WO2025035512A1 - 基于正交孪生的油气管道裂纹量化方法及装置、存储介质 - Google Patents

基于正交孪生的油气管道裂纹量化方法及装置、存储介质 Download PDF

Info

Publication number
WO2025035512A1
WO2025035512A1 PCT/CN2023/116749 CN2023116749W WO2025035512A1 WO 2025035512 A1 WO2025035512 A1 WO 2025035512A1 CN 2023116749 W CN2023116749 W CN 2023116749W WO 2025035512 A1 WO2025035512 A1 WO 2025035512A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
crack
magnetic
orthogonal
axis
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2023/116749
Other languages
English (en)
French (fr)
Inventor
郭静波
王艺钊
胡铁华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Publication of WO2025035512A1 publication Critical patent/WO2025035512A1/zh
Anticipated expiration legal-status Critical
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
    • G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
    • G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
    • G01N27/83—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws by investigating stray magnetic fields
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B7/00—Measuring arrangements characterised by the use of electric or magnetic techniques
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C9/00—Measuring inclination, e.g. by clinometers, by levels
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
    • G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
    • G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
    • G01N27/90—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws using eddy currents
    • G01N27/9046—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws using eddy currents by analysing electrical signals

Definitions

  • the embodiments of the present disclosure relate to but are not limited to the field of crack detection technology, and in particular to a method and device for quantifying cracks in oil and gas pipelines based on orthogonal twinning, and a storage medium.
  • the applicable objects of twinning include but are not limited to pipe body crack signals, elbow crack signals, girth weld crack signals of oil and gas pipelines or hydrogen pipelines, high-speed railway rail crack signals, tank bottom plate crack signals and other ferromagnetic metal material crack signals.
  • Ferromagnetic metal materials such as oil and gas pipelines, hydrogen pipelines, high-speed rails, and oil storage tanks have been operating in complex natural environments and pressure loads for a long time. Stress cracks, fatigue cracks, brittle cracks, hydrogen-induced cracks and other tiny cracks appear inside the materials or on the inner and outer walls, and then develop into visible cracks or larger metal losses. Cracks are one of the main sources of various defects in the service process of metal materials.
  • cracks are usually small in size, especially when the detection speed is greater than 2m/s. If the crack is in the initial stage of occurrence or in the early development stage during service, it is more difficult to detect, identify, and quantify early tiny cracks.
  • the crack response signal in crack detection is closely related to the excitation signal, excitation direction, sensing sensor, etc.
  • the unidirectional magnetic flux leakage (axial MFL or circumferential MFL) detection of oil and gas pipelines based on the excitation direction, for cracks of the same size, the larger the inclination angle modulus, the larger the amplitude of its magnetic flux leakage response signal; the smaller the inclination angle modulus, the smaller the amplitude of its magnetic flux leakage response signal. Therefore, the axial crack or crack signal under axial excitation conditions is very weak, and it is even considered by the industry that industrial-grade detection cannot be achieved.
  • the disclosed embodiment provides a method and device for quantifying cracks in an oil and gas pipeline based on orthogonal twinning, and a storage medium.
  • orthogonal twinning here is defined as, under the condition of the original response signal of unidirectional DC excitation being known, a response signal corresponding to a twin magnetic field in the same plane, equal in size and perpendicular in direction as the unidirectional DC excitation is obtained.
  • This response signal is defined as a "twin response signal” (such as a twin leakage magnetic response signal or a twin dynamic magnetic/eddy current response signal), which also forms a twin relationship with the response signal of the original unidirectional DC excitation, that is, the "original response signal” (such as the original leakage magnetic response signal or the original dynamic magnetic or eddy current response signal).
  • the quantization accuracy of the "small tilt angle crack” of the oil and gas pipeline can be greatly improved, especially the quantization accuracy of the cracks distributed parallel to the excitation direction, including the length accuracy, width accuracy, depth accuracy, tilt angle accuracy, etc. of the crack.
  • the "small tilt angle crack” here refers to a crack with a positive and negative deviation of no more than 20° tilt angle based on the excitation direction.
  • the methods of "orthogonal twinning” include but are not limited to machine learning model mapping method, mathematical analysis method, circuit (or electromagnetic) simulation method, hardware device real signal generation method, etc.
  • the process of transforming the “original response signal” into the “twin response signal” is defined as “orthogonal twin transformation”.
  • the disclosed embodiment provides a method for quantifying cracks in oil and gas pipelines based on orthogonal twinning, comprising: obtaining a three-axis leakage magnetic measurement signal modulus of an inner wall crack or an outer wall crack under a unidirectional DC excitation condition, and obtaining a dynamic magnetic signal or an eddy current signal under a dynamic magnetic field excitation condition orthogonal to the DC excitation, wherein the unidirectional DC excitation refers to a DC excitation method with a single direction; inputting the three-axis leakage magnetic measurement signal modulus into a crack signal orthogonal twinning model to obtain a leakage magnetic enhancement estimation signal, wherein the leakage magnetic enhancement estimation signal is a function of the simulated orthogonal twin three-axis leakage magnetic signal modulus, wherein the orthogonal twin three-axis leakage magnetic signal modulus is the three-axis leakage magnetic response signal modulus under a virtual orthogonal twin DC excitation condition, wherein the virtual orthogonal twin DC excitation is in the same plane as the
  • extracting a characteristic vector from the leakage magnetic enhancement estimation signal and the dynamic magnetic signal or the eddy current signal includes: obtaining at least one first eigenvalue based on the leakage magnetic enhancement estimation signal; obtaining at least one second eigenvalue based on the dynamic magnetic signal or the eddy current signal; and combining the first eigenvalue and the second eigenvalue into a characteristic vector.
