A multi-wheel incomplete magnetic leakage data fusion method

Abstract

The application provides a multi-rotation incomplete magnetic flux leakage data fusion method, and relates to the technical field of data fusion, comprising: acquiring multi-rotation magnetic flux leakage data; extracting magnetic flux leakage data features; inputting the magnetic flux leakage data features into a random forest model to determine effective magnetic flux leakage data features, invalid magnetic flux leakage data features and channel types corresponding to the magnetic flux leakage data; extracting spatial information of part of effective channels and complete incomplete channels; determining reference wheel data; based on the reference wheel data, performing circumferential rotation adjustment and axial mileage adjustment on all signal points of the remaining rotation magnetic flux leakage data; acquiring effective magnetic flux leakage data features in the non-reference wheel data located in the incomplete area interval; and fusing the effective magnetic flux leakage data features into the reference wheel data to obtain magnetic flux leakage fusion data, so as to solve the problems of repeated or missing axial or circumferential fusion boundaries, excessive signal stretching and extrusion, and signal jumping caused by non-uniform signal reference values when the current incomplete magnetic flux leakage data is fused.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, and in particular to a method for fusion of multi-round residual magnetic leakage data. Background Technology

[0002] In the energy and chemical industries, pipelines serve as the core "lifeline" for the cross-regional transportation of energy and chemical raw materials. Their safe and stable operation is of paramount importance, directly impacting energy security, supply chain stability, and public safety. Because pipelines operate in complex environments, various defects inevitably occur. To prevent these defects from worsening and to avoid leaks, regular internal pipeline inspections are crucial. Magnetic flux leakage (MFL) testing technology, with its non-destructive and precise advantages, has become the mainstream choice for pipeline defect identification, meeting the need for accurate defect identification to ensure safe pipeline operation.

[0003] Currently, the primary method for pipeline inspection is magnetic flux leakage (MF) testing. This involves deploying testing equipment and, after the pipeline is shut down, a specialized team collaboratively inserts the detector into the pipeline for testing. Considering the high costs involved in each round of pipeline inspection—including equipment deployment, pipeline shutdown, and team collaboration—as well as the significant time commitment, multiple rounds of testing are typically performed to improve the validity of the data. Furthermore, because the detector rotates slowly within the pipeline, multiple rounds of testing ensure that even if data in a particular clock direction is lost in one round, valid data may still be available in other rounds, thus aiming to obtain more comprehensive data.

[0004] However, the accumulation of impurities in the pipeline, numerous bends, and pipe deformation can easily damage the detector probe, resulting in incomplete data from a single round of testing. Even after two or more rounds of testing, the data may still fail to meet the standards due to probe damage. Even with multiple rounds of testing data and the completion of preliminary multi-round data analysis and feature alignment, direct data fusion can still encounter problems such as repeated or missing axial or circumferential fusion boundaries, excessive signal stretching and compression, and signal jumps caused by inconsistent signal reference values, failing to meet the requirements for high-precision defect assessment. Summary of the Invention

[0005] This application provides a multi-round residual magnetic flux leakage data fusion method to solve the technical problems that occur when fusing residual magnetic flux leakage data, such as repeated or missing axial or circumferential fusion boundaries, excessive stretching and compression of signals, and signal jumps caused by inconsistent signal reference values.

[0006] This application provides a method for fusing multi-round residual magnetic flux leakage data, including: Multiple rounds of magnetic flux leakage data are acquired; the magnetic flux leakage data are the magnetic flux leakage signal data of each channel collected by the magnetic flux leakage detector during at least two rounds of detection of the pipeline under test; the channels correspond to the pipeline under test in different clock directions; Based on the leakage magnetic field data, a sliding window is established; Extract the magnetic flux leakage data features within the sliding window; the magnetic flux leakage data features include: signal missing rate, noise variance, amplitude over-range percentage, and signal stability. The magnetic flux leakage data features are input into a random forest model to determine the valid and invalid magnetic flux leakage data features within each sliding window. The random forest model is trained from several sets of historical magnetic flux leakage data with the same diameter as the pipe under test. The historical magnetic flux leakage data is labeled with valid and invalid magnetic flux leakage data features. The magnetic flux leakage data features corresponding to the valid magnetic flux leakage data features are within a preset normal range, while the magnetic flux leakage data features corresponding to the invalid magnetic flux leakage data features are not within the preset normal range. Based on the characteristics of the effective and invalid magnetic flux leakage data, the channel type corresponding to the magnetic flux leakage data is determined; the channel type includes: effective channel, partially effective channel, and completely incomplete channel; Spatial information of the partially valid channels and the completely incomplete channels is extracted and integrated to form a list of incomplete regions. The list of incomplete regions is used to display the channel number, axial mileage range, circumferential clock azimuth range, degree of incompleteness, and incomplete features corresponding to the partially valid channels and the completely incomplete channels. The degree of incompleteness includes: slight incompleteness, moderate incompleteness, and severe incompleteness. The incomplete features include: missing information, excessive noise, and abnormal amplitude. Determine the baseline data; the baseline data is the magnetic flux leakage data of the round with the highest proportion of effective channels and the lowest signal loss rate among the multiple rounds of magnetic flux leakage data; Based on the reference wheel data, all signal points of the remaining rounds of magnetic flux leakage data are adjusted by circumferential rotation and axial mileage, so that the circumferential clock orientation of the signal points at the same physical location in the multiple rounds of magnetic flux leakage data is unified, and the error between the axial mileage of the signal points corresponding to the remaining rounds of magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference wheel data is less than or equal to a preset number of magnetic flux leakage data sampling points. Determine the range of incomplete regions containing incomplete features in the reference wheel data; Obtain the effective magnetic flux leakage data features located within the missing region interval in the non-reference wheel data; The effective magnetic flux leakage data features are fused into the reference wheel data, and the incomplete region intervals are filled in to obtain the magnetic flux leakage fusion data.

[0007] In some embodiments, the method further includes: Extract the circumferential weld features from the magnetic flux leakage data; Based on the characteristics of the circumferential weld, the corresponding pipe sections in the multiple rounds of magnetic flux leakage data are aligned.

[0008] In some embodiments, the step of extracting the magnetic flux leakage data features within the sliding window includes: Obtain the missing data within the sliding window; the missing data is the leakage magnetic field data corresponding to a signal amplitude of 0 or a signal exceeding its range; Based on the missing data, the signal missing rate is determined; The noise variance is obtained by calculating the sum of squared deviations and average values ​​of the signal amplitude and mean within the sliding window. Obtain abnormal magnetic leakage data where the signal amplitude within the sliding window exceeds the range of the magnetic leakage sensor. Based on the abnormal magnetic leakage data, determine the percentage of amplitude exceeding the range; The signal stability is obtained by calculating the absolute mean of the rate of change of amplitude of adjacent data points within the sliding window.

