Hydrogen conveying pipeline defect detection method and system, equipment and storage medium

Abstract

The invention provides a hydrogen conveying pipeline defect detection method and system, equipment and a storage medium, and belongs to the technical field of defect detection.The method comprises the steps that phased array ultrasonic detection data, digital ray imaging data and hydrogen concentration data of a hydrogen conveying pipeline are obtained; defect boundary features are extracted based on phased array ultrasonic detection data, defect microstructure features are extracted based on digital ray imaging data, and hydrogen concentration space-time distribution features are extracted based on hydrogen concentration data; the defect boundary features, the defect microstructure features and the hydrogen concentration space-time distribution features serve as nodes to construct a modal association graph, and the edge weight between every two nodes in the modal association graph is determined by mutual information of the two nodes; generating a multi-modal association feature vector based on the modal association graph; and obtaining a defect detection result of the hydrogen conveying pipeline based on the multi-modal correlation feature vector. According to the invention, the accuracy of hydrogen conveying pipeline defect detection can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of defect detection, and more particularly relates to a hydrogen pipeline defect detection method and system, equipment and a storage medium. BACKGROUND

[0002] As a core facility for hydrogen energy storage and transportation, the safe operation of a hydrogen pipeline is directly related to the reliability of the hydrogen energy industry chain. Hydrogen has strong permeability and hydrogen embrittlement effect, which can easily lead to the deterioration of the microstructure of the pipeline material and cause defects such as crack propagation. However, the existing hydrogen pipeline defect detection technology mainly relies on a single detection method, and the limited information dimension can easily lead to defect misjudgment or missed detection, which is difficult to meet the demand for comprehensive and accurate defect detection of hydrogen pipelines, and a more accurate and comprehensive hydrogen pipeline defect detection method is urgently needed. SUMMARY

[0003] The application aims to provide a hydrogen pipeline defect detection method and system, equipment and a storage medium to improve the accuracy of hydrogen pipeline defect detection.

[0004] The first aspect of the embodiment of the application provides a hydrogen pipeline defect detection method, which comprises the following steps: acquiring phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of a hydrogen pipeline; extracting defect boundary features based on the phased array ultrasonic detection data, extracting defect microstructure features based on the digital radiographic imaging data, and extracting hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration data; constructing a modal correlation graph by taking the defect boundary features, the defect microstructure features and the hydrogen concentration spatiotemporal distribution features as nodes, determining the edge weight between each two nodes in the modal correlation graph according to the mutual information of the two nodes, and generating a multi-modal correlation feature vector based on the modal correlation graph; obtaining a defect detection result of the hydrogen pipeline based on the multi-modal correlation feature vector; the defect detection result comprises a defect type and a defect position.

[0005] The second aspect of the embodiment of the application provides a hydrogen pipeline defect detection system, which comprises the following modules: a data acquisition module configured to acquire phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of a hydrogen pipeline; a feature extraction module configured to extract defect boundary features based on the phased array ultrasonic detection data, extract defect microstructure features based on the digital radiographic imaging data, and extract hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration data; a feature fusion module configured to construct a modality correlation graph by taking the defect boundary feature, the defect microstructure feature, and the hydrogen concentration spatiotemporal distribution feature as nodes, and determine the edge weight between each two nodes in the modality correlation graph according to the mutual information of the two nodes; and generate a multi-modal correlation feature vector based on the modality correlation graph; a defect detection module configured to obtain a defect detection result of the hydrogen pipeline based on the multi-modal correlation feature vector; the defect detection result including a defect type and a defect location.

[0006] In a third aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the hydrogen pipeline defect detection method when executing the computer program.

[0007] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the hydrogen pipeline defect detection method.

[0008] The hydrogen pipeline defect detection method and system, the device, and the storage medium provided by the embodiments of the present application have the following advantages: the traditional hydrogen pipeline defect detection method relies on a single detection means and has the problem of one-sided feature information. The embodiments of the present application realize the all-around characterization of defects from macro to micro and from static form to dynamic evolution by extracting the defect boundary feature, the microstructure feature, and the hydrogen concentration spatiotemporal distribution feature respectively. The embodiments of the present application construct a modality correlation graph and quantify the correlation between features by mutual information, solve the problem of multi-source data heterogeneity, make the scattered features form an organic whole, and improve the completeness of feature representation. The multi-modal correlation feature vector of the embodiments of the present application fuses the complementary information of three types of sample data, has a more comprehensive learning basis than a single modal model, can effectively reduce the single data noise interference, and improve the accuracy of defect type and location judgment.

[0009] In summary, the present application significantly improves the comprehensiveness and accuracy of hydrogen pipeline defect detection through multi-modal feature fusion and correlation modeling, and provides more reliable technical support for hydrogen safety transportation. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort.

[0011] Figure 1 A flowchart of the hydrogen pipeline defect detection method provided by an embodiment of the present application is shown in FIG. 1. Figure 2 A schematic diagram of a sample tube with artificial preset defects is provided for an embodiment of the present application. Figure 3 A structural block diagram of a hydrogen pipeline defect detection system is provided for an embodiment of the present application. Figure 4 A schematic block diagram of an electronic device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0012] In the following description, for the purposes of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0013] In order to make the objects, technical solutions and advantages of the present application clearer, the following will be described with specific embodiments in conjunction with the accompanying drawings.

[0014] Reference will be made to Figure 1 , Figure 1 A flowchart of a hydrogen pipeline defect detection method is provided for an embodiment of the present application. The method can be executed by an electronic device. Specifically, the method can include S101-S104.

[0015] S101: Obtain phased array ultrasonic detection data, digital radiographic imaging data, and hydrogen concentration data of the hydrogen pipeline.

[0016] In the present embodiment, the phased array ultrasonic detection data refers to a structured echo signal matrix set containing pipeline matrix and defect reflection information, which is collected by multi-angle and multi-element collaborative scanning of the hydrogen pipeline by a phased array ultrasonic probe array. Each matrix element corresponds to the defect / matrix reflection signal amplitude and timing information under a specific element, scanning angle, and depth. The digital radiographic imaging data refers to the gray-scale image data received by the detector after high-energy rays penetrate the pipeline. The hydrogen concentration data refers to a set of hydrogen concentration values along the hydrogen pipeline collected by a sensor array.

[0017] In this embodiment, considering that the hydrogen transmission pipeline is prone to hydrogen embrittlement due to strong hydrogen permeability, and existing single detection technology is difficult to fully characterize defects (such as macro morphology, micro damage and hydrogen-induced correlation), this embodiment obtains multi-modal data to capture defect features from different dimensions. For the acquisition of phased array ultrasonic testing data, specifically, phased array ultrasonic testing can generate a controllable acoustic beam by controlling the probe element delay excitation. When the sound wave propagates inside the pipeline, the defect interface will reflect the echo. After the echo (i.e. phased array ultrasonic testing data) is obtained in this embodiment, the defect position and morphology can be obtained by analyzing the time and amplitude of the echo. For the acquisition of digital radiographic imaging data, specifically, digital radiographic imaging utilizes the attenuation difference of different materials (defects and matrix) when the ray penetrates the pipeline to form a gray scale contrast image on the detector. Because the ray can penetrate the material, the microstructure such as grain boundary and inclusion can be presented. For the acquisition of hydrogen concentration data, specifically, hydrogen concentration detection can be realized by pre-installed sensor array adsorbing hydrogen to produce physical signals (such as resistance change). Because hydrogen is easy to permeate and related to defect evolution, this embodiment can correlate the pipeline defects by analyzing the hydrogen concentration data.

[0018] For example, this embodiment can periodically obtain phased array ultrasonic testing data through a phased array probe installed along the circumferential direction of the pipeline. The probe can be coupled with the outer wall of the pipeline, and the scanning parameters (such as frequency 5 MHz, array element 128, and angle step 0.5°) are set in advance. The probe can collect echo signals (i.e. phased array ultrasonic testing data) in real time and store them during the movement along the axial direction of the pipeline.

