Underground pipeline damage identification and positioning method and system based on feature hierarchical model

Abstract

The present invention discloses a method and system for identifying and locating underground pipeline damage based on a feature hierarchy model, and the method includes: S1, arranging a plurality of Raspberry Pi ultrasonic detectors along the direction of the underground pipeline, and constructing ultrasonic pipe wall damage sample data containing labels; S2, constructing a sample feature hierarchy label learning model to extract features from the ultrasonic pipe wall damage sample data to obtain a label feature vector and a weight coefficient vector through an attention mechanism module; S3, relying on the Raspberry Pi ultrasonic detectors arranged on the underground pipeline wall to collect underground pipeline data at the sampling point, input the sample feature hierarchy label learning model, and output the damage identification result, and use the ultrasonic detector to calculate the location information of the sampling point and store it in association with the damage identification result, and filter and output it. The present invention realizes the output of the sampling point positioning position information and the sampling point damage identification association data, which is convenient for management personnel to efficiently record and maintain work.

Description

Technical Field

[0001] The present invention relates to the field of nondestructive testing of underground pipelines, and in particular to a method and system for identifying and locating underground pipeline damage based on a feature hierarchy model. Background Art

[0002] With the rapid advancement of urbanization, the underground pipeline system, as the "neural network" of urban infrastructure, carries key functions such as water supply, drainage, gas, electricity, and communications. Its stability and safety are directly related to the normal operation of the city and the quality of life of residents. However, since most underground pipelines are buried underground, with diverse materials and complex layouts, their maintenance and management face many challenges. In particular, under the influence of factors such as pipeline aging, corrosion, and external force damage, pipeline damage occurs from time to time. If the damaged position cannot be discovered and accurately located in time, it may cause serious safety accidents and economic losses.

[0003] At present, a variety of technical means have been developed in the field of underground pipeline damage identification and positioning, such as electromagnetic method, ground penetrating radar, infrared thermal imaging, etc. These methods have their own advantages and disadvantages, but they generally have some limitations; for example, although the electromagnetic method is good for locating metal pipelines, it is not effective for non-metallic pipelines or complex geological environments; although ground penetrating radar can penetrate a certain thickness of the covering layer, its resolution and depth are subject to certain restrictions, and its ability to identify minor damage is limited; infrared thermal imaging is limited by the ambient temperature and pipeline material, and its application range is relatively narrow.

[0004] As a non-destructive testing technology, ultrasonic technology has shown unique advantages in the detection of internal defects in materials. When ultrasonic waves propagate in materials, they will reflect, refract, and other phenomena when encountering defects (such as cracks, holes, etc.). By measuring the changes in parameters such as the propagation time, frequency, and amplitude of ultrasonic waves, it is possible to accurately identify and locate internal defects in materials. However, in the field of underground pipeline damage identification and positioning, traditional ultrasonic detection methods are often limited by the complexity and difficulty of detection equipment, making it difficult to achieve efficient and accurate automated detection, and the data processing capabilities are weak, making it difficult to process and analyze a large amount of detection data in real time and accurately. As a small, high-performance single-board computer, Raspberry Pi has powerful data processing capabilities and flexible configuration methods, and can realize real-time collection, processing, and analysis of ultrasonic signals. Therefore, the detection, identification, and positioning of underground pipeline damage are still technical difficulties in existing technologies. How to use Raspberry Pi ultrasonic waves and combine feature extraction and recognition to achieve underground pipeline damage identification and positioning is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problems pointed out by the background technology, and to provide a method and system for identifying and locating underground pipeline damage based on a feature hierarchical model. In combination with a Raspberry Pi ultrasonic detector, the damage of underground pipelines can be identified without prior knowledge and the damage conditions can be hierarchically classified. The sample feature hierarchical label learning model also performs feature hierarchical extraction accordingly, which improves the classification efficiency of underground pipelines, reduces recognition errors, and provides decision-making references for the maintenance and management of underground pipelines.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for identifying and locating underground pipeline damage based on a feature hierarchy model, the method comprising:

[0008] S1. Arrange several Raspberry Pi ultrasonic detectors along the underground pipelines and on the underground pipeline walls, prepare damaged and healthy samples of the underground pipeline walls, and collect ultrasonic wall damaged sample data containing labels by transmitting waves and echoes from ultrasonic detectors according to the underground pipelines. The labels include damaged labels and healthy labels.

