A method and equipment for flaw detection of water conservancy project pipelines

Abstract

The present invention relates to the field of defect detection technology, and specifically to a method and device for flaw detection of water conservancy engineering pipelines. Collect a pipeline ultrasonic data set containing multiple known defect types; divide the signal segments by calculating the reflection probability of the data value in a local range, perform acoustic path analysis in each signal segment, and determine the pipeline echo probability in combination with the reflection probability; consider the influence of materials such as concrete on the propagation of sound waves, use the time-frequency difference between signal segments to adjust the pipeline echo probability, and obtain the corrected probability, thereby determining the true and effective pipeline echo signal; then, determine the characterization degree value of each characteristic parameter according to the difference and fluctuation characteristics of the characteristic parameters under different defect types; randomly select a sample subset, select the characteristic parameters according to the quantity characteristics and the characterization degree value to construct a decision tree, and train to obtain a random forest model; finally, perform flaw detection on the pipeline to be tested based on the model, and obtain more accurate pipeline detection results.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method and equipment for flaw detection of pipelines in water conservancy projects. Background Art

[0002] Water conservancy project pipelines are an important part of the water delivery system. Their long-term operation is easily affected by factors such as water erosion, corrosion, and pressure fluctuations, which can lead to structural defects such as cracks and wall thickness reduction inside the pipeline. These defects will not only reduce the service life of the pipeline, but may also cause safety accidents such as leakage and structural damage, causing serious economic losses and environmental impacts.

[0003] In recent years, with the rapid development of ultrasonic detection technology, its application in pipeline defect detection has become increasingly widespread. Ultrasonic detection technology transmits ultrasonic waves into the inside of the pipeline, receives the reflected signals, and analyzes the defects inside the pipeline based on the characteristics of the signals. However, in actual applications, since the outside of water conservancy pipelines is often wrapped with a concrete layer, there will be a lot of noise in the echo signal, which not only masks the real and effective pipeline echo signal, but also increases the difficulty of feature extraction. Therefore, if the real and effective signals in the ultrasonic data are not separated for feature extraction, the final pipeline detection results will be less accurate. Summary of the invention

[0004] In order to solve the technical problem of defect detection, the purpose of the present invention is to provide a method and equipment for flaw detection of water conservancy project pipelines. The technical scheme adopted is as follows:

[0005] Acquire a pipeline ultrasonic data set, wherein the ultrasonic data set includes pipeline ultrasonic data of various defect types;

[0006] In each pipeline ultrasonic data, the reflection probability at each moment is determined according to the significance of each data value in the local range; each pipeline ultrasonic data is segmented based on the reflection probability at each moment to obtain multiple signal segments;

[0007] Perform acoustic path analysis in each signal segment and determine the pipeline echo probability of each signal segment in combination with the reflection probability; adjust the pipeline echo probability of each signal segment using the time-frequency difference between the signal segments to obtain a corrected probability; determine the pipeline echo signal in all signal segments of each pipeline ultrasonic data based on the corrected probability;

[0008] Obtain multiple characteristic parameters of each pipeline echo signal; determine the characterization degree value of each characteristic parameter under each defect type according to the numerical differences and numerical fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters;

[0009] A sample subset is randomly selected from all pipeline echo signals. In each sample subset, based on the quantity characteristics of pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various characteristic parameters, characteristic parameters are selected to construct a decision tree, thereby obtaining a trained random forest model; multiple characteristic parameters of the pipeline echo signal of the pipeline to be tested are used as inputs of the trained random forest model, and the pipeline detection results are output.

[0010] Furthermore, the method for obtaining the reflection probability includes:

[0011] Determine the preset neighborhood corresponding to each moment;

[0012] The mean of the data values ​​of all neighboring moments in the pipeline ultrasonic data at each moment is taken as the mean eigenvalue;

[0013] The normalized value of the difference between the data value in the pipeline ultrasonic data at each moment and the mean characteristic value is used as the reflection probability at each moment.

[0014] Furthermore, each pipeline ultrasonic data is segmented based on the reflection probability at each time to obtain multiple signal segments, including:

[0015] The reflection probabilities at all times are smoothly connected in time series to obtain a reflection probability curve;

[0016] On the reflection probability curve, a maximum point is obtained, and a data segment corresponding to a curve segment between any two adjacent maximum points in the pipeline ultrasonic data is used as a signal segment.

[0017] Furthermore, the method for obtaining the pipeline echo probability includes:

[0018] In each signal segment, the mean of the reflection probabilities at the two endpoints is taken as the intensity characteristic value;

[0019] The product of the time interval between the two endpoints and the propagation speed of the sound wave in the pipeline material is taken as the actual pipe wall thickness, and the absolute value of the difference between the actual pipe wall thickness and the preset pipe wall thickness is negatively correlated and normalized to obtain a similarity factor;

[0020] The product of the intensity characteristic value corresponding to each signal segment and the similarity factor is normalized to obtain the value, which is used as the pipeline echo probability of each signal segment.