[0019] This embodiment can obtain digital radiographic imaging data by arranging a ray source (such as an X-ray machine) on one side of the pipeline and a flat panel detector installed at the corresponding position on the other side. The energy of the ray source can be 200 keV, and the exposure time can be 1 s-5 s to ensure clear image gray scale. The ray source flat panel detector can collect a cross-sectional image every 0.5 m along the axial direction of the hydrogen transmission pipeline.

[0020] This embodiment can obtain hydrogen concentration data in real time at a fixed frequency by installing hydrogen concentration sensors every 1 m along the axial direction of the outer wall of the pipeline. The hydrogen concentration sensors can be in close contact with the surface of the pipeline, and continuously record the hydrogen concentration values at each position. This embodiment can align the acquisition times of the three types of equipment through a time marker synchronizer to ensure that the multi-modal data at the same time and the same position can be correlated.

[0021] S102: Extract defect boundary features based on phased array ultrasonic testing data, extract defect microstructure features based on digital radiographic imaging data, and extract hydrogen concentration spatiotemporal distribution features based on hydrogen concentration data.

[0022] In the embodiment, the steps of extracting defect boundary features based on phased array ultrasonic detection data, extracting defect microstructure features based on digital radiographic imaging data, and extracting hydrogen concentration spatiotemporal distribution features based on hydrogen concentration data are not strictly ordered, and any possible execution order is within the protection scope of the embodiment. For example, in the embodiment, the defect boundary features, the defect microstructure features, and the hydrogen concentration spatiotemporal distribution features can be extracted simultaneously.

[0023] In the embodiment, the phased array ultrasonic detection data includes a raw echo signal matrix obtained by circumferential scanning of an ultrasonic probe along a hydrogen conveying pipeline. The defect boundary features are extracted based on the phased array ultrasonic detection data, specifically including: performing noise reduction processing on the raw echo signal matrix to obtain a denoised echo signal matrix; generating a binary segmentation map based on the echo signal matrix, and extracting a closed defect boundary based on the binary segmentation map; calculating defect geometric features, defect topological features, and defect texture features based on the closed defect boundary, and splicing the defect geometric features, the defect topological features, and the defect texture features to obtain the defect boundary features.

[0024] In the embodiment, the raw echo signal matrix refers to a set of reflection signals received by an array element of a phased array probe. The binary segmentation map refers to a black-and-white binary image converted from the echo signal matrix. The closed defect boundary refers to a continuous and unbroken defect contour line. The defect geometric features refer to parameters describing the shape of the defect, which can include length, maximum width, and aspect ratio, etc. The defect topological features refer to boundary spatial distribution features, which can include the number of concave-convex points and average curvature, etc. The defect texture features refer to the gray scale distribution features of the boundary region, which can include the mean gray value and the gradient standard deviation, etc.

[0025] In the embodiment, the phased array ultrasonic detection utilizes the reflection characteristics of sound waves at the defect interface. The raw echo signal matrix contains the reflection differences between the defect and the background, but due to the influence of electronic noise and material scattering, there will be interference in the raw echo signal matrix. The noise reduction processing of the embodiment can suppress irrelevant signals in the raw echo signal matrix, retain the true amplitude and timing characteristics of the defect echo, and provide more accurate and reliable data for subsequent segmentation. The embodiment converts the continuous signal into a black-and-white binary image through binary segmentation. The combination of the defect geometric features, the defect topological features, and the defect texture features can comprehensively describe the spatial properties of the defect and support the differentiation of defect types.

[0026] For example, the extraction process of the defect boundary features can include: Noise reduction processing: In the embodiment, a wavelet noise reduction algorithm can be used to decompose the raw echo signal matrix by 3-5 layers, and a soft threshold processing (for example, the threshold can be set to 1.5 times the standard deviation of the noise) is performed on the high-frequency coefficients obtained by the decomposition to reconstruct the denoised echo signal matrix, so that the signal-to-noise ratio is improved to more than 20 dB.

[0027] Generate the binary segmentation map: In this embodiment, the adaptive threshold method can be used on the denoised echo signal matrix, and the local threshold is calculated in units of 5x5 pixel windows. The pixels in the window whose echo amplitude exceeds the threshold are marked as 1 (defects), and the rest are 0 (no defects), to obtain the binary segmentation map.

[0028] Extract the closed defect boundary: In this embodiment, 8-neighbor contour tracing is performed on the segmentation map, morphological closing operation (3x3 structure element) is used to connect the broken edges, short boundaries with a length less than 0.5 mm are deleted, and the contour line with a closed degree higher than 95% is retained as the closed defect boundary.

[0029] Calculate the features and splice: In this embodiment, the geometric features can be calculated by the boundary pixel coordinates, for example, the length is the straight-line distance between the starting point and the ending point of the boundary, and the maximum width is the maximum distance perpendicular to the length direction. The length and the maximum width are taken as the geometric features. In this embodiment, the topological features can be calculated by the curvature formula, for example, the curvature is calculated every 10 pixel points, and the mean value and the standard deviation are taken as the topological features. In this embodiment, the texture features can be obtained by the gray level histogram statistics of the boundary region, for example, the pixel proportion of 10 gray interval is extracted, and the pixel proportion data of the 10 gray intervals are taken as the texture features. In this embodiment, the three types of features are spliced in the order of geometry, topology and texture to form the defect boundary features.

[0030] In this embodiment, the digital radiographic imaging data includes a gray-scale image of a hydrogen transmission pipeline cross section; based on the digital radiographic imaging data, defect microstructure features are extracted, specifically including: dividing the gray-scale image into a plurality of pixel groups, and extracting gray-scale gradient data corresponding to each pixel group respectively; based on the gray-scale gradient data of the plurality of pixel groups, grain distribution features of the hydrogen transmission pipeline are extracted; a high gray abnormal discrete area with a gray value exceeding a first gray threshold in the gray-scale image is determined, and a defect density feature is determined based on the high gray abnormal discrete area; and the grain distribution features and the defect density features are taken as the defect microstructure features.

[0031] In this embodiment, the gray-scale image refers to a two-dimensional image generated after the ray penetrates the pipeline. A pixel group corresponds to a local area after image division. The gray-scale gradient data refers to the gray-scale change rate of adjacent pixels in the pixel group, which can include gradient direction and amplitude, etc. The grain distribution features refer to the spatial distribution parameters of the material grains, which can include grain equivalent diameter, orientation angle, and distribution uniformity, etc. The first gray threshold is a gray critical value for distinguishing normal areas from abnormal areas. The high gray abnormal discrete area refers to an isolated pixel group with a gray value exceeding the first gray threshold, and the parameters of the high gray abnormal discrete area can include area and circularity, etc. The defect density feature refers to the number and equivalent area proportion of high gray abnormal areas per unit area, etc.

[0032] In the present embodiment, digital radiography utilizes the difference in attenuation of different materials to rays, and the attenuation coefficients of grains and defects in the gray-scale image are different, which is manifested as the difference in gray value. The present embodiment can reduce the computational complexity by dividing the image into pixel groups. The present embodiment can identify grain boundaries (i.e., regions of sudden change in gradient) by analyzing the gray-scale gradient within the group, and then extract the distribution characteristics of grain size and orientation. The discrete high gray abnormal region corresponds to the dense defect (such as inclusion and hydrogen blister) that is difficult for the ray to penetrate, and the corresponding defect density feature is used to characterize the degree of micro-damage of the material (hydrogen-induced defects usually exhibit high-density distribution). Through the correlation analysis of grain distribution and defect density, the present embodiment can distinguish hydrogen-induced micro-damage from conventional material defects, and provide a basis for judging the cause of the defects.