[0009] S2. Construct a sample feature hierarchical label learning model to extract features from ultrasonic pipe wall damage sample data to obtain label feature vector v ij , V ij Represents the feature vector of label i and feature j, and then adds the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij It represents the weight coefficient of label i and feature j after normalization according to label i; the sample feature hierarchical label learning model inputs the ultrasonic pipe wall damage sample data for model iterative training;

[0010] S3. Relying on the Raspberry Pi ultrasonic detectors arranged on the walls of underground pipelines, the underground pipeline data of the sampling points are collected and input into the sample feature hierarchical label learning model. Then, the damage recognition results of the sampling points are output. The ultrasonic detectors are used to calculate the location information of the sampling points and store the data in association with the damage recognition results. Then, the data with the damage recognition results as damaged are screened out for associated output.

[0011] In order to better implement the present invention, the method for ultrasonic detector to calculate the sampling point position information is as follows: locate the position information of the ultrasonic detector, the position information of the underground pipeline wall sampling point is the relative position of the sampling point and the ultrasonic detector, the horizontal distance of the underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, the height of the underground pipeline wall sampling point from the ground is h = S × sinθ, Where S represents the distance from the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector transmitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave transmitting direction and the plumb line.

[0012] In order to better implement the present invention, a number of ultrasonic detectors arranged in the direction of the underground pipeline are used to collect underground pipeline data at each sampling point, and the position information of each sampling point and the corresponding damage identification result are obtained; or the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data at each sampling point on the underground pipeline wall in turn, and the position information of each sampling point and the corresponding damage identification result are obtained.

[0013] Preferably, step S3 also includes the following method:

[0014] A position depth coordinate system is constructed, with the ultrasonic detector at the starting point as the origin and the direction of the underground pipeline as the x-axis; the ultrasonic detector at the starting point collects underground pipeline data from the sampling points on the wall of the underground pipeline, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result, the position information includes the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data in sequence from the sampling points on the wall of the underground pipeline, and obtains the position information of each sampling point and the corresponding damage identification result, and obtains complete damage identification and positioning data of the underground pipeline wall, the damage identification and positioning data includes the position information of the sampling point and the corresponding damage identification result.

[0015] Preferably, positioning points whose damage identification results are damage types are screened out from the damage identification and positioning data, and the screened damage identification and positioning data are output.

[0016] Preferably, in step S2, the i-th tag contains L j features, then the i-th label feature attention data expression is as follows:

[0017] Preferably, the features include multi-level key features including time interval features, amplitude features, frequency features, angle features, and propagation speed features between the transmitted wave and the echo; and the attention mechanism module is allocated according to the correlation between the features and the label results.

[0018] Preferably, the damage identification results or labels include healthy and undamaged, slightly damaged, and severely damaged.

[0019] An ultrasonic identification and positioning system for underground pipeline damage based on a feature hierarchical learning model comprises an ultrasonic pipe wall damage sample database, an ultrasonic detection system and a sample feature hierarchical label learning model, wherein the sample feature hierarchical label learning model comprises an attention mechanism module, the ultrasonic detection system comprises a plurality of Raspberry Pi ultrasonic detectors arranged along the direction of the underground pipeline for the underground pipeline wall, the ultrasonic pipe wall damage sample database stores ultrasonic pipe wall damage sample data containing labels, the ultrasonic pipe wall damage sample data is data collected by arranging underground pipeline wall damage samples and healthy samples according to the underground pipeline and transmitting waves and echoes of ultrasonic detectors, and the labels comprise damage labels and healthy labels; the sample feature hierarchical label learning model performs feature extraction to obtain a label feature vector v ij , v ij The feature vector representing label i and feature j is then passed through the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij It represents the weight coefficient of label i and feature j after normalization according to label i; the sample feature hierarchical label learning model inputs the ultrasonic pipe wall damage sample data for model iterative training; the Raspberry Pi ultrasonic detector arranged on the underground pipeline wall is used to detect the underground pipeline data of the sampling point and input it into the sample feature hierarchical label learning model, and the damage recognition result of the sampling point is output; the location information of the sampling point is calculated by the ultrasonic detector and stored in association with the damage recognition result, and then the data with the damage recognition result is screened out for association output.