[0021] Furthermore, the method for obtaining the modified probability includes:

[0022] Perform short-time Fourier transform on each pipeline ultrasonic data to obtain a time-frequency spectrum diagram;

[0023] For any signal segment in each pipeline ultrasonic data, the mean of the amplitude values ​​at the same frequency at all times in the signal segment is taken as the amplitude mean;

[0024] Combine any two signal segments to obtain all non-repeating signal segment combinations;

[0025] In each signal segment combination, the absolute value of the difference between the amplitude means of the two signal segments at the same frequency is calculated as the deviation factor, and the sum of the deviation factors of the two signal segments at all frequencies is normalized to obtain the value as the frequency domain amplitude difference between the two signal segments;

[0026] The mean value of the frequency domain amplitude difference between each signal segment and all other signal segments is multiplied by the pipeline echo probability of each signal segment, and the obtained product is normalized as the corrected probability of each signal segment.

[0027] Furthermore, the method for acquiring the pipeline echo signal includes:

[0028] Among all signal segments of each pipeline ultrasonic data, the signal segment with the largest correction probability is taken as the pipeline echo signal.

[0029] Furthermore, the method for obtaining the characterization degree value includes:

[0030] Classify all pipeline echo signals according to their defect categories;

[0031] In each pipeline echo signal, the standard deviation of the value of each characteristic parameter in all pipeline echo signals is taken as the fluctuation factor;

[0032] Calculate the numerical mean of all pipeline echo signals in each pipeline echo signal under each characteristic parameter as the characteristic value;

[0033] For any pipeline echo signal, under the same characteristic parameters, the absolute value of the difference between the characteristic value corresponding to the pipeline echo signal and the characteristic value of each other pipeline echo signal is used as the difference factor, and the mean value of all the difference factors corresponding to the pipeline echo signal under each characteristic parameter is used as the difference characteristic value;

[0034] According to the difference characteristic value and fluctuation factor of each characteristic parameter under each pipeline echo signal, the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs is obtained, and the characterization degree value is negatively correlated with the fluctuation factor, and the characterization degree value is positively correlated with the difference characteristic value.

[0035] Furthermore, based on the quantitative characteristics of the pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various characteristic parameters, characteristic parameters are selected to construct a decision tree, thereby obtaining a trained random forest model, including:

[0036] In each sample subset, the quantitative characteristics of the pipeline echo signals of different defect types and the characterization degree value of each characteristic parameter under each defect type are analyzed to determine the candidate degree value of each characteristic parameter in each sample subset;

[0037] In each sample subset, all types of feature parameters are arranged in descending order according to the candidate degree values ​​to obtain a sorted sequence;

[0038] In each sample subset, feature parameters are selected according to the arrangement of the feature parameters in the sorting sequence to construct a decision tree, and multiple decision trees are obtained, thereby obtaining a trained random forest model.

[0039] Furthermore, the method for obtaining the candidate degree value includes:

[0040] In each sample subset, the ratio of the number of pipeline echo signals of each defect type to the number of all pipeline echo signals is taken as the number ratio of pipeline echo signals of each defect type;

[0041] The product of the quantity proportion of each pipeline echo signal and the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs is taken as the candidate factor of each characteristic parameter in each sample subset;

[0042] In each sample subset, the normalized value of the mean of all candidate factors of each feature parameter is used as the candidate degree value of each feature parameter in each sample subset.

[0043] A water conservancy project pipeline flaw detection device comprises a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and when the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor, the steps of a water conservancy project pipeline flaw detection method are implemented.

[0044] The present invention has the following beneficial effects:

[0045] In the present invention, the ultrasonic data of the pipeline with known defect types is used to train the random forest model, so as to detect defects in the pipeline to be tested. First, a pipeline ultrasonic data set is obtained, which includes pipeline ultrasonic data of multiple known defect types. Since the surface of the water conservancy pipeline will be wrapped with concrete or other materials, there will be interference data in the pipeline ultrasonic data, which is not a real and effective pipeline echo signal. Therefore, in the pipeline ultrasonic data, the reflection probability is determined based on the significant situation of the data value in the local range, and the signal segment is divided according to the reflection probability. Further, the sound path analysis is performed in each signal segment and combined with the reflection probability to determine the pipeline echo probability of each signal segment; since concrete is a porous and inhomogeneous material, the sound wave will be scattered in large quantities during propagation, so the random fluctuation is large, and the echo generated when the pipeline surface is reflected is relatively small. Therefore, there will be differences in time-frequency characteristics between different signal segments, so the time-frequency difference between the signal segments is used to adjust the pipeline echo probability, obtain the corrected probability, and determine the real and effective pipeline echo signal in all signal segments based on this indicator, avoiding the interference of signals such as noise. Furthermore, the characteristic parameters of each pipeline echo signal are obtained. Since the characterization degree, that is, the importance, of different characteristic parameters is different under different defect types, in the present invention, according to the difference and fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters, the characterization degree value of each characteristic parameter under each defect type is determined to reflect the weight. Then, several sample subsets are randomly selected, and characteristic parameters are selected according to the quantity characteristics of pipeline echo signals of different defect types and the characterization degree values ​​of characteristic parameters to construct a decision tree, so as to obtain a trained random forest model. At this time, the trained random forest model effectively eliminates the interference of noise signals on pipeline echo signals, and the representativeness of the selected characteristic parameters is strong enough, so the reliability of the model will be higher. Finally, based on the model, the pipeline to be tested is inspected to obtain a more accurate pipeline detection result. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A method flow chart of a method for flaw detection of a water conservancy project pipeline provided by one embodiment of the present invention;

[0048] Figure 2 The present invention is a schematic diagram of the equipment structure of a water conservancy project pipeline flaw detection equipment provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a method and device for flaw detection of a water conservancy project pipeline proposed by the present invention, its specific implementation method, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0050] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0051] The specific scheme of a water conservancy project pipeline flaw detection method and equipment provided by the present invention is described in detail below with reference to the accompanying drawings.

