A quantitative method for detecting cracks on the inner wall of steel pipelines

Abstract

The present application relates to the technical field of computer systems based on specific calculation models, and specifically to a method for quantifying crack detection on the inner wall of a steel pipeline, including: obtaining a set of eddy current multi-channel detection signals of a cracked pipeline; obtaining a defect signal data set; converting the defect signal data set into a signal image data set; inputting the signal image data set into a YOLOv8 model for training to obtain a crack defect annotation model; calculating crack defect signal range data of signals in the signal image data set; obtaining a crack defect signal characteristic value data set; obtaining a crack depth quantification value and a crack length quantification value; determining the number of channels; and obtaining a crack width quantification value. The present invention solves the problems of poor consistency of manual annotation in the prior art, and large limitations in the use process and scope of application.

Description

Technical Field

[0001] The present application relates to the technical field of computer systems based on specific calculation models, and in particular, to a method for quantifying the detection of cracks on the inner wall of a steel pipeline. Background Art

[0002] Oil and gas pipelines are the "lifeline" of energy transportation, and their inherent safety is related to the national economy and people's livelihood. Cracks are the most common failure mode of pipeline systems. Cracks expand from small damage to large ones, gradually reducing the pressure-bearing capacity of the pipeline, and eventually leading to pipeline failure and serious accidents. Among them, the detection of small cracks in pipelines is a difficult problem.

[0003] In the prior art, a GA-BP neural network model based on pulsed eddy current technology is proposed, which can fit the nonlinear trend between magnetic induction intensity and crack depth and width, so as to quantitatively characterize cracks in ferromagnetic steel. However, this method needs to achieve quantification by fitting the nonlinear trend between magnetic induction intensity and crack depth and width. This fitting requires targeted analysis of the detection signal, requires an in-depth understanding of the detection principle, and has high requirements for the staff. On the other hand, the processing of the detection signal is manual. The quality and efficiency of manual defect annotation are affected by the professional knowledge and working status of the staff, and the annotation consistency is poor. In addition, it is impossible to use multi-channel detection signals, and the use process and scope of application are greatly limited. Summary of the invention

[0004] The purpose of the present application is to provide a method for quantifying the detection of cracks on the inner wall of a steel pipeline, which solves the problems in the prior art of poor consistency of manual marking and large limitations in the use process and scope of application.

[0005] The technical solution of this application:

[0006] The present application provides a method for detecting and quantifying cracks on the inner wall of a steel pipeline, which is executed by a processor and includes:

[0007] Obtaining a set of eddy current multi-channel detection signals of a cracked pipe;

[0008] Annotate the eddy current multi-channel detection signal set to obtain the defect signal data set;

[0009] Perform data cleaning on the defect signal data set, and convert the defect signal data set into a signal image data set;

[0010] The signal image data set is input into the YOLOv8 model for training to obtain a crack defect annotation model;

[0011] Based on the signal image data set and the crack defect annotation model, the crack defect signal range data of the signal in the signal image data set is calculated;

[0012] Based on the crack defect signal range data, a crack defect signal characteristic value data set is obtained;

[0013] The crack defect signal characteristic value dataset is input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighting method to obtain the quantified values ​​of crack depth and crack length;

[0014] Determine the number of channels based on the eddy current multi-channel detection signal set;

[0015] The crack defect signal eigenvalue dataset and channel number are input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighted method to obtain the quantified value of the crack width.

[0016] Furthermore, the method of obtaining the eddy current multi-channel detection signal set of the cracked pipe is specifically as follows: scanning the cracks of the target pipe under the conditions of fixed lift-off value and fixed parameters, obtaining the eddy current multi-channel detection signal, and forming the eddy current multi-channel detection signal set.

[0017] Furthermore, the eddy current multi-channel detection signal set is labeled to obtain a defect signal data set, specifically:

[0018] Manually annotate crack defect signal segments on eddy current multi-channel detection signals to obtain defect annotation data;

[0019] Data enhancement is performed on the defect annotation data to obtain defect signal data and form a defect signal dataset.