[0020] In order to better realize the ultrasonic identification and positioning system for underground pipeline damage of the present invention, the method for the ultrasonic detector to calculate the sampling point position information is as follows: the position information of the positioning ultrasonic detector, the position information of the underground pipeline wall sampling point is the relative position of the sampling point and the ultrasonic detector, the horizontal distance of the underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, the height of the underground pipeline wall sampling point from the ground is h = S × sinθ, Wherein, S represents the distance measured by the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector emitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave emission direction and the plumb line; the sample feature hierarchical label learning model also includes a position depth coordinate system with the ultrasonic detector at the starting point as the origin and the direction of the underground pipeline as the x-axis. The ultrasonic detector at the starting point collects underground pipeline data from the sampling point of the underground pipeline wall, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result. The position information includes the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data from the sampling points of the underground pipeline wall in turn, and the position information of each sampling point and the corresponding damage identification result are obtained, so as to obtain complete damage identification and positioning data of the underground pipeline wall. The damage identification and positioning data includes the position information of the sampling point and the corresponding damage identification result.

[0021] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0022] (1) The present invention provides a method for identifying and locating underground pipeline damage. Combined with a Raspberry Pi ultrasonic detector, the method can identify damage to underground pipelines without prior knowledge and classify the damage conditions in a hierarchical manner. The sample feature hierarchical label learning model also performs feature hierarchical extraction accordingly, thereby improving the classification efficiency of underground pipelines, reducing recognition errors, and providing decision-making references for the maintenance and management of underground pipelines.

[0023] (2) The present invention constructs a sample feature hierarchical label learning model, and constructs a relationship matrix between label results and feature hierarchies by inputting sample data for model training. The relationship matrix includes weight coefficients. The deployed Raspberry Pi ultrasonic detector collects data of transmitted waves and echoes from the sampling points on the wall of the underground pipeline. The sample feature hierarchical label learning model is used to identify whether the sampling point is damaged. Then, based on the position information of the ultrasonic detector, the position information of the sampling point relative to the ultrasonic detector is calculated. At the same time, the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point on the wall of the underground pipeline from the ground (i.e., the depth of the sampling point on the wall of the underground pipeline) are obtained, so as to output the associated data of the sampling point positioning information and the sampling point damage identification.

[0024] (3) The present invention can adjust the detection angle of the ultrasonic detector or move the ultrasonic detector to detect data with different sampling. When the ultrasonic detector is moved, a position depth coordinate system is constructed with the ultrasonic detector at the starting position as the origin and the direction of the underground pipeline as the x-axis. In this way, the position information of different sampling points can be collected and damage identified, and then expressed in the position depth coordinate system to achieve accurate data output, which is convenient for management personnel to efficiently record and maintain underground pipelines.

[0025] (4) The ultrasonic detector of the Raspberry Pi of the present invention can be flexibly configured and developed according to different underground pipeline scenarios and actual engineering needs. Its high-performance processor and rich interfaces can be expanded to realize remote monitoring and transportation, thereby improving the efficiency of non-destructive testing of underground pipelines and the damage identification effect. The sample feature hierarchical label learning model of the present invention adds an attention mechanism module, which focuses on key information from the classified multiple labeled samples and extracts key features through the contribution weights of each feature, thereby reducing the complexity of model training, improving the accuracy of recognition and the generalization of the model, making the model more practical and reliable, and performing positioning calculation functions according to the damage situation and degree. The results are more intuitive and easy to operate, which is convenient for managers to efficiently record and maintain underground pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the method flow of the underground pipeline damage identification and positioning method of the present invention;