  • the first eigenvalue includes: a major axis, a minor axis, a major axis inclination angle and a peak value
  • obtaining the first eigenvalue based on the leakage magnetic enhancement estimation signal includes: binarizing the leakage magnetic enhancement estimation signal according to a preset binarization threshold to obtain a binary leakage magnetic signal; calculating the major axis, minor axis, major axis inclination angle of the edge contour of the binary leakage magnetic signal and the peak value of the leakage magnetic enhancement estimation signal, and using the obtained major axis, minor axis, major axis inclination angle and peak value as the first eigenvalue;
  • the dynamic magnetic signal or eddy current signal includes a front and rear coil differential signal and a left and right coil differential signal, and the second eigenvalue includes: the peak value and peak-to-peak spacing of the front and rear coil differential signal, and the peak value and peak spacing of the left and right coil differential signal.
  • the crack signal orthogonal twin model is an autoencoder model implemented by a convolutional neural network, and the autoencoder model includes an encoder part and a decoder part; and the crack scale estimation model is a fully connected neural network.
  • the leakage magnetic enhancement estimation signal is the orthogonal twin of the simulated optimal three-axis leakage magnetic signal modulus S Ao and the simulated a function f( ⁇ S Ao , ⁇ S ⁇ ) of a three-axis magnetic flux leakage signal modulus S ⁇ , wherein the simulated optimal three-axis magnetic flux leakage signal modulus S Ao is a simulated three-axis magnetic flux leakage signal modulus that minimizes
  • the method further includes: for multiple crack samples, using the modulus of the three-axis magnetic leakage measurement signal under unidirectional DC excitation conditions and the corresponding magnetic leakage enhancement estimation signal to train the crack signal orthogonal twin model, wherein the magnetic leakage enhancement estimation signal corresponding to each crack sample is generated by the following method: taking the measured scale of the crack sample as the initial crack scale; dividing the maximum element of the MA matrix by the maximum element of the initial matrix to obtain an adjustment factor ⁇ , MA is the modulus of the three-axis magnetic leakage measurement signal; calculating
  • F wherein SA is the simulated three-axis magnetic leakage signal modulus obtained by substituting the crack scale into the magnetic dipole model, and ⁇ ⁇ F is the Frobenius norm of the matrix; repeatedly adjusting the crack scale and inputting the crack scale into the crack magnetic dipole model until the optimal crack scale that minimizes
  • the crack signal orthogonal twin model is trained using the modulus of the three-axis leakage magnetic measurement signal under unidirectional DC excitation conditions and the corresponding leakage magnetic enhancement estimation signal, including: marking the modulus of the three-axis leakage magnetic measurement signal under unidirectional DC excitation conditions and the corresponding leakage magnetic enhancement estimation signal of each crack sample as a group of orthogonal twin mapping pairs; randomly splitting the N groups of orthogonal twin mapping pairs into training sets and test sets according to the ratio k:(1-k), where 0 ⁇ k ⁇ 1; using k ⁇ N groups of the orthogonal twin mapping pairs to train the crack signal orthogonal twin model, and using (1-k) ⁇ N groups of the orthogonal twin mapping pairs to test the crack signal orthogonal twin model.
  • the crack size is trained using the feature vector.
  • the invention discloses a crack scale estimation model, comprising: marking the first eigenvalue and the second eigenvalue of each crack sample and the corresponding true value of the crack size and the inclination angle as a set of crack scale estimation mapping pairs; using k ⁇ N sets of the orthogonal twin mapping pairs corresponding to the crack scale estimation mapping pairs to train the crack scale estimation model, and using (1-k) ⁇ N sets of the orthogonal twin mapping pairs corresponding to the crack scale estimation mapping pairs to test the crack scale estimation model.
  • the disclosed embodiment also provides an oil and gas pipeline crack quantification device based on orthogonal twinning, comprising: a magnetic leakage sensor probe under unidirectional DC excitation conditions, a dynamic magnetic or eddy current sensor probe under unidirectional DC excitation and dynamic magnetic field excitation conditions orthogonal to DC excitation, a memory for storing instructions, algorithms, and models, a processor for executing the oil and gas pipeline crack quantification method, and a bus system connecting various units.
  • the embodiments of the present disclosure also provide a storage medium on which a computer program is stored.
  • the program is executed by a processor, the method for quantifying oil and gas pipeline cracks based on orthogonal twinning described in any embodiment of the present disclosure is implemented.
  • the oil and gas pipeline crack quantification method and device and storage medium based on orthogonal twinning of the disclosed embodiment determine the objective existence of cracks through crack response signals based on multiple detection principles.
  • orthogonal twinning is performed on the crack signal to obtain an enhanced signal corresponding to the crack, and then based on the enhanced crack signal of the orthogonal twinning, a machine learning model is applied to achieve high-precision size quantification of the original measured weak crack signal.