[0052] See also Figure 1 , which shows a method flow chart of a method for flaw detection of a water conservancy project pipeline provided by an embodiment of the present invention, the method comprising the following steps:

[0053] Step S1: Obtain a pipeline ultrasonic data set, where the ultrasonic data set contains pipeline ultrasonic data of various defect types.

[0054] In water conservancy projects, pipelines are an important part of water delivery, drainage and irrigation systems, and the monitoring and maintenance of their operating status are particularly important. However, since pipelines are buried underground or underwater for a long time, they are susceptible to environmental corrosion, geological changes, construction quality and other factors, resulting in various defects inside the pipelines, such as cracks, corrosion, deformation, etc. These defects will not only reduce the water delivery efficiency of the pipeline, but may also cause serious accidents such as leakage and burst, posing a serious threat to the normal operation of water conservancy projects. Ultrasonic detection technology emits ultrasonic waves into the pipeline and receives the reflected signals, so that the defects inside the pipeline can be analyzed according to the characteristics of the signal, so ultrasonic detection technology is usually used for pipeline flaw detection.

[0055] In an embodiment of the present invention, pipeline samples with various known defect types, such as pipeline corrosion, pipeline cracks, etc., are obtained, and then an ultrasonic flaw detector is placed in a detection area of ​​the pipeline sample to obtain pipeline ultrasonic data, and the pipeline ultrasonic data of the pipeline samples with all known defect types are taken as a pipeline ultrasonic data set.

[0056] It should be noted that when conducting ultrasonic testing, the position and angle of the probe need to be adjusted according to the implementation scenario to ensure effective reception of the signal; the sampling frequency can be set to 1MHz, and the specific values ​​and resolution and other parameters can be adjusted according to the implementation scenario, and no limitation or elaboration is made here.

[0057] After obtaining the pipeline ultrasonic data set, the pipeline ultrasonic data can be analyzed to construct a random forest model.

[0058] Step S2: In each pipeline ultrasonic data, the reflection probability at each moment is determined according to the significance of each data value in the local range; each pipeline ultrasonic data is segmented based on the reflection probability at each moment to obtain multiple signal segments.

[0059] Since the surface of the water conservancy project pipeline is usually covered with concrete or other materials, the ultrasonic data of the water conservancy project pipeline will contain a large amount of noise and interference signals, which makes it impossible for the characteristic parameters directly extracted from the original pipeline ultrasonic data to accurately reflect the characteristics of the pipeline echo signal, which will cause the accuracy of pipeline flaw detection to decrease. Therefore, in the embodiment of the present invention, it is necessary to screen out the real and effective pipeline echo signal from the pipeline ultrasonic data. When performing pipeline flaw detection, the ultrasonic wave will reflect when it contacts the object, and the reflection characteristics of different media are different. Therefore, the embodiment of the present invention first determines the reflection probability at each moment according to the significance of each data value in the local range in the pipeline ultrasonic data; then the pipeline ultrasonic data can be segmented based on the reflection probability at each moment to obtain multiple signal segments; to prepare for the subsequent screening of pipeline echo signals.

[0060] Preferably, in one embodiment of the present invention, the method for obtaining the reflection probability includes:

[0061] First, in each pipeline ultrasonic data, the preset neighborhood corresponding to each moment is determined, with the aim of finding a local range for each moment to facilitate the analysis of local significant features.

[0062] Then, the mean of the data values ​​of all the neighboring moments in the pipeline ultrasonic data at each moment is taken as the mean eigenvalue. The mean eigenvalue represents the average ultrasonic signal strength in the local range at each moment. When the data value in the pipeline ultrasonic data at each moment is greater than the corresponding mean eigenvalue, it is considered that the ultrasonic signal at each moment significantly exceeds the local average level, otherwise it is lower than the local average level. Therefore, the difference between the data value in the pipeline ultrasonic data at each moment and the mean eigenvalue is normalized and used as the reflection probability at each moment. The greater the reflection probability, the more likely it is that the ultrasonic wave encounters a strong reflection when it encounters an obvious interface between two media, which means that the ultrasonic wave encounters a change in the properties of the medium during propagation. Since the difference here may be positive or negative, it can be used The function normalizes it.

[0063] It should be noted that the preset neighborhood is set to be centered on each moment, with 10 moments before and after in time sequence, that is, a total of 11 moments are included in the neighborhood. When the number of moments on one side is less than 10, the total number can be supplemented on the other side; the size of the preset neighborhood can be adjusted according to the implementation scenario and is not limited here.

[0064] At this point, in each pipeline ultrasonic data, the reflection probability at each moment can be obtained, and then the pipeline ultrasonic data can be divided based on the reflection probability to obtain multiple signal segments.

[0065] Preferably, in one embodiment of the present invention, the method for acquiring a signal segment includes:

[0066] The reflection probability at each moment is a discrete data value. In the embodiment of the present invention, the reflection probabilities at all moments are smoothly connected in time series to obtain a reflection probability curve.