[0020] Furthermore, the data cleaning is performed on the defect signal data set to convert the defect signal data set into a signal image data set, specifically:

[0021] Convert defect signal segments in defect annotation data into signal raster images;

[0022] The signal grid image is subjected to equal length correspondence processing and scaling processing to obtain signal image data and form a signal image data set.

[0023] Furthermore, the YOLOv8 model is obtained based on the pre-trained model yolov8n.pt.

[0024] Furthermore, the crack defect signal range data of the signal in the signal image data set is calculated based on the signal image data set and the crack defect annotation model, specifically:

[0025] Input the signal image data set into the crack defect annotation model to perform defect signal annotation to obtain signal defect annotation data;

[0026] Inputting the eddy current multi-channel detection signal set into the crack defect annotation model to perform defect signal annotation and obtain automatic annotation data;

[0027] Acquire conversion ratio data based on the signal image data;

[0028] Based on the conversion ratio data and the automatic annotation data, the crack defect signal range data of the signal defect annotation data is calculated.

[0029] Furthermore, the crack defect signal characteristic value data set is obtained based on the crack defect signal range data, specifically:

[0030] Performing normalization processing on crack defect signal range data;

[0031] Merge multi-channel signals into a single-channel signal;

[0032] The eigenvalue calculation is performed on the single-channel signal to obtain the crack defect signal eigenvalue data set.

[0033] Furthermore, the crack defect signal characteristic value data set is input into the LightGBM model for training to obtain the crack depth quantization value and the crack length quantization value, specifically:

[0034] Based on the LightGBM model, the five-fold cross-validation method and parameter search method are used to train and test the crack defect signal eigenvalue data set to obtain the primary depth quantization model and the primary width quantization model.

[0035] Perform parameter adjustment on the primary depth quantization model and the primary width quantization model to obtain an optimal depth quantization model and an optimal width quantization model;

[0036] The crack defect signal characteristic value data set is input into the optimal depth quantization model and the optimal width quantization model to perform quantization, and the crack depth quantization value and the crack length quantization value are obtained.

[0037] Furthermore, the crack defect signal characteristic value data set and the number of channels are input into the LightGBM model for training to obtain the crack width quantization value, specifically:

[0038] Based on the crack defect signal characteristic value data set, the width quantitative characteristic value is calculated;

[0039] Calculate the quantized value based on the width quantization feature value ;

[0040] The quantized value is input into the LightGBM model to perform quantization and obtain the quantized value of the crack width; where:

[0041]

[0042] In the formula, is the true width of the crack defect, is the multi-channel sensor spacing.

[0043] The technical solution of the present application has at least the following advantages and beneficial effects:

[0044] The present application provides a method for quantifying the detection of cracks on the inner wall of a steel pipeline, by providing a method for converting a defect signal data set into a signal image data set, and then using a YOLOv8 model to automatically annotate the crack defects in the signal image data set. After obtaining the annotated defect signal segment, a crack defect signal eigenvalue data set is obtained. Finally, the crack defect signal eigenvalue data set is input into a LightGBM model to quantify the defect length, width and depth. Compared with the methods of the prior art, the method provided by the present application can realize automatic annotation of crack defects, full extraction of signal features and utilization of multiple data channels. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the detection method of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Application Overview

[0048] Common existing technologies for steel pipeline crack detection include eddy current, electromagnetic ultrasound, ACFM, etc. These detection technologies have their own applicable scenarios, but they still have certain limitations in some cases. The existing detection quantification methods are generally as follows: Step 1: Use electromagnetic detection technologies such as eddy current, ACFM, alternating excitation, etc. to detect defects such as cracks; Step 2: Build a steel pipeline simulation model on the simulation software or build a test platform for simulating steel pipeline internal detection in the laboratory, and test the specimens containing cracks through simulation or laboratory tests. Each detection requires human intervention, and the crack detection data set is obtained through manual processing; Step 3: Based on this data set, quantify the length, width or depth of the crack. However, this type of method requires staff to extract signal features based on the understanding of the principles of detection technology, combined with machine learning methods, and realize signal quantification. There are high professional requirements for the staff performing the detection, resulting in a small scope of application, and frequent human intervention is required to complete, and the implementation process is relatively complicated.