  • the oil and gas pipeline crack quantification method of the disclosed embodiment through the orthogonal twin model of crack signals, based on unidirectional DC excitation conditions, proposes a new method and technology for accurately detecting, identifying, and quantifying tiny cracks, especially cracks distributed parallel to the excitation direction, which has important practical significance for the safe operation of major infrastructure such as oil and gas pipelines, hydrogen pipelines, high-speed railway rails, and oil storage tanks.
  • FIG1 is a schematic diagram of an ellipsoidal profile and orthogonal magnetic field of a crack in an oil and gas pipeline according to an exemplary embodiment of the present disclosure
  • FIG2 is a flow chart of an oil and gas pipeline crack quantification method and device based on orthogonal twinning and a storage medium according to an exemplary embodiment of the present disclosure
  • 3A is a framework diagram of an oil and gas pipeline crack quantification method and device based on orthogonal twinning and a storage medium in a model training stage according to an exemplary embodiment of the present disclosure
  • FIG3B is a framework diagram of an oil and gas pipeline crack quantification method and device based on orthogonal twinning and a storage medium in a model use stage according to an exemplary embodiment of the present disclosure
  • FIG4A is a loss function iteration process of a “crack signal orthogonal twin model” (ML OT ) according to an exemplary embodiment of the present disclosure
  • FIG4B is a determination coefficient iteration process of a “crack signal orthogonal twin model” (ML OT ) according to an exemplary embodiment of the present disclosure
  • FIG5A is a loss function iteration process of a “crack size estimation model” (ML size ) according to an exemplary embodiment of the present disclosure
  • FIG5B is an iterative process of determining coefficients of a “crack size estimation model” (ML size ) according to an exemplary embodiment of the present disclosure
  • FIG6A is a 3-D effect diagram of signal enhancement for a crack with a 0° tilt angle according to an exemplary embodiment of the present disclosure
  • FIG6B is a top view of the signal enhancement effect of a crack with a tilt angle of 0° according to an exemplary embodiment of the present disclosure
  • FIG. 7A is a 3-D effect diagram of signal enhancement for a crack with a 3° tilt angle according to an exemplary embodiment of the present disclosure
  • FIG7B is a top view of the signal enhancement effect of a crack with an inclined angle of 3° according to an exemplary embodiment of the present disclosure
  • FIG8A is a 3-D effect diagram of signal enhancement for a crack with an inclined angle of 18° according to an exemplary embodiment of the present disclosure
  • FIG8B is a top view of the signal enhancement effect of a crack with an inclined angle of 18° according to an exemplary embodiment of the present disclosure
  • FIG9 is a schematic structural diagram of an oil and gas pipeline crack quantification device based on orthogonal twinning according to an exemplary embodiment of the present disclosure.
  • the present disclosure proposes the concept of "orthogonal twinning", as shown in Figure 1, which aims to virtually twin the leakage magnetic response signal generated by the excitation magnetic field H ⁇ in the same plane, equal in size and perpendicular to the DC excitation magnetic field under the condition of only single-direction DC excitation HA .
  • the leakage magnetic enhancement estimation signal is obtained by jointly processing the original response signal and the twin response signal, and then the leakage magnetic enhancement estimation signal is used to estimate the crack length, width, depth and inclination angle, and finally achieve high-precision quantification of cracks, especially cracks with small inclination angles.
  • the inclination angle of a crack is defined as the angle between the major axis direction of the crack ellipsoid contour and the original excitation direction HA , ranging from (-90°, 90°].
  • the crack quantification method disclosed in the present invention is applicable to cracks with small inclination angles but is not limited to cracks with small inclination angles.
  • "cracks with small inclination angles” refer to cracks with an inclination angle of no more than 20° with a positive or negative deviation based on the excitation direction.
  • unidirectional DC excitation refers to a DC excitation method with a single direction, including but not limited to the existing axial DC excitation method, annular DC excitation method, and spiral DC excitation method.
  • a spatial rectangular coordinate system is established on the inner wall of the oil and gas pipeline.
  • the geometric center of the crack ellipsoid outline is taken as the coordinate origin, and the excitation direction (HA direction) is defined as the X-axis direction;
  • the direction perpendicular to the X-axis direction in the plane of the pipeline wall is the Y-axis direction;
  • the direction perpendicular to the pipeline wall and pointing to the inside of the pipeline is the Z-axis direction;
  • the opposite direction of the Y-axis is the H ⁇ direction;
  • the major axis direction of the crack ellipsoid outline is the X′-axis direction, and the inclination angle relative to the X-axis direction is ⁇ ;
  • the direction perpendicular to the X′axis in the plane of the pipeline wall is the Y′-axis direction, and its inclination angle relative to the Y-axis direction is ⁇ ;
  • the direction coinciding with the Z-axis is
  • the embodiment of the present disclosure provides an oil and gas pipeline crack quantification method and device based on orthogonal twinning, and a storage medium, comprising the following steps:
  • the crack quantification method of the embodiment of the present disclosure is described by taking an oil and gas pipeline as an example.
  • the implementation steps in other ferromagnetic metal materials can refer to this method and will not be described in detail in the present disclosure.
  • step 201 may include the following steps:
  • a weak matrix signal of crack response with a scale of (L, W, D, ⁇ ) is obtained under unidirectional DC excitation conditions.
