Intelligent maintenance decision method and device for buried pipelines

Abstract

The present invention discloses a method and device for intelligent maintenance decision-making of buried pipelines, the method comprising: obtaining a target magnetic stress image of a pipeline to be repaired; inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; the maintenance decision result includes a defect pixel position and a maintenance level; the pre-trained maintenance decision model is obtained by training and testing a neural network model using a sample magnetic stress image with a predetermined defect pixel position and a maintenance level as the input of the model, and using the defect pixel position and the maintenance level in the sample magnetic stress image as the output of the model. Based on magnetic induction signal image data, the present invention proposes an intelligent maintenance decision method that is highly adaptable and not affected by human subjective factors, improves the accuracy of the analysis of the detection results, avoids blind maintenance, and can greatly improve economy and reduce labor.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline safety, and in particular to a method and device for intelligent maintenance decision-making of buried pipelines. Background Art

[0002] The pipelines in the plant are the lifeblood of the production of refineries and stations, and are the link between various facilities. The buried pipelines in the plant are difficult to detect, are severely corroded, and often leak, causing huge losses to the environment, safety and economy. At present, the main method for excavation and detection of buried pipelines in the plant is magnetic stress detection.

[0003] The existing technology uses magnetic induction signals to make pipeline maintenance decisions. The detection company uses magnetic induction images and manual experience to check for interference signals, and then classifies the magnetic stress detection signal anomalies (mild, moderate, and severe) according to relevant classification standards. Pipeline companies make maintenance decisions based on the classification results provided by the detection company and excavate and repair serious anomalies. When the pipeline is undergoing magnetic stress detection, because the detection method is non-contact, it is easily interfered with by trees, metals, etc. in the surrounding environment. When analyzing the signal, due to the large differences in personnel levels and the huge workload, the accuracy of manual troubleshooting of interference signals is low, the work efficiency is low, and misjudgment and omission of signals often occur, which not only causes companies to blindly repair, but also causes economic losses and social impacts due to shutdowns. In addition, for pipeline companies, the amount of information in the detection signal image data is huge, but the data potential cannot be effectively mined, resulting in a waste of data resources. Summary of the invention

[0004] In view of the problems existing in the prior art, the embodiments of the present invention provide a method and device for intelligent maintenance decision-making of buried pipelines.

[0005] Specifically, the embodiment of the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for intelligent maintenance decision-making of a buried pipeline, comprising:

[0007] Obtaining a target magnetic stress image of the pipeline for which a maintenance decision is to be made; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline for which a maintenance decision is to be made;

[0008] Inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level;

[0009] Among them, the pre-trained maintenance decision model is obtained by training and testing the neural network model by taking the sample magnetic stress image with predetermined defective pixel positions and maintenance levels as the input of the model, and taking the defective pixel positions and maintenance levels in the sample magnetic stress image as the output of the model.

[0010] Furthermore, the buried pipeline intelligent maintenance decision method further includes: a training process of the maintenance decision model;

[0011] The training process of the maintenance decision model includes:

[0012] Collecting a preset number of sample magnetic stress images; wherein the preset number of sample magnetic stress images need to cover various defective pixel positions and various repair levels;

[0013] According to the detection signal analysis expert, the defect pixel locations and repair levels of the preset number of sample magnetic stress images are determined by experience or excavation;

[0014] The sample magnetic stress images with predetermined defective pixel positions and repair levels are used as the input of the model, and the defective pixel positions and repair levels in the sample magnetic stress images are used as the output of the model. The neural network model is trained and tested based on the intelligent learning algorithm to obtain a maintenance decision model.

[0015] Furthermore, after collecting a preset number of sample magnetic stress images, the method further comprises the following steps:

[0016] The sample magnetic stress images are uniformly scaled, specifically including:

[0017] First, count the inspection mileage X of each inspected pipe section image max , average inspection mileage Pixel length L src and the average original pixel length Average Width Round to the nearest integer;

[0018] To ensure that the detection mileage ratio of all images is unified, firstly, the pixel length L of all images is src Adjust so that the adjusted sample image length Count the average pixel value of the image length L0 Round to an integer, and then scale the image:

[0019] Case 1: For detection mileage The sample magnetic stress image is scaled from the original pixel length L0 to a length of The original width W0 is scaled to The coordinates of the pixels in the scaled image are obtained by coordinate interpolation:

[0020]

[0021]

[0022] in, is the coordinate of the pixel after transformation, L0, W0 are the length and width of the sample image;

[0023] The coordinates (x, y) of the transformed pixel point belong to the pixel point coordinate set P:

[0024]

[0025] in,

[0026] When the pixel point coordinates (x, y) are integers, the pixel value f(x, y) of the point after processing is the same as the pixel value of the corresponding point in the original image; when the pixel point coordinates (x, y) are not integers, the pixel value f(x, y) of the point after processing is obtained by performing bilinear interpolation and rounding the pixels of the four points surrounding the point in the original image;

[0027] Case 2: For detection mileage Average sample magnetic stress images, copy and splice the images until the sum of the detected mileage of the spliced ​​images is Within the interval, execute situation 1 again;

[0028] Case 3: For detection mileage The sample magnetic stress image is first split manually. The splitting principle is to ensure that the image detection mileage after splitting is within The abnormal signal area will not be split into two pictures, and the case 1 will be executed for the split pictures.