[0027] Figure 2 This is a principle structure diagram of the underground pipeline damage identification and positioning system of the present invention;

[0028] Figure 3 This is a schematic diagram of the principle of the Raspberry Pi ultrasonic detector in the embodiment;

[0029] Figure 4 A schematic diagram of the principle structure of the ultrasonic detector of the Raspberry Pi in the embodiment;

[0030] Figure 5 This is a schematic diagram of the principle of obtaining features by performing similarity analysis based on sample data as an example of a sample feature hierarchical label learning model in an embodiment;

[0031] Figure 6 It is a schematic diagram of calculating the sampling point position using the ultrasonic detector position information in the embodiment. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below in conjunction with embodiments:

[0033] Example

[0034] like Figure 1 , Figure 3 to Figure 6 As shown, a method for identifying and locating underground pipeline damage based on a feature hierarchy model includes:

[0035] S1. Arrange several Raspberry Pi ultrasonic detectors along the underground pipeline wall. The present invention builds an ultrasonic detection system of Raspberry Pi. The ultrasonic detection system includes several ultrasonic detectors arranged by Raspberry Pi. The sample feature hierarchical label learning model is placed in the computer PC. The ultrasonic detection system is connected to the sample feature hierarchical label learning model in communication. Figure 3As shown, the ultrasonic transmission frequency and other parameters of the ultrasonic detector (the transmission frequency and other parameters must ensure the filling layer between the underground pipeline and the ground, and at the same time be able to transmit the received echo. The filling layer generally uses the main material layer. If there are two main material layers between the underground pipeline and the ground, such as a concrete layer on the soil layer, the concrete layer can be removed to leave only the main soil layer, or the concrete layer can be removed and the concrete layer can be removed. Two transmission echo detections are performed respectively, and the location of the underground pipeline is determined by comparison. The corresponding echo at the location of the underground pipeline is determined based on the two echoes of removing the concrete layer and not removing the concrete layer. For example, the two echoes before and after, one Generally, the second echo is the reflected echo of the underground pipeline wall) and the emission angle of the oblique longitudinal wave probe (the angle can obtain the angle θ between the ultrasonic emission direction and the plumb line) are set respectively. The ultrasonic wave emitted by the ultrasonic detector is amplified by the circuit and the ultrasonic signal intensity is increased (the circuit amplification and excitation transducer inside the ultrasonic detector reduce the attenuation loss of the ultrasonic transmission process) and is emitted to the underground pipeline wall, and then hits the underground pipeline wall to reflect to form an ultrasonic echo and is received by the ultrasonic detector (internal receiving transducer). The timer of the ultrasonic detection system of the Raspberry Pi can calculate the time difference between the transmission wave and the echo propagation. Prepare the damaged samples and healthy samples of the underground pipeline wall (the present invention is based on the damaged samples and healthy samples arranged in the underground pipeline for no prior classification or sample marking classification, and the damaged position is marked at the corresponding position of the underground pipeline wall. The material of the underground pipeline should be selected from metal, plastic and other classifications to facilitate accurate classification and identification) according to the underground pipeline layout, the ultrasonic detector transmits the wave and the echo to collect the ultrasonic wall damaged sample data containing labels, and the labels include damaged labels and healthy labels. In some embodiments, the sample feature hierarchical label learning model may also integrate data such as service life and maintenance records before sample input. The ultrasonic pipe wall damage sample data may also be classified and integrated with service life and maintenance records (such as historical damage time, fault type, maintenance frequency, etc.). When inputting the sampled underground pipeline data into the sample feature hierarchical label learning model, these data of the underground pipeline are also input synchronously (to facilitate matching feature extraction and analysis). Before the underground pipeline samples and the above data of the underground pipeline to be tested are input, data cleaning and normalization processing are required to remove duplicate records and records lacking key information (such as pipeline material, etc.). The collected pipeline data are classified and uniformly encoded according to the pipe material, historical damage records, etc. to ensure that the ultrasonic data has a unified data format, convert the time data format into seconds, eliminate the dimensional differences between the data, and perform denoising and filtering processing. A bandpass filter is used to set the passband range to eliminate noise, eliminate abnormal data, and retain the required ultrasonic data.