  • a 3-axis magnetic flux leakage and 2-axis dynamic magnetic or eddy current sensor ultra-high resolution integrated probe is used to obtain a weak matrix signal of crack response with a scale of (L, W, D, ⁇ ) under unidirectional DC excitation conditions, such as obtaining the modulus of the three-axis magnetic flux leakage measurement signal under the conditions of the DC excitation field HA in the A direction.
  • (L, W, D, ⁇ ) are the actual measured length, width, depth, and inclination angle of the crack; They respectively represent the X-axis component, Y-axis component and Z-axis component of the three-axis magnetic flux leakage measurement signal; Respectively represent the dynamic magnetic or eddy current signals of the front and rear coil differential (hereinafter referred to as "front and rear differential") of the ultra-high resolution integrated probe circuit board of the 3-axis leakage magnetic and 2-axis dynamic magnetic or eddy current sensor and the dynamic magnetic or eddy current signals of the left and right coil differential (hereinafter referred to as "left and right differential").
  • the above MA and DA are matrix signals. At least include the indication signal of the existence of cracks, such as and The crack existence indication signal includes the boundary information of the crack. The signal can distinguish whether the crack is an inner wall crack (ID) or an outer wall crack (OD).
  • the inner wall crack magnetic dipole model or the outer wall crack magnetic dipole model is applied to generate the simulated triaxial magnetic leakage signal modulus of the virtual crack with scale ( Ls , Ws , Ds , ⁇ s ).
  • SA is a matrix signal
  • ( Ls , Ws , Ds , ⁇ s ) are the length, width, depth and inclination angle of the virtual crack, respectively, which are intermediate variables, and their initial values can correspond to the known measured crack scales (L, W, D, ⁇ ); They are the X-axis component, Y-axis component and Z-axis component of the simulated three-axis magnetic leakage signal.
  • fxX ′ ( ⁇ ), fyX ′ ( ⁇ ), fzX ′ ( ⁇ ) are respectively the three-axis (X′ axis, Y′ axis, Z′ axis) integration operators of the magnetic dipole under the excitation of the X′ axis component of HA ;
  • fxY ′ ( ⁇ ), fyY ′ ( ⁇ ), fzY′ ( ⁇ ) are respectively the three-axis (X′ axis, Y′ axis, Z′ axis) integration operators of the magnetic dipole under the excitation of the Y′ axis component of HA ;
  • ⁇ 0 4 ⁇ 10-7 (H/m) is the vacuum magnetic permeability, ⁇ r is the relative magnetic permeability of the pipe wall under saturation conditions;
  • ⁇ x ′ is the magnetic charge surface density of the magnetic dipole in the X ′ axis direction
  • ⁇ y ′ is the magnetic charge surface density of the magnetic dipole in the Y′ axis direction
  • ⁇ F is the Frobenius norm of the matrix
  • the adjustment factor: ⁇ max( MA )/max( SA_initial ), which is equal to the maximum element of the MA matrix divided by the maximum element of the initial matrix
  • the initial matrix ( SA_initial ) is the simulated triaxial magnetic flux leakage signal modulus obtained by substituting the measured crack scale (L, W, D, ⁇ ) into the magnetic dipole model.
  • Complex crack shapes can be decomposed into a combination of simple strip contours, and the strip contours can be abstracted into simple geometric shapes such as cuboids, ellipsoids, and elliptical cylinders.
  • the crack contour is approximated as an ellipsoid.
  • the crack contour may also be approximated as a rectangular parallelepiped, an elliptical cylinder, etc., which is not limited in the embodiment of the present disclosure.
  • ⁇ xOT′ is the surface density of the orthogonal magnetic charge of the magnetic dipole in the X′ axis direction
  • ⁇ yOT′ is the surface density of the orthogonal magnetic charge of the magnetic dipole in the Y′ axis direction.
  • N groups of orthogonal twins are divided into The raw mapping pairs and measured scale sets are randomly split into training sets and test sets, with the number of k ⁇ N and (1-k) ⁇ N, respectively.
  • a part of the k ⁇ N training sets is randomly selected as the validation set for model training.
  • the edge contour is usually an ellipse.
  • the major axis of the edge profile is a i
  • the minor axis is b i
  • the inclination angle of the major axis is The peak value is p i
  • step 202 may include the following steps:
  • the weak matrix signal of crack response under unidirectional DC excitation conditions is obtained.
  • the modulus of the triaxial magnetic flux leakage measurement signal under the condition of DC excitation field HA in direction A is obtained.
  • the dynamic magnetic or eddy current signal under the condition of DC excitation magnetic field HA in direction A At least include indicators of crack presence, such as and
  • the crack existence indication signal includes the boundary information of the crack. The signal can distinguish whether the crack is an inner wall crack (ID) or an outer wall crack (OD).
  • the MA signal is input into the MLOT ("orthogonal twin model of inner wall crack signal” or “orthogonal twin model of outer wall crack signal") trained in the first stage, and the leakage magnetic enhancement estimation signal is output.
