Pipeline integrity evaluation method and device based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based pipeline integrity assessment method and device. The method comprises: acquiring magnetic induction detection signal data from the pipeline and processing the magnetic induction detection signal data; inputting the magnetic induction detection signal data into an integrity assessment model to determine the pipeline defect location, repair level, defect type, and whether there is metal or tree interference; the integrity assessment model is pre-trained and tested. Based on the magnetic induction detection signal data, the present invention proposes an intelligent pipeline integrity assessment method that is highly accurate, adaptable, and unaffected by human subjective factors. This method effectively improves the accuracy of pipeline assessments, effectively avoids blind repairs, and significantly improves economic efficiency, reduces labor costs, and enhances pipeline safety.

Description

Technical Field

[0001] The present invention relates to the field of pipeline safety technology, and in particular to an artificial intelligence-based pipeline integrity evaluation method and device. Background Art

[0002] In-plant pipelines are the lifeblood of refineries, gathering and transportation networks, and stations, serving as the nexus between all facilities. Buried pipelines within refineries are difficult to inspect, suffer from severe corrosion, and frequently leak, resulting in significant environmental, safety, and economic losses. Currently, the primary method for excavation inspection of buried pipelines within refineries is magnetic stress testing.

[0003] Existing magnetic induction detection signal interpretation technology mainly judges the signal graph directly based on manual experience and classifies magnetic stress detection signal anomalies according to relevant grading standards. However, this method cannot determine the impact of environmental factors such as trees and metal objects on the signal, nor can it predict the type of defect. At the same time, too many subjective factors are introduced into the judgment of the severity of the signal, and the accuracy of the evaluation cannot be guaranteed. In addition, the labor cost is high and the economic efficiency is poor. Summary of the Invention

[0004] In response to the problems existing in the prior art, the embodiments of the present invention provide a pipeline integrity evaluation method and device based on artificial intelligence.

[0005] Specifically, the embodiments of the present invention provide the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a pipeline integrity assessment method based on artificial intelligence, comprising:

[0007] After acquiring the magnetic induction detection signal data of the pipeline, the magnetic induction detection signal data is processed;

[0008] Inputting the magnetic induction detection signal data into the integrity assessment model to obtain the defect location, repair level, defect type, and whether there is metal or tree interference in the pipeline;

[0009] The integrity evaluation model is obtained by processing the sample magnetic induction detection signal data of the pipeline, which has been previously determined by expert experience or excavation verification to determine whether there are abnormalities and the location of abnormal defects, defect types, maintenance levels, and whether there is interference from metal or trees, as input, and taking the starting and ending positions of the defects on the original signal data and the corresponding maintenance levels, defect types, and whether there is interference from metal or trees as output, and training and testing the initial neural network model.

[0010] Furthermore, the method further comprises: a process of processing the sample set of the integrity assessment model;

[0011] The processing of the sample set of the integrity evaluation model includes:

[0012] Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference;

[0013] The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience;

[0014] The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio.

[0015] The training set and the test set are used to train the network model using the training set and to test the network effect using the test set.

[0016] Furthermore, after collecting a preset number of sample magnetic induction detection signal data, the sample set processing process further includes the following steps:

[0017] The length normalization processing is performed on the sample magnetic induction detection signal data, specifically including:

[0018] The magnetic induction detection signal of each section of pipeline is a two-dimensional array in the form of:

[0019]

[0020] Where m is the number of sampling times, and the magnetic induction detection signal obtained at each sampling point is:

[0021]

[0022] Among them, n is the dimension of magnetic induction detection signal, h ij Indicates the value of the j-th signal feature of the magnetic induction detection signal;

[0023] The sampling frequency of the magnetic induction detection signal data is unified. The length of H represents the length of the detected pipe section. The magnetic induction detection signal data is standardized so that the length of each detection signal vector is the same, which is convenient for network training.

[0024] First, the average length of the magnetic induction detection signal data of all inspected pipe sections is counted. Round to an integer, then perform Cases 1 to 5 below:

[0025] Case 1: For array length Signal data, use interpolation to scale the array length to Specifically, first get the position of the sampling points of the new array in the original numerical coordinates:

[0026]

[0027] Where, X i It is the position information set of the new array after transformation;

[0028] According to the new location information X i , use linear interpolation to calculate the detection signal at the new position The interpolation method is:

[0029]

[0030] Where, j = [X i ] is X i The integer part of is the magnetic induction detection signal data at the corresponding position in the original data, and the new magnetic induction detection signal data H is obtained. * The length of the vector is

[0031] Case 2: For array length Signal data, for The arrays are concatenated to obtain a new signal:

[0032]

[0033] The signal length becomes N*l until Then follow the method in case 1 to scale the length to

[0034] Case 3: For array length The data without abnormal signals are divided into the opposite sides from the starting position and the end position of the array Length, get two new data, namely:

[0035]