[0033] For example, the grain distribution characteristics are extracted, which specifically includes: the present embodiment can divide the gray-scale image into non-overlapping pixel groups according to 5*5 pixels, and calculate the gray difference value of adjacent pixels in each group to obtain the gray gradient data (reflecting the degree of change in gray value between pixels). Further, the present embodiment can uniformly cluster the global gray gradient data of the entire image, and mark the pixels with a gradient value greater than 50 as grain boundaries (the atomic arrangement at the grain boundary is irregular, and the gray value changes significantly); the present embodiment can merge adjacent non-grain boundary pixels through a region growing algorithm to form independent grain regions, and then calculate the equivalent diameter (area equivalent circle diameter) and orientation angle (the angle between the longest axis of the grain boundary and the horizontal direction) of each grain, and count the proportion of grains in different size intervals, and finally form the grain distribution characteristics to comprehensively describe the size, arrangement and distribution of normal grains.

[0034] For example, the defect density characteristics are extracted, which specifically includes: in the same gray-scale image, for the abnormal regions that are significantly different from normal grains, the present embodiment can use a first gray threshold to screen high gray pixels (defect regions attenuate more strongly to rays, and have higher gray values); the present embodiment can merge discrete high gray pixels through a region growing algorithm (the gray difference between adjacent pixels is less than 30) to obtain high gray abnormal discrete regions with an area greater than 5 pixels; the present embodiment can count the number of such regions per unit area and the average distance between regions to form the defect density characteristics to quantify the distribution density of micro-defects.

[0035] After obtaining the grain distribution characteristics and the defect density characteristics, the present embodiment can splice the grain distribution characteristics and the defect density characteristics in order to form a defect microstructure feature vector. The grain distribution characteristics reflect the microstructure state of the material matrix, the defect density characteristics depict the abnormal damage in the matrix, and the two are complementary, which can comprehensively characterize the microstructure health state of the hydrogen transport pipeline, and provide a key basis for distinguishing hydrogen-induced defects (often accompanied by grain refinement and high-density micro-defects) from conventional material defects.

[0036] In this embodiment, the extraction of the hydrogen concentration spatiotemporal distribution feature is based on the dynamic diffusion law of hydrogen leakage. Hydrogen-induced defects can cause local hydrogen concentration to fluctuate over time (time dimension) and form a spatial concentration gradient along the axial direction of the pipeline (spatial dimension). By capturing the trend changes in the time series and the gradient differences in the spatial distribution, and combining the coupling relationship between the two, the dynamic evolution law of the hydrogen concentration can be fully characterized, providing a basis for correlating the defect location (concentration gradient peak) and activity (diffusion intensity).

[0037] For example, to extract the time sequence feature of the hydrogen concentration data at a single location, the following steps can be included: In this embodiment, the hydrogen concentration data at each pipeline location can be fitted into a hydrogen concentration fitting curve (trend) and a change rate fitting curve (fluctuation speed), and then fused to form a time sequence feature. Then, according to the order of the axial position of the pipeline, the time sequence features of the locations are arranged into a hydrogen concentration spatial distribution sequence, while the location identifiers are kept. In this embodiment, the difference between the time sequence features of adjacent locations can be calculated, and then divided by the interval to obtain a spatial hydrogen concentration gradient sequence (quantifying spatial differences). Then, the spatial hydrogen concentration gradient sequence is divided into multiple subsequences, and the concentration change rate of adjacent subsequences at the same location is calculated to obtain a spatial propagation feature. Finally, in this embodiment, the spatial hydrogen concentration gradient sequence and the spatial propagation feature can be fused to form a hydrogen concentration spatiotemporal distribution feature.

[0038] S103: Construct a modal correlation graph with the defect boundary feature, the defect microstructure feature, and the hydrogen concentration spatiotemporal distribution feature as nodes. The edge weight between any two nodes in the modal correlation graph is determined by the mutual information of the two nodes. Generate a multi-modal correlation feature vector based on the modal correlation graph.

[0039] In this embodiment, the modal correlation graph refers to a correlation network graph structure with three types of features as nodes, including the defect boundary feature, the defect microstructure feature, and the hydrogen concentration spatiotemporal distribution feature. The edge weight refers to the correlation strength value between nodes, which is calculated by mutual information and ranges from 0 to 1, with 0 indicating no correlation and 1 indicating complete correlation. The multi-modal correlation feature vector refers to a comprehensive vector that fuses the three types of features and their correlation relationships, with a dimension equal to the sum of the dimensions of the three types of features and the sum of the number of edge weights.

[0040] In this embodiment, the calculation process of the mutual information of two nodes includes: determining the node feature vectors corresponding to the two nodes respectively, and calculating the mutual information of the two nodes through the joint probability distribution based on the node feature vectors corresponding to the two nodes (any two of the defect boundary feature, the defect microstructure feature, and the hydrogen concentration spatiotemporal distribution feature).

[0041] In this embodiment, the formation and evolution of hydrogen pipeline defects are jointly influenced by macro-morphology, microstructure and hydrogen concentration variation, and there is an inherent correlation among the three types of features, such as the correlation between hydrogen concentration gradient and crack boundary expansion. The interdependence between features can be quantified by mutual information in this embodiment, and the higher the mutual information value, the tighter the correlation. This embodiment constructs a modal correlation graph by taking mutual information as an edge weight, which can intuitively present the coupling law between features. When fusing features based on graph structure, the original information of each modality can be retained, and the correlation can also be included, so that the multi-modal correlation feature vector is closer to the nature of the defect, and the problem of ignoring feature correlation in traditional simple splicing is solved, providing a more comprehensive input for subsequent model recognition.

[0042] For example, in the first step, this embodiment can determine node features, specifically including: taking the defect boundary feature vector, the defect microstructure feature vector and the hydrogen concentration spatiotemporal distribution feature vector as three nodes.

[0043] In the second step, this embodiment can calculate mutual information, specifically including: this embodiment first determines the feature vectors corresponding to the two node pairs to be calculated. For example, select the defect boundary feature vector X (dimension M) and the defect microstructure feature vector Y (dimension N), where the elements of X are defect geometry, topology and texture feature parameters, and the elements of Y are grain distribution and defect density feature parameters.

[0044] This embodiment performs discretization processing on the feature vectors corresponding to the two determined nodes, specifically: each dimension of X is divided into 5-10 intervals according to the numerical range (such as length features are divided into 0-1mm, 1-3mm intervals, etc.), and similarly, each dimension of Y is divided into intervals, so that continuous features are converted into discrete symbols (such as using 0-9 to represent different intervals), ensuring the feasibility of subsequent probability calculation.

[0045] This embodiment can calculate the edge probability distribution and joint probability distribution of the two nodes after discretization processing, specifically including: counting the frequency of occurrence of each discrete symbol in X to obtain the edge probability distribution P(X) of X; similarly, the edge probability distribution P(Y) of Y is obtained. Count the frequency of occurrence of the combination of discrete symbols in X and Y to obtain the joint probability distribution P(X, Y).

[0046] This embodiment can calculate the mutual information value based on the obtained joint probability distribution, specifically including: according to the definition of mutual information, the ratio of P(X, Y) to P(X)P(Y) for all discrete symbol combinations is taken logarithm, then multiplied by P(X, Y) and accumulated to obtain the original mutual information value.

[0047] Finally, the original mutual information value obtained in the embodiment can be normalized: the original mutual information value is divided by max(H(X), H(Y)) (H is the information entropy), to obtain a normalized mutual information value in the range of 0-1, which is the edge weight between the two nodes, representing the correlation strength between the features.

[0048] Through the above steps, the edge weights between the defect boundary feature vector and the defect microstructure feature vector, the defect microstructure feature vector and the hydrogen concentration spatiotemporal distribution feature vector, and the defect boundary feature vector and the hydrogen concentration spatiotemporal distribution feature vector can be finally obtained.

[0049] Finally, the embodiment can construct a modal correlation graph according to the determined nodes and the calculated edge weights, specifically including: taking the nodes as vertices and the edge weights as connection strengths to generate a graph structure containing the node feature values and the edge weights. The embodiment can generate a multi-modal correlation feature vector, specifically including: splicing the three types of feature vectors in the order of nodes, and then merging the three groups of edge weight values to form a fusion feature vector.

[0050] S104: obtaining a defect detection result of the hydrogen pipeline based on the multi-modal correlation feature vector; the defect detection result includes a defect type and a defect position.