  • the major axis of the edge profile is a
  • the minor axis is b
  • the inclination angle of the major axis is The peak value is p
  • the peak-to-peak value is
  • the peak-to-peak spacing is
  • the peak value is The peak spacing is This set of feature vectors
  • This embodiment uses an ultra-high resolution integrated probe based on 3-axis magnetic leakage and 2-axis dynamic magnetic or eddy current sensors to collect weak crack signals.
  • 3-axis magnetic leakage refers to the axial component, radial component and circumferential component of the leakage magnetic field;
  • 2-axis dynamic magnetic or eddy current means that the probe circuit board can simultaneously collect the front and rear differential dynamic magnetic or eddy current signals and the left and right differential dynamic magnetic or eddy current signals.
  • the front and rear differential dynamic magnetic or eddy current signals can detect and identify 360° omnidirectional cracks inside and outside, but as the crack tilt angle modulus decreases, the front and rear differential dynamic magnetic or eddy current signal amplitude gradually decreases; while the left and right differential dynamic magnetic or eddy current signal amplitude increases with the decrease of the crack tilt angle modulus; the two complement each other and make up for each other's shortcomings. Therefore, the use of 3-axis magnetic leakage and 2-axis dynamic magnetic or eddy current sensor ultra-high resolution integrated probes can determine whether cracks exist, distinguish internal cracks from external cracks, and determine suspicious crack areas.
  • the entire implementation process is divided into two stages.
  • the first stage is the establishment stage of the "crack quantification model” (model training stage); the second stage is the use stage of the "crack quantification model” (model testing stage).
  • step 1) use 3-axis leakage magnetic field and 2-axis dynamic magnetic field or eddy current detectors with axial excitation to conduct (X80 steel, 15.30mm wall thickness) artificial crack oil and gas pipeline pulling experiment, thereby obtaining the pulling data of artificial cracks.
  • axial excitation means that the excitation direction (X-axis direction) coincides with the axial direction at this time, and the established coordinate system is as shown in Figure 1.
  • 2160 artificial cracks of different sizes and different inclination angles are processed on the pulling pipeline as samples for model training, including 1080 inner wall cracks (ID) and 1080 outer wall cracks (OD); for inner wall cracks and outer wall cracks, 75% of the samples (810) are randomly assigned as training sets (marked with train), and the other 25% of the samples (270) are test sets (marked with test), and 25% of the samples (203) are randomly selected from the training set and defined as the validation set (marked with val), and their scale distribution ranges are shown in Table 1. Import the above pulling data into the "Crack Original Sample Calibration Software", select and export the crack original sample data file in the known area of the processed crack, and save it in txt format.
  • Each crack original sample data file consists of five matrix data, namely, leakage magnetic axial data matrix MFL Radial Data Matrix Flux Leakage Toroidal Data Matrix Dynamic magnetic or eddy current before and after differential data matrix and dynamic magnetic or eddy current left and right differential data matrix and,
  • 810 samples in the training set are prepared for the first stage training of the "crack quantification model”
  • 270 samples in the test set are prepared for the second stage testing of the "crack quantification model”.
  • step 2) the magnitude of the axial excitation field HA of the pulling test is measured in advance. Then, under the same axial excitation field conditions, for each of the 1080 artificial crack samples with different sizes and different inclination angles, the measured scales recorded in their file names are used.
  • f xX′ ( ⁇ ), f yX′ ( ⁇ ), f zX′ ( ⁇ ) are the three-axis (X′ axis, Y′ axis, Z′ axis) integration operators of the magnetic dipole under the excitation of the X′ axis component of HA ;
  • f xY′ ( ⁇ ), f yY′ ( ⁇ ), f zY′ ( ⁇ ) are the three-axis (X′ axis, Y′ axis, Z′ axis) integration operators of the magnetic dipole under the excitation of the Y′ axis component of HA ;
  • ⁇ 0 4 ⁇ 10 -7 (H/m)
  • ⁇ r is the relative permeability of the tube wall under saturation conditions is the magnetic charge surface density of the magnetic dipole in the X′ axis direction corresponding to the i-th artificial crack sample, is the magnetic charge surface density
  • step 3 in order to make the effect of crack simulation consistent with the effect of actual crack measurement, for the inner wall crack and outer wall crack, the objective function and constraint conditions are established one by one for 1080 artificial crack samples of different sizes and different inclination angles:
  • the simulated three-axis magnetic leakage signal modulus is obtained by substituting the measured crack scale (L i , W i , D i , ⁇ i ) into the magnetic dipole model.
  • the above objective function (18) is satisfied.
  • the optimal virtual crack size that satisfies (18) is denoted as The corresponding simulated three-axis magnetic leakage signal is recorded as
  • the orthogonal magnetic dipole model of the inner wall crack and the orthogonal magnetic dipole model of the outer wall crack are used to calculate the orthogonal twin triaxial magnetic leakage signal modulus of 2160 artificial crack samples of different sizes and different inclination angles.
  • the surface density of the orthogonal magnetic charge in the X′ direction of the magnetic dipole corresponding to the i-th artificial crack sample is the surface density of the magnetic charge orthogonal to the Y′ axis of the magnetic dipole corresponding to the i-th artificial crack sample.
  • step 5 for the inner wall crack and the outer wall crack, 1080 orthogonal twin mapping pairs of different sizes and different inclination angles and the measured size sets are obtained respectively.
  • This embodiment specifically uses the convolutional neural network autoencoder (CNN Autoencoder) algorithm to build the crack signal orthogonal twin model.