[0036] Case 4: For array length The non-abnormal signal data H is from the array Starting point position Start cutting at length and get a length of signal until the length of the remaining part Then, according to the method of case 1-case 3, the length of the remaining part is also changed to

[0037] Case 5: For array length There is abnormal signal data H, then the abnormal region set G is expressed as

[0038] G={[g i ,g i +D i ],i=1,2,…,q}

[0039] Where q is the number of abnormal regions in H, g i is the starting point of the anomaly at the i-th location, D i is the length of the abnormal position at the i-th location. First, for each abnormal area, divide it from the start and end points of the abnormal signal to both sides. Length, if or If the interval completely contains other signal abnormal intervals, the abnormal intervals are merged and trimmed. Specifically: When , the abnormal areas are merged into the same signal and trimmed. Similarly, when When , the abnormal areas are merged into the same signal and trimmed. The length of the trimmed array is like or The interval does not completely contain other abnormal signals. From the abnormal signal [g i ,g i +D i ] are extended to both sides, and the length is interval If this interval contains the endpoint of other abnormal signals, then the endpoint is used as a clipping point and then extended to the other side to ensure that the length of the part containing the abnormal signal is To include endpoint g i-1 +D i-1 For example, get the clipping interval If the right endpoint falls within other abnormal intervals, it will shrink to the left to the point without abnormalities, and the final trimming interval [g i-1 +D i-1 ,g i+1 ], repeat Case 1-Case 2 for the final trimmed signal array; for signal, directly The signal abnormal area is cropped into a signal array, and then the remaining part is processed by executing case 1 to case 5; for the signal-free area remaining after the above cropping, case 1 to case 4 are executed cyclically.

[0040] Furthermore, the process of establishing the integrity evaluation model includes:

[0041] The normalized magnetic induction signal is regarded as a two-dimensional array, the two-dimensional array is vectorized, and the vector is extracted through the grayscale VGG convolutional neural network to obtain a feature map vector;

[0042] The feature map vector is extracted by the convolution head and then fed into the region generation network (RPN) to obtain the target region, which refers to the starting and ending mileage of the abnormal area.

[0043] The target area is used as the input layer of the subsequent network. After the RPN network, four fully connected neural networks are established. The output of the network is the probability of the pipeline defect type, maintenance level, and whether there is metal or tree interference. The maintenance level and the presence of tree or metal interference are calculated using the cross entropy loss function, and the defect type is calculated using the multi-label Sigmod loss function. The probability of the signal sample being predicted as each type of defect is obtained.

[0044]

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

[0046] Furthermore, the established network model was trained using a training set, where the model input was the normalized signal data, and the model output was the starting and ending positions of the defect on the original signal data, as well as the corresponding repair level, defect type, and the presence or absence of interference from metal and tree factors; a test set was then used to test the network effect.

[0047] Furthermore, the method includes: a process of applying the model;

[0048] The application process of the model includes:

[0049] For the magnetic induction detection signal data, firstly, the array length ratio of the magnetic induction detection signal data is unified;

[0050] If the signal length Then, the signal array is scaled to the standard size according to the methods of Case 1 and Case 2 above;

[0051] If the signal length Then, starting from the left end point of the signal array, create a Sliding window, each The signal is trimmed once until the right side of the sliding window reaches the boundary of the signal array, and the result is Signal arrays are input into the network for prediction. According to the network prediction results, the interval of the abnormal signal position on the data and the defect type, maintenance level, whether it is affected by trees, and whether it is affected by metal are output.

[0052] Furthermore, 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 network detection accuracy. In addition, after further accumulation of signal data and pipeline excavation data, the influence of the pipeline's ontological attributes on the model can be further considered. Specifically, this includes: One-hot encoding of the pipeline's ontological attribute variables, or vectorization of continuous numerical encoding, that is, statistically analyzing the pipeline attribute data factors during excavation, recording the data of these factors, and obtaining new feature vectors.

[0053] Pipeline attribute data (pipeline material, diameter, wall thickness) are classified into one-hot codes and continuous codes in the following format:

[0054]

[0055] Where m′ represents the pipe type, n′ represents the type of pipe diameter, and r represents the different thickness values ​​of the pipe. The feature map vector corresponding to the region obtained by the RPN network is concatenated with the pipe attribute vector. The resulting new vector is used as the input of the classification network. The network is reconstructed and trained according to the original loss function to further investigate the impact of pipe attributes on the detection signal and improve network performance.

[0056] In a second aspect, an embodiment of the present invention further provides an artificial intelligence-based pipeline integrity assessment device, comprising:

[0057] An acquisition module, used for acquiring magnetic induction detection signal data of the pipeline and processing the magnetic induction detection signal data;

[0058] An integrity evaluation module is used to input the magnetic induction detection signal data into an integrity evaluation model to obtain the defect location, repair level, defect type, and whether there is metal or tree interference on the pipeline;

[0059] The integrity evaluation model is obtained by processing the sample magnetic induction detection signal data of the pipeline, which has been previously determined by expert experience or excavation verification to determine whether there are abnormalities and the location, defect type, maintenance level and whether there is metal or tree interference, as input, and taking the starting and ending positions of the defects on the original signal data and the corresponding maintenance level, defect type and whether there is metal or tree interference as output, and training and testing the initial neural network model.