[0029] Furthermore, the sample magnetic stress image is used as the input of the model, the defect pixel position and the maintenance level in the sample magnetic stress image are used as the output of the model, and the neural network model is trained and tested based on the intelligent learning algorithm to obtain a maintenance decision model, including:

[0030] According to the statistical results of the magnetic stress images obtained historically, the ratio of the number of magnetic stress images containing defects to the number of magnetic stress images not containing defects is determined;

[0031] According to the ratio of the number of pictures, the samples with determined defective pixel locations and repair levels are randomly divided into a training set and a test set to train and test the repair decision model.

[0032] Furthermore, before inputting the target magnetic stress image into a pre-trained maintenance decision model, the method further includes:

[0033] When applying the model, for the target magnetic stress image, the image is firstly normalized in terms of the detection mileage ratio;

[0034] If the mileage is checked Then scale the image to the standard size according to the method described in the above situation 1;

[0035] If the mileage is checked Then, starting from the left end point of the image, create a The sliding window of The image is trimmed once until the right side of the sliding window reaches the boundary of the image, and we get pictures, input these pictures into the network for prediction, and output the defective pixel location and repair level based on the network prediction results.

[0036] Furthermore, the buried pipeline intelligent maintenance decision method further includes: a process of adjusting and migrating the maintenance decision model;

[0037] The process of adjusting and migrating the maintenance decision model includes:

[0038] After accumulating pipeline excavation data, the surrounding environmental factors that affect magnetic induction, including metal, trees and pipeline properties, are One-Hot encoded or continuously encoded to obtain the first feature vector; the feature vector corresponding to the feature map obtained by the convolutional neural network of the target area selected by the region generation network RPN is spliced ​​with the first feature vector to obtain the second feature vector; the second feature vector is calculated by the fully connected neural network layer, and then the Softmax function is used to calculate the probability L of the sample magnetic stress image belonging to each maintenance level i , use the stochastic gradient descent method SGD to train the fully connected network, the network output L i In the results, the maximum L i The most likely maintenance level for the abnormality is shown in the figure below.

[0039] Among them, the probability L of the signal sample belonging to each maintenance level i The calculation method is:

[0040]

[0041] Among them, p i is the output of the fully connected network layer, and N represents the number of maintenance levels.

[0042] Furthermore, the buried pipeline intelligent maintenance decision method further includes: optimizing the maintenance decision model so that the maintenance decision model outputs a defect type;

[0043] Add a multi-label Sigmod loss function after the fully connected layer of the network to obtain the probability of the sample magnetic stress image being predicted as each type of defect

[0044]

[0045] in, is the column vector output by the last fully connected layer of the network, M is the number of defect types,

[0046] Take the cross entropy loss function and use the stochastic gradient descent method SGD to train the network. The network output The defect types corresponding to all positions in the sentence that are greater than the threshold;

[0047] After completing the above processing, the network outputs the defective pixel location, defect type and corresponding repair level during prediction.

[0048] In a second aspect, an embodiment of the present invention further provides an intelligent maintenance decision-making device for buried pipelines, comprising:

[0049] An acquisition module, used to acquire a target magnetic stress image of the pipeline for which a maintenance decision is to be made; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline for which a maintenance decision is to be made;

[0050] An intelligent maintenance decision module, used for inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level;

[0051] Among them, the pre-trained maintenance decision model is obtained by training and testing the neural network model by taking the sample magnetic stress image with predetermined defective pixel positions and maintenance levels as the input of the model, and taking the defective pixel positions and maintenance levels in the sample magnetic stress image as the output of the model.

[0052] Furthermore, the buried pipeline intelligent maintenance decision device further includes: a training module for executing a training process on the maintenance decision model;

[0053] The training process of the maintenance decision model includes:

[0054] Collecting a preset number of sample magnetic stress images; wherein the preset number of sample magnetic stress images need to cover various defective pixel positions and various repair levels;

[0055] According to the detection signal analysis expert, the defect pixel locations and repair levels of the preset number of sample magnetic stress images are determined by experience or excavation;

[0056] The sample magnetic stress images with predetermined defective pixel positions and repair levels are used as the input of the model, and the defective pixel positions and repair levels in the sample magnetic stress images are used as the output of the model. The neural network model is trained and tested based on the intelligent learning algorithm to obtain a maintenance decision model.

[0057] Furthermore, after collecting a preset number of sample magnetic stress images, the training module is further configured to perform the following steps:

[0058] The sample magnetic stress images are uniformly scaled, specifically including:

[0059] First, count the inspection mileage X of each inspected pipe section image max , average inspection mileage Pixel length L src and the average original pixel length Average Width Round to the nearest integer;

[0060] To ensure that the detection mileage ratio of all images is unified, firstly, the pixel length L of all images is src Adjust so that the adjusted sample image length Count the average pixel value of the image length L0 Round to an integer, and then scale the image:

[0061] Case 1: For detection mileage Average sample magnetic stress image, scale the original pixel length L0 of the image to a length of The original width W0 is scaled to The coordinates of the pixels in the scaled image are obtained by coordinate interpolation:

[0062]

[0063]