[0009] In some embodiments, the step of determining the channel type corresponding to the magnetic flux leakage data based on the effective magnetic flux leakage data characteristics and the invalid magnetic flux leakage data characteristics includes: Based on the effective and invalid magnetic flux leakage data features, the effective window ratio in the magnetic flux leakage signal data corresponding to each channel is determined; the effective window is defined as the sliding window in which all magnetic flux leakage data features are effective magnetic flux leakage data features. If the effective window ratio is equal to the first preset value, then the channel type corresponding to the magnetic flux leakage data is an effective channel; If the effective window ratio is less than the first preset value and greater than or equal to the second preset value, then the channel type corresponding to the leakage magnetic data is a partially effective channel. If the effective window ratio is smaller than the second preset value, then the channel type corresponding to the magnetic flux leakage data is a completely incomplete channel.

[0010] In some embodiments, the step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes: Based on the sampling timestamps of the sliding windows in the partially effective channels and the completely incomplete channels, as well as the mileage counter data, the axial mileage range of the sliding window is determined; Based on the circumferential installation position of the leakage magnetic field sensor within the channel, determine the circumferential clock orientation range of the partially effective channel and the completely incomplete channel; If the proportion of effective magnetic flux leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the first preset proportion and greater than or equal to the second preset proportion, then the sliding window is slightly incomplete. If the proportion of effective magnetic flux leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the second preset proportion and greater than or equal to the third preset proportion, then the sliding window is moderately incomplete. If the proportion of effective magnetic leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than a third preset proportion, then the sliding window is severely incomplete.

[0011] In some embodiments, the step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes: If the signal loss rate within the sliding window in the partially effective channel and the completely incomplete channel is greater than a third preset value, then the incomplete feature corresponding to the sliding window is missing. If the noise variance within the sliding window in the partially effective channel and the completely incomplete channel is greater than the preset quantile of the noise variance of the baseline of the defect-free pipeline under the same working conditions, then the defect feature corresponding to the sliding window is noise exceeding the standard. If the signal stability within the sliding window in the partially effective channel and the completely incomplete channel is greater than a fourth preset value, then the incomplete feature corresponding to the sliding window is an amplitude anomaly.

[0012] In some embodiments, the step of circumferentially rotating and adjusting all signal points of the remaining rounds of magnetic flux leakage data based on the reference round data includes: Based on the reference wheel data, circumferential feature points are determined; the circumferential feature points are the first peak points of the circumferential weld seam signals corresponding to the effective channels in the reference wheel data. Extract the first circumferential clock azimuth angle of the mileage where the circumferential feature point is located; Determine the second peak point of the circumferential weld signal corresponding to the effective channel in the remaining rounds of magnetic flux leakage data, and extract the second circumferential clock azimuth angle of the mileage where the second peak point is located; Based on the first circumferential clock azimuth angle and the second circumferential clock azimuth angle, calculate the azimuth deviation; According to the azimuth deviation, all signal points of the remaining rounds of magnetic flux leakage data are circumferentially rotated and adjusted so that the circumferential clock azimuth of signal points at the same physical location in the multiple rounds of magnetic flux leakage data is unified.

[0013] In some embodiments, the step of adjusting the axial mileage of all signal points of the remaining wheel leakage magnetic data based on the reference wheel data includes: Based on the pipe section alignment results, anchor points are determined; the anchor points are weld seams and defect features with a depth greater than a preset thickness in the magnetic flux leakage data. Based on the anchor point, the remaining round magnetic flux leakage data is subjected to mileage scaling processing so that the error between the pipe section length corresponding to the remaining round magnetic flux leakage data and the pipe section length corresponding to the reference round data is less than a preset number of magnetic flux leakage data sampling points. Based on the valid channel signals in the reference wheel data, the target mileage offset is determined; Based on the target mileage offset, the axial mileage of all signal points of the remaining round magnetic flux leakage data is adjusted so that the error between the axial mileage of the signal points corresponding to the remaining round magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference round data is less than or equal to a preset number of magnetic flux leakage data sampling points.

[0014] In some embodiments, the step of obtaining the effective magnetic flux leakage data features located within the incomplete region interval in the non-reference wheel data includes: Determine the degree of incompleteness in the non-baseline wheel data located within the incomplete region interval; If the proportion of effective magnetic flux leakage data features within the incomplete region is 100%, then all effective magnetic flux leakage data features are extracted. If the degree of incompleteness is mild, then the effective magnetic flux leakage data feature portion within the incomplete region is extracted; If the degree of incompleteness is moderate, then extract the effective magnetic flux leakage data feature portion within the incomplete region and the adjacent effective magnetic flux leakage data feature of the ineffective magnetic flux leakage data feature. If the degree of incompleteness is severe, then the extraction of effective magnetic flux leakage data features within the incomplete region will be abandoned. The magnetic flux leakage fusion data includes information tags; the information tags include: mileage, reference wheel channel number, clock orientation, amplitude, source type, corresponding non-reference wheel channel number, and incomplete annotation field; the incomplete annotation field includes: original real data, multi-wheel splicing data, extremely small gap interpolation data, and incomplete data.

[0015] In some embodiments, the method further includes: The magnetic flux leakage fusion data is corrected to obtain the target magnetic flux leakage fusion data; The step of correcting the magnetic flux leakage fusion data includes: The original data region, the fusion region, and the fusion edge region of the magnetic flux leakage fusion data are determined; the fusion edge region is the junction between the original data region and the fusion region. Unify the amplitude of the fused data within the fusion region to the amplitude of the reference wheel data in the same channel; In addition, a preset number of data sampling points are obtained in the fusion edge region, and the data sampling points are smoothed using a moving average smoothing algorithm; In addition, additional regional information is added to the fusion region; the additional regional information includes: timestamp, leakage magnetic field internal sensor status.