[0060] Furthermore, the artificial intelligence-based pipeline integrity assessment device further includes: a processing process of the sample set of the integrity assessment model;

[0061] The processing of the sample set of the integrity evaluation model includes:

[0062] Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference;

[0063] The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience;

[0064] The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio.

[0065] The training set and the test set are used to train the network model using the training set and to test the network effect using the test set.

[0066] As can be seen from the above technical solution, the artificial intelligence-based pipeline integrity assessment method provided in this embodiment first obtains magnetic induction detection signal data from the pipeline, then inputs this magnetic induction detection signal data into an integrity assessment model to obtain the pipeline's defect location, maintenance level, defect type, and presence of metal or tree interference. The integrity assessment model processes sample magnetic induction detection signal data, previously determined through expert experience or excavation verification, as a sample set, and uses it as input. The starting and ending positions of the defects on the raw signal data, as well as the corresponding maintenance level, defect type, and presence of metal or tree interference, are used as outputs. This is obtained by training and testing an initial neural network model. Based on magnetic induction detection signal data, this embodiment of the present invention employs artificial intelligence to replace the existing manual evaluation methods that are easily influenced by subjective factors. This makes the evaluation results more objective, accurate, and reliable, effectively avoiding blind repairs while significantly improving economic efficiency and reducing labor. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0068] Figure 1 This is a flow chart of a pipeline integrity assessment method based on artificial intelligence provided by one embodiment of the present invention;

[0069] Figure 2 1 is a schematic diagram of a pipeline integrity evaluation process based on magnetic induction detection signal data and a neural network according to an embodiment of the present invention;

[0070] Figure 3 1 is a schematic structural diagram of an artificial intelligence-based pipeline integrity evaluation device provided by one embodiment of the present invention;

[0071] Figure 4 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The following embodiments of the present invention are further described in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0073] Figure 1 FIG. 1 shows a flow chart of a pipeline integrity evaluation method based on artificial intelligence provided by an embodiment of the present invention, as shown in FIG. Figure 1 As shown, the pipeline integrity evaluation method based on artificial intelligence provided by the embodiment of the present invention specifically includes the following contents:

[0074] Step 101: After obtaining the magnetic induction detection signal data of the pipeline, the magnetic induction detection signal data is processed;

[0075] Step 102: Inputting the magnetic induction detection signal data into an integrity assessment model to obtain the pipeline defect location, repair level, defect type, and whether there is metal or tree interference;

[0076] Among them, the integrity evaluation model is obtained by training and testing the initial neural network model after processing the sample magnetic induction detection signal data of whether there are abnormalities and the location of abnormal pipeline defects, defect type, maintenance level, and whether there is metal or tree interference, which are determined in advance through expert experience or excavation verification, as input, and the starting and ending positions of the defects on the original signal data and the corresponding maintenance level, defect type, and whether there is interference from metal or tree factors as output.

[0077] In this embodiment, it can be understood that buried pipelines such as those in refineries, gathering and transportation networks, and stations are the link connecting various facilities within the factory. Since such pipelines cannot be inspected internally, and since they are buried underground, observation is difficult, excavation costs are high, and they are more susceptible to corrosion from the soil and other environments. With the increase in operating cycles, pipeline leakage accidents occur frequently, seriously affecting production operations and maintenance, and threatening the safety of the environment, people, and property. At present, the maintenance decision-making method for buried pipelines that cannot be inspected internally is mainly based on maintenance decisions based on non-contact detection. Currently, magnetic stress detection is one of the most applicable and universal non-contact detection methods and is widely used by enterprises. In actual on-site detection, the magnetic induction detection signal data of the pipeline can be detected by a magnetic induction detection tool. In this embodiment, the magnetic induction detection signal data is a two-dimensional array signal obtained by magnetic induction detection.

[0078] In this embodiment, the defect location refers to the mileage corresponding to the defect on the pipeline or the abnormal interval corresponding to the defect on the pipeline, the starting position and the ending position of the defect on the original signal data (i.e., the number of columns in the two-dimensional array), etc. The maintenance level refers to the severity of the damage to the pipeline, such as immediate repair, planned repair, monitoring and use, etc. The defect type refers to the type of defect corresponding to the abnormal area on the pipeline, such as dents, metal loss, weld abnormalities, etc. In addition, since the magnetic induction signal is more sensitive to the surrounding environment of the buried pipeline (such as trees, metal, wires, and the properties of the pipeline itself), further obtaining whether there is metal or tree interference around the pipeline can more reasonably or referentially determine whether the defect type, defect location and maintenance level in the evaluation results are accurate enough, thereby making the entire evaluation more objective and reasonable.