[0064] in, is the coordinate of the pixel after transformation, L0, W0 are the length and width of the sample image;

[0065] The coordinates (x, y) of the transformed pixel point belong to the pixel point coordinate set P:

[0066]

[0067] in,

[0068] When the pixel point coordinates (x, y) are integers, the pixel value f(x, y) of the point after processing is the same as the pixel value of the corresponding point in the original image; when the pixel point coordinates (x, y) are not integers, the pixel value f(x, y) of the point after processing is obtained by performing bilinear interpolation and rounding the pixels of the four points surrounding the point in the original image;

[0069] Case 2: For detection mileage Average sample magnetic stress images, copy and splice the images until the sum of the detected mileage of the spliced ​​images is Within the interval, execute situation 1 again;

[0070] Case 3: For detection mileage The sample magnetic stress image is first split manually. The splitting principle is to ensure that the image detection mileage after splitting is within The abnormal signal area will not be split into two pictures, and the case 1 will be executed for the split pictures.

[0071] It can be seen from the above technical scheme that the buried pipeline intelligent maintenance decision method and device provided by the embodiment of the present invention first obtains the target magnetic stress image of the pipeline to be repaired, and then inputs the target magnetic stress image into a pre-trained maintenance decision model to obtain the maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes the defect pixel position and the maintenance level; the pre-trained maintenance decision model is obtained by training and testing the neural network model using a sample magnetic stress image with a predetermined defect pixel position and maintenance level as input, and the defect pixel position and maintenance level in the sample magnetic stress image as output. The embodiment of the present invention proposes an intelligent maintenance decision method that is highly adaptable and not affected by human subjective factors based on magnetic induction signal image data, improves the accuracy of the analysis of the detection results, avoids blind maintenance, can greatly improve the economy and reduce labor, and at the same time improves the safety of the pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0073] Figure 1 is a flow chart of a method for intelligent maintenance decision-making of underground pipelines provided by an embodiment of the present invention;

[0074] Figure 2 This is the intention of the intelligent pipeline maintenance decision-making process based on the target magnetic stress image and the neural network provided by one embodiment of the present invention;

[0075] Figure 3 It is a structural schematic diagram of an intelligent maintenance decision-making device for underground pipelines provided by an embodiment of the present invention;

[0076] Figure 4 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0078] Figure 1 FIG. 1 is a flow chart showing a method for intelligent maintenance decision-making of underground pipelines provided by an embodiment of the present invention. Figure 1 As shown, the buried pipeline intelligent maintenance decision method provided by the embodiment of the present invention specifically includes the following contents:

[0079] Step 101: obtaining a target magnetic stress image of a pipeline for which a maintenance decision is to be made; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline for which a maintenance decision is to be made;

[0080] In this step, the detection signal graph is usually a two-dimensional image. In the magnetic induction detection signal result of the two-dimensional image, the horizontal axis X is the detection mileage, in meters (m), and the vertical axis H is the magnetic field strength, in amperes per meter (A / m).

[0081] Step 102: inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level;

[0082] Among them, the pre-trained maintenance decision model is obtained by training and testing the neural network model by taking the sample magnetic stress image with predetermined defective pixel positions and maintenance levels as the input of the model, and taking the defective pixel positions and maintenance levels in the sample magnetic stress image as the output of the model.

[0083] In this step, it should be noted that this embodiment, with the help of magnetic induction signal image data, proposes an intelligent maintenance decision-making method that is highly adaptable and not affected by human subjective factors based on the magnetic induction signal image data, which improves the accuracy of the analysis of the detection results, avoids blind maintenance, can greatly improve economy and reduce labor, and at the same time improves the safety of the pipeline.

[0084] It can be seen from the above technical scheme that the buried pipeline intelligent maintenance decision method provided by the embodiment of the present invention first obtains the target magnetic stress image of the pipeline to be repaired, and then inputs the target magnetic stress image into a pre-trained maintenance decision model to obtain the maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes the defect pixel position and the maintenance level; the pre-trained maintenance decision model is obtained by training and testing the neural network model by taking the sample magnetic stress image with the defect pixel position and maintenance level determined in advance as input, and the defect pixel position and maintenance level in the sample magnetic stress image as output. The embodiment of the present invention proposes an intelligent maintenance decision method with strong adaptability and not affected by human subjective factors based on magnetic induction signal image data, which improves the accuracy of the analysis of the detection results, avoids blind maintenance, can greatly improve the economy and reduce labor, and at the same time improves the safety of the pipeline.

[0085] Based on the content of the above embodiment, in this embodiment, the intelligent maintenance decision method for buried pipelines further includes: a training process of the maintenance decision model;

[0086] The training process of the maintenance decision model includes:

[0087] Collecting a preset number of sample magnetic stress images; wherein the preset number of sample magnetic stress images need to cover various defective pixel positions and various repair levels;

[0088] According to the detection signal analysis expert, the defect pixel locations and repair levels of the preset number of sample magnetic stress images are determined by experience or excavation;

[0089] A sample magnetic stress image with predetermined defective pixel positions and repair levels is used as input, and the defective pixel positions and repair levels in the sample magnetic stress image are used as output. The neural network model is trained and tested based on an intelligent learning algorithm to obtain a maintenance decision model.

