original guided wave response signal to extract the set of independent wave mode components directly related to the pipeline defect. The set of independent wave mode components includes longitudinal guided wave components propagating along the pipeline axis, torsional guided wave components distributed along the pipeline circumferentially, and surface guided wave components attenuating along the pipeline radially. Based on the propagation delay and energy distribution of each component in the set of independent wave mode components, a multidimensional defect feature tensor is constructed to characterize the spatial features of the defect. The multi-dimensional defect feature tensor is input into a trained pipeline defect discrimination network for layer-by-layer feature mapping and abstraction, and the output is a preliminary diagnostic report containing defect category labels and defect geometric contours. The preliminary diagnostic report is subjected to physical field coupling correction based on pipeline service environment parameters to generate the final diagnostic result.

2. The pipeline defect diagnosis method based on guided wave technology according to claim 1, characterized in that, The step of performing multi-level signal analysis and separation operations on the original guided wave response signal to extract a set of independent wave mode components directly related to pipeline defects includes: The original guided wave response signal is decomposed in the time-frequency domain based on wavelet packet transform to obtain a set of sub-signals covering different frequency bands; For each sub-signal, calculate its coherence coefficient with a preset standard guided wave mode template, and identify and separate the noise components and multipath reflection components mixed in the sub-signal based on the strength of the coherence coefficient. The denoised sub-signals are processed using blind source separation technology. Based on the principle of signal statistical independence, the time-domain waveforms of the longitudinal guided wave component, the torsional guided wave component, and the surface guided wave component are iteratively decoupled. Envelope detection and phase synchronization analysis are performed on each decoupled time-domain waveform to extract the complete parameters of arrival time, group velocity, and energy decay curve of each wave mode component, and these parameters are combined to form the set of independent wave mode components.

3. The pipeline defect diagnosis method based on guided wave technology according to claim 2, characterized in that, Based on the propagation delay and energy distribution of each component in the independent wave mode component set, a multi-dimensional defect feature tensor for characterizing the spatial features of defects is constructed, including: Based on the arrival time sequence of the longitudinal guided wave component, the wave velocity disturbance value at different positions along the axial direction of the pipeline is calculated, and an axial wave velocity anomaly distribution map in the first dimension is generated based on the wave velocity disturbance value. Analyze the energy attenuation curve of the torsional guided wave component and, in conjunction with its circumferential propagation path length, calculate the circumferential energy loss rate spectrum of the pipeline, which serves as the second dimension of the circumferential energy loss spectrum. The attenuation characteristics of the surface guided wave components are quantified, and based on their differences in penetration ability in the radial depth, the equivalent scattering intensity profile in the pipe wall thickness direction is derived, thus forming a third-dimensional radial scattering intensity profile. The axial wave velocity anomaly distribution map, the circumferential energy defect map, and the radial scattering intensity profile are aligned and superimposed on spatial coordinates to form a multidimensional data volume with a spatial grid structure, namely the multidimensional defect feature tensor.

4. The pipeline defect diagnosis method based on guided wave technology according to claim 3, characterized in that, The step of inputting the multi-dimensional defect feature tensor into a trained pipeline defect discrimination network for layer-by-layer feature mapping and abstraction includes: The first feature extraction layer of the pipeline defect discrimination network receives the multi-dimensional defect feature tensor and extracts the local spatial pattern features of the multi-dimensional data volume by scanning with a three-dimensional convolutional kernel to generate a primary feature map. The second feature abstraction layer of the pipeline defect discrimination network performs pooling and nonlinear transformation on the primary feature map, compressing the data dimension while enhancing the rotation and translation invariance of the features, and generating intermediate abstract features. The third classification decision layer of the pipeline defect discrimination network performs a fully connected operation on the intermediate abstract features, maps the high-dimensional features to a preset defect category space, and outputs the defect category label. In parallel, the fourth contour generation layer of the pipeline defect discrimination network reconstructs the three-dimensional geometric shape of the defect in the pipeline space step by step through deconvolution operation based on the intermediate abstract features, and outputs the geometric contour of the defect.

5. The pipeline defect diagnosis method based on guided wave technology according to claim 4, characterized in that, The physical field coupling correction performed on the preliminary diagnostic report based on pipeline service environment parameters includes: The pressure fluctuation data, temperature gradient distribution data, and external soil stress data of the medium inside the pipeline are obtained, which together constitute the service environment parameters of the pipeline. The defect geometry profile is coupled with the medium pressure fluctuation data for analysis to calculate the stress concentration factor at the defect edge under dynamic pressure load, and the boundary sharpness of the defect geometry profile is adjusted accordingly. The temperature gradient distribution data is correlated with the circumferential energy loss map to correct the influence of material sound velocity changes caused by temperature inhomogeneity on the energy loss rate calculation, and a temperature-compensated circumferential energy loss map is generated. By combining the soil stress data and the axial wave velocity anomaly distribution map, the additional effect of external mechanical constraints on wave velocity disturbance is evaluated, and stress field correction is performed on the wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map.