[0079] In this embodiment, it should be noted that this embodiment uses magnetic induction detection signal data and proposes a pipeline integrity intelligent evaluation method that is highly adaptable and not affected by human subjective factors based on the magnetic induction detection signal data, thereby improving the accuracy of the analysis of the detection results, avoiding blind repairs, greatly improving economy and reducing labor, and at the same time improving the safety of the pipeline.

[0080] In this embodiment, it should be noted that the pipeline inspection signal data is generally a data pattern of signal strength corresponding to pipeline mileage. This data is usually displayed and analyzed in the form of a signal graph. For example, based on the obtained magnetic stress signal intensity value, a corresponding magnetic stress signal image can be drawn, and then an intelligent algorithm can be used to predict the severity of pipeline defects based on the signal image. However, since digital signals lose some precision when converted into images, the accuracy of the algorithm has certain limitations. Processing and calculating the image requires a large amount of computing resources, resulting in slow calculation speed and very high requirements for computer hardware. Therefore, this embodiment proposes a pipeline integrity intelligent evaluation method that directly targets data signals. This method is not only applicable to the intelligent evaluation of pipeline integrity based on signal data from non-contact pipeline inspection, such as magnetic stress inspection, but can also be used for direct intelligent evaluation of other types of inspection data. On the one hand, it has a wide range of applicability. On the other hand, while improving the accuracy and efficiency of pipeline integrity evaluation, it also reduces labor costs, improves economic benefits and pipeline safety, and fills the technical gap in existing pipeline integrity intelligent evaluation.

[0081] As can be seen from the above technical solution, the artificial intelligence-based pipeline integrity assessment method provided in this embodiment first obtains magnetic induction detection signal data from the pipeline, then inputs this magnetic induction detection signal data into an integrity assessment model to obtain the pipeline's defect location, maintenance level, defect type, and presence of metal or tree interference. The integrity assessment model processes sample magnetic induction detection signal data, previously determined through expert experience or excavation verification, as a sample set, and uses it as input. The starting and ending positions of the defects on the raw signal data, as well as the corresponding maintenance level, defect type, and presence of metal or tree interference, are used as outputs. This is obtained by training and testing an initial neural network model. Based on magnetic induction detection signal data, this embodiment of the present invention employs artificial intelligence to replace the existing manual evaluation methods that are easily influenced by subjective factors. This makes the evaluation results more objective, accurate, and reliable, effectively avoiding blind repairs while significantly improving economic efficiency and reducing labor.

[0082] Based on the content of the above embodiment, in this embodiment, the pipeline integrity evaluation method based on artificial intelligence further includes: a processing process of the sample set of the integrity evaluation model;

[0083] The processing of the sample set of the integrity evaluation model includes:

[0084] Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference;

[0085] The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience;

[0086] The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio.

[0087] The training set and the test set are used to train the network model using the training set and to test the network effect using the test set.

[0088] In this embodiment, the preset number of sample magnetic induction detection signal data is required to cover various defect locations, defect types, and repair levels, as well as situations with and without tree and metal interference. This allows the integrity assessment model obtained after model training using the preset number of sample magnetic induction detection signal data to accurately identify the defect location, defect type, and repair level, as well as the presence of tree and metal interference, in the pipeline magnetic induction detection signal data. Furthermore, this embodiment labels the repair level, defect location, defect type, and presence of tree and metal interference of abnormal points in the sample magnetic induction detection signal data as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio. This ensures a more accurate and reasonable training process, while also making the testing process more effective.

[0089] Based on the content of the above embodiment, in this embodiment, after collecting a preset number of sample magnetic induction detection signal data, the method further includes the following steps:

[0090] The length normalization processing is performed on the sample magnetic induction detection signal data, specifically including:

[0091] The magnetic induction detection signal of each pipeline section is usually a two-dimensional array in the form of:

[0092]

[0093] Where m is the number of sampling times, and the magnetic induction detection signal obtained at each sampling point is:

[0094]

[0095] Among them, n is the dimension of magnetic induction detection signal, h ij Indicates the value of the j-th signal feature of the magnetic induction detection signal;

[0096] The sampling frequency of the magnetic induction detection signal data is unified. The length of H represents the length of the detected pipe section. The magnetic induction detection signal data is standardized so that the length of each detection signal vector is the same, which is convenient for network training.