[0090] In this embodiment, the preset number of sample magnetic stress images cover various defective pixel positions and various maintenance levels, so that the maintenance decision model obtained after model training using the preset number of sample magnetic stress images can accurately identify the defective pixel positions and maintenance levels of the target magnetic stress images of the pipeline for which maintenance decisions are to be made.

[0091] Based on the content of the above embodiment, in this embodiment, after collecting a preset number of sample magnetic stress images, the method further includes the following steps:

[0092] The sample magnetic stress images are uniformly scaled, specifically including:

[0093] First, count the inspection mileage X of each inspected pipe section image max , average inspection mileage Pixel length L src and the average original pixel length Average Width Round to the nearest integer;

[0094] To ensure that the detection mileage ratio of all images is unified, firstly, the pixel length L of all images is src Adjust so that the adjusted sample image length Count the average pixel value of the image length L0 Round to an integer, and then scale the image:

[0095] Case 1: For detection mileage The sample magnetic stress image is scaled from the original pixel length L0 to a length of The original width W0 is scaled to The coordinates of the pixels in the scaled image are obtained by coordinate interpolation:

[0096]

[0097]

[0098] in, is the coordinate of the pixel after transformation, L0, W0 are the length and width of the sample image;

[0099] The coordinates (x, y) of the transformed pixel point belong to the pixel point coordinate set P:

[0100]

[0101] in,

[0102] When the pixel point coordinates (x, y) are integers, the pixel value f(x, y) of the point after processing is the same as the pixel value of the corresponding point in the original image; when the pixel point coordinates (x, y) are not integers, the pixel value f(x, y) of the point after processing is obtained by performing bilinear interpolation and rounding the pixels of the four points surrounding the point in the original image;

[0103] Case 2: For detection mileage Average sample magnetic stress images, copy and splice the images until the sum of the detected mileage of the spliced ​​images is Within the interval, execute situation 1 again;

[0104] Case 3: For detection mileage The average sample magnetic stress image is first split manually. The splitting principle is to ensure that the image detection mileage after splitting is within The abnormal signal area will not be split into two pictures, and the case 1 will be executed for the split pictures.

[0105] In this embodiment, by first determining the relationship between the detection mileage of the target magnetic stress image and the average detection mileage of the magnetic stress image of the inspected pipe section, and then uniformly scaling the target magnetic stress image based on the relationship between the detection mileage of the target magnetic stress image and the average detection mileage of the magnetic stress image of the inspected pipe section, the problem that the horizontal coordinate scale of the target magnetic stress image obtained each time is different due to the different mileage measured each time, and thus the target magnetic stress image cannot be directly used for intelligent identification of pipeline maintenance level based on neural network, can be solved.

[0106] Based on the content of the above embodiment, in this embodiment, when the pixel point coordinate (x, y) is not an integer, the pixels of the four points surrounding the point in the original image are bilinearly interpolated and then rounded to obtain the pixel value f (x, y) of the point after processing, including:

[0107] When the pixel coordinates (x, y) are not integers, the pixels of the four points surrounding the point in the original image are bilinearly interpolated and rounded to get the pixel value f(x, y) of the point after processing. The bilinear interpolation is:

[0108]

[0109] Among them, Q 11 , Q 12 , Q 21 , Q 22 The four pixels surrounding the point in the original image are (x1, y1), (x1, y2), (x2, y1), (x2, y2), and the pixel values ​​in the original image are f(Q 11 ), f(Q 12 ), f(Q 21 ), f(Q 22 ).

[0110] Based on the content of the above embodiment, in this embodiment, a sample magnetic stress image whose defective pixel position and maintenance level have been determined in advance is used as input, and the defective pixel position and maintenance level in the sample magnetic stress image are used as output. The neural network model is trained and tested based on the intelligent learning algorithm to obtain a maintenance decision model, including:

[0111] According to the statistical results of the magnetic stress images obtained historically, the ratio of the number of magnetic stress images containing defects to the number of magnetic stress images not containing defects is determined;

[0112] According to the ratio of the number of pictures, the samples with determined defective pixel locations and repair levels are randomly divided into a training set and a test set to train and test the repair decision model.

[0113] In this embodiment, according to the statistical results of the magnetic stress images obtained historically, the ratio of the number of magnetic stress images containing defects and the number of magnetic stress images not containing defects is determined, and then according to the ratio of the number of images, the samples with determined defective pixel positions and maintenance levels are randomly divided into a training set and a test set, and then the training set is used to train the neural network model, and the test set is used to test the effect of the trained maintenance decision model. By randomly dividing the samples with determined defective pixel positions and maintenance levels into a training set and a test set according to the ratio of the number of images, the effect of model training and the effect of subsequent testing can be improved, so that both model training and model testing can achieve corresponding expectations.

[0114] Based on the contents of the above embodiments, in this embodiment, in the process of training the neural network model, the stochastic gradient descent method SGD is used to update the network parameters and train the neural network model; wherein, in the selection of the anchor, according to the characteristics of the magnetic induction signal, an anchor of a size with an aspect ratio of 1:2 is selected.