[0067] Since the maximum value on the reflection probability curve usually corresponds to the reflection signal when the ultrasonic wave encounters an obvious interface or defect inside the pipeline, the maximum point is obtained on the reflection probability curve, and the curve segment between any two adjacent maximum points in the pipeline ultrasonic data corresponds to a data segment as a signal segment. At this time, each signal segment may contain a specific reflection event, which is convenient for accurately screening out the pipeline echo signal in subsequent analysis.

[0068] It should be noted that, in other embodiments of the present invention, the least square method may also be used to perform curve fitting on the reflection probability at all times to obtain a reflection probability curve. The least square method is a well-known technology, and the specific process is not described here.

[0069] At this point, each pipeline ultrasonic data can be divided to obtain all signal segments.

[0070] Step S3: Perform acoustic path analysis in each signal segment, and determine the pipeline echo probability of each signal segment in combination with the reflection probability; adjust the pipeline echo probability of each signal segment using the time-frequency difference between the signal segments to obtain a corrected probability; determine the pipeline echo signal in all signal segments of each pipeline ultrasonic data based on the corrected probability.

[0071] By performing acoustic path analysis in each signal segment, we can understand the propagation distance of the sound wave in the pipeline, which is the basis for identifying the pipeline echo signal. At the same time, combined with the reflection probability, the pipeline echo probability of each signal segment can be obtained, which is used to preliminarily judge the possibility that each signal segment is a pipeline echo. Since the concrete wrapped on the pipeline surface is a porous and heterogeneous material, the sound wave will be scattered in large quantities during propagation, so the random fluctuation is large, and the echo generated by the reflection on the pipeline surface has relatively small volatility. Therefore, there will be differences in time-frequency characteristics between different signal segments. Therefore, the time-frequency difference between signal segments is used to adjust the pipeline echo probability, and the corrected probability is obtained. Based on this indicator, the real and effective pipeline echo signal is determined in all signal segments, avoiding the interference of noise and other signals, and providing strong support for the accuracy of subsequent feature extraction.

[0072] Preferably, in one embodiment of the present invention, the method for obtaining the pipeline echo probability includes:

[0073] In each signal segment, the mean of the reflection probabilities at the two endpoint moments is taken as the intensity characteristic value, which represents the overall reflection intensity of each signal segment.

[0074] The propagation speed of sound waves in the pipeline material is known, so the distance the sound waves propagate can be calculated through the time interval as the actual pipe wall thickness. This value is then compared with the preset pipe wall thickness (normal pipe wall thickness) to obtain an indicator that reflects the similarity between the two. The more similar they are, the more likely the signal segment is to be a pipe echo signal.

[0075] Therefore, the product of the time interval between the two endpoints and the propagation speed of the sound wave in the pipeline material is taken as the actual pipe wall thickness, and the absolute value of the difference between the actual pipe wall thickness and the preset pipe wall thickness is negatively correlated and normalized to achieve logical relationship correction and obtain the similarity factor. At this time, the larger the similarity factor of a certain signal segment, the more likely the signal segment is a real and effective pipeline echo signal. The negative correlation mapping and normalization here can be done using the formula ,in, It represents an exponential function with the natural constant e as the base, and x represents the independent variable.

[0076] Finally, the product of the intensity characteristic value and the similarity factor corresponding to each signal segment is normalized to obtain the value as the pipeline echo probability of each signal segment. The intensity characteristic value provides information about the reflection intensity, while the similarity factor takes into account the influence of the pipe wall thickness on the echo signal. The product of the two is normalized to obtain a more comprehensive and accurate pipeline echo probability value. The larger the pipeline echo probability value of a certain signal segment, the more likely the signal segment is a real and effective pipeline echo signal. Normalization is a technical means well known to those skilled in the art, and the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0077] It should be noted that the preset pipe wall thickness can be determined according to the construction requirements of the pipeline in the implementation scenario, and the propagation speed of the sound wave in the pipeline material is known after the pipeline material is determined.

[0078] At this point, the pipeline echo probability of each signal segment in each pipeline ultrasonic data can be obtained.

[0079] Considering that when ultrasonic waves propagate in different media, the corresponding echo signals fluctuate differently, in the embodiment of the present invention, the echo signals in non-pipeline materials are considered to be noise, so among all the signal segments, there is only one signal segment that represents the pipeline echo signal, and the remaining signal segments are noise signals; since the concrete on the pipeline surface is a porous and heterogeneous material, the sound waves will be scattered in large quantities during propagation, and the volatility is large, while the echo signal generated by the pipeline reflection has a small volatility. Time-frequency analysis can simultaneously expand the signal in time and frequency, and can reveal the behavior of the signal at specific time points and frequency points, which is helpful to deeply extract the key information in the signal. Therefore, the embodiment of the present invention analyzes the time-frequency differences between different signal segments, thereby adjusting the pipeline echo probability of each signal segment to obtain a corrected probability.

[0080] Preferably, in one embodiment of the present invention, the method for obtaining the modified probability includes:

[0081] First, each pipeline ultrasonic data is subjected to short-time Fourier transform to obtain a time-frequency spectrum. The time-frequency spectrum can simultaneously display the distribution of the signal in time and frequency, which helps to reveal the time-varying characteristics of the signal.