6. The pipeline defect diagnosis method based on guided wave technology according to claim 5, characterized in that, The calculation of the stress concentration factor at the defect edge under dynamic pressure load, and the adjustment of the boundary sharpness of the defect geometry accordingly, includes: A finite element model of the pipeline containing the geometric contour of the defect is established, and the pressure fluctuation data of the medium is applied as a time-varying boundary condition to the inner wall of the model. The equivalent stress cloud map of the defect region during the complete pressure cycle is calculated by finite element method, and the curve of the maximum principal stress at the defect edge changing with time is extracted from the equivalent stress cloud map. Based on the maximum principal stress variation curve and the fatigue strength limit of the pipeline material, the value of the stress concentration factor is quantified. Based on the value of the stress concentration factor, the boundary nodes with large curvature in the defect geometric profile are subjected to smoothing iteration. The goal of the iteration is to match the local curvature of the boundary nodes with the calculated local stress gradient, thereby outputting the defect geometric profile with adjusted boundary sharpness.

7. The pipeline defect diagnosis method based on guided wave technology according to claim 5, characterized in that, The correction for the impact of material sound velocity variations caused by temperature inhomogeneity on the energy loss rate calculation includes: Based on the temperature gradient distribution data, a lookup table is established to show the correspondence between temperature and material sound velocity in each region of the pipe circumference. The original energy loss rate value is read from the circumferential energy loss map, and the corresponding local temperature value is obtained from the temperature gradient distribution data according to the signal propagation path. Using the aforementioned lookup table, the sound speed correction coefficient corresponding to the local temperature value is found, and the original energy loss rate value is recalibrated using the sound speed correction coefficient. The recalibrated energy loss rate values ​​are filled back into the corresponding positions of the circumferential energy deficit map to generate the temperature-compensated circumferential energy deficit map.

8. The pipeline defect diagnosis method based on guided wave technology according to claim 5, characterized in that, The stress field correction of the wave velocity disturbance values ​​in the axial wave velocity anomaly distribution map includes: The soil stress data is decomposed into components acting on the pipe axis and components perpendicular to the axis. An empirical correlation model between wave velocity disturbance and axial stress component is constructed, and the wave velocity reference offset caused by axial stress component in the soil stress data is calculated using the empirical correlation model. Subtract the wave velocity reference offset from the original wave velocity disturbance value of the axial wave velocity anomaly distribution map to eliminate the systematic deviation caused by external axial stress. Considering the micro-deformation of the pipe cross section caused by the stress component perpendicular to the axial direction, a cross section stiffness factor is introduced to perform a secondary fine-tuning of the wave velocity disturbance value after eliminating the systematic deviation caused by the external axial stress, so as to obtain the final wave velocity disturbance value after stress field correction and update the axial wave velocity anomaly distribution map.

9. The pipeline defect diagnosis method based on guided wave technology according to claim 2, characterized in that, The calculation of its coherence coefficient with the preset standard guided wave mode template includes: Each sub-signal is preprocessed with zero mean and then windowed using the Hanning window function to suppress spectral leakage. The windowed sub-signal is cross-correlated with the preset standard guided wave mode templates corresponding to the same excitation frequency and pipe specifications stored in the database to obtain a cross-correlation sequence; Calculate the maximum value of the cross-correlation coefficient sequence, normalize it, and use it as the coherence coefficient between the windowed sub-signal and the corresponding standard guided wave mode template; A coherence coefficient threshold is set, and sub-signal components with coherence coefficients lower than the threshold are marked as noise components and multipath reflection components, and are removed from subsequent processing.

10. The pipeline defect diagnosis method based on guided wave technology according to claim 3, characterized in that, The step of calculating wave velocity disturbance values ​​at different locations along the pipe axis based on the arrival time sequence of the longitudinal guided wave components, and generating a first-dimensional axial wave velocity anomaly distribution map based on the wave velocity disturbance values, includes: The arrival time sequence is formed by extracting the arrival time of a series of known discrete detection points along the pipe axis from the complete parameters of the longitudinal guided wave component. Based on the known axial distance between adjacent detection points and the corresponding arrival time difference, the measured longitudinal guided wave propagation group velocity of the section between the adjacent detection points is calculated. Obtain the longitudinal guided wave reference group velocity of the corresponding pipe segment under defect-free conditions; Calculate the percentage of the relative deviation between the measured group velocity and the reference group velocity, and define the percentage value as the wave velocity disturbance value at the corresponding axial position; Using the axial position of the pipeline as the abscissa and the calculated wave velocity disturbance value as the ordinate, data interpolation and gridding are performed to generate a continuous axial wave velocity anomaly distribution map that reflects the spatial distribution.

Images