[0097] First, the average length of the magnetic induction detection signal data of all inspected pipe sections is counted. Round to an integer, then perform Cases 1 to 5 below:

[0098] Case 1: For array length Signal data, use interpolation to scale the array length to Specifically, first get the position of the sampling points of the new array in the original numerical coordinates:

[0099]

[0100] Where, X i It is the position information set of the new array after transformation;

[0101] According to the new location information X i , use linear interpolation to calculate the detection signal at the new position The interpolation method is:

[0102]

[0103] Where, j = [X i ] is X i The integer part of is the magnetic induction detection signal data at the corresponding position in the original data, and the new magnetic induction detection signal data H is obtained. * The length of the vector is

[0104] Case 2: For array length Signal data, for The arrays are concatenated to obtain a new signal:

[0105]

[0106] The signal length becomes N*l until Then follow the method in case 1 to scale the length to

[0107] Case 3: For array length The data without abnormal signals are divided into the opposite sides from the starting position and the end position of the array Length, get two new data, namely:

[0108]

[0109] Case 4: For array length The non-abnormal signal data H is from the array Starting point position Start cutting at length and get a length of signal until the length of the remaining part Then, according to the method of case 1-case 3, the length of the remaining part is also changed to

[0110] Case 5: For array length There is abnormal signal data H, then the abnormal region set G is expressed as

[0111] G={[g i ,g i +D i ],i=1,2,…,q}

[0112] Where q is the number of abnormal regions in H, g i is the starting point of the anomaly at the i-th location, D i is the length of the abnormal position at the i-th location. First, for each abnormal area, divide it from the start and end points of the abnormal signal to both sides. Length, if or If the interval completely contains other signal abnormal intervals, the abnormal intervals are merged and trimmed. Specifically: When , the abnormal areas are merged into the same signal and trimmed. Similarly, when When , the abnormal areas are merged into the same signal and trimmed. The length of the trimmed array is like or The interval does not completely contain other abnormal signals. From the abnormal signal [g i ,g i +D i ] are extended to both sides, and the length is interval If this interval contains the endpoint of other abnormal signals, then the endpoint is used as a clipping point and then extended to the other side to ensure that the length of the part containing the abnormal signal is To include endpoint g i-1 +D i-1 For example, get the clipping interval If the right endpoint falls within other abnormal intervals, it will shrink to the left to the point without abnormalities, and the final trimming interval [g i-1 +D i-1 ,g i+1 ], repeat Case 1-Case 2 for the final trimmed signal array; for signal, directly The signal abnormal area is cropped into a signal array, and then the remaining part is processed by executing case 1 to case 5; for the signal-free area remaining after the above cropping, case 1 to case 4 are executed cyclically.

[0113] In this embodiment, the magnetic induction detection signal data is uniformly scaled based on the relationship between the array length of the magnetic induction detection signal data and the average array length of the inspected pipe section. This can solve the problem that the horizontal axis scale of the magnetic induction detection signal data obtained each time is different due to the different mileage measured each time, which in turn makes it impossible to directly use the obtained magnetic induction detection signal data for intelligent identification of pipeline maintenance levels based on neural networks.

[0114] The following combination Figure 2 Introduce the model network establishment and optimization process:

[0115] 1. Establishment of model network

[0116] (1) Sample set processing

[0117] After normalization, the abnormal magnetic stress signal data is labeled based on the location of the pipeline anomaly, determined by signal analysis experts based on experience or excavation, the repair level for the abnormal signal (magnetic induction abnormal signal repair levels are divided into three categories: immediate repair, planned repair, and monitoring and use), the defect type (such as dents, metal loss, and weld anomalies), the presence of tree interference (yes or no), and the presence of metal interference (yes or no). The ratio of abnormal to normal signals is calculated, and the existing data is randomly divided into training and test sets based on this ratio.

[0118] (2) Network structure establishment

[0119] The normalized magnetic induction signal is regarded as a two-dimensional array, the two-dimensional array is vectorized, and the vector is passed through the grayscale VGG convolutional neural network to extract features and obtain a feature map vector.

[0120] After being extracted by the convolutional head, the feature map vector is fed into the RPN (Region Proposal Network) to obtain the target region (the starting and ending mileage of the abnormal region). The target region serves as the input layer of the subsequent network. After the RPN network, four fully connected neural networks are established. The network outputs are the probability of pipeline defect type, maintenance level, and whether there is metal interference, tree interference, and other fault types. The maintenance level and the presence of tree and metal interference use the cross entropy loss function, and the defect type uses the multi-label Sigmod loss function to obtain the probability of the signal sample being predicted as each type of defect.

[0121]

[0122] in, is the column vector output by the last fully connected layer of the network, M is the number of defect types, M is the number of defect types. To reduce the loss function, SGD (Stochastic Gradient Descent) is used to update network parameters and train the object detection network. To speed up network training, anchor selection can be based on the characteristics of the magnetic induction signal, choosing only one anchor size with an aspect ratio of 1:2.

[0123] (3) Model training

[0124] The network model was trained using the training set. The model input was normalized signal data, and the model output was the starting and ending locations of the defects in the original signal data (i.e., the number of columns in a two-dimensional array), along with the corresponding repair level, defect type, and the presence of metal and tree interference. The network's performance was then tested using the test set.