[0016] This application provides a method for fusing multi-round residual magnetic flux leakage data, including: acquiring multi-round magnetic flux leakage data; the magnetic flux leakage data is magnetic flux leakage signal data collected by an internal magnetic flux leakage detector during at least two rounds of detection of the pipeline under test, corresponding to each channel; the channels correspond to the pipeline under test in different clock directions; establishing a sliding window based on the magnetic flux leakage data; extracting magnetic flux leakage data features within the sliding window; the magnetic flux leakage data features include: signal missing rate, noise variance, amplitude over-range ratio, and signal stability; inputting the magnetic flux leakage data features into a random forest model to determine the effective and invalid magnetic flux leakage data features within each sliding window; the random forest model... The machine forest model is trained from several sets of historical magnetic flux leakage data with the same diameter as the pipe under test. These historical magnetic flux leakage data are labeled with valid and invalid magnetic flux leakage data features. The magnetic flux leakage data features corresponding to the valid features are within a preset normal range, while the magnetic flux leakage data features corresponding to the invalid features are not within the preset normal range. Based on these valid and invalid magnetic flux leakage data features, the channel type corresponding to the magnetic flux leakage data is determined. The channel type includes: valid channel, partially valid channel, and completely incomplete channel. The spatial information of the partially valid channels and the completely incomplete channels is extracted and integrated to form a cleared incomplete region. The list of incomplete areas is used to display the channel number, axial mileage range, circumferential clock azimuth range, degree of incompleteness, and incompleteness characteristics corresponding to the partially valid channels and the completely incomplete channels. The degree of incompleteness includes: mild incompleteness, moderate incompleteness, and severe incompleteness. The incompleteness characteristics include: missing data, excessive noise, and abnormal amplitude. A reference wheel data is determined. The reference wheel data is the magnetic flux leakage data from the round with the highest proportion of valid channels and the lowest signal loss rate among the multiple rounds of magnetic flux leakage data. Based on the reference wheel data, circumferential rotation and axial mileage adjustments are made to all signal points of the remaining rounds of magnetic flux leakage data to ensure that the multiple rounds of magnetic flux leakage data... The circumferential clock orientation of signal points at the same physical location is unified, and the error between the axial mileage of the signal points corresponding to the remaining rounds of magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference round data is less than or equal to a preset number of magnetic flux leakage data sampling points; the missing region interval containing missing features in the reference round data is determined; the effective magnetic flux leakage data features located in the missing region interval in the non-reference round data are obtained; the effective magnetic flux leakage data features are fused into the reference round data to complete the missing region interval, thereby obtaining magnetic flux leakage fusion data, so as to achieve accurate identification of effective channels and missing regions and improve the matching accuracy and reliability of circumferential and axial directions of multi-round data. Attached Figure Description

[0017] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the multi-round residual magnetic flux leakage data fusion method in this application; Figure 2 This is a schematic diagram of the fusion of multiple rounds of residual magnetic flux leakage data in this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0020] In some techniques, data fusion of incomplete magnetic flux leakage data can result in issues such as repeated or missing axial or circumferential fusion boundaries, excessive signal stretching and compression, and signal jumps caused by inconsistent signal reference values. To address these problems, this application provides a multi-round incomplete magnetic flux leakage data fusion method, which is described below:

[0021] like Figure 1 The diagram shown is a flowchart of the multi-round residual magnetic flux leakage data fusion method in this application. The detector enters the pipeline for detection. Due to probe damage, the data from the damaged probe is invalid (including missing data, excessive noise, and abnormal amplitude). Based on the percentage of valid data, it is divided into three levels: mildly incomplete (valid data percentage [97%, 100%)), moderately incomplete (valid data percentage [60%, 97%)), and severely incomplete (valid data percentage <60%).

[0022] This application provides a method for fusing multi-round residual magnetic flux leakage data, including the following steps: S100: Acquire multi-round magnetic flux leakage data; the magnetic flux leakage data refers to the magnetic flux leakage signal data collected by the magnetic flux leakage detector during at least two rounds of internal testing of the pipeline under test, corresponding to each channel; the channels correspond to the pipeline under test in different clock directions; wherein, the multi-round magnetic flux leakage data is a series of detection data obtained by conducting internal testing on the same pipeline at different times or during different operating stages. The clock direction is the direction of oil flow towards the pipeline under test, with 12:00 above the pipeline, 6:00 below the pipeline, 9:00 on the left side of the pipeline, and 3:00 on the right side of the pipeline. The magnetic flux leakage data generated after the pipeline detector enters the pipeline under test is recorded with the clock direction of the pipeline under test as the reference.

[0023] Specifically, the pipeline under test is subjected to at least two rounds of internal testing using a pipeline magnetic flux leakage detector. Multi-channel raw magnetic flux leakage signals are collected in each round of testing, the data is downloaded to a portable hard drive, and data analysis software is used to analyze the data.

[0024] like Figure 2 The diagram shown is a schematic diagram of multi-round residual magnetic leakage data fusion in this application.

[0025] The method further includes the following steps: S110: Extract the circumferential weld features from the magnetic flux leakage data.

[0026] S120: Based on the characteristics of the circumferential weld, the corresponding pipe sections in the multiple rounds of magnetic flux leakage data are aligned. After analyzing the two rounds of magnetic flux leakage detection data of the pipeline under test, the two rounds of magnetic flux leakage data are matched section by section according to the pipe section length, providing a basic spatial reference for subsequent data fusion.

[0027] S200: Based on the leakage magnetic field data, establish a sliding window. A sliding window is established based on 500 sampling points of the leakage magnetic field data signal from each channel.

[0028] S300: Extract the magnetic flux leakage data features within the sliding window; the magnetic flux leakage data features include: signal missing rate, noise variance, amplitude over-range ratio, and signal stability.

[0029] The step of extracting the magnetic flux leakage data features within the sliding window includes the following sub-steps: S310: Obtain missing data within the sliding window; the missing data is the leakage magnetic field data corresponding to a signal amplitude of 0 or a signal exceeding its range.

[0030] S320: Based on the missing data, determine the signal missing rate; the signal missing rate is the percentage of missing data (leakage magnetic data signal sampling points corresponding to amplitude of 0 or over-range) within the sliding window; the signal missing rate is: .

[0031] S330: Calculate the average of the squared deviations between the signal amplitude and the mean within the sliding window to obtain the noise variance; the noise variance is the average of the squared deviations between the leakage magnetic data signal amplitude and the mean within the sliding window, and the calculation formula is: ; In the formula, Let x be the noise variance. i The amplitude of a single sampling point is μ, the mean of the window is n=500.

[0032] S340: Obtain abnormal magnetic leakage data where the signal amplitude within the sliding window is greater than the range of the magnetic leakage detector; S350: Based on the abnormal magnetic leakage data, determine the percentage of amplitude exceeding the range; the percentage of amplitude exceeding the range is the percentage of the number of signals whose amplitude exceeds the range of the sensor within the sliding window.

[0033] S360: Calculate the absolute mean of the amplitude change rates of adjacent data points within the sliding window to obtain the signal stability. The signal stability is the absolute mean of the amplitude change rates of adjacent sampling points within the sliding window. The original stability calculation formula is: .

[0034] In the formula, Characterized as the original stationarity; It is represented by the amplitude of adjacent sampling points within the sliding window.

[0035] After obtaining the original stationarity, the signal stationarity is obtained through normalization; the normalization calculation formula is: .

[0036] In the formula, Characterized by the maximum original stationarity, It is characterized as the minimum original stationarity.