[0036] S2. Construct a sample feature hierarchical label learning model (see the model structure for details). Figure 4Preferably, the sample feature hierarchical label learning model adopts a BP neural network, whose topological structure includes an input layer, a hidden layer and an output layer, and there is no coupling between neurons of the same level in the network) to extract features from the ultrasonic pipe wall damage sample data to obtain a label feature vector v ij , v ij Represents the feature vector of label i and feature j, and then adds the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij Represents the weight coefficient of label i and feature j after normalization according to label i. The sample feature hierarchical label learning model inputs ultrasonic pipe wall damage sample data for model iterative training. Features include multi-level key features including time interval features, amplitude features, frequency features, angle features, and propagation speed features between the transmitted wave and the echo (may also include pipeline material, etc.); the attention mechanism module is allocated according to the correlation between the features and the label results. Through multiple iterations and gradient descent method to optimize the relevant parameters (such as weights) of the model optimization processing, retain the recognition feature library, and assign labels based on the similarity between the new data input by the test set and the features in the library, so as to identify whether the underground pipeline is damaged and the degree of damage. In this embodiment, the Adam optimizer can be used to adjust the network weights, set the maximum number of training times to 1000, and the learning rate to 0.01 for model training. According to the actual situation and the collected data, the training output labels are adjusted for secondary adjustment to assist iteration. After the model training is completed, the underground pipelines that need to be identified are input into the model for damage identification. Labels are assigned based on the similarity between the extracted features and the feature library in the model. Only three labels, namely healthy and undamaged, slightly damaged, and severely damaged, are output. Other embodiments can make other distinctions based on the conditions of the underground pipelines.

[0037] In some embodiments, the sample feature hierarchical label learning model can construct a template point including key features and a classifier, such as Figure 5 As shown in the figure, the sample data is divided into similar sets and dissimilar sets by calculating the similarity of key features between sample points and template points. For dissimilar sets, threshold conditions are set, and dissimilar sets are classified into difference hierarchies and sorted by dissimilarity. Labels are marked according to domain knowledge and the actual situation of underground pipelines. Finally, the sample data is classified into 1 similar set (i.e., healthy set, marked as HD) and k-1 dissimilar sets (i.e., abnormal sets, marked as AD-1, AD-2...AD-(k-1) in sequence according to the degree of similarity). The similarity between sample points and template points is calculated using the Gaussian kernel function. The specific formula is as follows:

[0038] Where x is a randomly selected template point in the sample data, x′ is the remaining sample point, ||xx′|| is the Euclidean distance between the template point and the sample point, σ is the width parameter of the Gaussian kernel function, and T is the similarity, with a value range of (0, 1].

[0039] Then the infonce loss function is introduced to constrain the loss. The specific calculation formula is as follows:

[0040]

[0041] Among them, K is the number of samples, f(HD) is the similarity of the similar set composed of sample points similar to the template point, and f(AD-j) is the similarity of the different set composed of sample points different from the template point; L represents the error between the predicted result and the actual situation during the learning process. When the error is at a suitable minimum value, the model training is completed. In this way, samples are classified into similar sets and different sets based on the similarity of sample data, which is convenient for feature extraction for classification.

[0042] In this embodiment, in step S2, the data set of the sample feature hierarchical label learning model (the data set contains multi-dimensional features, which will be used for model feature extraction and damage identification) is ultrasonic pipe wall damage sample data, which is divided into training set, test set and validation set according to the ratio of 7:2:1 for model training, testing and validation.