[0049] Based on the foregoing, the detection method provided in this application converts the defect signal dataset into a signal image dataset, and then uses the YOLOv8 model to automatically annotate the crack defects in the signal image dataset. After obtaining the annotated defect signal segment, the crack defect signal eigenvalue dataset is obtained. Finally, the crack defect signal eigenvalue dataset is input into the LightGBM model to quantify the defect length, width and depth.

[0050] Example

[0051] Please refer to Figure 1 The present application provides a method for detecting and quantifying cracks on the inner wall of a steel pipeline. The method is executed by a processor and includes:

[0052] Obtaining a set of eddy current multi-channel detection signals of a cracked pipe;

[0053] Annotate the eddy current multi-channel detection signal set to obtain the defect signal data set;

[0054] Perform data cleaning on the defect signal data set, and convert the defect signal data set into a signal image data set;

[0055] The signal image data set is input into the YOLOv8 model for training to obtain a crack defect annotation model;

[0056] Based on the signal image data set and the crack defect annotation model, the crack defect signal range data of the signal in the signal image data set is calculated;

[0057] Based on the crack defect signal range data, a crack defect signal characteristic value data set is obtained;

[0058] The crack defect signal characteristic value dataset is input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighting method to obtain the quantified values ​​of crack depth and crack length;

[0059] Determine the number of channels based on the eddy current multi-channel detection signal set;

[0060] The crack defect signal eigenvalue dataset and channel number are input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighted method to obtain the quantified value of the crack width.

[0061] It should be noted that, compared with the methods of the prior art, the method provided in the present application can realize automatic marking of crack defects, sufficient extraction of signal features and utilization of multiple data channels, so that the entire implementation process can reduce the professional requirements for staff and does not require excessive human participation or interference, thus solving the problems of poor consistency of manual marking in the prior art and large limitations in usage process and scope of application.

[0062] It should be noted that this embodiment applies digital signal features to pipeline crack signals. The difference from the existing technical solution, that is, the traditional manual feature method, is that this embodiment can directly use digital signal features and no longer requires manual analysis of signal principles. At the same time, the feature extraction method is more universal.

[0063] In detail, the method described is based on micro-crack weighting to perform parameter tuning on the LightGBM model, wherein a micro-crack weighted simulated annealing algorithm is used for parameter tuning.

[0064] The following is a detailed description of the detection and quantification method provided in this application:

[0065] In some embodiments, before obtaining the eddy current multi-channel detection signal, it is necessary to build a laboratory steel pipeline internal detection simulation scanning test platform based on differential eddy current detection technology extracted by incremental magnetic permeability.

[0066] In some embodiments, the method of obtaining a set of eddy current multi-channel detection signals of a cracked pipe is as follows: scanning the cracks of the target pipe under the conditions of fixed lift-off value and fixed parameters, obtaining eddy current multi-channel detection signals, and forming an eddy current multi-channel detection signal set. In detail, the eddy current multi-channel detection signals are obtained through laboratory standard parts testing, simulation experiments, pulling tests, etc., and at least 204 eddy current multi-channel detection signals are obtained.

[0067] In some embodiments, the eddy current multi-channel detection signal set is labeled to obtain a defect signal data set, specifically:

[0068] Manually annotate crack defect signal segments on eddy current multi-channel detection signals to obtain defect annotation data;

[0069] Data enhancement is performed on the defect annotation data to obtain defect signal data and form a defect signal dataset.

[0070] Optionally, the enhancement method is to shift the defect signal range left or right along the sampling point direction / time direction on the premise of including the defect, and the movement cannot exceed the actual defect signal segment, thereby increasing the data volume on the premise of including the defect signal segment and obtaining defect signal data.