  • CNN Autoencoder convolutional neural network autoencoder
  • the physical step exceeds 0.5mm, cubic spline interpolation can be performed. If the matrix dimension does not meet the requirements, the crack center is used as the matrix center, and the edge part exceeding 200 dimensions is deleted, and the edge part less than 200 dimensions is padded with zeros.
  • the convolutional neural network autoencoder designed in this embodiment has a total of 37 layers, of which the encoding part contains 18 layers and the decoding part contains 19 layers.
  • the network architecture built using Python language is as follows:
  • Conv2D represents a two-dimensional convolution layer
  • SpatialDropout2D represents a two-dimensional spatial random zeroing layer
  • BatchNormalization represents a data batch normalization layer
  • ReLU represents an activation function layer
  • MaxPooling2D represents a two-dimensional pooling layer
  • UpSampling2D represents a two-dimensional upsampling layer
  • Activation represents an activation layer.
  • Each line of Python represents a layer of structure, and the output of each layer of structure is used as the input of the next layer of structure, and the output of the next layer of structure is used as the input of the next layer of structure.
  • the edge contour is usually an ellipse.
  • the major axis of the edge profile is a i
  • the minor axis is b i
  • the inclination angle of the major axis is The peak value is p i
  • the peak-to-peak value is
  • This embodiment specifically uses a fully connected neural network (DNN) algorithm to build a crack size estimation model.
  • the fully connected neural network designed in this embodiment has a total of 5 layers, namely 1 input layer, 3 hidden layers, and 1 output layer.
  • the network architecture built using Python language is as follows:
  • Dense is used to implement the fully connected layer
  • the middle layer has 10 neurons
  • the output layer has 4 neurons
  • kernel_initializer is the keyword to specify the initialization method
  • activation is the keyword to specify the activation function.
  • the iteration number of model training is set to 500, and the iteration process of model training is shown in Figures 4A, 4B, 5A, and 5B.
  • Figure 4A is the iterative process of the loss function of the "orthogonal twin model of crack signal" (ML OT )
  • Figure 4B is the iterative process of the coefficient of determination (i.e., R 2 ) of the "orthogonal twin model of crack signal" (ML OT ).
  • the evaluation index of the "orthogonal twin model of crack signal" (ML OT ) is shown in Table 2.
  • the loss function of the training set and the validation set is finally reduced to 0.0002, indicating that the final mean square error (MSE) of the ML OT model is very small and there is no overfitting.
  • MSE mean square error
  • the determination coefficient of the ML OT model training set finally reached 0.9371
  • the determination coefficient of the validation set finally reached 0.9281, indicating that the final fitting effect of the ML OT model was good and basically met the requirements of the orthogonal twin transformation.
  • Figure 5A is the iterative process of the loss function of the "crack size estimation model” (ML size )
  • Figure 5B is the iterative process of the coefficient of determination (i.e., R 2 ) of the "crack size estimation model” (ML size ).
  • the evaluation index of the "crack size estimation model” (ML size ) is shown in Table 3. It can be seen that after the ML size model converges, the loss function of the training set is finally reduced to 0.0044, and the loss function of the validation set is finally reduced to 0.0027, indicating that the final mean square error (MSE) of the ML size model is also very small, and there is no overfitting.
  • MSE mean square error
  • the determination coefficient of the ML size model training set finally reaches 0.9907
  • the determination coefficient of the validation set finally reaches 0.9940, indicating that the final fitting effect of the ML size model is very good, and the estimation accuracy of the crack length, width, depth, and inclination angle is very high.
  • model testing In the second stage, the model was tested on 270 test set samples that were not trained in the first stage.
  • the steps for using the model (model testing) are:
  • step 1) weak matrix signals of crack response under unidirectional DC excitation are obtained based on different detection principles.
  • Each crack original sample data file consists of five matrix data, namely, leakage magnetic axial data matrix MFL Radial Data Matrix Flux Leakage Toroidal Data Matrix Dynamic magnetic or eddy current before and after differential data matrix and dynamic magnetic or eddy current left and right differential data matrix At least include indicators of crack presence, such as and
  • the crack existence indication signal includes the boundary information of the crack. The signal can distinguish whether the crack is an inner wall crack (ID) or an outer wall crack (OD).
  • the signal is input into ML OT ("orthogonal twin model of inner wall crack signal” or “orthogonal twin model of outer wall crack signal”), and the leakage magnetic enhancement estimation signal is output.
  • the edge contour is usually an ellipse.
  • the major axis of the edge profile is a i
  • the minor axis is b i
  • the inclination angle of the major axis is The peak value is p i
  • the peak-to-peak value is
  • step 6 for inner wall cracks or outer wall cracks, ensure that all 270 cracks in the first phase test set are quantified.
  • Figures 6A, 6B, 7A, 7B, 8A, and 8B are orthogonal twinning effect diagrams of three groups of inner wall crack signals of different scales in the test set samples. From left to right, the three figures are the original input leakage magnetic measurement signal modulus MA , the picture title is "Input Signal”, the leakage magnetic enhancement estimation signal output by MLOT The picture title is “Predicted output signal”; and the ideal signal f( ⁇ S Ao , ⁇ S ⁇ ) output by ML OT , the picture title is “True output signal”.