[0051] In the embodiment, the defect detection result of the hydrogen pipeline can be generated by a target defect recognition model based on the phased array ultrasonic detection data, the digital radiographic imaging data, and the hydrogen concentration data of the hydrogen pipeline; the target defect recognition model is trained based on a target training sample set. Each training sample in the target training sample set includes phased array ultrasonic detection sample data, digital radiographic imaging sample data, hydrogen concentration sample data, and a defect label corresponding to the sample data.

[0052] In the embodiment, the target defect recognition model is a machine learning model for hydrogen pipeline defect detection, and the model parameters of the target defect recognition model can include 3-5 layers of network layers, 64-256 hidden layer nodes, activation function types, and 100-500 iterations. The defect type can include hydrogen-induced cracks, hydrogen bubbling, and mechanical damage, and each defect type corresponds to a unique identifier. The defect position refers to the three-dimensional coordinates of the defect on the pipeline, and the parameters include an axial mileage of 0-1000 m, a circumferential angle of 0-360°, and a radial depth of 0-50 mm. The target training sample set refers to a data set used for training the target defect recognition model, containing multiple training samples, each training sample containing phased array ultrasonic, digital radiographic, and hydrogen concentration data and corresponding labels. The defect label refers to the defect information corresponding to the sample data, which can include a defect type label 1-5 (corresponding to different defects) and a defect position coordinate.

[0053] In the embodiment, the target defect identification model learns the mapping relationship between the phased array ultrasonic detection sample data, the digital radiographic imaging sample data, the hydrogen concentration sample data and the defect label in the training sample, and establishes a nonlinear mapping from the input data to the output result. During the model training process, the difference between the predicted result and the true label is minimized by optimizing the parameters, so that the model has the generalization ability to new samples. When detecting defects in the hydrogen conveying pipeline, the phased array ultrasonic detection data, the digital radiographic imaging data and the hydrogen concentration data obtained are input into the model. The model can extract features from the input data, fuse the extracted features to obtain a multi-modal correlation feature vector, and output the defect type and position through the multi-modal correlation feature vector and the defect rule based on training and learning. The problem of large recognition error and inaccurate positioning caused by incomplete information in the traditional method is solved, and the embodiment is especially suitable for complex defect detection of hydrogen-induced defects affected by multiple factors.

[0054] For example, the embodiment can prepare a target training sample set, specifically including: collecting phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data sample data of the hydrogen conveying pipeline with preset defects, labeling the corresponding defect type and position as a label, and dividing the data into a training set and a validation set in a ratio of 8:2.

[0055] The embodiment can train the model, specifically including: initializing the structure of the target defect identification model, inputting the multi-modal correlation feature vector of the training set into the model, calculating the difference between the predicted label and the true label with the cross-entropy loss function, iteratively optimizing the model parameters with the gradient descent method, evaluating the accuracy with the validation set every iteration, and saving the model parameters until the accuracy improvement of the last 5 iterations is less than 0.5%.

[0056] The embodiment can use the model to detect defects, specifically including: obtaining three types of data of the hydrogen conveying pipeline to be detected and inputting them into the trained model, the model outputting a defect type identifier and a three-dimensional coordinate, verifying the output result (such as whether the position coordinate is within the range of the pipeline), and finally outputting the defect detection result.

[0057] As can be seen from the above, the embodiment fully captures the defect features from three dimensions of macroscopic morphology, microscopic structure and hydrogen-induced correlation by fusing phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data, solving the problem that single detection technology cannot completely represent the defects of the hydrogen conveying pipeline. The defect boundary features of phased array ultrasonic, the microscopic structure features of digital radiographic imaging and the spatiotemporal distribution features of hydrogen concentration are complementary to each other, which can reflect the evolution law of the defects in all directions.

[0058] The embodiment constructs a modal correlation graph and generates a multi-modal correlation feature vector by mutual information, retains the internal correlation between different features, and avoids information loss caused by simple splicing of traditional features. This deep fusion method makes the feature representation closer to the nature of the defect, significantly improves the distinguishing ability of the target defect recognition model for complex defects, and significantly improves the recognition accuracy of easily confused defects such as hydrogen-induced cracks and mechanical damage.

[0059] The target defect recognition model is trained based on a multi-modal training sample set, can capture the hydrogen embrittlement effect in combination with the hydrogen concentration spatiotemporal distribution feature, can accurately output the defect type and three-dimensional position, and effectively meets the high safety detection requirements of the hydrogen conveying pipeline, thereby providing reliable protection for hydrogen energy storage and transportation safety.

[0060] In an embodiment of the present application, the hydrogen concentration data includes hydrogen concentration data of multiple pipeline positions in the hydrogen conveying pipeline within a target time period; the hydrogen concentration spatiotemporal distribution feature is extracted based on the hydrogen concentration data, including: for each pipeline position in the hydrogen conveying pipeline, the hydrogen concentration time sequence feature of the pipeline position within the target time period is extracted based on the hydrogen concentration data of the pipeline position within the target time period; the hydrogen concentration spatial distribution sequence is generated based on all pipeline positions in the hydrogen conveying pipeline and the hydrogen concentration time sequence features of each pipeline position within the corresponding target time period; the hydrogen concentration spatial distribution sequence includes hydrogen concentration time sequence features arranged in order according to the spatial position relationship, and each hydrogen concentration time sequence feature corresponds to a position identifier; the hydrogen concentration spatiotemporal distribution feature is extracted based on the hydrogen concentration spatial distribution sequence.

[0061] In the embodiment, the target time period refers to a continuous time interval of hydrogen concentration monitoring. The pipeline position refers to a monitoring point arranged axially along the hydrogen conveying pipeline, and the position identifier is an axial mileage (such as 0m, 1m…10m, etc.) or a number (1-10, etc.). The hydrogen concentration data refers to a set of concentration values of each position within the target time period.

[0062] In the embodiment, the hydrogen concentration time sequence feature of the pipeline position within the target time period is extracted based on the hydrogen concentration data of the pipeline position within the target time period, specifically including: determining a first time window based on the data variation span of the hydrogen concentration data of the pipeline position within the target time period; the data variation span is determined by the maximum hydrogen concentration value and the minimum hydrogen concentration value in the hydrogen concentration data, and the data variation span is negatively correlated with the first time window; calculating the mean sequence and the standard deviation sequence of the hydrogen concentration data of the pipeline position within the target time period according to the first time window; generating a hydrogen concentration fitting curve of the pipeline position within the target time period based on the mean sequence, and generating a hydrogen concentration rate of change fitting curve of the pipeline position within the target time period based on the standard deviation sequence; determining the hydrogen concentration time sequence feature of the pipeline position within the target time period based on the hydrogen concentration fitting curve and the hydrogen concentration rate of change fitting curve.

[0063] In the embodiment, the mean sequence and the standard deviation sequence of the hydrogen concentration data of the pipeline position in the target time period are calculated according to the first time window, specifically including: dividing the target time period into a plurality of time blocks according to the first time window; calculating the mean of the hydrogen concentration data corresponding to each time block, and splicing the means of the hydrogen concentration data corresponding to each time block in time sequence into the mean sequence of the hydrogen concentration data; calculating the standard deviation of the hydrogen concentration data corresponding to each time block, and splicing the standard deviations of the hydrogen concentration data corresponding to each time block in time sequence into the standard deviation sequence of the hydrogen concentration data.

[0064] In the embodiment, the data variation span refers to the fluctuation range of the concentration of a single position, which can be specifically the difference between the maximum concentration value and the minimum concentration value. The first time window refers to the time interval for calculating the statistical characteristics, and the parameters can include the window length such as 1 minute. The first time window is negatively correlated with the variation span, and the larger the span is, the smaller the window time is. The mean sequence refers to a sequence composed of the average values of the concentrations in each time window, and the length of the mean sequence is the target time period / window length (such as 24 hours / 30 minutes=48 points). The standard deviation sequence refers to a sequence composed of the standard deviations of the concentrations in each time window, and the length of the standard deviation sequence is consistent with that of the mean sequence. The hydrogen concentration fitting curve refers to a curve obtained by smoothing fitting the mean sequence. The hydrogen concentration change rate fitting curve refers to a curve obtained by fitting the change rate of the standard deviation sequence, wherein the negative value is the decline rate and the positive value is the rise rate. The hydrogen concentration time sequence feature refers to a feature vector fused with the fitting curve and the change rate curve.