[0071] In some embodiments, the data cleaning of the defect signal data set is performed to convert the defect signal data set into a signal image data set, specifically:

[0072] Convert defect signal segments in defect annotation data into signal raster images;

[0073] The signal grid image is subjected to equal length correspondence processing and scaling processing to obtain signal image data and form a signal image data set.

[0074] In detail, the defect signal segments in the defect signal data set are converted into signal grid images, so that the defect signal segment sampling interval, the signal reference value of each channel and the grid image side length correspond to each other in a certain ratio, and the signal image data set is obtained.

[0075] In some embodiments, the YOLOv8 model is obtained based on the pre-trained model yolov8n.pt.

[0076] In detail, the signal image data set is randomly divided into a training set, a validation set, and a test set. A YOLOv8 model is obtained based on the pre-trained model yolov8n.pt, and then the training set is sequentially input into the YOLOv8 model for training. The output content includes: the coordinates (x, y), width (w), and height (h) of the center point of the bounding box of the image normalized relative to the size of the input image; category information; confidence information. The deviation between the output of the pre-trained model during the training process and the true value of the training set is calculated using the default settings of YOLOv8 for loss value calculation, error feedback, and parameter optimization. The validation set is used to evaluate the YOLOv8 model during the training process. After the model converges, the performance of the model is evaluated using the test set, and finally a crack defect annotation model is obtained. The crack defect annotation model is used to automatically annotate eddy current multi-channel detection signals. It should be noted that, compared with other pre-trained models, the pre-trained model yolov8n.pt selected in this embodiment has fewer parameters and faster training speed.

[0077] In some embodiments, the calculation of the crack defect signal range data of the signal in the signal image data set based on the signal image data set and the crack defect annotation model is specifically:

[0078] Input the signal image data set into the crack defect annotation model to perform defect signal annotation to obtain signal defect annotation data;

[0079] Inputting the eddy current multi-channel detection signal set into the crack defect annotation model to perform defect signal annotation and obtain automatic annotation data;

[0080] Acquire conversion ratio data based on the signal image data;

[0081] Based on the conversion ratio data and the automatic annotation data, the crack defect signal range data of the signal defect annotation data is calculated.

[0082] In detail, the annotated image is used to obtain the start and end points of each defect in the original one-dimensional signal according to the annotation information, the image pixel value and the signal sampling point interval ratio, and then the defect signal segment is obtained to realize the automatic extraction of defects and features. The range refers to the defect start and end points in the time unit, and the channel number containing the defect.

[0083] In some embodiments, the method of obtaining a crack defect signal characteristic value data set based on the crack defect signal range data is specifically as follows:

[0084] Performing normalization processing on crack defect signal range data;

[0085] Merge multi-channel signals into a single-channel signal;

[0086] The eigenvalue calculation is performed on the single-channel signal to obtain the crack defect signal eigenvalue data set.

[0087] Specifically, each channel signal within the range of the grain defect signal data is normalized, and then a calculation method among the maximum, minimum and average is selected to merge the multi-channel signals into a single-channel signal, and then the single-channel signal mean is calculated. ,variance , RMS , mean square value , Standard Deviation , peak value, peak value , Kurtosis factor , skewness factor , Form Factor , Peak Factor , Pulse Factor , Margin Factor And energy value The 14 eigenvalues ​​are calculated to obtain the crack defect signal eigenvalue data set. In the above, the maximum is the signal of the channel with the largest peak-to-peak value, the minimum is the signal of the channel with the smallest peak-to-peak value, and the average is the average value of the sampling points at the corresponding time of each channel. Finally, a one-dimensional signal is obtained, where:

[0088] Mean for:

[0089]

[0090] variance for:

[0091]

[0092] RMS for:

[0093]

[0094] Mean Square Value for:

[0095]

[0096] Standard Deviation for:

[0097]

[0098] The peak value is:

[0099]

[0100] Peak value for:

[0101]

[0102] Kurtosis factor for:

[0103]

[0104] Skewness Factor for:

[0105]

[0106] Form Factor for:

[0107]

[0108] Crest Factor for:

[0109]

[0110] Pulse Factor for:

[0111]

[0112] Margin Factor for:

[0113]

[0114]

[0115] Energy value for:

[0116]

[0117] in, f It should be noted that the above formulas are all prior art.