[0043] S3. The Raspberry Pi ultrasonic detector arranged on the wall of the underground pipeline collects the underground pipeline data of the sampling point and inputs the sample feature hierarchical label learning model, then outputs the damage recognition result of the sampling point, uses the ultrasonic detector to calculate the location information of the sampling point (the location information of the sampling point is calculated by the location information of the ultrasonic detector) and associates the data with the damage recognition result for storage, and then selects the data with the damage recognition result as damaged for association output. In some embodiments, several ultrasonic detectors arranged in the direction of the underground pipeline collect underground pipeline data at each sampling point respectively, and obtain the location information of each sampling point and the corresponding damage recognition result; or the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data at each sampling point of the underground pipeline wall in turn, and obtain the location information of each sampling point and the corresponding damage recognition result (the damage recognition result or label includes healthy and undamaged, slightly damaged, and severely damaged). The positioning points (the location points with the damage recognition result as the damaged type) are selected from the damage recognition positioning data, and the selected damage recognition positioning data are output.

[0044] In some embodiments, step S3 further includes the following method:

[0045] Construct a location depth coordinate system, with the ultrasonic detector at the starting point as the origin, the direction of the underground pipeline as the x-axis (here the horizontal projection line of the underground pipeline direction is taken), and the y-axis perpendicular to the underground pipeline direction as the plumb line. The ultrasonic detector at the starting point collects underground pipeline data from the sampling points of the underground pipeline wall, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result, the position information includes the horizontal distance of the sampling point on the ultrasonic detector plane and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data in sequence from the sampling points of the underground pipeline wall, and the position information of each sampling point and the corresponding damage identification result are obtained, and the complete damage identification and positioning data of the underground pipeline wall are obtained, the damage identification and positioning data include the position information of the sampling point and the corresponding damage identification result (the damage identification result or label includes healthy and undamaged, slightly damaged, and seriously damaged); the position information of the mobile ultrasonic detector is mapped to the position depth coordinate system, and the position information of the sampling point detected by the mobile ultrasonic detector is also mapped to the position depth coordinate system, so that as the ultrasonic detector moves, the position information of the sampling point can be expressed in the position depth coordinate system, so that the position information collection and damage identification of different sampling points can be realized, which is convenient for management personnel to efficiently record and maintain underground pipelines. The damage identification and positioning data is filtered out to find positioning points whose damage identification results are damage types (position points whose damage identification results are damage types), and the filtered damage identification and positioning data is output.

[0046] In some embodiments, in step S3, the ultrasonic detector measures the sampling point position information in the following manner: locate the ultrasonic detector position information, the underground pipeline wall sampling point position information is the relative position of the sampling point and the ultrasonic detector (if a longitude and latitude coordinate system or a coordinate system is established, the ultrasonic detector position information can be obtained, and the position information determined by the sampling point can also be obtained), the horizontal distance of the underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, the height of the underground pipeline wall sampling point from the ground is h = S × sinθ, Among them, s represents the distance measured by the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector transmitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave transmitting direction and the plumb line. The horizontal distance D of the sampling point on the plane of the ultrasonic detector and the height h of the underground pipeline wall sampling point from the ground are the location information of the underground pipeline wall sampling point, which can obtain the specific location of the sampling point. If the coordinate system is established, the coordinate position of the sampling point can be directly obtained.

[0047] In some embodiments, in step S2, the i-th tag contains L j features, then the i-th label feature attention data expression is as follows:

[0048] like Figure 2 to Figure 6 As shown, an ultrasonic identification and positioning system for underground pipeline damage based on a feature hierarchical learning model includes an ultrasonic pipe wall damage sample database, an ultrasonic detection system, and a sample feature hierarchical label learning model. The sample feature hierarchical label learning model includes an attention mechanism module. The ultrasonic detection system includes a plurality of Raspberry Pi ultrasonic detectors arranged along the direction of the underground pipeline for the underground pipeline wall. The ultrasonic pipe wall damage sample database stores ultrasonic pipe wall damage sample data containing labels. The ultrasonic pipe wall damage sample data is the data collected by arranging underground pipelines and healthy samples and transmitting waves and echoes of ultrasonic detectors. The labels include damage labels and healthy labels. The sample feature hierarchical label learning model performs feature extraction to obtain a label feature vector v ij , v ij The feature vector representing label i and feature j is then passed through the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij Represents the weight coefficient of label i and feature j after normalization according to label i. The sample feature hierarchical label learning model inputs the ultrasonic pipe wall damage sample data for model iterative training. The Raspberry Pi ultrasonic detector arranged on the underground pipeline wall is used to detect the underground pipeline data of the sampling point and input into the sample feature hierarchical label learning model, and the damage identification result of the sampling point is output. The location information of the sampling point is calculated by the ultrasonic detector and stored in association with the damage identification result, and then the data with the damage identification result is screened out for association output.