[0115] Based on the content of the above embodiment, in this embodiment, before inputting the target magnetic stress image into the pre-trained maintenance decision model, the method further includes:

[0116] When applying the model, for the target magnetic stress image, the image is firstly normalized in terms of the detection mileage ratio;

[0117] If the mileage is checked Then scale the image to the standard size according to the method described in the above situation 1;

[0118] If the mileage is checked Then, starting from the left end point of the image, create a The sliding window of The image is trimmed once until the right side of the sliding window reaches the boundary of the image, and we get pictures, input these pictures into the network for prediction, and output the defective pixel location and repair level based on the network prediction results.

[0119] In this embodiment, it should be noted that when the network model is applied, for a magnetic induction signal, if the mileage is detected Then scale the signal array to the standard size; if the mileage is detected This ensures that the potential fault area can be fully detected by the network at least once and reduces the distortion of the data signal to improve the network accuracy. At this time, no scaling is performed but starting from the left end point of the signal array, a length of The sliding window of The signal is trimmed once until the right side of the sliding window reaches the boundary of the signal array, and we get (Round down) signal arrays, input these signal arrays into the network for prediction, and output the interval of the abnormal signal position on the data and the abnormal signal maintenance level according to the network prediction results.

[0120] Based on the content of the above embodiment, in this embodiment, the intelligent maintenance decision method for buried pipelines further includes: a process of adjusting and migrating the maintenance decision model;

[0121] The process of adjusting and migrating the maintenance decision model includes:

[0122] After accumulating pipeline excavation data, the surrounding environmental factors that affect magnetic induction, including metal, trees and pipeline properties, are One-Hot encoded or continuously encoded to obtain the first feature vector; the feature vector corresponding to the feature map obtained by the convolutional neural network of the target area selected by the region generation network RPN is spliced ​​with the first feature vector to obtain the second feature vector; the second feature vector is calculated by the fully connected neural network layer, and then the Softmax function is used to calculate the probability L of the sample magnetic stress image belonging to each maintenance level i , use the stochastic gradient descent method SGD to train the fully connected network, the network output L i In the results, the maximum L i The most likely maintenance level for the abnormality is shown in the figure below.

[0123] Among them, the probability L of the signal sample belonging to each maintenance level i The calculation method is:

[0124]

[0125] Among them, p i is the output of the fully connected network layer, and N represents the number of maintenance levels.

[0126] In this embodiment, through such processing, the maintenance level finally predicted by the maintenance decision model is more accurate and has greater reference significance, thereby effectively solving the problem of low accuracy and low work efficiency in manual troubleshooting of interference signals due to large differences in personnel levels and huge workload. The misjudgment and omission of signals not only cause the company to carry out blind maintenance, but also cause economic losses and social impacts due to shutdowns.

[0127] Based on the content of the above embodiment, in this embodiment, the intelligent maintenance decision method for buried pipelines further includes: optimizing the maintenance decision model so that the maintenance decision model outputs a defect type;

[0128] Add a multi-label Sigmod loss function after the fully connected layer of the network to obtain the probability of the sample magnetic stress image being predicted as each type of defect

[0129]

[0130] in, is the column vector output by the last fully connected layer of the network, M is the number of defect types,

[0131] Take the cross entropy loss function and use the stochastic gradient descent method SGD to train the network. The network output The defect types corresponding to all positions in the vector that are greater than the threshold;

[0132] After completing the above processing, the network outputs the defective pixel location, defect type and corresponding repair level during prediction.

[0133] In this embodiment, after further accumulation of data, the output of the network can also migrate to more accurate defect types, such as dents, metal corrosion, weld defects, etc. Through the above processing process provided by this embodiment, when using the model for prediction, accurate defect type information can be further output, thereby providing more effective guidance for subsequent maintenance.

[0134] According to the above two embodiments, by collecting the defect data and surrounding environment data of pipeline excavation and the maintenance decision data obtained through empirical formulas, the accumulated actual defect maintenance decision results and their corresponding data signals are substituted into the training module for model tuning and further accurate prediction of the defect type.

[0135] It can be understood that the intelligent maintenance decision-making process based on a neural network provided in this embodiment specifically includes the following processing contents:

[0136] First, anomalies in the magnetic stress image and their maintenance levels (determined by expert experience analysis or excavation) are marked, and the anomalies in the magnetic stress image with each maintenance level are used as data labels; the collected data set is used to build a neural network model, the data is standardized and preprocessed, and then the data is randomly divided into a training set and a test set in proportion, the model is trained by the training set data, and then the model effect is tested with the test set data, and the maintenance decision results are output. In addition, the defect data verified by pipeline excavation can be further collected, and the standard method is used to make maintenance decisions for the defects obtained by excavation. The accumulated actual defect maintenance decision results and their corresponding image signals are substituted into the training module for tuning processing. When the amount of data accumulates to a certain extent, the model is directly migrated to further improve the accuracy of the network through actual data. In addition, the model can be optimized by adding a layer of multi-label Sigmod loss function, so that the network migrates to more accurate defect types, such as dents, metal corrosion, weld defects, etc.

[0137] The following will be combined Figure 2 The intelligent maintenance decision-making process based on a neural network provided by an embodiment of the present invention is described in detail:

[0138] First, the image is normalized into vectors of the same size according to pixel values, and the vectors are passed through the grayscale VGG convolutional neural network to extract features to obtain feature map vectors. After the feature map is extracted by the convolution head, it is sent to the RPN (Region Proposal Network) to obtain the target region, and then the target region is predicted.