[0118] In some embodiments, the crack defect signal characteristic value data set is input into the LightGBM model for training to obtain the crack depth quantization value and the crack length quantization value, specifically:

[0119] Based on the LightGBM model, the five-fold cross-validation method and parameter search method are used to train and test the crack defect signal eigenvalue data set to obtain the primary depth quantization model and the primary width quantization model.

[0120] Perform parameter adjustment on the primary depth quantization model and the primary width quantization model to obtain an optimal depth quantization model and an optimal width quantization model;

[0121] The crack defect signal characteristic value data set is input into the optimal depth quantization model and the optimal width quantization model to perform quantization, and the crack depth quantization value and the crack length quantization value are obtained.

[0122] In detail, the crack defect depth and width are quantified based on the crack defect signal characteristic value data set and the LightGBM model to obtain the primary depth quantization model and the primary width quantization model. The primary depth quantization model and the primary width quantization model are trained and tested by the 5-fold cross-validation method. The value range of the optional parameters is set. The parameter search is used to carry out the parameter tuning of the primary depth quantization model and the primary width quantization model while using the 5-fold cross-validation. The mean absolute error (MAE) is used to measure the difference between the predicted value and the true value to obtain the optimal depth quantization model and the optimal width quantization model. Finally, the crack defect signal characteristic value data set is input into the optimal depth quantization model and the optimal width quantization model for quantization to obtain the crack depth quantization value and the crack length quantization value. The simulated annealing algorithm can accept the performance degradation of the model within a certain range until the performance rises again, which can avoid the model from falling into the local optimum. The advantage of weighted adjustment is that it can make the parameter tolerance of the poor performance of small cracks in parameter adjustment greater, and the parameter tolerance of the poor performance of large cracks smaller. In detail, five-fold cross validation and parameter search are performed simultaneously, the data is divided into five parts and one of them is used for validation in turn, and parameter search is performed on each part in the process of rotation. Optionally, the method for implementing parameter search can use grid search or simulated annealing algorithm based on small crack weighting.

[0123] In some embodiments, the specific steps of optimizing the parameters of the improved simulated annealing based on the simulated annealing algorithm and the weighting of small cracks to obtain the quantized value of the crack depth, the quantized value of the crack length, and the quantized value of the crack width are as follows:

[0124] S1. Initialization parameter combination: randomly initialize a set of LightGBM parameters as the initial solution and set the initial temperature , Temperature drop rate and termination temperature ;

[0125] S2, fitness evaluation: calculate the performance of the LightGBM model corresponding to the initial parameter combination on the training set and the validation set, and use the performance as the fitness of the initial solution;

[0126] S3, generate new solutions: by making small random changes based on the current parameter combination, for example, adding a small random value to each parameter, thereby generating a new parameter combination;

[0127] S4. Accept the new solution:

[0128] When the crack depth and crack length are quantified: the LightGBM model performance corresponding to the new solution is calculated. If the fitness of the new solution is better than the current solution, the new solution is accepted; if the fitness of the new solution is worse than the current solution, the new solution is accepted with probability. Accept the new interpretation, which is is the difference between the fitness of the new solution and the current solution, is the current temperature, L is all the values ​​of the defect length, and each value is , D is all the values ​​of defect depth, each value is ,but is the maximum value in length, is the maximum value in depth;

[0129] When the crack width is quantified: calculate the LightGBM model performance corresponding to the new solution. If the fitness of the new solution is better than the current solution, accept the new solution; if the fitness of the new solution is worse than the current solution, Accept the new interpretation, which is is the difference between the fitness of the new solution and the current solution, is the current temperature, W is all the values ​​of the defect width, and each value is ,but is the maximum value of width;

[0130] S5, temperature drop: If the new solution is accepted, the new solution is used as the current solution; then the temperature drops at a certain rate. Lower the temperature, i.e. ;

[0131] S6, iterative search: Repeat the generation of new solutions, acceptance of new solution judgments and temperature reduction operations until the temperature drops to the termination temperature T end .