[0065] For example, the process of extracting the hydrogen concentration time sequence feature based on the hydrogen concentration data can include: (1) determining the target time period and the pipeline position: in the embodiment, 24 consecutive hours can be selected as the target time period, and a total of 50 monitoring points (position identifiers 1-50) can be arranged every 1 meter along the pipeline axis, and the hydrogen concentration data of each position can be collected, wherein the sampling frequency is set to 5 Hz, and the mean value is taken every 1 minute for downsampling.

[0066] (2) calculating the data variation span and the first time window: for each position, the maximum concentration value and the minimum concentration value in the target time period can be counted in the embodiment, and the variation span (for example, the maximum concentration value of position 10 is 800 ppm, and the minimum concentration value is 300 ppm, and the span is 500 ppm) can be calculated; the first time window can be determined according to the negative correlation rule in the embodiment, such as setting the first time window to 10 minutes when the span is greater than 300 ppm, setting the first time window to 20 minutes when the span is between 100-300 ppm, and setting the first time window to 30 minutes when the span is less than 100 ppm.

[0067] (3) Generating mean sequence and standard deviation sequence: The embodiment can divide the concentration data of the position according to the first time window sliding, calculate the concentration mean and standard deviation of each window, and form the mean sequence and standard deviation sequence.

[0068] (4) Fitting curve and extracting time sequence features: The embodiment can perform second-order polynomial fitting on the mean sequence to obtain a hydrogen concentration fitting curve, and extract the peak value, valley value and average slope of the curve; the embodiment can calculate the difference (divided by the window interval) of adjacent windows of the standard deviation sequence, perform linear fitting to obtain a change rate curve, and extract the maximum value and minimum value of the change rate; the embodiment can splice the two types of features such as the peak value, valley value and average slope of the curve, and the maximum value and minimum value of the change rate into a hydrogen concentration time sequence feature vector.

[0069] In the embodiment, the hydrogen concentration spatiotemporal distribution features are extracted based on the hydrogen concentration spatial distribution sequence, specifically including: calculating the difference of the hydrogen concentration time sequence features corresponding to each two adjacent positions in the hydrogen concentration spatial distribution sequence to obtain a spatial hydrogen concentration difference sequence; generating a spatial hydrogen concentration gradient sequence based on the spatial hydrogen concentration difference sequence; dividing the hydrogen concentration spatial distribution sequence into hydrogen concentration spatial distribution subsequences of multiple time periods according to a second time window; the second time window is larger than the first time window; calculating the hydrogen concentration change rate of the same position identifier in each two adjacent hydrogen concentration spatial distribution subsequences; extracting the spatial propagation features of the hydrogen concentration change rate based on the hydrogen concentration change rates of the same position identifier in all hydrogen concentration spatial distribution subsequences of the time periods; the spatial propagation features include spatial propagation direction and spatial propagation intensity; taking the spatial hydrogen concentration gradient sequence and the spatial propagation features as the hydrogen concentration spatiotemporal distribution features.

[0070] In the embodiment, the hydrogen concentration spatial distribution sequence refers to a set of time sequence features arranged in order of position identifiers, the sequence length of the hydrogen concentration spatial distribution sequence is the total number of positions, and each element of the hydrogen concentration spatial distribution sequence is a time sequence feature vector of the corresponding position. The spatial hydrogen concentration difference sequence refers to a set of differences of adjacent position time sequence features, such as the difference of the mean sequence of positions i and i+1. The spatial hydrogen concentration gradient sequence refers to a set of gradient values obtained by dividing the difference sequence by the position interval. The second time window refers to the time interval for dividing the spatial distribution sequence. The second time window is an integer multiple of the first time window. The first time window is used for hydrogen concentration time sequence feature extraction of a single pipeline position, and is used to capture fine-grained fluctuations of local concentration (such as sudden concentration rise / drop in a short time), which requires a small window to achieve high time resolution to adapt to scenarios of rapid hydrogen concentration change (such as initial stage of defect leakage).

[0071] In the embodiment, the second time window is used to divide the hydrogen concentration spatial distribution sequence for analyzing the spatial propagation law of concentration change (such as the trend of diffusion from a certain position to the adjacent area), which needs to cover a long enough time to reflect the cumulative effect of spatial propagation. If the window is too small, the propagation characteristics cannot be shown due to insufficient time span; if it is an integer multiple of the first time window, multiple fine-grained time sequence characteristics can be integrated into a spatial sub-sequence, which not only retains the time sequence details, but also ensures that the spatial analysis has enough time dimension support. For example, when the first time window is 5 minutes, the second time window can be 30 minutes, that is, it contains data of 6 first time windows. This size relationship design can make the fine-grained time sequence characteristics of the first time window provide basic data for the spatial propagation analysis of the second time window, and at the same time, the larger span of the second time window ensures the recognizability of the spatial law, which adapts to the hierarchical needs of hydrogen concentration spatiotemporal distribution feature extraction.

[0072] In the embodiment, the hydrogen concentration spatial distribution sub-sequence refers to the spatial distribution sequence within a single second time window. The hydrogen concentration change rate refers to the ratio of the concentration change amount of the same position in adjacent sub-sequences to the time interval. The spatial propagation characteristics include the propagation direction (axial positive direction / negative direction / undirectional) and the propagation intensity (cumulative sum of hydrogen concentration change rate). The hydrogen concentration spatiotemporal distribution feature refers to the vector that integrates the gradient sequence and the propagation characteristics.

[0073] In the embodiment, the hydrogen concentration at a single pipe position changes dynamically over time, and its fluctuation law (such as rising / descending trend, fluctuation amplitude) is closely related to the defect activity (such as hydrogen-induced defect leakage, which will cause the concentration to continue to rise). The data change span reflects the degree of concentration fluctuation: when the fluctuation is large (such as in the early stage of leakage), a short time window (high time resolution) is needed to capture the details, and when the fluctuation is small (such as in the stable diffusion stage), a long time window (strong noise suppression ability) is used. This negative correlation design can balance the time resolution and noise suppression. The mean sequence and the standard deviation sequence are used to represent the concentration average level and the fluctuation stability respectively, and the fitting curve can smooth the random noise and highlight the trend change; the change rate curve quantifies the fluctuation speed, and the time sequence characteristics formed by the combination of the two can completely describe the time dynamic law of the unit position.

[0074] In this embodiment, the hydrogen leakage of the hydrogen pipeline defect has spatial propagation characteristics (such as axial diffusion), and needs to be combined with multi-position data to capture the spatial law. The spatial hydrogen concentration difference sequence reflects the concentration difference of adjacent positions, and the gradient sequence further quantifies the concentration change per unit distance (the gradient sudden increase point usually corresponds to the vicinity of the defect). According to the second time window, the time-space problem can be converted into an analysis mode of time slices combined with spatial distribution, which is convenient for tracking the spatial propagation process of concentration change. This embodiment can identify the propagation direction (such as continuous diffusion from position i to i+1) and the propagation intensity of the change trend by calculating the same position change rate of adjacent subsequences. The higher the cumulative value of the change rate, the more significant the hydrogen diffusion. This embodiment combines the spatial hydrogen concentration gradient sequence and the spatial propagation characteristics of the hydrogen concentration change rate, which can simultaneously represent the spatial non-uniformity and dynamic diffusion law of hydrogen concentration, and provides key evidence for correlating the defect position (gradient peak) and activity (high activity corresponding to high propagation intensity).

[0075] For example, the process of extracting the hydrogen concentration space-time distribution characteristics based on the hydrogen concentration space distribution sequence can include: (1) Generate a hydrogen concentration space distribution sequence: this embodiment can arrange the time sequence feature vectors of each position in order according to the position identifier 1-50 to form a spatial distribution sequence.