[0082] For any signal segment in each pipeline ultrasonic data, the mean of the amplitude values ​​at the same frequency at all times in the signal segment is taken as the amplitude mean. The amplitude mean represents the average intensity of the signal segment at a certain frequency. Then the time-frequency difference between any two signal segments can be analyzed: any two signal segments are combined to obtain all non-repetitive signal segment combinations.

[0083] In each signal segment combination, the absolute value of the difference between the amplitude means of the two signal segments at the same frequency is calculated as the deviation factor, and the sum of the deviation factors of the two signal segments at all frequencies is normalized as the frequency domain amplitude difference between the two signal segments. Since the pipeline echo signal and the noise signal have different characteristics in the frequency domain, the larger the frequency domain amplitude difference, the more it indicates that there is a pipeline echo signal in the two signal segments. Normalization is a technical means well known to those skilled in the art, and the selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0084] Finally, the mean of the frequency domain amplitude difference between each signal segment and all other signal segments is calculated as the adjustment weight. The greater the frequency domain amplitude difference between a certain signal segment and other signal segments, the more likely the signal segment is to be a real and effective pipeline echo signal, so the adjustment weight of the signal segment is greater. Therefore, the product of the adjustment weight of each signal segment and the pipeline echo probability of each signal segment is normalized to obtain the value as the corrected probability of each signal segment. At this time, the greater the corrected probability of a certain signal segment, the more likely the signal segment is to be a real and effective pipeline echo signal. Normalization is a technical means well known to those skilled in the art, and the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0085] It should be noted that short-time Fourier transform is a well-known technology, and the specific process will not be described here in detail.

[0086] At this point, the correction probability of each signal segment can be obtained, which represents the possibility that each signal segment is a real and valid pipeline echo signal, so that the pipeline echo signal can be determined in all signal segments of each pipeline ultrasonic data.

[0087] Preferably, in one embodiment of the present invention, the method for acquiring the pipeline echo signal includes:

[0088] Based on the above analysis, it can be seen that the greater the correction probability of a signal segment, the more likely it is that the signal segment is a real and valid pipeline echo signal. Therefore, among all the signal segments of each pipeline ultrasonic data, the signal segment with the largest correction probability is taken as the pipeline echo signal.

[0089] At this point, the pipeline echo signal in each pipeline ultrasonic data can be obtained.

[0090] Step S4: obtaining multiple characteristic parameters of each pipeline echo signal; determining the characterization degree value of each characteristic parameter under each defect type according to the numerical differences and numerical fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters.

[0091] By processing the pipeline echo signal, a variety of characteristic parameters reflecting the signal characteristics can be extracted, such as: maximum amplitude, average amplitude, amplitude standard deviation, pipeline echo signal time length, echo main frequency, main frequency bandwidth and other multi-dimensional characteristics, these characteristic parameters can describe the characteristics of the pipeline echo signal from different angles. It should be noted that various characteristic parameters can be calculated based on the pipeline echo signal, and the specific process is not repeated. In other embodiments of the present invention, the types of characteristic parameters can also be added or reduced, and the specific number and type are not limited or repeated here.

[0092] Furthermore, since the characteristics of the pipeline echo signals corresponding to pipelines of different defect types are different, different characteristic parameters should have different weights, that is, characterization degrees, for pipeline echo signals of different defect types. Therefore, in an embodiment of the present invention, according to the differences and fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters, the characterization degree value of each characteristic parameter under each defect type is determined to provide a weight for the subsequent decision algorithm.

[0093] Preferably, in one embodiment of the present invention, the method for obtaining the characterization degree value includes:

[0094] First, all pipeline echo signals are classified according to their defect categories.

[0095] In each pipeline echo signal, the standard deviation of the values ​​of each characteristic parameter in all pipeline echo signals is taken as the fluctuation factor. The fluctuation factor reflects the stability or variability of the same type of pipeline echo signals under certain characteristic parameters. In a certain pipeline echo signal, when the fluctuation factor corresponding to a certain characteristic parameter is larger, it means that the performance characteristics of this characteristic parameter in the pipeline echo signal are more unstable, then the representation degree of this characteristic parameter in this pipeline echo signal, that is, the weight should be reduced; conversely, when the fluctuation factor is smaller, it means that the performance characteristics of this characteristic parameter in this pipeline echo signal are more stable, and the representation degree, that is, the weight should be increased.

[0096] Furthermore, the differences of the same characteristic parameters in different types of pipeline echo signals can be further compared.

[0097] The numerical mean of all pipeline echo signals in each pipeline echo signal under each characteristic parameter is calculated as the characteristic value, which reflects the average performance of each characteristic parameter in each pipeline echo signal.

[0098] For any kind of pipeline echo signal, under the same characteristic parameters, the absolute value of the difference between the characteristic value corresponding to this kind of pipeline echo signal and the characteristic value of each other pipeline echo signal is used as the difference factor. The larger the difference factor is, the more obvious the performance of this kind of pipeline echo signal is compared with other pipeline echo signals under certain characteristic parameters. Therefore, in this pipeline echo signal, the weight of this characteristic parameter should be increased.

[0099] Then, under each characteristic parameter, the mean of all difference factors corresponding to each pipeline echo signal is taken as the difference characteristic value. The difference characteristic value reflects the difference between the performance characteristics of each characteristic parameter in each pipeline echo signal and other pipeline echo signals. The larger the value, the more obvious the performance of each characteristic parameter in each pipeline echo signal is compared with other pipeline echo signals.