[0037] S400: Input the magnetic flux leakage data features into a random forest model to determine the valid and invalid magnetic flux leakage data features within each sliding window; the random forest model is trained from several sets of historical magnetic flux leakage data with the same diameter as the pipe under test; the historical magnetic flux leakage data is labeled with valid and invalid magnetic flux leakage data features, the magnetic flux leakage data features corresponding to the valid magnetic flux leakage data features are within a preset normal range, and the magnetic flux leakage data features corresponding to the invalid magnetic flux leakage data features are not within the preset normal range; Specifically, 1000 sets of historical data on magnetic flux leakage detection of pipelines of the same diameter were collected, and manually labeled as "effective channels" (the magnetic flux leakage data features corresponding to the effective magnetic flux leakage data features are within the preset normal range) or "incomplete channels" to form a training set; the model parameters were adjusted through 5-fold cross-validation, and the number of decision trees was finally set to 100, the maximum depth of each tree was 8, and the minimum number of samples in the leaf nodes was 5. After the model was trained, the classification accuracy was ≥95% and the misclassification rate was ≤3%, and the final random forest model was obtained to determine the effective and ineffective magnetic flux leakage data features.

[0038] S500: Based on the characteristics of the effective magnetic flux leakage data and the characteristics of the invalid magnetic flux leakage data, determine the channel type corresponding to the magnetic flux leakage data; the channel type includes: effective channel, partially effective channel, and completely incomplete channel.

[0039] The step of determining the channel type corresponding to the magnetic flux leakage data based on the effective and invalid magnetic flux leakage data characteristics includes the following sub-steps: S510: Based on the effective and invalid magnetic flux leakage data features, determine the effective window ratio in the magnetic flux leakage signal data corresponding to each channel; the effective window is defined as the magnetic flux leakage data features within the sliding window being all effective magnetic flux leakage data features.

[0040] S520: If the effective window ratio is equal to the first preset value, then the channel type corresponding to the leakage magnetic data is an effective channel.

[0041] S530: If the effective window ratio is less than the first preset value and greater than or equal to the second preset value, then the channel type corresponding to the leakage magnetic data is a partially effective channel.

[0042] S540: If the effective window ratio is smaller than the second preset value, then the channel type corresponding to the magnetic leakage data is a completely incomplete channel.

[0043] Specifically, for each channel in each round of detection, a 4D feature is calculated using a sliding window of 500 sampling points (step size 100 points). The "effective channel and incomplete channel" determination results for each sliding window are output using a random forest model. The effective window percentage for each channel is calculated: if the percentage is 100%, it is determined as an "effective channel," and all magnetic flux leakage data signals are retained; if the percentage is [60, 100), it is determined as a "partially effective channel," and the magnetic flux leakage data signals corresponding to the effective window are retained; if the percentage is less than 60%, it is determined as a "completely incomplete channel," and all magnetic flux leakage data signals for that channel are removed. The effective channel signal set D for each round is output. i ={d1,d2,…,d m (m is the total number of leakage magnetic data channels).

[0044] S600: Extract the spatial information of the partially valid channels and the completely incomplete channels and integrate them to form a list of incomplete regions; the list of incomplete regions is used to display the channel number, axial mileage range, circumferential clock azimuth range, degree of incompleteness, and incomplete features corresponding to the partially valid channels and the completely incomplete channels; the degree of incompleteness includes: slight incompleteness, moderate incompleteness, and severe incompleteness; the incomplete features include: missing, excessive noise, and abnormal amplitude.

[0045] The step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes the following sub-steps: S610: Determine the axial mileage range of the sliding window based on the sampling timestamps of the sliding window in the partially effective channel and the completely incomplete channel, as well as the mileage counter data.

[0046] S620: Determine the circumferential clock orientation range of the partially effective channel and the completely incomplete channel based on the circumferential installation position of the internal leakage magnetic sensor in the channel.

[0047] S630: If the proportion of effective magnetic leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the first preset proportion and greater than or equal to the second preset proportion, then the sliding window is slightly incomplete.

[0048] S640: If the proportion of effective magnetic leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the second preset proportion and greater than or equal to the third preset proportion, then the sliding window is moderately incomplete.

[0049] S650: If the proportion of effective magnetic leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the third preset proportion, then the sliding window is severely incomplete.

[0050] Specifically, for both "partially valid channels" and "completely incomplete channels," the spatial information corresponding to the incomplete windows is extracted: Axial mileage range: Based on the sampling timestamps corresponding to the incomplete window and combined with the odometer data, the axial mileage range of the window is calculated. .

[0051] Circumferential clock azimuth range: Determine the circumferential clock azimuth range of the channel signal based on the circumferential installation position of the leakage magnetic field sensor corresponding to the channel. (For example, Channel 1 corresponds to 0°-5°, i.e., 0:00-0:10 azimuth).

[0052] Incompleteness level: Based on the percentage of valid data, if the percentage is [97%, 100%), the incompleteness level is slightly incomplete; if the percentage is [60%, 97%), the incompleteness level is moderately incomplete; if the percentage is less than 60%, the incompleteness level is severely incomplete.

[0053] The step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes the following sub-steps: S660: If the signal loss rate within the sliding window in the partially effective channel and the completely incomplete channel is greater than a third preset value, then the incomplete feature corresponding to the sliding window is missing.

[0054] S670: If the noise variance within the sliding window in the partially effective channel and the completely incomplete channel is greater than the preset quantile of the baseline noise variance of the defect-free pipeline under the same working conditions, then the defect feature corresponding to the sliding window is noise exceeding the standard.

[0055] S680: If the signal stability within the sliding window of the partially effective channel and the completely incomplete channel is greater than the fourth preset value, then the incomplete feature corresponding to the sliding window is an amplitude anomaly.

[0056] Specifically, the incomplete features are as follows: Missing: Based on the signal missing rate, when the missing rate is greater than 80%, the missing feature of the missing window is defined as missing.

[0057] Excessive noise: Calculate the noise variance σ of the signal within the incomplete window. 2 When this parameter is greater than the 99th percentile of the noise variance of the baseline of a defect-free pipeline under the same working conditions, the defective feature of the defective window is defined as noise exceeding the standard.

[0058] Amplitude anomalies: Based on signal stability determination, when the normalized signal stability S within the window... norm When the value is greater than 0.8, the incomplete feature of the incomplete window is defined as an amplitude anomaly.

[0059] Integrate to form a list of incomplete areas Each item z i Includes: Channel number, axial mileage range Circumferential clock azimuth range , Degree of incompleteness, and characteristics of incompleteness (missing, excessive noise, abnormal amplitude).

[0060] S700: Determine the baseline data; the baseline data is the magnetic flux leakage data of the round with the highest proportion of effective channels and the lowest signal loss rate among multiple rounds of magnetic flux leakage data. The baseline data is selected from the rounds with the highest proportion of effective channels and the lowest signal loss rate among multiple rounds of magnetic flux leakage data; wherein, if the proportion of effective channels is the same in multiple rounds, the magnetic flux leakage data corresponding to the round with the earliest detection time is selected.

[0061] S800: Based on the reference wheel data, perform circumferential rotation adjustment and axial mileage adjustment on all signal points of the remaining rounds of magnetic leakage data, so that the circumferential clock orientation of the signal points at the same physical location in the multiple rounds of magnetic leakage data is unified, and the error between the axial mileage of the signal points corresponding to the remaining rounds of magnetic leakage data and the axial mileage of the signal points corresponding to the reference wheel data is less than or equal to a preset number of magnetic leakage data sampling points.