[0076] (2) Calculate the spatial hydrogen concentration difference and the spatial hydrogen concentration gradient sequence: this embodiment can perform dimension-by-dimension difference calculation (such as the difference of the peak value of the mean sequence) on the time sequence feature vectors of adjacent positions (such as 1 and 2, 2 and 3…49 and 50) to obtain a spatial hydrogen concentration difference sequence containing 49 elements; divide each difference by the position interval (1 meter) to generate a spatial hydrogen concentration gradient sequence.

[0077] (3) Divide the hydrogen concentration space distribution subsequence: this embodiment can divide the target time period (24 hours) into 24 time periods according to the second time window (such as 1 hour), and each time period corresponds to a spatial distribution subsequence.

[0078] (4) Calculate the hydrogen concentration change rate and the spatial propagation characteristics: this embodiment can calculate the time sequence feature change amount (such as the peak value change) of the same position of adjacent time period subsequences (such as the 1st and the 2nd, the 2nd and the 3rd…the 11th and the 12th), and divide by the time interval (1 hour) to obtain the change rate; this embodiment can count the signs (positive / negative) of the change rates of each position, if the change rates of the continuous 3 adjacent positions are all positive, the propagation direction is the positive direction of the axial direction (and vice versa); this embodiment can take the cumulative sum of the change rate in the propagation direction as the propagation intensity.

[0079] (5) Integration of spatiotemporal distribution characteristics: In this embodiment, the spatial hydrogen concentration gradient sequence can be spliced ​​with the propagation direction (coded as 1 / -1) and the propagation intensity to form a spatiotemporal distribution feature vector of hydrogen concentration for subsequent modal association diagram construction.

[0080] This embodiment extracts the time series characteristics of hydrogen concentration through a dynamically adaptive time window design, adjusts the first time window according to the span of data changes, captures details with high time resolution when the concentration fluctuates violently, and suppresses noise through a longer window in the stable stage, effectively balancing the accuracy and anti-interference ability of the time series characteristics, and fully depicts the dynamic change law of the hydrogen concentration at a single location.

[0081] At the same time, this embodiment combines spatial dimension analysis and calculates the differences and gradients of adjacent positions to accurately locate concentration mutation areas, providing a key basis for defect location identification; a second time window that is larger than the first time window and an integer multiple of it is used to divide the spatial distribution sequence, which not only retains fine-grained time series information but also ensures the identifiability of spatial propagation characteristics. By analyzing the spatial propagation direction and intensity of the change rate, it can effectively capture the diffusion trend of hydrogen leakage and quantify defect activity.

[0082] Overall, this embodiment integrates the spatiotemporal distribution characteristics of spatial gradients and propagation characteristics, which can comprehensively characterize the spatiotemporal dynamic laws of hydrogen concentration, provide a reliable basis for the precise positioning and activity assessment of hydrogen pipeline defects, significantly improve the scientificity and practicality of hydrogen concentration monitoring, and ensure the safe operation of the hydrogen transmission system.

[0083] In one embodiment of the present application, a method for acquiring a target training sample set includes: acquiring phased array ultrasonic testing data, digital radiographic imaging data, and hydrogen concentration data of a hydrogen transmission pipeline with a preset defect; using the phased array ultrasonic testing data, digital radiographic imaging data, hydrogen concentration data, and a defect label of the hydrogen transmission pipeline with the preset defect as a first training sample set; the defect label includes a defect type and a defect location; For each training sample in the first training sample set, data expansion is performed on at least one of phased array ultrasonic testing data, digital radiographic imaging data, and hydrogen concentration data in the training sample based on a defect label of the training sample to obtain an expanded second training sample set; The first training sample set and the second training sample set are used as target training sample sets.

[0084] In the present embodiment, the pre-set defective hydrogen pipeline refers to a pipeline specimen artificially processed with known defects, the defect types include hydrogen-induced cracks, hydrogen blistering and mechanical damage, etc., the defect parameters include length of 0.5-10 mm, width of 0.1-2 mm and depth of 0.2-5 mm, etc., the pipeline material is consistent with the actual hydrogen pipeline (such as X65 steel). The first training sample set refers to a data set containing original detection data and labels, each sample contains phased array ultrasonic data, digital ray data and hydrogen concentration data and defect labels, and the sample amount can be 100-500 groups. Data augmentation refers to the process of generating new samples based on original samples. The second training sample set refers to the sample set generated by augmentation, and the target training sample set refers to the union of the first sample set and the second sample set.

[0085] In the present embodiment, although the original pre-set defect sample can provide basic data, the number is limited and it is difficult to cover all working conditions. Based on the defect label, the data augmentation can simulate the variation scene (such as probe angle deviation and ray energy fluctuation) in actual detection under the premise of retaining the defect core features (such as crack direction, hydrogen concentration gradient and the relevance of the defect), and increase the sample diversity. For example, for hydrogen-induced crack samples, the crack boundary topology and the spatial correlation of high hydrogen concentration area are maintained during augmentation, and the ultrasonic echo intensity is adjusted to simulate different detection distances, so that the model learns the essential features of the defect rather than accidental noise, and the generalization ability to real scenes is improved.

[0086] For example, the acquisition process of the target training sample set specifically includes: (1) Constructing the first training sample set: in the present embodiment, the phased array ultrasonic probe, the ray source and the hydrogen concentration sensor can be arranged on the pre-set defective pipeline, and three types of data can be collected. The historical defect detection data of the hydrogen pipeline can also be obtained; then the defect type and the defect position are manually labeled to form the first training sample set.

[0087] (2) Generating the second training sample set: in the present embodiment, each sample can be augmented based on the defect label: for example, for ultrasonic data, the echo amplitude can be adjusted according to the defect depth (the amplitude is attenuated by 0.5 dB for every 1 mm increase in depth); for ray data, the gray scale can be adjusted according to the defect density (the gray scale value of the high-density defect area is ±8%); for hydrogen concentration data, the diffusion rate can be simulated according to the defect activity (the concentration increase rate of high-activity defects is increased by 10%), and the augmented data and the initial sample data are randomly combined to form the second training sample set.

[0088] (3) Finally, the first training sample set and the second training sample set can be combined to form the target training sample set, which is used for training the target defect recognition model.

[0089] For example, as shown in Figure 2 ​Figure 2 Schematic diagram of the sample tube with artificial pre-set defects provided in this embodiment. The N5 longitudinal outer groove refers to an N-type groove located on the outer surface of the sample tube, with its length parallel to the sample tube axis and a depth of 5% of the nominal wall thickness of the sample tube. In the figure, A and C are outer edge grooves located on the parent material adjacent to the weld edge, and B is an outer center groove located in the center of the weld. The N5 longitudinal inner groove refers to an N-type groove located on the inner surface of the sample tube, with its length parallel to the sample tube axis and a depth of 5% of the nominal wall thickness of the sample tube. In the figure, E and G are inner edge grooves located on the parent material adjacent to the weld edge, and F is an inner center groove located in the center of the weld. Transverse inner and outer grooves refer to N-type grooves with their length perpendicular to the sample tube axis, spanning the weld, and a depth of 5% of the nominal wall thickness of the sample tube. In the figure, H and I are inner and outer transverse grooves, respectively. D is a 1.6mm through hole, and L, M, N, and O are four groups of 3.0mm flat-bottom holes, perpendicular to the groove surface. J, K, P and Q are four groups of 3.0mm flat-bottom holes, perpendicular to the groove surface. S and T are 3.2mm pipe end blind holes, and U and V are 6.0mm hot zone separation holes.