[0132] In some embodiments, the crack defect signal characteristic value data set and the number of channels are input into the LightGBM model for training to obtain the crack width quantization value, specifically:

[0133] Based on the crack defect signal characteristic value data set, the width quantitative characteristic value is calculated;

[0134] Calculate the quantized value based on the width quantization feature value ;

[0135] The quantized value is input into the LightGBM model to perform quantization and obtain the quantized value of the crack width; where:

[0136]

[0137] In the formula, is the true width of the crack defect, is the multi-channel sensor spacing.

[0138] In detail, the length quantization eigenvalue is calculated based on the aforementioned 14 eigenvalues (i=1, 2, 3, ...14), then calculate the quantized value Then, the LightGBM model is used to quantify the defect length. The parameter setting and training method in the quantization are the same as the process of obtaining the crack width quantization value.

[0139]

[0140] In the formula, the eigenvalue corresponding to each channel is (i=1, 2, 3, ...14), the channel number containing the crack defect is c (i=1,2,3,…14).

[0141] In some embodiments, other advanced lightweight electromagnetic detection technologies suitable for detecting tiny cracks in steel pipes can be used to replace the eddy current detection technology in this article to complete the detection inside the steel pipe.

[0142] In some embodiments, other machine learning algorithms, such as random forests, neural networks, etc., can be used to input feature values ​​into the algorithm for crack quantification.

[0143] In some embodiments, eigenvalues ​​may be further combined, eigenvalues ​​may be screened, eigenvalues ​​may be PCA transformed, etc. to achieve optimization of eigenvalues.

[0144] It is worth noting that the prior art designs an improved eddy current probe, which reconstructs the crack length and depth based on the ANN model and the detection of the normal magnetic component. However, this method achieves quantification by studying the detection of the normal magnetic component, which requires a deep understanding of the detection principle, and the processing of the detection signal is manual. It still has high professional requirements for the staff, and human interference is frequent. It is also impossible to automatically extract defects and features, and it is impossible to use multi-channel detection signals. Another prior art is based on spatial magnetic signal extraction technology and machine learning algorithms, and proposes an online detection and characterization model for submarine pipeline cracks, which extracts multiple signal features to characterize the tiny defects of submarine pipelines. However, this method also processes the detection signal manually, and it is also impossible to automatically extract defects and features, and it is impossible to use multi-channel detection signals. There is also an existing technology that uses the leakage magnetic detection signal to calculate the radial and axial characteristic quantities (seven characteristic quantities: radial component peak-to-valley value, radial component peak-to-valley distance, axial component peak-to-valley value, axial component span value, axial differential peak-to-valley distance, axial component signal waveform area and axial component signal waveform energy) and inputs them into the PSO-RBF neural network model improved by the particle swarm algorithm for defect quantification. However, this method also processes the detection signal manually and cannot realize automatic extraction of defects and features. There is also the problem of incomplete extraction of defect signal characteristic values. The detection method provided in this embodiment provides the application of multiple one-dimensional signal time-frequency features to the quantification of crack detection in pipelines, so that the signal feature extraction in the defect quantification work is more comprehensive, and uses LightGBM to quantify the length, width and depth of the crack to realize the automatic extraction of more features, which will greatly reduce human participation or interference, solve the problem of insufficient utilization of existing defect signal information and the need for principle analysis, and also solve the problem of insufficient utilization of defect signal information; by converting the detection signal into a raster image and using a crack defect annotation model based on YOLOv8, the defect signal segment of the original detection signal is obtained, and the automatic annotation and extraction of the new defect segment is realized, solving the problem of manual annotation of crack defects in the detection signal.