[0139] According to the position of the abnormal area of ​​the manually annotated image (which can be understood as the position of the defective pixel) and the type of pipeline signal abnormality (which can be understood as the maintenance level), the loss function of the sample in the network is calculated. After counting the ratio of the number of abnormal and normal signals, the existing data is randomly divided into a training set and a test set according to the ratio value. The training set is used to train the network model. The input of the model is a normalized image, and the output of the model is the defective pixel position and the maintenance level. The maintenance level of the magnetic stress abnormal signal is divided into three situations: immediate repair, planned repair, and monitoring and use. In order to reduce the loss function, SGD (Stochastic Gradient Descent) is used to update the network parameters and train the target detection network. In order to speed up the network training speed, in the anchor selection, only one size of anchor can be selected according to the shape characteristics of the magnetic induction signal image, with an aspect ratio of 1:2. The test set is used to test the network effect. After further data collection and accumulation, in order to improve the accuracy of the network, the newly added data is substituted into the training module for tuning to further improve the accuracy of network detection.

[0140] When applying the network model, for a magnetic stress signal image, the image is first normalized to the detection mileage ratio. Then scale the image to the standard size according to the method described in Case 1; if Then, starting from the left end point of the image, create a The sliding window of The image is trimmed once until the right side of the sliding window reaches the boundary of the image, and we get (Round down) pictures. These pictures are input into the network for prediction, and the pixel position of the abnormal signal and the abnormal signal repair level are output according to the network prediction results.

[0141] After further accumulation of signal image data and pipeline excavation data, the variables of pipeline influencing factors such as metal, trees, pipeline attributes, etc. can be One-Hot encoded, or continuous numerical encoding can be vectorized, that is, during excavation, the factors that may affect magnetic stress such as trees, wires, and pipeline attribute data around the pipeline are counted, and the data of these factors are recorded to obtain new feature vectors. For example, the pipeline attribute data (pipeline material, diameter, wall thickness) is classified into One-Hot encoding and continuous encoding, as follows:

[0142]

[0143] Among them, m represents the type of pipe, n represents the type of pipe diameter, and r represents the different thickness values ​​of the pipe. The feature map of the special signal area selected by the RPN network is spliced ​​with the feature vector of the surrounding environmental factors. The spliced ​​feature vector is calculated by the fully connected neural network layer, and then the Softmax function is used to calculate the probability L of the signal sample belonging to each maintenance level. i , the SGD method is also used to train the network, and the network output L i In the results, the maximum L i The most likely maintenance level for the abnormality is L. The probability that the signal sample belongs to each maintenance level is L i The calculation is as follows:

[0144]

[0145] In the formula, p i It is the output of the fully connected network layer, and N represents the number of maintenance levels, such as divided into 3 levels: immediate repair, planned repair, and monitored use.

[0146] After further accumulation of data, the output of the network can also migrate to more accurate defect types, such as dents, metal corrosion, weld defects, etc. Add a layer of multi-label Sigmod loss function after the fully connected layer of the network to obtain the probability of the signal sample being predicted as each type of defect

[0147]

[0148] in, is the column vector output by the last fully connected layer of the network, M is the number of abnormal types,

[0149] Use the cross entropy loss function and SGD method to train the network. The network output The defect types corresponding to all positions in the vector that are greater than a threshold (usually 0.5).

[0150] After completing the above steps, the network can output the defect type and the corresponding maintenance level during prediction, such as "corrosion defect, monitoring and use".

[0151] It can be seen that this embodiment proposes an intelligent maintenance decision-making method based on magnetic induction signal image data that is highly adaptable and free of human subjective factors, so as to improve the accuracy of the analysis of the detection results, avoid blind maintenance, greatly improve the economy and work efficiency, reduce labor, and thus improve the safety level of the pipeline. This embodiment pre-processes the historically accumulated image signals, then trains the target detection convolutional neural network to make decisions on the processing results, and adds surrounding environmental factors when making decisions to continuously improve the recognition accuracy.

[0152] Figure 3 FIG. 1 shows a schematic diagram of the structure of an intelligent maintenance decision-making device for underground pipelines provided by an embodiment of the present invention. Figure 3 As shown, the buried pipeline intelligent maintenance decision-making device provided by the embodiment of the present invention includes:

[0153] The acquisition module 201 is used to acquire a target magnetic stress image of the pipeline to be repaired; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline to be repaired;

[0154] The intelligent maintenance decision module 202 is used to input the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level;

[0155] Among them, the pre-trained maintenance decision model is obtained by training and testing the neural network model by taking the sample magnetic stress image with predetermined defective pixel positions and maintenance levels as the input of the model, and taking the defective pixel positions and maintenance levels in the sample magnetic stress image as the output of the model.

[0156] Since the buried pipeline intelligent maintenance decision-making device provided in this embodiment can be used to execute the buried pipeline intelligent maintenance decision-making method provided in the above embodiments, its working principle and beneficial effects are similar and will not be described in detail here.