[0090] In this embodiment, the first training sample set is constructed based on real inspection data from pre-defective pipelines, ensuring accurate correlation between original features and defect labels, providing a reliable foundation for the model. This implementation generates a second training sample set through targeted expansion of defect labels. This preserves core defect characteristics (such as the correlation between crack topology and hydrogen concentration) while covering more actual inspection scenario variations (such as probe deviation and energy fluctuations), effectively addressing issues such as insufficient original sample size and incomplete operating condition coverage. The combined target training sample set enables the model to fully learn the essential characteristics of defects and the patterns observed in diverse scenarios, significantly improving its ability to generalize and identify various defects in hydrogen pipelines and providing strong support for the accuracy of inspection results.

[0091] Corresponding to the hydrogen pipeline defect detection method of the above embodiment, Figure 3 This is a structural block diagram of a hydrogen pipeline defect detection system provided by an embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 3 The hydrogen pipeline defect detection system 20 includes: a data acquisition module 21, a feature extraction module 22, a feature fusion module 23 and a defect detection module 24.

[0092] The data acquisition module 21 is used to acquire phased array ultrasonic testing data, digital radiographic imaging data and hydrogen concentration data of the hydrogen pipeline; A feature extraction module 22 is used to extract defect boundary features based on phased array ultrasonic testing data, extract defect microstructure features based on digital radiographic imaging data, and extract hydrogen concentration spatiotemporal distribution features based on hydrogen concentration data; The feature fusion module 23 is configured to construct a modal correlation graph by taking the defect boundary feature, the defect microstructure feature, and the hydrogen concentration spatiotemporal distribution feature as nodes, and determine an edge weight between each two nodes in the modal correlation graph according to mutual information of the two nodes; and generate a multi-modal correlation feature vector based on the modal correlation graph. The defect detection module 24 is configured to obtain a defect detection result of the hydrogen conveying pipeline based on the multi-modal correlation feature vector; and the defect detection result includes a defect type and a defect location.

[0093] In an embodiment of the present application, the phased array ultrasonic detection data includes an original echo signal matrix obtained by circular scanning of the hydrogen conveying pipeline by an ultrasonic probe; and the feature extraction module 22 is specifically configured to: perform noise reduction processing on the original echo signal matrix to obtain a denoised echo signal matrix; generate a binary segmentation graph based on the echo signal matrix, and extract a closed defect boundary based on the binary segmentation graph; the closed defect boundary is used to represent a spatial profile of the defect in the hydrogen conveying pipeline, so as to reflect spatial distribution and morphological properties of the defect; calculate a defect geometric feature, a defect topological feature, and a defect texture feature based on the closed defect boundary, and splice the defect geometric feature, the defect topological feature, and the defect texture feature to obtain a defect boundary feature.

[0094] In an embodiment of the present application, the digital ray imaging data includes a gray-scale image of a cross section of the hydrogen conveying pipeline; and the feature extraction module 22 is specifically further configured to: divide the gray-scale image into a plurality of pixel groups, and extract gray-scale gradient data corresponding to each pixel group; extract a grain distribution feature of the hydrogen conveying pipeline based on the gray-scale gradient data of the plurality of pixel groups; determine a high gray-scale abnormal discrete area in which a gray-scale value exceeds a first gray-scale threshold in the gray-scale image, and determine a defect density feature based on the high gray-scale abnormal discrete area; and take the grain distribution feature and the defect density feature as a defect microstructure feature.

[0095] In an embodiment of the present application, the hydrogen concentration data includes hydrogen concentration data of a plurality of pipeline positions in the hydrogen conveying pipeline within a target time period; and the feature extraction module 22 is specifically further configured to: for each pipeline position in the hydrogen conveying pipeline, extract a hydrogen concentration time sequence feature of the pipeline position within a target time period based on hydrogen concentration data of the pipeline position within the target time period; generate a hydrogen concentration spatial distribution sequence based on all pipeline positions in the hydrogen conveying pipeline and the hydrogen concentration time sequence features of the target time period corresponding to each pipeline position; the hydrogen concentration spatial distribution sequence includes hydrogen concentration time sequence features arranged in order according to a spatial position relationship, and each hydrogen concentration time sequence feature corresponds to a position identifier; Extract the hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration spatial distribution sequence.

[0096] In an embodiment of the present application, the feature extraction module 22 is further configured to: determine a first time window based on the data variation span of the hydrogen concentration data in the target time period of the pipeline location; the data variation span is determined by the maximum hydrogen concentration value and the minimum hydrogen concentration value in the hydrogen concentration data, and the data variation span is negatively correlated with the first time window; calculate the mean sequence and the standard deviation sequence of the hydrogen concentration data of the pipeline location in the target time period according to the first time window; generate a hydrogen concentration fitting curve of the pipeline location in the target time period based on the mean sequence, and generate a hydrogen concentration rate fitting curve of the pipeline location in the target time period based on the standard deviation sequence; determine the hydrogen concentration time sequence features of the pipeline location in the target time period based on the hydrogen concentration fitting curve and the hydrogen concentration rate fitting curve.

[0097] In an embodiment of the present application, the feature extraction module 22 is further configured to: perform difference calculation on the hydrogen concentration time sequence features corresponding to each two adjacent positions in the hydrogen concentration spatial distribution sequence to obtain a spatial hydrogen concentration difference sequence; generate a spatial hydrogen concentration gradient sequence based on the spatial hydrogen concentration difference sequence; divide the hydrogen concentration spatial distribution sequence into hydrogen concentration spatial distribution subsequences of multiple time periods according to a second time window; the second time window is greater than the first time window; calculate the hydrogen concentration rate of the same position identifier in the hydrogen concentration spatial distribution subsequences of each two adjacent time periods; extract the spatial propagation features of the hydrogen concentration rate based on the hydrogen concentration rates of the same position identifier in the hydrogen concentration spatial distribution subsequences of all time periods; the spatial propagation features include the spatial propagation direction and the spatial propagation intensity; take the spatial hydrogen concentration gradient sequence and the spatial propagation features as the hydrogen concentration spatiotemporal distribution features.

[0098] In an embodiment of the present application, the defect detection result of the hydrogen conveying pipeline is generated based on the phased array ultrasonic detection data, the digital radiographic imaging data and the hydrogen concentration data of the hydrogen conveying pipeline, and through a target defect recognition model; the target defect recognition model is obtained based on a target training sample set; The target training sample set is obtained in the following manner: Acquiring phased array ultrasonic testing data, digital radiographic imaging data, and hydrogen concentration data of a hydrogen transmission pipeline with a preset defect; using the phased array ultrasonic testing data, digital radiographic imaging data, hydrogen concentration data, and defect labels of the hydrogen transmission pipeline with the preset defect as a first training sample set; the defect labels include defect type and defect location; For each training sample in the first training sample set, data expansion is performed on at least one of phased array ultrasonic testing data, digital radiographic imaging data, and hydrogen concentration data in the training sample based on a defect label of the training sample to obtain an expanded second training sample set; The first training sample set and the second training sample set are used as target training sample sets.

[0099] See also Figure 4 , Figure 4 This is a schematic block diagram of an electronic device provided in one embodiment of the present application. Figure 4 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned system embodiments, such as Figure 3 The functions of the data acquisition module 21, feature extraction module 22, feature fusion module 23 and defect detection module 24 are shown.

[0100] It should be understood that in the embodiment of the present application, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0101] The input device 302 can include a touchpad, a fingerprint collection sensor (for collecting fingerprint information and direction information of a user's fingerprint), a microphone, etc., and the output device 303 can include a display (LCD, etc.), a speaker, etc.

[0102] The memory 304 can include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A part of the memory 304 can also include a non-volatile random access memory. For example, the memory 304 can also store information of hydrogen concentration data.

[0103] In specific implementations, the processor 301, the input device 302, and the output device 303 described in the embodiments of the present application can perform the implementation manners described in the embodiments of the hydrogen pipeline defect detection method provided by the embodiments of the present application, and can also perform the implementation manners of the electronic device 300 described in the embodiments of the present application, which will not be described here.