[0062] The step of circumferentially rotating and adjusting all signal points of the remaining rounds of magnetic flux leakage data based on the reference round data includes the following sub-steps: S810: Based on the reference wheel data, determine the circumferential feature point; the circumferential feature point is the first peak point of the circumferential weld signal corresponding to the effective channel in the reference wheel data; in the effective channels of the reference wheel data, select the first peak point of the circumferential weld signal of each channel as the circumferential feature point.

[0063] S820: Extract the first circumferential clock azimuth angle of the mileage where the circumferential feature point is located; extract the first circumferential clock azimuth angle of the mileage where the circumferential feature point is located. (j is the circumferential feature point number, the azimuth angle is taken as the top of the pipe as 0 point, and increases clockwise, with an accuracy of ±0.1°).

[0064] S830: Determine the second peak point of the circumferential weld signal corresponding to the effective channel in the remaining rounds of magnetic flux leakage data, and extract the second circumferential clock azimuth angle of the mileage where the second peak point is located.

[0065] S840: Calculate the azimuth deviation based on the first circumferential clock azimuth and the second circumferential clock azimuth; for the effective channels of the remaining rounds (target rounds), extract the second peak point of the corresponding circumferential weld and calculate its second circumferential clock azimuth. This leads to the azimuth deviation. .

[0066] S850: Adjust the circumferential rotation of all signal points in the remaining rounds of magnetic flux leakage data according to the stated azimuth deviation, ensuring that the circumferential clock azimuth of signal points at the same physical location in the multiple rounds of magnetic flux leakage data is unified. For all magnetic flux leakage data signal points in the target round, adjust the azimuth deviation Δα... j Perform circumferential rotation adjustment using the following formula: This ensures that the circumferential clock orientation at the same physical location is completely consistent across multiple data rounds, with the final circumferential orientation error ≤ 5°.

[0067] The step of adjusting the axial mileage of all signal points of the remaining wheel leakage magnetic data based on the reference wheel data includes the following sub-steps: S860: Based on the pipe section alignment results, determine the anchor point; the anchor point is the weld or defect feature with a depth greater than the preset thickness in the leakage magnetic data; based on the weld and defect alignment results in the previous data analysis work, select the weld or defect with a depth greater than 10%t (t is the pipe wall thickness) as the anchor point.

[0068] S870: Based on the anchor point, perform mileage scaling on the remaining round magnetic flux leakage data so that the error between the tube segment length corresponding to the remaining round magnetic flux leakage data and the tube segment length corresponding to the reference round data is less than a preset number of magnetic flux leakage data sampling points; perform mileage scaling on the target round so that the error between the tube segment length of the target round's magnetic flux leakage detection data and the tube segment length of the reference round is less than or equal to 5 sampling points.

[0069] S880: Determine the target mileage offset based on the valid channel signal in the reference wheel data.

[0070] S890: Based on the target mileage offset, adjust the axial mileage of all signal points of the remaining round magnetic flux leakage data so that the error between the axial mileage of the signal points corresponding to the remaining round magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference round data is less than or equal to a preset number of magnetic flux leakage data sampling points.

[0071] Specifically, this is achieved through reinforcement learning model construction: State space: contains 12 parameters, namely “feature position deviation, signal cosine similarity, offset change rate, circumferential weld spacing deviation, defect peak position deviation, signal stability deviation (calculated based on stability within the normalized range of [0,1]), axial tensile rate, axial extrusion rate, effective feature matching number, noise variance ratio, amplitude mean deviation, and sampling point density deviation”.

[0072] Action space: Axial mileage offset adjustment value, with a range limited to [-5,5] sampling points.

[0073] Reward function: R = 0.7 × S + 0.3 × T (S is the feature similarity, ranging from [0,1]; T is the offset smoothing coefficient, T = 1 when the offset change is less than or equal to 5 sampling points, otherwise T = 0.5).

[0074] The optimization process is as follows: Using the effective channel signal of the baseline wheel as the target, the initial offset is 0, and the process is iterated 1000 times. In each iteration, an action (adjusting the offset) is selected based on the current state, the reward value is calculated, and the model parameters are updated. After iterating until the reward value stabilizes (the reward value fluctuation is less than or equal to 0.01 for 50 consecutive iterations), the target mileage offset Δ is output. Lopt .

[0075] Gradient descent accelerates convergence: In the reinforcement learning iteration process, gradient descent is introduced to adjust the offset in the direction of the negative gradient of the reward value. The learning rate is set to 0.01, which improves the iteration convergence speed by 30%.

[0076] Fine alignment execution: based on target mileage offset Δ Lopt Adjust the axial mileage of the target wheel signal to ultimately achieve an axial alignment error of less than or equal to 5 sampling points.

[0077] For example, this application also provides alignment secondary verification (edge ​​defect similarity verification): First, extract real defect data: From the baseline and target data that have completed the two-dimensional matching, extract the real defect signal segments located at the fusion edge. The extraction rule is: the defect quantization depth is greater than 5%t (t is the pipe wall thickness, and the defect depth is quantized using the proprietary equipment quantization model).

[0078] Constructing a two-round real-world defect dataset (Base wheel) and (Target round), data for each defect Includes spatial information (axial distance L, circumferential clock orientation α, defect length) Defect width ), signal feature vector (axial peak amplitude A, axial peak-to-valley difference) Radial peak-to-valley difference Circumferential peak-valley difference ).

[0079] Next, a two-dimensional matching verification is performed: Spatial range matching: Combining data analysis results, for Each defect in ,exist Search for candidate defects that meet the criteria of "axial mileage deviation less than or equal to 10 sampling points and circumferential clock azimuth deviation less than or equal to 5°". , filter out defect pairs that meet the space matching criteria.

[0080] Signal feature matching: The feature similarity between acceptable and defective pairs is calculated using the cosine similarity formula, which is as follows: .

[0081] By setting a feature similarity threshold greater than or equal to 0.85, defect pairs that meet the feature matching criteria are retained.

[0082] Finally, the alignment validity is determined: The number of defective pairs that are successfully matched. and Total number of defects The ratio of these values ​​yields the matching rate. .

[0083] Set a matching rate threshold ≥85%: If ≥ If the alignment is deemed valid, proceed to the subsequent incompleteness completion stage; if Return to the two-dimensional space matching stage, adjust the weight of the reinforcement learning reward function (increase the feature similarity weight to 0.8) or the sliding window size (adjust to 256 points), and re-execute the above matching and verification process until the target threshold or the upper limit of the iteration threshold is reached.