[0157] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, see Figure 4 , the electronic device specifically includes the following contents: a processor 301, a memory 302, a communication interface 303 and a communication bus 304;

[0158] The processor 301, the memory 302, and the communication interface 303 communicate with each other through the communication bus 304; the communication interface 303 is used to realize information transmission between various devices;

[0159] The processor 301 is used to call the computer program in the memory 302. When the processor executes the computer program, all steps of the above-mentioned intelligent maintenance decision method for buried pipelines are implemented. For example, when the processor executes the computer program, the following steps are implemented: obtaining a target magnetic stress image of the pipeline to be subject to maintenance decision; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline to be subject to maintenance decision; inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes defective pixel positions and maintenance levels; wherein the pre-trained maintenance decision model is obtained by training and testing a neural network model by using a sample magnetic stress image with predetermined defective pixel positions and maintenance levels as the input of the model and the defective pixel positions and maintenance levels in the sample magnetic stress image as the output of the model.

[0160] Based on the same inventive concept, another embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, all the steps of the above-mentioned intelligent maintenance decision method for buried pipelines are implemented. For example, when the processor executes the computer program, the following steps are implemented: obtaining a target magnetic stress image of the pipeline to be subject to maintenance decision; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline to be subject to maintenance decision; inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level; wherein the pre-trained maintenance decision model is obtained by training and testing a neural network model by using a sample magnetic stress image with a predetermined defective pixel position and a maintenance level as the input of the model, and using the defective pixel position and the maintenance level in the sample magnetic stress image as the output of the model.

[0161] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0162] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention. Those of ordinary skill in the art may understand and implement them without creative effort.

[0163] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the buried pipeline intelligent maintenance decision method described in each embodiment or some parts of the embodiment.

[0164] In addition, in the present invention, terms such as "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0165] In addition, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0166] In addition, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent maintenance decision-making of buried pipelines, characterized in that: include: Obtaining a target magnetic stress image of the pipeline for which a maintenance decision is to be made; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline for which a maintenance decision is to be made; Inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level; The pre-trained maintenance decision model is obtained by training and testing a neural network model by taking a sample magnetic stress image with a predetermined defect pixel position and maintenance level as the input of the model and taking the defect pixel position and maintenance level in the sample magnetic stress image as the output of the model; It also includes: a training process of the maintenance decision model; The training process of the maintenance decision model includes: Collecting a preset number of sample magnetic stress images; wherein the preset number of sample magnetic stress images need to cover various defective pixel positions and various repair levels; According to the detection signal analysis expert, the defect pixel locations and repair levels of the preset number of sample magnetic stress images are determined by experience or excavation; The sample magnetic stress images with predetermined defect pixel positions and repair levels are used as the input of the model, the defect pixel positions and repair levels in the sample magnetic stress images are used as the output of the model, and the neural network model is trained and tested based on the intelligent learning algorithm to obtain a repair decision model; After collecting a preset number of sample magnetic stress images, the method further includes the following steps: The sample magnetic stress images are uniformly scaled, specifically including: First, count the inspection mileage X of each inspected pipe section image max , average inspection mileage Pixel length L src and the average original pixel length Average Width Round to an integer; To ensure that the detection mileage ratio of all images is unified, firstly, the pixel length L of all images is src Adjust so that the adjusted sample image length Count the average pixel value of the image length L0 Round to an integer, and then scale the image: Case 1: For detection mileage The sample magnetic stress image is scaled from the original pixel length L0 to a length of The original width W0 is scaled to The coordinates of the pixels in the scaled image are obtained by coordinate interpolation: in, is the coordinate of the pixel after transformation, L0, W0 is the length and width of the sample image; The coordinates (x, y) of the transformed pixel point belong to the pixel point coordinate set P: Among them, P ij =(x i ,y j );i=1,2,…,m,j=1,2,…,n; When the pixel point coordinates (x, y) are integers, the pixel value f(x, y) of the point after processing is the same as the pixel value of the corresponding point in the original image; when the pixel point coordinates (x, y) are not integers, the pixel value f(x, y) of the point after processing is obtained by bilinear interpolation and rounding with the pixels of the four points surrounding the point in the original image; Case 2: For detection mileage Sample magnetic stress images are copied and spliced ​​until the sum of the detected mileage of the spliced ​​images is Within the interval, execute situation 1 again; Case 3: For detection mileage The sample magnetic stress image is first split manually. The splitting principle is to ensure that the image detection mileage after splitting is within The abnormal signal area will not be split into two pictures, and the case 1 will be executed for the split pictures.

2. The intelligent maintenance decision method for buried pipelines according to claim 1 is characterized in that: The sample magnetic stress image is used as the input of the model, the defect pixel position and the maintenance level in the sample magnetic stress image are used as the output of the model, and the neural network model is trained and tested based on the intelligent learning algorithm to obtain a maintenance decision model, including: According to the statistical results of the magnetic stress images obtained historically, the ratio of the number of magnetic stress images containing defects to the number of magnetic stress images not containing defects is determined; According to the ratio of the number of pictures, the samples with determined defective pixel locations and repair levels are randomly divided into a training set and a test set to train and test the repair decision model.