[0125] (4) Model application

[0126] When applying the network model, for a magnetic induction signal, if the signal length Then scale the signal array to the standard size according to the method in Part 1; if the signal length Ensure that the potential fault area can be fully detected by the network at least once and reduce 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 Sliding window, each The signal is trimmed once until the right side of the sliding window reaches the boundary of the signal array, and the result is (Round down) signal arrays. These signal arrays are input into the network for prediction. Based on the network prediction results, the interval of the abnormal signal location in the data, the defect type of the abnormal signal, the repair level, and whether it is affected by trees or metal are output.

[0127] 2. Model Optimization

[0128] After further data collection and accumulation, in order to improve the accuracy of the network, the newly added data will be substituted into the training module for tuning to further improve the network detection accuracy. In addition, after further accumulation of signal data and pipeline excavation data, the impact of the pipeline's ontological attributes on the model can be further considered. Specifically, this includes: One-hot encoding of pipeline ontological attribute variables, or vectorization of continuous numerical encoding, that is, statistically analyzing pipeline attribute data factors during excavation, recording the data of these factors, and obtaining new feature vectors;

[0129] Pipeline attribute data (pipeline material, diameter, wall thickness) are classified into one-hot codes and continuous codes in the following format:

[0130]

[0131] Where m′ represents the pipe type, n′ represents the type of pipe diameter, and r represents the different thickness values ​​of the pipe. The feature map vector corresponding to the region obtained by the RPN network is concatenated with the pipe attribute vector. The resulting new vector is used as the input of the classification network. The network is reconstructed and trained according to the original loss function to further investigate the impact of pipe attributes on the detection signal and improve network performance.

[0132] Figure 3 FIG. 1 shows a schematic diagram of a pipeline integrity evaluation device based on artificial intelligence according to an embodiment of the present invention. Figure 3 As shown, the pipeline integrity evaluation device based on artificial intelligence provided by the embodiment of the present invention includes:

[0133] An acquisition module 201 is used to acquire magnetic induction detection signal data of a pipeline;

[0134] The integrity evaluation module 202 is used to input the magnetic induction detection signal data into the integrity evaluation model to obtain the defect location, repair level, defect type and whether there is metal or tree interference on the pipeline;

[0135] Among them, the integrity evaluation model is obtained by processing sample magnetic induction detection signal data such as whether there are abnormalities and the location of abnormal pipeline defects, defect type, maintenance level, and whether there is metal or tree interference, which are determined in advance through expert experience or excavation verification, as input, and the starting and ending positions of the defects on the original signal data and the corresponding maintenance level, defect type, and whether there is metal or tree interference as output, and training and testing the initial neural network model.

[0136] Based on the content of the above embodiment, in this embodiment, the pipeline integrity evaluation device based on artificial intelligence further includes: a processing process of the sample set of the integrity evaluation model;

[0137] The processing of the sample set of the integrity evaluation model includes:

[0138] Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference;

[0139] The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience;

[0140] The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio.

[0141] The training set and the test set are used to train the network model using the training set and to test the network effect using the test set.

[0142] Since the artificial intelligence-based pipeline integrity evaluation device provided in this embodiment can be used to execute the artificial intelligence-based pipeline integrity evaluation method provided in the above embodiment, its working principle and beneficial effects are similar and will not be described in detail here.

[0143] 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;

[0144] The processor 301, memory 302, and communication interface 303 communicate with each other via the communication bus 304; the communication interface 303 is used to implement information transmission between devices;

[0145] 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 pipeline integrity evaluation method based on artificial intelligence are implemented. For example, when the processor executes the computer program, the following steps are implemented: after obtaining the magnetic induction detection signal data of the pipeline, the magnetic induction detection signal data is processed; the magnetic induction detection signal data is input into the integrity evaluation model to obtain the defect location, maintenance level, defect type, and whether there is metal or tree interference of the pipeline; wherein the integrity evaluation model is obtained by processing sample magnetic induction detection signal data such as whether there is an abnormality and the location, defect type, maintenance level, and whether there is metal or tree interference of the pipeline abnormal defect, which are previously determined through expert experience or excavation verification, as a sample set as input, and the starting position and ending position of the defect on the original signal data and the corresponding maintenance level, defect type, and whether there is metal or tree interference as output, and the initial neural network model is trained and tested.

[0146] Based on the same inventive concept, another embodiment of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, all steps of the above-mentioned artificial intelligence-based pipeline integrity evaluation method are implemented. For example, when the processor executes the computer program, the following steps are implemented: obtaining magnetic induction detection signal data of the pipeline and processing the magnetic induction detection signal data; inputting the magnetic induction detection signal data into an integrity evaluation model to obtain the defect location, maintenance level, defect type, and whether there is metal or tree interference of the pipeline; wherein the integrity evaluation model is obtained by processing sample magnetic induction detection signal data of whether there is an abnormality and the abnormal defect location, defect type, maintenance level, and whether there is metal or tree interference of the pipeline, which are previously determined through expert experience or excavation verification, as a sample set as input, and taking the starting position and ending position of the defect on the original signal data and the corresponding maintenance level, defect type, and whether there is metal or tree interference as output, and training and testing the initial neural network model.