[0084] S900: Determine the range of incomplete regions containing incomplete features in the reference wheel data; based on the list of incomplete regions Z, determine the range of incomplete regions using only the spatial coordinates (axial direction) of the reference wheel. Zhou Xiang Anchor the damaged area, clarify the degree and characteristics of the damage, and break away from the dependence on probe numbering.

[0085] For example, for all non-reference wheel data, ensure that two-dimensional spatial matching has been completed, satisfying that the circumferential error is less than or equal to 5° and the axial error is less than or equal to 5 sampling points, so as to ensure that the retrieved data corresponds to the same physical location.

[0086] S1000: Obtain the effective magnetic leakage data features located within the incomplete region interval in the non-reference wheel data.

[0087] The step of obtaining the effective magnetic flux leakage data characteristics located within the incomplete region interval in the non-reference wheel data includes the following sub-steps: S1010: Determine the degree of incompleteness in the non-baseline wheel data located within the incomplete region interval.

[0088] S1020: If the proportion of effective magnetic flux leakage data features within the incomplete region is 100%, then all effective magnetic flux leakage data features are extracted.

[0089] S1030: If the degree of incompleteness is mild, then extract the effective leakage magnetic data feature portion within the incomplete region.

[0090] S1040: If the degree of incompleteness is moderate, then extract the effective magnetic flux leakage data feature portion within the incomplete region and the adjacent effective magnetic flux leakage data feature of the ineffective magnetic flux leakage data feature.

[0091] S1050: If the degree of incompleteness is severe incompleteness, then the extraction of effective magnetic leakage data features within the incomplete region is abandoned.

[0092] Specifically, priority retrieval of real data at the same location: for a single incomplete region Z i Traverse the aligned non-reference wheel data, only search for regions with overlapping spatial coordinates, and filter according to the following priority (stop when a higher priority region is found): Level 1 (100% valid data): 100% of the valid data in the same position as the baseline round are directly extracted for completion.

[0093] Level 2 (slightly incomplete data): The percentage of valid data at the same position as the non-baseline wheel is [97%, 100%), and the true valid segments are extracted.

[0094] Level 3 (moderately incomplete data): The proportion of valid data at the same position as the non-baseline wheel is [60%, 97%), and the true valid segments are extracted.

[0095] Level 4 (Severely Incomplete Data): Valid data percentage <60%, search abandoned.

[0096] S1100: The effective magnetic leakage data features are fused into the reference wheel data to complete the incomplete region intervals and obtain the magnetic leakage fusion data.

[0097] Specifically, a layered completion strategy is adopted: Slightly incomplete (validity percentage [97%, 100%)): Only real data is stitched together, without interpolation. If Level 1 data is found, the missing part of the baseline wheel is directly replaced; if only Level 2 data is found, it is stitched together with the real segment of the baseline wheel, and the data with the smaller noise variance is taken in the overlapping area; if no data is found, the original data of the baseline wheel is retained and marked "incomplete - slightly incomplete".

[0098] Moderate incompleteness (validity ratio [60%, 97%)): Primarily based on real data stitching, supplemented by interpolation for extremely small gaps. First, stitch together multiple rounds of real and valid segments. Only for gaps of 5 or fewer sampling points in a single segment, use linear interpolation of real data from adjacent channels to label the corresponding state.

[0099] Severely incomplete (<60% effective): Directly abandon the completion, only retain the existing real data fragments of the baseline wheel, and mark "severely incomplete - not completed" and the spatial range for manual review.

[0100] The magnetic flux leakage fusion data includes information tags; these tags include: mileage, reference wheel channel number, clock orientation, amplitude, source type, corresponding non-reference wheel channel number, and incomplete annotation field; the incomplete annotation field includes: original true data, multi-wheel splicing data, minimal gap interpolation data, and incomplete data. Data is integrated according to the spatial coordinates of the reference wheel, and each data point is labeled with "original true data, multi-wheel splicing data, minimal gap interpolation data, and incomplete data," outputting a CSV format data processing record containing mileage, reference wheel channel number, clock orientation, amplitude, source type, corresponding target wheel channel number, and incomplete annotation field.

[0101] The method further includes the following steps: S1200: Correct the magnetic flux leakage fusion data to obtain the target magnetic flux leakage fusion data.

[0102] The step of correcting the magnetic flux leakage fusion data includes: S1210: Determine the original data region, fusion region, and fusion edge region of the magnetic flux leakage fusion data; the fusion edge region is the junction between the original data region and the fusion region.

[0103] Specifically, a self-supervised learning model (3 layers of convolution in the encoder and 3 layers of deconvolution in the decoder) is constructed. Normal region signals with 100% effective channels and no defects are input into the baseline and target rounds of data, respectively. The training objective is to minimize the signal reconstruction error (loss function MSE, 500 iterations, learning rate 0.001). After training convergence (loss fluctuation less than or equal to 0.0001 for 20 consecutive rounds), the baseline arrays for each channel of the baseline round are output. Target wheel channel reference array .

[0104] With reference array set Using the original data region, the fusion region, and the fusion edge region as the sole benchmark, the initial magnetic flux leakage fusion data are partitioned and a differential correction strategy is executed to determine the original data region, the fusion region, and the fusion edge region of the magnetic flux leakage fusion data.

[0105] S1220: Unify the amplitude of the fused data within the fusion region to the amplitude of the reference wheel data in the same channel; perform one-to-one quantitative reference correction for each channel, unifying the amplitude reference of the fused data to the channel reference value of the reference wheel through the reference wheel channel number and the fusion region. The correction formula is: ; In the formula, The amplitude of the Kth sampling point in the fusion region after correction is given. The unprocessed amplitude of the Kth sampling point in the fusion region. The reference value is the channel number corresponding to the reference round data of the k-th sampling point. This represents the original baseline value corresponding to the Kth sampling point in the channel where the target wheel data is located. Minimum threshold (take) This avoids calculation errors when the channel amplitude is 0.

[0106] S1230: And, obtain a preset number of data sampling points in the fusion edge region, and use a moving average smoothing algorithm to perform signal smoothing processing on the data sampling points; for the fusion boundary region with 100 sampling points before and after the junction of the fusion region and the original data region, use a moving average smoothing algorithm to perform signal smoothing processing on the data sampling points in the fusion boundary region: the sliding window size is 10 sampling points, the sliding step size is 1 sampling point, and signal smoothing transition correction is performed.

[0107] S1240: And, add additional regional information to the fused region; the additional regional information includes: timestamp, leakage magnetic field sensor status.

[0108] This application provides a method for fusing multi-round residual magnetic flux leakage data, with the following specific objectives: (1) Achieve accurate identification of effective channels and incomplete areas: By using a dynamic feature-driven intelligent screening model to replace the fixed threshold judgment mode, it adapts to the signal differences in different detection scenarios, improves the utilization rate of effective information, and provides a precise spatial positioning basis for subsequent fusion.