3. The intelligent maintenance decision method for buried pipelines according to claim 1 is characterized in that: Before inputting the target magnetic stress image into a pre-trained maintenance decision model, the method further includes: When applying the model, for the target magnetic stress image, the image is firstly normalized in terms of the detection mileage ratio; If the mileage is checked Then scaling the image to a standard size according to the method of situation 1 as described in claim 1; If the mileage is checked Then, starting from the left end point of the image, create a The sliding window of The image is trimmed once until the right side of the sliding window reaches the boundary of the image, and we get pictures, input these pictures into the network for prediction, and output the defective pixel location and repair level based on the network prediction results.

4. The intelligent maintenance decision method for buried pipelines according to claim 1 is characterized in that: Also includes: A process of adjusting and migrating the maintenance decision model; The process of adjusting and migrating the maintenance decision model includes: After accumulating pipeline excavation data, the surrounding environmental factors that affect magnetic induction, including metal, trees and pipeline properties, are One-Hot encoded or continuously encoded to obtain the first feature vector; the feature vector corresponding to the feature map obtained by the convolutional neural network of the target area selected by the region generation network RPN is spliced ​​with the first feature vector to obtain the second feature vector; the second feature vector is calculated by the fully connected neural network layer, and then the Softmax function is used to calculate the probability L of the sample magnetic stress image belonging to each maintenance level i , use the stochastic gradient descent method SGD to train the fully connected network, the network output L i In the results, the maximum L i The most likely maintenance level for the abnormality is shown in the figure below. Among them, the probability L of the signal sample belonging to each maintenance level i The calculation method is: Among them, p i is the output of the fully connected network layer, and N represents the number of maintenance levels.

5. The intelligent maintenance decision method for buried pipelines according to claim 4 is characterized in that: Also includes: Optimizing the maintenance decision model so that the maintenance decision model outputs a defect type; Add a multi-label Sigmod loss function after the fully connected layer of the network to obtain the probability of the sample magnetic stress image being predicted as each type of defect in, is the column vector output by the last fully connected layer of the network, M is the number of defect types, Take the cross entropy loss function and use the stochastic gradient descent method SGD to train the network. The network output The defect types corresponding to all positions in the vector that are greater than the threshold; After completing the above processing, the network outputs the defective pixel location, defect type and corresponding repair level during prediction.

6. An intelligent maintenance decision-making device for buried pipelines, characterized in that: include: An acquisition module, used to acquire a target magnetic stress image of the pipeline for which a maintenance decision is to be made; wherein the target magnetic stress image is a magnetic stress detection signal image of the pipeline for which a maintenance decision is to be made; An intelligent maintenance decision module, used for inputting the target magnetic stress image into a pre-trained maintenance decision model to obtain a maintenance decision result of the target magnetic stress image; wherein the maintenance decision result includes a defective pixel position and a maintenance level; The pre-trained maintenance decision model is obtained by training and testing a neural network model by taking a sample magnetic stress image with a predetermined defect pixel position and maintenance level as the input of the model and taking the defect pixel position and maintenance level in the sample magnetic stress image as the output of the model; It also includes: a training module, used to perform a training process on the maintenance decision model; The training process of the maintenance decision model includes: Collecting a preset number of sample magnetic stress images; wherein the preset number of sample magnetic stress images need to cover various defective pixel positions and various repair levels; According to the detection signal analysis expert, the defect pixel locations and repair levels of the preset number of sample magnetic stress images are determined by experience or excavation; The sample magnetic stress images with predetermined defect pixel positions and repair levels are used as the input of the model, the defect pixel positions and repair levels in the sample magnetic stress images are used as the output of the model, and the neural network model is trained and tested based on the intelligent learning algorithm to obtain a repair decision model; After collecting a preset number of sample magnetic stress images, the training module is further used to perform the following steps: The sample magnetic stress images are uniformly scaled, specifically including: First, count the inspection mileage X of each inspected pipe section image max , average inspection mileage Pixel length L src and the average original pixel length Average Width Round to an integer; To ensure that the detection mileage ratio of all images is unified, firstly, the pixel length L of all images is src Adjust so that the adjusted sample image length Count the average pixel value of the image length L0 Round to an integer, and then scale the image: Case 1: For detection mileage The sample magnetic stress image is scaled from the original pixel length L0 to a length of The original width W0 is scaled to The coordinates of the pixels in the scaled image are obtained by coordinate interpolation: in, is the coordinate of the pixel after transformation, L0, W0 is the length and width of the sample image; The transformed pixel coordinates (x, t) belong to the pixel coordinate set P: Among them, P ij =(x i ,y j );i=1,2,…,m,j=1,2,…,n; When the pixel point coordinates (x, y) are integers, the pixel value f(x, y) of the point after processing is the same as the pixel value of the corresponding point in the original image; when the pixel point coordinates (x, y) are not integers, the pixel value f(x, y) of the point after processing is obtained by bilinear interpolation and rounding with the pixels of the four points surrounding the point in the original image; Case 2: For detection mileage Sample magnetic stress images are copied and spliced ​​until the sum of the detected mileage of the spliced ​​images is Within the interval, execute situation 1 again; Case 3: For detection mileage The sample magnetic stress image is first split manually. The splitting principle is to ensure that the image detection mileage after splitting is within The abnormal signal area will not be split into two pictures, and the case 1 will be executed for the split pictures.

Images