[0100] Finally, according to the difference characteristic value and fluctuation factor of each characteristic parameter under each pipeline echo signal, the characterization degree value of each characteristic parameter under each defect type of pipeline echo signal is obtained. The formula model of the characterization degree value includes:

[0101]

[0102] in, It represents the characterization degree value of each characteristic parameter under the defect type to which the i-th pipeline echo signal belongs; It represents the fluctuation factor of each characteristic parameter under the i-th pipeline echo signal; Indicates the number of types of pipeline echo signals, that is, the number of defect types; represents the characteristic value of each characteristic parameter in the i-th pipeline echo signal; represents the characteristic value of each characteristic parameter in the jth pipeline echo signal; Indicates the preset first parameter; Represents the normalization function.

[0103] In the formula model that represents the degree value, represents the difference factor of each characteristic parameter between the i-th pipeline echo signal and the j-th pipeline echo signal; It represents the difference characteristic value of each characteristic parameter under the i-th pipeline echo signal. Based on the above analysis, it can be seen that when the difference characteristic value is larger, it means that the performance of each characteristic parameter in each pipeline echo signal is more obvious than that of other pipeline echo signals, so the characterization degree of this characteristic parameter is greater; when the fluctuation factor of a certain characteristic parameter in a certain pipeline echo signal is smaller, it means that the performance characteristics of this characteristic parameter in this pipeline echo signal are more stable, and the characterization degree should be improved. Therefore, the characterization degree value is negatively correlated with the fluctuation factor, and the characterization degree value is positively correlated with the difference characteristic value. In this embodiment of the present invention, the fluctuation factor is used as the denominator and the difference characteristic value is used as the numerator, so as to normalize the obtained ratio to obtain the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs, and the larger the value, the stronger the performance ability, that is, the more this characteristic parameter can represent the signal characteristics of this defect type.

[0104] It should be noted that the first parameter is preset The function of is to prevent the denominator from being 0. The value can be 0.001. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0105] At this point, through the above method, the characterization degree value of each characteristic parameter under each defect type can be obtained, which reflects the representativeness of each characteristic parameter for each defect type. Therefore, in the subsequent process, a decision tree can be constructed based on the characterization degree value of the characteristic parameter, thereby improving the reliability of the decision tree and ensuring the accuracy of the final detection result.

[0106] Step S5: randomly select a sample subset from all pipeline echo signals. In each sample subset, based on the quantity characteristics of pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various characteristic parameters, select characteristic parameters to construct a decision tree, so as to obtain a trained random forest model; use the various characteristic parameters of the pipeline echo signal of the pipeline to be tested as input of the trained random forest model, and output the pipeline detection result.

[0107] Since there are numerous pipeline echo signals and they contain a large number of characteristic parameters and complex data patterns, and random forests can process such high-dimensional data and extract useful information from them, in an embodiment of the present invention, a random forest model is selected for pipeline flaw detection.

[0108] By randomly selecting several sample subsets from all pipeline echo signals, the model's dependence on specific training data can be reduced, thereby enhancing the model's generalization ability and enabling it to make accurate predictions when faced with new data; in each sample subset, in order to ensure that the feature parameters used to construct the decision tree are the most discriminative and representative, the embodiment of the present invention selects feature parameters based on the quantitative characteristics of pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various feature parameters to construct a decision tree, so that the trained random forest model obtained by integrating all decision trees has higher accuracy.

[0109] Preferably, in one embodiment of the present invention, based on the quantitative characteristics of pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various characteristic parameters, characteristic parameters are selected to construct a decision tree, thereby obtaining a trained random forest model, including:

[0110] In each randomly selected sample subset, the ratio of the number of pipeline echo signals of each defect type to the number of all pipeline echo signals is taken as the proportion of the number of pipeline echo signals of each defect type. The purpose of this step is to understand the data distribution in each sample subset. The larger the proportion of the number of pipeline echo signals of a certain defect type, the more pipeline echo signals of this defect type there are in the sample subset. In this case, the weight of the feature parameters that perform more prominently under the pipeline echo signal of this defect type should be increased, that is, the candidate degree value should be larger.

[0111] Therefore, the product of the quantity proportion of each pipeline echo signal and the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs is taken as the candidate factor of each characteristic parameter in each sample subset. At this time, when the quantity proportion is larger and the characterization degree value is larger, the candidate factor is larger, which means that the corresponding characteristic parameter is more representative in the sample subset and can better distinguish the differences between defect types.

[0112] At this point, the candidate factors of each characteristic parameter under each pipeline echo signal can be obtained in each sample subset, and then the mean of all candidate factors of each characteristic parameter is normalized as the candidate degree value of each characteristic parameter in each sample subset. The larger the candidate degree value, the stronger the representativeness of the characteristic parameter in the sample subset, and the better the distinguishability of the characteristic parameter for the sample subset. Normalization is a technical means well known to those skilled in the art, and the selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0113] The sorting step helps to identify the final feature parameters, which should be given priority when building decision trees and random forest models. Therefore, in each sample subset, all types of feature parameters are arranged in descending order according to the candidate degree values ​​to obtain the sorted sequence.