[0109] (2) Improve the accuracy and reliability of multi-round data two-dimensional matching: Construct a circumferential-axial collaborative matching system, and combine the secondary verification mechanism of typical defects at the fusion edge in the analysis results to ensure accurate spatial correspondence of multi-round data.

[0110] (3) Achieve targeted and accurate completion of incomplete areas and data authenticity: Based on multiple rounds of real and effective data, design a hierarchical completion strategy for different degrees of incompleteness, eliminate the risk of distortion introduced by fictitious data, and maximize the effective value of multiple rounds of data.

[0111] (4) Taking into account signal consistency: Adopt a differentiated correction strategy that distinguishes regions, unify the signal reference, and ensure that the fused data has consistency.

[0112] This application provides a method for fusing multi-round residual magnetic flux leakage data. Through the collaborative integration of real data, it reliably improves data validity and solves the following core problems: (1) Overcoming the problem of inaccurate identification of effective channels and incomplete areas: By dynamically filtering effective channels and accurately marking incomplete areas through multi-dimensional features, the adaptability of different detection scenarios and the utilization rate of effective information are improved.

[0113] (2) To make up for the lack of accuracy of two-dimensional spatial matching and the lack of verification in the data fusion process, a circumferential-axial collaborative matching mechanism is constructed. Combined with the secondary verification of edge defects, the reliability of multi-round data spatial correspondence is ensured, and the problems of data duplication, missing data and signal distortion in the fusion area are solved.

[0114] (3) Overcoming the limitations of insufficient targeting in filling in missing areas: Targeted extraction of multiple rounds of real and effective data, and the adoption of a layered filling strategy for different degrees of incompleteness, to achieve accurate filling of missing areas and eliminate the distortion risk introduced by existing interpolation or fictitious data practices.

[0115] (4) Improve the poor signal correction effect: adopt a differentiated correction strategy that distinguishes between normal areas and defective areas, and combine multiple rounds of cross-validation to ensure that the corrected data is true and reliable while taking into account both signal consistency and defect feature integrity.

[0116] This application provides a method for fusing multi-round residual magnetic flux leakage data, which has the following advantages: (1) Data utilization dimension: Upgrading from "single-round reconstruction" to "multi-round fusion" to explore complementary value. Currently, reconstruction is only performed on single-round data with missing data, without utilizing the complementarity of multi-round detection data. This application focuses on "multi-round incomplete data". By fusing effective information from multiple rounds, it can not only fill in the missing areas of single-round data, but also eliminate abnormal interference through cross-validation of multi-round data, ensuring the authenticity and integrity of the fused data, and solving the risk of generating non-authentic data in single-round reconstruction.

[0117] (2) Spatial Matching Dimension: From "no precise matching" to "dual-dimensional collaboration and secondary verification", the matching accuracy is greatly improved. Currently, data is only processed through "matrix block segmentation and normalization", which does not involve the spatial correspondence of multi-round data. The default data is aligned by mileage, lacking collaborative matching of the pipeline circumferential (clock position) and axial (mileage) directions, which is prone to data misalignment due to detection errors. This application constructs a "circumferential clock-axial mileage dual-dimensional spatial matching system", combined with a secondary verification mechanism for fusion edge defects, to ensure accurate spatial correspondence of multi-round data.

[0118] (3) Incompleteness Completion Dimension: From "single generative completion" to "hierarchical targeted completion", it is more targeted. Currently, GAN generative reconstruction is used, which adopts a uniform generation strategy for all incomplete regions without distinguishing the degree of incompleteness. This application designs a "hierarchical completion strategy" for different degrees of incompleteness, and performs targeted completion based on multiple rounds of real and valid data, eliminating the distortion risk introduced by fictitious data.

[0119] The above detailed embodiments further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A method for fusing multi-round residual magnetic flux leakage data, characterized in that, include: Acquire multiple rounds of magnetic flux leakage data; The magnetic flux leakage data refers to the magnetic flux leakage signal data of each channel collected by the magnetic flux leakage internal detector during at least two rounds of internal detection of the pipeline under test. The channels correspond to the pipes under test in different clock directions; Based on the leakage magnetic field data, a sliding window is established; Extract the magnetic flux leakage data features within the sliding window; the magnetic flux leakage data features include: signal missing rate, noise variance, amplitude over-range percentage, and signal stability. The magnetic flux leakage data features are input into a random forest model to determine the valid and invalid magnetic flux leakage data features within each sliding window. The random forest model is trained from several sets of historical magnetic flux leakage data with the same diameter as the pipe under test. The historical magnetic flux leakage data is labeled with valid and invalid magnetic flux leakage data features. The magnetic flux leakage data features corresponding to the valid magnetic flux leakage data features are within a preset normal range, while the magnetic flux leakage data features corresponding to the invalid magnetic flux leakage data features are not within the preset normal range. Based on the characteristics of the effective and invalid magnetic flux leakage data, the channel type corresponding to the magnetic flux leakage data is determined; the channel type includes: effective channel, partially effective channel, and completely incomplete channel; Spatial information of the partially valid channels and the completely incomplete channels is extracted and integrated to form a list of incomplete regions. The list of incomplete regions is used to display the channel number, axial mileage range, circumferential clock azimuth range, degree of incompleteness, and incomplete features corresponding to the partially valid channels and the completely incomplete channels. The degree of incompleteness includes: slight incompleteness, moderate incompleteness, and severe incompleteness. The incomplete features include: missing information, excessive noise, and abnormal amplitude. Determine the baseline data; the baseline data is the magnetic flux leakage data of the round with the highest proportion of effective channels and the lowest signal loss rate among the multiple rounds of magnetic flux leakage data; Based on the reference wheel data, all signal points of the remaining rounds of magnetic flux leakage data are adjusted by circumferential rotation and axial mileage, so that the circumferential clock orientation of the signal points at the same physical location in the multiple rounds of magnetic flux leakage data is unified, and the error between the axial mileage of the signal points corresponding to the remaining rounds of magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference wheel data is less than or equal to a preset number of magnetic flux leakage data sampling points. Determine the range of incomplete regions containing incomplete features in the reference wheel data; Obtain the effective magnetic flux leakage data features located within the missing region interval in the non-reference wheel data; The effective magnetic flux leakage data features are fused into the reference wheel data, and the incomplete region intervals are filled in to obtain the magnetic flux leakage fusion data.

2. The multi-round residual magnetic flux leakage data fusion method according to claim 1, characterized in that, The method further includes: Extract the circumferential weld features from the magnetic flux leakage data; Based on the characteristics of the circumferential weld, the corresponding pipe sections in the multiple rounds of magnetic flux leakage data are aligned.