[0145] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. For those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic features of the present application. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present application is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0146] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for detecting and quantifying cracks on the inner wall of a steel pipeline, the method being executed by a processor, characterized in that: include: Obtaining a set of eddy current multi-channel detection signals of a cracked pipe; Annotate the eddy current multi-channel detection signal set to obtain the defect signal data set; Perform data cleaning on the defect signal dataset and convert it into a signal image dataset. Specifically: Convert defect signal segments in defect annotation data into signal raster images; Performing equal-length correspondence processing and scaling processing on the signal grid image to obtain signal image data and form a signal image data set; Input the signal image data set into the YOLOv8 model for training to obtain a crack defect annotation model, wherein the YOLOv8 model is obtained based on the pre-trained model yolov8n.pt; Based on the signal image data set and the crack defect annotation model, the crack defect signal range data of the signal in the signal image data set is calculated, specifically: Input the signal image data set into the crack defect annotation model to perform defect signal annotation to obtain signal defect annotation data; Inputting the eddy current multi-channel detection signal set into the crack defect annotation model to perform defect signal annotation and obtain automatic annotation data; Acquire conversion ratio data based on the signal image data; Calculate crack defect signal range data of signal defect annotation data based on conversion ratio data and automatic annotation data; Based on the crack defect signal range data, a crack defect signal characteristic value data set is obtained; The crack defect signal characteristic value dataset is input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighting method to obtain the quantified values ​​of crack depth and crack length; Determine the number of channels based on the eddy current multi-channel detection signal set; The crack defect signal eigenvalue dataset and channel number are input into the LightGBM model for training, and the parameters of the LightGBM model are tuned based on the micro-crack weighted method to obtain the quantified value of the crack width.

2. The detection and quantification method according to claim 1, characterized in that: The method of obtaining the eddy current multi-channel detection signal set of the cracked pipe is specifically as follows: scanning the cracks of the target pipe under the conditions of fixed lift-off value and fixed parameters, obtaining the eddy current multi-channel detection signal, and forming the eddy current multi-channel detection signal set.

3. The detection and quantification method according to claim 2, characterized in that: The method of labeling the eddy current multi-channel detection signal set to obtain the defect signal data set is specifically as follows: Manually annotate crack defect signal segments on eddy current multi-channel detection signals to obtain defect annotation data; Data enhancement is performed on the defect annotation data to obtain defect signal data and form a defect signal dataset.

4. The detection and quantification method according to claim 1, characterized in that: The method of obtaining the crack defect signal characteristic value data set based on the crack defect signal range data is specifically as follows: Performing normalization processing on crack defect signal range data; Merge multi-channel signals into a single-channel signal; The eigenvalue calculation is performed on the single-channel signal to obtain the crack defect signal eigenvalue data set.

5. The detection and quantification method according to claim 1, characterized in that: The crack defect signal characteristic value data set is input into the LightGBM model for training to obtain the crack depth quantization value and the crack length quantization value. Specifically: Based on the LightGBM model, the five-fold cross-validation method and parameter search method are used to train and test the crack defect signal eigenvalue data set to obtain the primary depth quantization model and the primary length quantization model. Perform parameter adjustment on the primary depth quantization model and the primary length quantization model to obtain an optimal depth quantization model and an optimal length quantization model; The crack defect signal characteristic value data set is input into the optimal depth quantization model and the optimal length quantization model to perform quantization, and the crack depth quantization value and the crack length quantization value are obtained.

6. The detection and quantification method according to claim 1, characterized in that: The crack defect signal characteristic value data set and the number of channels are input into the LightGBM model for training to obtain the crack width quantization value, specifically: Based on the crack defect signal characteristic value data set, the width quantitative characteristic value is calculated; Calculate the quantized value based on the width quantization feature value ; The quantized value is input into the LightGBM model to perform quantization and obtain the quantized value of the crack width; where: ; In the formula, is the true width of the crack defect, is the multi-channel sensor spacing.

Images