[0049] The method for ultrasonic detector to calculate the sampling point location information is as follows: Figure 6 As shown, the location information of the positioning ultrasonic detector, the location information of the underground pipeline wall sampling point is the relative position of the sampling point and the ultrasonic detector, the horizontal distance of the underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, and the height of the underground pipeline wall sampling point from the ground is h = S × sinθ. Among them, S represents the distance measured by the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector transmitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave transmitting direction and the plumb line. The horizontal distance D of the sampling point on the plane of the ultrasonic detector and the height h of the underground pipeline wall sampling point from the ground are the location information of the underground pipeline wall sampling point, which can obtain the specific location of the sampling point. If the coordinate system is established, the coordinate position of the sampling point can be directly obtained. The sample feature hierarchical label learning model also includes a position depth coordinate system with the ultrasonic detector at the starting point as the origin, the direction of the underground pipeline as the x-axis, and the direction perpendicular to the underground pipeline as the y-axis (underground depth). The ultrasonic detector at the starting point collects underground pipeline data from the sampling points of the underground pipeline wall, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result. The position information includes the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data from the sampling points of the underground pipeline wall in turn, and the position information of each sampling point and the corresponding damage identification result are obtained, so as to obtain complete damage identification and positioning data of the underground pipeline wall. The damage identification and positioning data includes the position information of the sampling point and the corresponding damage identification result.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying and locating underground pipeline damage based on a feature hierarchy model, characterized in that: The methods include: S1. Arrange several Raspberry Pi ultrasonic detectors along the underground pipelines and on the underground pipeline walls, prepare damaged and healthy samples of the underground pipeline walls, and collect ultrasonic wall damaged sample data containing labels by transmitting waves and echoes from ultrasonic detectors according to the underground pipelines. The labels include damaged labels and healthy labels. S2. Construct a sample feature hierarchical label learning model to extract features from ultrasonic pipe wall damage sample data to obtain label feature vector v ij , v ij Represents the feature vector of label i and feature j, and then adds the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij It represents the weight coefficient of label i and feature j after normalization according to label i; the sample feature hierarchical label learning model inputs the ultrasonic pipe wall damage sample data for model iterative training; S3. Relying on the Raspberry Pi ultrasonic detectors arranged on the walls of underground pipelines, the underground pipeline data of the sampling points are collected and input into the sample feature hierarchical label learning model. Then, the damage recognition results of the sampling points are output. The ultrasonic detectors are used to calculate the location information of the sampling points and store the data in association with the damage recognition results. Then, the data with the damage recognition results as damaged are screened out for associated output.

2. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 1 is characterized in that: The method for ultrasonic detector to calculate the sampling point location information is as follows: locate the location information of ultrasonic detector, the location information of underground pipeline wall sampling point is the relative position of sampling point and ultrasonic detector, the horizontal distance of underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, the height of underground pipeline wall sampling point from the ground is h = S × sinθ, Where S represents the distance from the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector transmitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave transmitting direction and the plumb line.

3. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 1 is characterized in that: Several ultrasonic detectors arranged in the direction of the underground pipeline collect underground pipeline data at each sampling point, and obtain the position information of each sampling point and the corresponding damage identification result; or move the ultrasonic detector along the direction of the underground pipeline to collect underground pipeline data at each sampling point on the underground pipeline wall in turn, and obtain the position information of each sampling point and the corresponding damage identification result.

4. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 2 is characterized in that: Step S3 also includes the following method: A position depth coordinate system is constructed, with the ultrasonic detector at the starting point as the origin and the direction of the underground pipeline as the x-axis; the ultrasonic detector at the starting point collects underground pipeline data from the sampling points on the wall of the underground pipeline, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result, the position information includes the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data in sequence from the sampling points on the wall of the underground pipeline, and obtains the position information of each sampling point and the corresponding damage identification result, and obtains complete damage identification and positioning data of the underground pipeline wall, the damage identification and positioning data includes the position information of the sampling point and the corresponding damage identification result.

5. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 4 is characterized in that: The damage identification and positioning data are filtered out to find the positioning points whose damage identification results are damage types, and the filtered damage identification and positioning data are output.

6. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 1 is characterized in that: In step S2, the i-th tag contains L j features, then the i-th label feature attention data expression is as follows:

7. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 1 is characterized in that: The characteristics include multi-level key characteristics including the time interval characteristics, amplitude characteristics, frequency characteristics, angle characteristics and propagation speed characteristics of the transmitted wave and the echo; The attention mechanism module is allocated according to the correlation between features and label results.

8. The underground pipeline damage identification and positioning method based on the feature hierarchy model according to claim 1 is characterized in that: The damage identification results or labels include healthy and undamaged, slightly damaged, and severely damaged.

9. An ultrasonic identification and location system for underground pipeline damage based on a feature hierarchical learning model, characterized in that: It includes an ultrasonic pipe wall damage sample database, an ultrasonic detection system and a sample feature hierarchical label learning model. The sample feature hierarchical label learning model includes an attention mechanism module. The ultrasonic detection system includes a plurality of Raspberry Pi ultrasonic detectors arranged along the direction of the underground pipeline for the underground pipeline wall. The ultrasonic pipe wall damage sample database stores ultrasonic pipe wall damage sample data containing labels. The ultrasonic pipe wall damage sample data is the data collected by arranging underground pipelines and healthy samples through the ultrasonic detector transmitting waves and echoes. The labels include damage labels and healthy labels. The sample feature hierarchical label learning model performs feature extraction to obtain the label feature vector v ij , v ij The feature vector representing label i and feature j is then passed through the attention mechanism module to obtain the weight coefficient vector W corresponding to the label feature vector ij , W ij Represents the weight coefficient of label i and feature j after normalization according to label i; The sample feature hierarchical label learning model inputs ultrasonic pipe wall damage sample data for model iterative training; the Raspberry Pi ultrasonic detector arranged on the underground pipeline wall collects the underground pipeline data of the sampling point and inputs it into the sample feature hierarchical label learning model, and outputs the damage recognition result of the sampling point; the ultrasonic detector is used to calculate the location information of the sampling point and store it in association with the damage recognition result, and then the data with the damage recognition result is screened out for associated output.

10. The underground pipeline damage identification and positioning system based on the feature hierarchy model according to claim 9 is characterized in that: The method for ultrasonic detector to calculate the sampling point location information is as follows: locate the location information of ultrasonic detector, the location information of underground pipeline wall sampling point is the relative position of sampling point and ultrasonic detector, the horizontal distance of underground pipeline wall sampling point on the ultrasonic detector plane is D = S × cosθ, the height of underground pipeline wall sampling point from the ground is h = S × sinθ, Where S represents the distance from the ultrasonic detector to the sampling point, V represents the propagation speed of the ultrasonic wave, t represents the time interval from the ultrasonic detector transmitting the ultrasonic wave to receiving the echo, and θ represents the angle between the ultrasonic wave transmitting direction and the plumb line; The sample feature hierarchical label learning model also includes a position depth coordinate system with the ultrasonic detector at the starting point as the origin and the direction of the underground pipeline as the x-axis. The ultrasonic detector at the starting point collects underground pipeline data from the sampling points of the underground pipeline wall, and obtains the position information of the sampling point at the starting point and the corresponding damage identification result. The position information includes the horizontal distance of the sampling point on the plane of the ultrasonic detector and the height of the sampling point from the ground; then the ultrasonic detector is moved along the direction of the underground pipeline to collect underground pipeline data in turn from the sampling points of the underground pipeline wall, and the position information of each sampling point and the corresponding damage identification result are obtained, so as to obtain complete damage identification and positioning data of the underground pipeline wall. The damage identification and positioning data includes the position information of the sampling point and the corresponding damage identification result.

Images