3. The method for fusing multi-round residual magnetic flux leakage data according to claim 1, characterized in that, The step of extracting the magnetic flux leakage data features within the sliding window includes: Obtain the missing data within the sliding window; the missing data is the leakage magnetic field data corresponding to a signal amplitude of 0 or a signal exceeding its range; Based on the missing data, the signal missing rate is determined; The noise variance is obtained by calculating the sum of squared deviations and average values ​​of the signal amplitude and mean within the sliding window. Obtain abnormal magnetic leakage data where the signal amplitude within the sliding window exceeds the range of the magnetic leakage sensor. Based on the abnormal magnetic leakage data, determine the percentage of amplitude exceeding the range; The signal stability is obtained by calculating the absolute mean of the rate of change of amplitude of adjacent data points within the sliding window.

4. The multi-round residual magnetic flux leakage data fusion method according to claim 1, characterized in that, The step of determining the channel type corresponding to the magnetic flux leakage data based on the effective and invalid magnetic flux leakage data characteristics includes: Based on the effective and invalid magnetic flux leakage data features, the effective window ratio in the magnetic flux leakage signal data corresponding to each channel is determined; the effective window is defined as the sliding window in which all magnetic flux leakage data features are effective magnetic flux leakage data features. If the effective window ratio is equal to the first preset value, then the channel type corresponding to the magnetic flux leakage data is an effective channel; If the effective window ratio is less than the first preset value and greater than or equal to the second preset value, then the channel type corresponding to the leakage magnetic data is a partially effective channel. If the effective window ratio is smaller than the second preset value, then the channel type corresponding to the magnetic flux leakage data is a completely incomplete channel.

5. The multi-round residual magnetic flux leakage data fusion method according to claim 1, characterized in that, The step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes: Based on the sampling timestamps of the sliding windows in the partially effective channels and the completely incomplete channels, as well as the mileage counter data, the axial mileage range of the sliding window is determined; Based on the circumferential installation position of the leakage magnetic field sensor within the channel, determine the circumferential clock orientation range of the partially effective channel and the completely incomplete channel; If the proportion of effective magnetic flux leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the first preset proportion and greater than or equal to the second preset proportion, then the sliding window is slightly incomplete. If the proportion of effective magnetic flux leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than the second preset proportion and greater than or equal to the third preset proportion, then the sliding window is moderately incomplete. If the proportion of effective magnetic leakage data features within the sliding window in the partially effective channel and the completely incomplete channel is less than a third preset proportion, then the sliding window is severely incomplete.

6. The multi-round residual magnetic flux leakage data fusion method according to claim 3, characterized in that, The step of extracting the spatial information of the partially valid channels and the completely incomplete channels includes: If the signal loss rate within the sliding window in the partially effective channel and the completely incomplete channel is greater than a third preset value, then the incomplete feature corresponding to the sliding window is missing. If the noise variance within the sliding window in the partially effective channel and the completely incomplete channel is greater than the preset quantile of the noise variance of the baseline of the defect-free pipeline under the same working conditions, then the defect feature corresponding to the sliding window is noise exceeding the standard. If the signal stability within the sliding window in the partially effective channel and the completely incomplete channel is greater than a fourth preset value, then the incomplete feature corresponding to the sliding window is an amplitude anomaly.

7. The multi-round residual magnetic flux leakage data fusion method according to claim 1, characterized in that, The step of circumferentially rotating and adjusting all signal points of the remaining rounds of magnetic flux leakage data based on the reference round data includes: Based on the reference wheel data, circumferential feature points are determined; the circumferential feature points are the first peak points of the circumferential weld seam signals corresponding to the effective channels in the reference wheel data. Extract the first circumferential clock azimuth angle of the mileage where the circumferential feature point is located; Determine the second peak point of the circumferential weld signal corresponding to the effective channel in the remaining rounds of magnetic flux leakage data, and extract the second circumferential clock azimuth angle of the mileage where the second peak point is located; Based on the first circumferential clock azimuth angle and the second circumferential clock azimuth angle, calculate the azimuth deviation; According to the azimuth deviation, all signal points of the remaining rounds of magnetic flux leakage data are circumferentially rotated and adjusted so that the circumferential clock azimuth of signal points at the same physical location in the multiple rounds of magnetic flux leakage data is unified.

8. The multi-round residual magnetic flux leakage data fusion method according to claim 2, characterized in that, The step of adjusting the axial mileage of all signal points of the remaining wheel leakage magnetic data based on the reference wheel data includes: Based on the pipe section alignment results, anchor points are determined; the anchor points are weld seams and defect features with a depth greater than a preset thickness in the magnetic flux leakage data. Based on the anchor point, the remaining round magnetic flux leakage data is subjected to mileage scaling processing so that the error between the pipe section length corresponding to the remaining round magnetic flux leakage data and the pipe section length corresponding to the reference round data is less than a preset number of magnetic flux leakage data sampling points. Based on the valid channel signals in the reference wheel data, the target mileage offset is determined; Based on the target mileage offset, the axial mileage of all signal points of the remaining round magnetic flux leakage data is adjusted so that the error between the axial mileage of the signal points corresponding to the remaining round magnetic flux leakage data and the axial mileage of the signal points corresponding to the reference round data is less than or equal to a preset number of magnetic flux leakage data sampling points.

9. The multi-round residual magnetic flux leakage data fusion method according to claim 1, characterized in that, The step of obtaining the effective magnetic flux leakage data features located within the incomplete region interval in the non-reference wheel data includes: Determine the degree of incompleteness in the non-baseline wheel data located within the incomplete region interval; If the proportion of effective magnetic flux leakage data features within the incomplete region is 100%, then all effective magnetic flux leakage data features are extracted. If the degree of incompleteness is mild, then the effective magnetic flux leakage data feature portion within the incomplete region is extracted; If the degree of incompleteness is moderate, then extract the effective magnetic flux leakage data feature portion within the incomplete region and the adjacent effective magnetic flux leakage data feature of the ineffective magnetic flux leakage data feature. If the degree of incompleteness is severe, then the extraction of effective magnetic flux leakage data features within the incomplete region will be abandoned. The magnetic flux leakage fusion data includes information tags; the information tags include: mileage, reference wheel channel number, clock orientation, amplitude, source type, corresponding non-reference wheel channel number, and incomplete annotation field; the incomplete annotation field includes: original real data, multi-wheel splicing data, extremely small gap interpolation data, and incomplete data.

10. The method for fusing multi-round residual magnetic flux leakage data according to claim 1, characterized in that, The method further includes: The magnetic flux leakage fusion data is corrected to obtain the target magnetic flux leakage fusion data; The step of correcting the magnetic flux leakage fusion data includes: The original data region, the fusion region, and the fusion edge region of the magnetic flux leakage fusion data are determined; the fusion edge region is the junction between the original data region and the fusion region. Unify the amplitude of the fused data within the fusion region to the amplitude of the reference wheel data in the same channel; In addition, a preset number of data sampling points are obtained in the fusion edge region, and the data sampling points are smoothed using a moving average smoothing algorithm; In addition, additional regional information is added to the fusion region; the additional regional information includes: timestamp, leakage magnetic field internal sensor status.

Images