[0104] In another embodiment of the present application, a computer readable storage medium is provided, which stores a computer program. The computer program includes program instructions, which are executed by a processor to implement all or part of the processes of the above-mentioned embodiments. The computer program can also be used to instruct related hardware to complete the implementation. The computer program can be stored in a computer readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0105] The computer readable storage medium can be an internal storage unit of the electronic device of any of the above-mentioned embodiments, such as a hard disk or a memory of the electronic device. The computer readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer readable storage medium can include both an internal storage unit and an external storage device of the electronic device. The computer readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0106] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device and the unit described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device and the unit described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0108] In several embodiments provided in the present application, it should be understood that the disclosed electronic device and method can be implemented in other ways. For example, the system embodiments described above are merely schematic, for example, the division of the module / unit is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules, units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces or modules / units, and can also be electrical, mechanical or other form of connection.

[0109] The module / unit described as a separate component can be or can not be physically separated, and the component displayed as a module / unit can be or can not be a physical module / unit, that is, can be located in one place, or can be distributed to a plurality of network modules / units. Part or all of the modules / units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0110] In addition, each functional module / unit in each embodiment of the present application can be integrated in one processing module / unit, or each module / unit can be physically present alone, or two or more modules / units can be integrated in one module / unit. The integrated module / unit can be realized in the form of hardware or in the form of a software functional module / unit.

[0111] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for detecting defects in a hydrogen pipeline, characterized by, The method comprises: acquiring phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of the hydrogen conveying pipeline; extracting defect boundary features based on the phased array ultrasonic detection data, defect microstructure features based on the digital radiographic imaging data, and hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration data; constructing a modal correlation graph by taking the defect boundary features, the defect microstructure features and the hydrogen concentration spatiotemporal distribution features as nodes, the edge weight between each two nodes in the modal correlation graph being determined by the mutual information of the two nodes; generating a multi-modal correlation feature vector based on the modal correlation graph; obtaining a defect detection result of the hydrogen conveying pipeline based on the multi-modal correlation feature vector; the defect detection result comprising a defect type and a defect position.

2. The hydrogen pipeline defect detection method of claim 1, wherein, The phased array ultrasonic detection data comprises a raw echo signal matrix obtained by circumferentially scanning the hydrogen conveying pipeline with an ultrasonic probe; The method of extracting defect boundary features based on the phased array ultrasonic detection data comprises: performing noise reduction processing on the raw echo signal matrix to obtain a denoised echo signal matrix; generating a binary segmentation map based on the echo signal matrix, and extracting a closed defect boundary based on the binary segmentation map; the closed defect boundary is used to represent the spatial profile of the defect in the hydrogen conveying pipeline, and reflects the spatial distribution and morphological properties of the defect; calculating defect geometric features, defect topological features and defect texture features based on the closed defect boundary, and splicing the defect geometric features, the defect topological features and the defect texture features to obtain defect boundary features.

3. The hydrogen pipeline defect detection method of claim 1, wherein, The digital radiographic imaging data comprises a gray-scale image of a cross section of the hydrogen conveying pipeline; The method of extracting defect microstructure features based on the digital radiographic imaging data comprises: dividing the gray-scale image into a plurality of pixel groups, and extracting gray-scale gradient data corresponding to each pixel group; extracting grain distribution features of the hydrogen conveying pipeline based on the gray-scale gradient data of the plurality of pixel groups; determining a high gray-scale abnormal discrete area in the gray-scale image where the gray-scale value exceeds a first gray-scale threshold, and determining defect density features based on the high gray-scale abnormal discrete area; taking the grain distribution features and the defect density features as the defect microstructure features.

4. The hydrogen pipeline defect detection method of claim 1, wherein, The hydrogen concentration data comprises hydrogen concentration data of a plurality of pipeline positions in the hydrogen conveying pipeline within a target time period; The method of extracting hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration data comprises: for each pipeline position in the hydrogen conveying pipeline, extracting hydrogen concentration time series features of the pipeline position within a target time period based on the hydrogen concentration data of the pipeline position within the target time period; generating a hydrogen concentration spatial distribution sequence based on all pipeline positions in the hydrogen conveying pipeline and hydrogen concentration time series features corresponding to each pipeline position within a target time period; the hydrogen concentration spatial distribution sequence comprises hydrogen concentration time series features arranged in order according to spatial position relationship, each hydrogen concentration time series feature corresponding to a position identifier; extracting hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration spatial distribution sequence.

5. The hydrogen pipeline defect detection method of claim 4, wherein, The method of extracting hydrogen concentration time series features of the pipeline position within the target time period based on the hydrogen concentration data of the pipeline position within the target time period comprises: determine a first time window based on a data variation span of the hydrogen concentration data of the pipeline position in the target time period; the data variation span is determined by a maximum hydrogen concentration value and a minimum hydrogen concentration value in the hydrogen concentration data, and the data variation span is negatively correlated with the first time window; calculate a mean sequence and a standard deviation sequence of the hydrogen concentration data of the pipeline position in the target time period according to the first time window; generate a hydrogen concentration fitting curve of the pipeline position in the target time period based on the mean sequence, and generate a hydrogen concentration rate fitting curve of the pipeline position in the target time period based on the standard deviation sequence; determine a hydrogen concentration time sequence feature of the pipeline position in the target time period based on the hydrogen concentration fitting curve and the hydrogen concentration rate fitting curve.

6. The hydrogen pipeline defect detection method of claim 5, wherein, The hydrogen concentration space-time distribution feature based on the hydrogen concentration space distribution sequence includes: difference calculation is performed on the hydrogen concentration time sequence feature corresponding to each two adjacent positions in the hydrogen concentration space distribution sequence to obtain a spatial hydrogen concentration difference sequence; generate a spatial hydrogen concentration gradient sequence based on the spatial hydrogen concentration difference sequence; divide the hydrogen concentration space distribution sequence into hydrogen concentration space distribution subsequences of multiple time periods according to a second time window; the second time window is greater than the first time window; calculate the hydrogen concentration rate of the same position identifier in the hydrogen concentration space distribution subsequences of each two adjacent time periods; extract a spatial propagation feature of the hydrogen concentration rate based on the hydrogen concentration rate of the same position identifier in the hydrogen concentration space distribution subsequences of all time periods; the spatial propagation feature includes a spatial propagation direction and a spatial propagation intensity; the spatial hydrogen concentration gradient sequence and the spatial propagation feature are taken as the hydrogen concentration space-time distribution feature.

7. The hydrogen pipeline defect detection method of claim 1, wherein, The defect detection result of the hydrogen conveying pipeline is generated based on phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of the hydrogen conveying pipeline, and by a target defect recognition model; The target defect recognition model is trained based on a target training sample set; The target training sample set is obtained in the following manner: obtain phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of a hydrogen conveying pipeline with a preset defect; take the phased array ultrasonic detection data, the digital radiographic imaging data, the hydrogen concentration data and a defect label of the hydrogen conveying pipeline with the preset defect as a first training sample set; the defect label includes a defect type and a defect position; for each training sample in the first training sample set, at least one of the phased array ultrasonic detection data, the digital radiographic imaging data and the hydrogen concentration data in the training sample is augmented based on the defect label of the training sample to obtain a second training sample set generated by augmentation; take the first training sample set and the second training sample set as the target training sample set.

8. A hydrogen pipeline defect detection system, comprising: It includes: a data acquisition module for acquiring phased array ultrasonic detection data, digital radiographic imaging data and hydrogen concentration data of a hydrogen conveying pipeline; a feature extraction module, configured to extract defect boundary features based on the phased array ultrasonic detection data, defect microstructure features based on the digital radiographic imaging data, and hydrogen concentration spatiotemporal distribution features based on the hydrogen concentration data; a feature fusion module, configured to construct a modality correlation graph by taking the defect boundary features, the defect microstructure features, and the hydrogen concentration spatiotemporal distribution features as nodes, and determine an edge weight between any two nodes in the modality correlation graph according to mutual information of the two nodes, and generate a multi-modal correlation feature vector based on the modality correlation graph; a defect detection module, configured to obtain a defect detection result of the hydrogen transmission pipeline based on the multi-modal correlation feature vector, wherein the defect detection result includes a defect type and a defect location.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 7.

Images