[0147] In addition, the logical instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, 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. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute 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.

[0148] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the embodiments of the present invention. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0149] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the artificial intelligence-based pipeline integrity assessment method described in various embodiments or certain portions of the embodiments.

[0150] Furthermore, in the present invention, terms such as "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0151] In addition, in the present invention, relational terms such as first and second, etc. are used only 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 "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the statement "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0152] In addition, in the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean 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 can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0153] 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 various embodiments of the present invention.

Claims

1. A pipeline integrity evaluation method based on artificial intelligence, characterized in that: include: After acquiring the magnetic induction detection signal data of the pipeline, the magnetic induction detection signal data is processed; Inputting the magnetic induction detection signal data into the integrity assessment model to obtain the defect location, repair level, defect type, and whether there is metal or tree interference in the pipeline; The integrity assessment model is obtained by training and testing an initial neural network model using magnetic induction detection signal data, which is pre-determined through expert experience or excavation verification to determine whether there are abnormalities and the location, type, repair level, and presence of metal or tree interference, as input. The starting and ending positions of the defects on the original signal data, as well as the corresponding repair level, defect type, and presence of metal or tree interference, are processed as output. It also includes: a process of processing the sample set of the integrity evaluation model; The processing of the sample set of the integrity evaluation model includes: Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference; The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience; The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio. The training set and the test set are used to train the network model using the training set and to test the network effect using the test set; After collecting a preset number of sample magnetic induction detection signal data, the sample set processing process further includes the following steps: The length normalization processing is performed on the sample magnetic induction detection signal data, specifically including: The magnetic induction detection signal of each section of pipeline is a two-dimensional array in the form of: ； in, is the sampling number, and the magnetic induction detection signal obtained at each sampling point is: ； in, is the magnetic induction detection signal dimension, Indicates the magnetic induction detection signal The numerical value of a signal feature.

2. The pipeline integrity evaluation method based on artificial intelligence according to claim 1 is characterized in that: The sampling frequency of the magnetic induction detection signal data is unified. The length of represents the length of the detected pipe section. The magnetic induction detection signal data is standardized so that the length of each detection signal vector is the same for network training. First, the average length of the magnetic induction detection signal data of all inspected pipe sections is counted. , round to an integer, and then perform the following cases 1 to 5: Case 1: For array length Signal data, use interpolation to scale the array length to Specifically, first get the position of the sampling points of the new array in the original numerical coordinates: ； Where, It is the position information set of the new array after transformation; Based on the new location information , use linear interpolation to calculate the detection signal at the new position , the interpolation method is: ； Where, for The integer part of is the magnetic induction detection signal data of the corresponding position in the original data, and the new magnetic induction detection signal data is obtained The length of the vector is ; Case 2: For array length Signal data, for The arrays are concatenated to obtain a new signal: ； The signal length becomes ,until , and then scale the length to ; Case 3: For array length The data without abnormal signals is divided into two new data from the starting position and the end position of the array to the opposite side, namely: ， ； Case 4: For array length No abnormal signal data , from the array Starting point position Start cutting at length and get a length of signal until the length of the remaining part Then, according to the method of case 1-case 3, the length of the remaining part is also changed to ; Case 5: For array length Abnormal signal data , then the abnormal region set G is expressed as ； Where, for The number of abnormal regions in For the At an unusual starting point, For the The length of the abnormal position, first, for each abnormal area, divide it from the abnormal signal starting point to both sides Length, if or If the interval completely contains other signal abnormal intervals, the abnormal intervals are merged and trimmed. Specifically: When , the abnormal areas are merged into the same signal and trimmed. Similarly, when When , the abnormal areas are merged into the same signal and trimmed. The length of the trimmed array is ;like or The interval does not completely contain other abnormal signals. The midpoint of the two sides is extended to obtain a length of interval If this interval contains the endpoint of other abnormal signals, then the endpoint is used as a clipping point and then extended to the other side to ensure that the length of the part containing the abnormal signal is , for the clipping interval If the right endpoint falls within other abnormal intervals, it will shrink to the left to the point without abnormalities to obtain the final trimming interval. , repeat Case 1-Case 2 for the final trimmed signal array; for signal, directly The signal abnormal area is cropped into a signal array, and then the remaining part is processed by executing case 1 to case 5; for the signal-free area remaining after the above cropping, case 1 to case 4 are executed cyclically.