[0114] Finally, in each sample subset, feature parameters with high candidate degree values ​​are preferentially selected, that is, feature parameters are selected according to the arrangement order of feature parameters in the sorting sequence to construct a decision tree, and multiple decision trees are obtained. This step ensures that the model gives priority to the most discriminative feature parameters during training, which helps to build a simpler and more accurate decision tree, thereby integrating the decision trees of all sample subsets to improve the robustness and accuracy of the model, and a trained random forest model can be obtained.

[0115] It should be noted that the construction of a decision tree and the process of integrating decision trees to obtain a random forest model are both well-known technologies and will not be described in detail here; the number of randomly selected sample subsets and the number of pipeline echo signals in each sample subset can be adjusted according to the implementation scenario and will not be described in detail here.

[0116] At this point, based on the above steps, a trained random forest model can be obtained, which can be used to perform defect detection on the pipeline to be tested.

[0117] Specifically, the pipeline ultrasonic data of the pipeline to be tested is obtained, and the pipeline echo signal corresponding to the pipeline to be tested can be extracted according to the process in steps S2 and S3, and then various characteristic parameters in the pipeline echo signal are obtained, and the obtained characteristic parameters are used as the input of the trained random forest model, and finally the detection result of the pipeline to be tested can be obtained. If there is a defect in the pipeline to be tested, the detection result is the type of the defect, and if there is no defect in the pipeline to be tested, the detection result is normal.

[0118] In summary, in an embodiment of the present invention, the ultrasonic data of a pipeline with known defect types is used to train a random forest model, thereby performing defect detection on the pipeline to be tested. First, a pipeline ultrasonic data set is obtained, which includes pipeline ultrasonic data of multiple known defect types. Since the surface of the water conservancy pipeline will be wrapped with concrete or other materials, there will be interference data in the pipeline ultrasonic data, which is not a real and effective pipeline echo signal. Therefore, in the pipeline ultrasonic data, the reflection probability is determined based on the significant situation of the data value in the local range, and the signal segment is divided according to the reflection probability. Further, in each signal segment, the sound path analysis is performed and the reflection probability is combined to determine the pipeline echo probability of each signal segment; since concrete is a porous and inhomogeneous material, the sound wave will be scattered in large quantities during propagation, so the random fluctuation is large, and the echo generated when the pipeline surface is reflected is relatively small. Therefore, there will be differences in time-frequency characteristics between different signal segments, so the time-frequency difference between the signal segments is used to adjust the pipeline echo probability, obtain the corrected probability, and determine the real and effective pipeline echo signal in all signal segments based on this indicator, avoiding the interference of signals such as noise. Furthermore, characteristic parameters of each pipeline echo signal are obtained. Since the characterization degree, that is, the importance, of different characteristic parameters is different under different defect types, in the embodiment of the present invention, the characterization degree value of each characteristic parameter under each defect type is determined according to the difference and fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters, which is used to reflect the weight. Then, several sample subsets are randomly selected, and characteristic parameters are selected according to the quantity characteristics of pipeline echo signals of different defect types and the characterization degree values ​​of characteristic parameters to construct a decision tree, so as to obtain a trained random forest model. At this time, the trained random forest model effectively eliminates the interference of noise signals on pipeline echo signals, and the representativeness of the selected characteristic parameters is strong enough, so the reliability of the model will be higher. Finally, based on the model, the pipeline to be tested is inspected to obtain more accurate pipeline detection results.

[0119] An embodiment of the present invention also provides a water conservancy project pipeline flaw detection device, including a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and when the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor, the steps of a water conservancy project pipeline flaw detection method are implemented.

[0120] See also Figure 2, which shows a schematic diagram of the device structure of a water conservancy project pipeline flaw detection device provided by an embodiment of the present invention, including a processor 200, a memory 201, a bus 202 and a communication interface 203, wherein the processor 200, the communication interface 203 and the memory 201 are connected via the bus 202; wherein the memory 201 may include a high-speed random access memory, the bus 202 may be an ISA bus, a PCI bus or an EISA bus, etc., and the processor 200 may be an integrated circuit chip with signal processing capabilities; the memory 201 stores at least one instruction, at least one program, a code set or an instruction set, and when the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor, a step in a water conservancy project pipeline flaw detection method is implemented.

[0121] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0122] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for flaw detection of water conservancy project pipelines, characterized in that: The method comprises: Acquire a pipeline ultrasonic data set, wherein the ultrasonic data set includes pipeline ultrasonic data of various defect types; In each pipeline ultrasonic data, the reflection probability at each moment is determined according to the significance of each data value in the local range; each pipeline ultrasonic data is segmented based on the reflection probability at each moment to obtain multiple signal segments; Perform acoustic path analysis in each signal segment and determine the pipeline echo probability of each signal segment in combination with the reflection probability; adjust the pipeline echo probability of each signal segment using the time-frequency difference between the signal segments to obtain a corrected probability; determine the pipeline echo signal in all signal segments of each pipeline ultrasonic data based on the corrected probability; Obtain multiple characteristic parameters of each pipeline echo signal; determine the characterization degree value of each characteristic parameter under each defect type according to the numerical differences and numerical fluctuation characteristics of pipeline echo signals of different defect types under the same characteristic parameters; A sample subset is randomly selected from all pipeline echo signals. In each sample subset, based on the quantity characteristics of pipeline echo signals of different defect types and the characterization degree values ​​corresponding to various characteristic parameters, characteristic parameters are selected to construct a decision tree, thereby obtaining a trained random forest model; multiple characteristic parameters of the pipeline echo signal of the pipeline to be tested are used as inputs of the trained random forest model, and the pipeline detection results are output.

2. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The method for obtaining the reflection probability includes: Determine the preset neighborhood corresponding to each moment; The mean of the data values ​​of all neighboring moments in the pipeline ultrasonic data at each moment is taken as the mean eigenvalue; The normalized value of the difference between the data value in the pipeline ultrasonic data at each moment and the mean characteristic value is used as the reflection probability at each moment.

3. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The ultrasonic data of each pipeline is segmented based on the reflection probability at each time to obtain multiple signal segments, including: The reflection probabilities at all times are smoothly connected in time series to obtain a reflection probability curve; On the reflection probability curve, a maximum point is obtained, and a data segment corresponding to a curve segment between any two adjacent maximum points in the pipeline ultrasonic data is used as a signal segment.

4. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The method for obtaining the pipeline echo probability includes: In each signal segment, the mean of the reflection probabilities at the two endpoints is taken as the intensity characteristic value; The product of the time interval between the two endpoints and the propagation speed of the sound wave in the pipeline material is taken as the actual pipe wall thickness, and the absolute value of the difference between the actual pipe wall thickness and the preset pipe wall thickness is negatively correlated and normalized to obtain a similarity factor; The product of the intensity characteristic value corresponding to each signal segment and the similarity factor is normalized to obtain the value, which is used as the pipeline echo probability of each signal segment.

5. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The method for obtaining the modified probability includes: Perform short-time Fourier transform on each pipeline ultrasonic data to obtain a time-frequency spectrum diagram; For any signal segment in each pipeline ultrasonic data, the mean of the amplitude values ​​at the same frequency at all times in the signal segment is taken as the amplitude mean; Combine any two signal segments to obtain all non-repeating signal segment combinations; In each signal segment combination, the absolute value of the difference between the amplitude means of the two signal segments at the same frequency is calculated as the deviation factor, and the sum of the deviation factors of the two signal segments at all frequencies is normalized to obtain the value as the frequency domain amplitude difference between the two signal segments; The mean value of the frequency domain amplitude difference between each signal segment and all other signal segments is multiplied by the pipeline echo probability of each signal segment, and the obtained product is normalized as the corrected probability of each signal segment.

6. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The method for acquiring the pipeline echo signal comprises: Among all signal segments of each pipeline ultrasonic data, the signal segment with the largest correction probability is taken as the pipeline echo signal.

7. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The method for obtaining the characterization degree value comprises: Classify all pipeline echo signals according to their defect categories; In each pipeline echo signal, the standard deviation of the value of each characteristic parameter in all pipeline echo signals is taken as the fluctuation factor; Calculate the numerical mean of all pipeline echo signals in each pipeline echo signal under each characteristic parameter as the characteristic value; For any pipeline echo signal, under the same characteristic parameters, the absolute value of the difference between the characteristic value corresponding to the pipeline echo signal and the characteristic value of each other pipeline echo signal is used as the difference factor, and the mean value of all the difference factors corresponding to the pipeline echo signal under each characteristic parameter is used as the difference characteristic value; According to the difference characteristic value and fluctuation factor of each characteristic parameter under each pipeline echo signal, the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs is obtained, and the characterization degree value is negatively correlated with the fluctuation factor, and the characterization degree value is positively correlated with the difference characteristic value.

8. A method for flaw detection of water conservancy engineering pipelines according to claim 1, characterized in that: The quantitative characteristics of the pipeline echo signals based on different defect types and the characterization degree values ​​corresponding to various characteristic parameters are selected to construct a decision tree, thereby obtaining a trained random forest model, including: In each sample subset, the quantitative characteristics of the pipeline echo signals of different defect types and the characterization degree value of each characteristic parameter under each defect type are analyzed to determine the candidate degree value of each characteristic parameter in each sample subset; In each sample subset, all types of feature parameters are arranged in descending order according to the candidate degree values ​​to obtain a sorted sequence; In each sample subset, feature parameters are selected according to the arrangement of the feature parameters in the sorting sequence to construct a decision tree, and multiple decision trees are obtained, thereby obtaining a trained random forest model.

9. A method for flaw detection of water conservancy engineering pipelines according to claim 8, characterized in that: The method for obtaining the candidate degree value includes: In each sample subset, the ratio of the number of pipeline echo signals of each defect type to the number of all pipeline echo signals is taken as the number ratio of pipeline echo signals of each defect type; The product of the quantity proportion of each pipeline echo signal and the characterization degree value of each characteristic parameter under the defect type to which each pipeline echo signal belongs is taken as the candidate factor of each characteristic parameter in each sample subset; In each sample subset, the normalized value of the mean of all candidate factors of each feature parameter is used as the candidate degree value of each feature parameter in each sample subset.

10. A water conservancy project pipeline flaw detection equipment, characterized in that: The method comprises a processor and a memory, wherein at least one instruction, at least one program, a code set or an instruction set is stored in the memory, and when the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor, the steps of a method for flaw detection in a water conservancy project pipeline as described in any one of claims 1 to 9 are implemented.

Images