3. The pipeline integrity evaluation method based on artificial intelligence according to claim 1 is characterized in that: The process of establishing the integrity evaluation model includes: The normalized magnetic induction signal is regarded as a two-dimensional array, the two-dimensional array is vectorized, and the vector is extracted through the grayscale VGG convolutional neural network to obtain a feature map vector; The feature map vector is extracted by the convolution head and then fed into the region generation network (RPN) to obtain the target region, which refers to the starting and ending mileage of the abnormal area. The target area is used as the input layer of the subsequent network. After the RPN network, four fully connected neural networks are established. The output of the network is the probability of the pipeline defect type, maintenance level, and whether there is metal or tree interference. The maintenance level and the presence of tree or metal interference are calculated using the cross entropy loss function, and the defect type is calculated using the multi-label Sigmod loss function. The probability of the signal sample being predicted as each type of defect is obtained. , ； in, is the column vector output by the last fully connected layer of the network, , M is the number of defect types.

4. The pipeline integrity evaluation method based on artificial intelligence according to claim 1 is characterized in that: The network model is established by training with the training set, where the input of the model is the normalized signal data, and the output of the model is the starting and ending positions of the defect on the original signal data, as well as the corresponding repair level, defect type, and the presence or absence of metal or tree interference; the test set is then used to test the network effect.

5. The pipeline integrity evaluation method based on artificial intelligence according to claim 2 is characterized in that: The method includes: a process of applying the model; The application process of the model includes: For the magnetic induction detection signal data, firstly, the array length ratio of the magnetic induction detection signal data is unified; If the signal length , then the signal array is scaled to a standard size according to the method of Case 1 and Case 2 as described in claim 2; If the signal length , then starting from the left end point of the signal array, a sliding window of length is established, and each The signal is trimmed once until the right side of the sliding window reaches the boundary of the signal array, and the result is Signal arrays are input into the network for prediction. According to the network prediction results, the interval of the abnormal signal position on the data and the defect type, maintenance level, whether it is affected by trees, and whether it is affected by metal are output.

6. The pipeline integrity evaluation method based on artificial intelligence according to claim 3 is characterized in that: 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 network detection accuracy; also, After further accumulation of signal data and pipeline excavation data, the impact of pipeline ontology attributes on the model is further considered. Specifically, the pipeline ontology attribute variables are one-hot encoded or vectorized using continuous numerical encoding. That is, during excavation, the pipeline attribute data factors are counted and the data of these factors are recorded to obtain new feature vectors. Pipeline attribute data, including pipe material, diameter, and wall thickness, are classified into one-hot codes and continuous codes in the following format: ； in Represents the pipe type, Represents the type of pipe diameter, Representing different thickness values ​​of the pipeline; the feature map vector corresponding to the area obtained by the RPN network is spliced ​​with the pipeline attribute vector, and the obtained new vector is used as the input of the classification network. The network is reconstructed and trained according to the original loss function to further investigate the impact of pipeline attributes on the detection signal and improve the network effect.

7. A pipeline integrity evaluation device based on artificial intelligence, characterized in that: include: An acquisition module, used for acquiring magnetic induction detection signal data of the pipeline and processing the magnetic induction detection signal data; An integrity evaluation module is used to input the magnetic induction detection signal data into an integrity evaluation model to obtain the defect location, repair level, defect type, and whether there is metal or tree interference on the pipeline; The integrity assessment model is obtained by training and testing an initial neural network model using sample magnetic induction detection signal data, which has been previously determined through expert experience or excavation verification to determine whether there are abnormalities and the location, type, repair level, and presence of metal or tree interference, as input, and using the starting and ending positions of the defects on the original signal data, as well as the corresponding repair level, defect type, and presence of metal or tree interference, as output; It also includes: a process of processing the sample set of the integrity evaluation model; The processing of the sample set of the integrity evaluation model includes: Collecting a preset number of sample magnetic induction detection signal data; wherein the abnormal signals of the preset number of sample magnetic induction detection signal data need to cover various defect locations, defect types and repair levels, and need to cover situations with and without tree and metal interference; The defect location, defect type, repair level and presence of tree or metal interference of abnormal signals in the preset number of sample magnetic induction detection signal data are determined based on excavation verification or expert experience; The maintenance level, defect location, defect type, and presence of tree or metal interference of abnormal points in the sample magnetic induction detection signal data are marked as sample labels. The ratio of abnormal to normal signals in the sample set is calculated, and the existing sample set data is randomly divided into training and test sets based on this ratio. The training set and the test set are used to train the network model using the training set and to test the network effect using the test set; After collecting a preset number of sample magnetic induction detection signal data, the sample set processing process further includes the following steps: The length normalization processing is performed on the sample magnetic induction detection signal data, specifically including: The magnetic induction detection signal of each section of pipeline is a two-dimensional array in the form of: ； in, is the sampling number, and the magnetic induction detection signal obtained at each sampling point is: ； in, is the magnetic induction detection signal dimension, Indicates the magnetic induction detection signal The numerical value of a signal feature.

Images