  • Figure 6B shows that the original input signal MA presents a double peak feature, and it is even impossible to distinguish it as an axial crack. However, after the orthogonal twin transformation, the ML OT output signal It clearly reflects its axial crack characteristics, so The signal provides favorable conditions for quantifying the size of the axial crack.
  • Figure 7B shows that after the orthogonal twin transformation, the ML OT output signal of the crack is The crack signal characteristics are enhanced, providing favorable conditions for high-precision quantification of its scale.
  • the calculation results are shown in Table 4 at a confidence level of 90%. It can be seen that the confidence interval of the length estimation error of the inner wall crack is (-1.046mm, 1.026mm), the confidence interval of the width estimation error is (-0.006mm, 0.005mm), the confidence interval of the depth estimation error is (-0.214mm, 0.193mm), and the confidence interval of the inclination angle estimation error is (-2.282°, 2.452°).
  • the oil and gas pipeline crack quantification method and device based on orthogonal twinning and the storage medium disclosed in the present invention have very high scale quantization accuracy.
  • the disclosed embodiment also provides an oil and gas pipeline crack quantification device based on orthogonal twinning, including: 1) 3-axis magnetic leakage and a 2-axis dynamic magnetic/eddy current sensor ultra-high resolution integrated probe, which can collect front and rear differential dynamic magnetic/eddy current signals and left and right differential dynamic magnetic/eddy current signals; 2) a memory for storing instructions, algorithms, and models; 3) a processor for executing an oil and gas pipeline crack quantification method based on orthogonal twinning; 4) a display for displaying quantification results; 5) a bus system connecting various units.
  • an orthogonal twin-based oil and gas pipeline crack quantification device may include: a 3-axis magnetic flux leakage and 2-axis dynamic magnetic/eddy current sensor ultra-high resolution integrated probe 910, a memory 920, a processor 930, a display 940 and a bus system 950, wherein the 3-axis magnetic flux leakage and 2-axis dynamic magnetic/eddy current sensor ultra-high resolution integrated probe 910, the memory 920, the processor 930, the display 940 are connected via the bus system 950; the 3-axis magnetic flux leakage and 2-axis dynamic magnetic/eddy current sensor ultra-high resolution integrated probe 910, the memory 920, the processor 930, the display 940
  • the ultra-high resolution integrated probe 910 is used to detect and obtain weak crack response signals with a scale of (L, W, D, ⁇ ) under unidirectional DC excitation conditions, including 3-axis magnetic leakage signals and front and rear differential dynamic magnetic/eddy current signals and left and right differential dynamic magnetic/eddy current signals; the
  • the processor 930 is used to execute the instructions stored in the memory 920 to quantify cracks through the oil and gas pipeline crack quantification method based on orthogonal twins. Specifically, the processor 930 can train the "crack signal orthogonal twin model” and the “crack scale estimation model” respectively; it can enhance weak crack signals with different inclination angles and perform scale estimation; and finally display the quantification results through the display 940.
  • the memory 920 may include a read-only memory and a random access memory, and provide instructions and data to the processor 930, including the "crack signal orthogonal twin model" and the “crack scale estimation model", etc.
  • a portion of the memory 920 may also include a non-volatile random access memory.
  • the memory 920 may also store information about the device type.
  • the processor 930 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • a general-purpose processor may be a microprocessor or the processor 930 may be any conventional processor, etc.
  • the display 940 can also display the crack detection data in the memory 920 through the crack analysis software.
  • the bus system 950 may also include a power bus, a control bus, a status signal bus, and the like.
  • the processing performed by the oil and gas pipeline crack quantification device based on orthogonal twins can be completed by the hardware integrated logic circuit in the processor 930 or the instructions in the form of software. That is, the steps of the oil and gas pipeline crack quantification device based on orthogonal twins in the embodiment of the present disclosure can be executed by a hardware processor, or by a combination of hardware and software modules in the processor 930.
  • the software module can be located in a storage medium such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc.
  • the storage medium is located in the memory 920, and the processor 930 reads the information in the memory 920 and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.
  • An embodiment of the present disclosure also provides a storage medium, which stores executable instructions.
  • the executable instructions When the executable instructions are executed by a processor, the oil and gas pipeline crack quantification method based on orthogonal twins provided by any of the above embodiments of the present disclosure can be implemented; in addition, the storage medium can also store crack quantization models, including "crack signal orthogonal twin model” and “crack scale estimation model”; the small inclination angle cracks are quantified using the trained “crack signal orthogonal twin model” and “crack scale estimation model” respectively.
  • Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium).
  • a computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data).
  • Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.
  • communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Chemical & Material Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Electrochemistry (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Investigating Or Analyzing Materials By The Use Of Magnetic Means (AREA)

Abstract

一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质，该方法包括：获取在单向直流励磁条件下内壁裂纹或者外壁裂纹的三轴漏磁测量信号模值，获取在与直流励磁正交的动态磁场激励条件下的动磁信号或涡流信号；将所述三轴漏磁测量信号模值输入裂纹信号正交孪生模型，得到漏磁增强估计信号，所述漏磁增强估计信号为正交孪生三轴漏磁信号模值的函数，所述正交孪生三轴漏磁信号模值为在虚拟正交孪生直流励磁条件下的漏磁响应信号模值，所述虚拟正交孪生直流励磁与所述单向直流励磁同平面、大小相等且方向垂直；从所述漏磁增强估计信号以及所述动磁信号或涡流信号中提取特征向量，将所述特征向量输入裂纹尺度估计模型得到裂纹的尺寸和倾斜角。

Description

基于正交孪生的油气管道裂纹量化方法及装置、存储介质
本申请要求于2023年8月16日提交中国专利局、申请号为2023110347949、发明名称为“基于正交孪生的油气管道裂纹量化方法及装置、存储介质”的中国专利申请的优先权，其内容应理解为通过引用的方式并入本申请中。
技术领域
本公开实施例涉及但不限于裂纹检测技术领域，尤其涉及一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质，孪生的适用对象包括但不限于油气管道或输氢管道的管体裂纹信号、弯头裂纹信号、环焊缝裂纹信号，高铁钢轨裂纹信号、储罐底板裂纹等铁磁性金属材料裂纹信号。
背景技术
油气管道、输氢管道、高铁钢轨、石油储罐等铁磁性金属材料长期运营在复杂的自然环境和压力负荷中，在材料内部或内外壁出现应力裂纹、疲劳裂纹、脆性裂纹、氢致裂纹等微小裂纹，进而发展成为可见的裂隙或尺度较大的金属损失。裂纹是金属材料服役过程中各类缺陷的主要源头因素之一。
金属损失相对裂纹而言由于体积较大而更容易被检测，目前已经有多种比较成熟的技术可以准确检测、识别、量化金属损失缺陷。但是裂纹通常尺度较小，尤其是在检测速度大于2m/s的条件下，如果裂纹处于发生的起始阶段或服役过程中裂纹处于的较早发展阶段，对早期微小裂纹的检测、识别、量化难度更高。
同时，裂纹检测中裂纹响应信号与激励信号、激励方向、感测传感器等密切相关。例如，在油气管道单向漏磁(轴向MFL或环向MFL)检测中，以励磁方向为基准，对于相同尺寸的裂纹，它的倾斜角模值越大，它的漏磁响应信号的幅值越大；它的倾斜角模值越小，它的漏磁响应信号的幅值越小。因此，轴向励磁条件下的轴向裂纹或裂隙信号非常微弱，甚至被业界认为无法实现工业级检测。改变轴向励磁为环向励磁可以提高轴向裂纹或裂隙信号的强度，但环向励磁检测器又对环向裂纹或裂缝的检测效果欠佳。为此，工业界出现了轴向励磁MFL检测器外加环向励磁MFL检测器的串行组合MFL检测器方案。很明显组合MFL检测器方案明显提升了设备制造成本和检测工程成本；同时传统MFL检测受其原理的制约，对微小裂纹的检测、识别、量化能力明显不足或根本无法实现。
发明内容
本公开实施例提供了一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质， 这里的“正交孪生”概念被定义为，在已知单向直流励磁的原始响应信号的条件下，得到与单向直流励磁同平面、大小相等、方向垂直的孪生磁场对应的响应信号，该响应信号被定义为“孪生响应信号”(例如孪生漏磁响应信号或孪生动磁/涡流响应信号)，它与原始单向直流励磁的响应信号，即“原始响应信号”(例如原始漏磁响应信号或原始动磁或涡流响应信号)也构成孪生关系。通过联合处理“孪生响应信号”和“原始响应信号”，可以大幅度提高油气管道“小倾斜角裂纹”的量化精度，尤其是与励磁方向平行分布的裂纹的量化精度，包括裂纹的长度精度、宽度精度、深度精度、倾斜角精度等等。这里的“小倾斜角裂纹”是指以励磁方向为基准，正负偏差不超过20°倾斜角的裂纹。“正交孪生”的方法包括但不限于机器学习模型映射法、数学解析法、电路(或电磁)仿真法、硬件设备真实信号产生法等。将“原始响应信号”变换为“孪生响应信号”的过程定义为“正交孪生变换”。
本公开实施例采用的技术方案为：
本公开实施例提供了一种基于正交孪生的油气管道裂纹量化方法，包括：获取在单向直流励磁条件下的内壁裂纹或者外壁裂纹三轴漏磁测量信号模值，获取在与直流励磁正交的动态磁场激励条件下的动磁信号或涡流信号，所述单向直流励磁指具有单一方向的直流励磁方式；将所述三轴漏磁测量信号模值输入裂纹信号正交孪生模型，得到漏磁增强估计信号，所述漏磁增强估计信号为仿真的正交孪生三轴漏磁信号模值的函数，所述正交孪生三轴漏磁信号模值为在虚拟正交孪生直流励磁条件下的三轴漏磁响应信号模值，所述虚拟正交孪生直流励磁与所述单向直流励磁同平面、大小相等且方向垂直；从所述漏磁增强估计信号以及所述动磁信号或涡流信号中提取特征向量，将所述特征向量输入裂纹尺度估计模型，得到裂纹的尺寸和倾斜角。
可选地，从所述漏磁增强估计信号以及所述动磁信号或涡流信号中提取特征向量，包括：根据所述漏磁增强估计信号得到至少一个第一特征值；根据所述动磁信号或涡流信号得到至少一个第二特征值；将所述第一特征值和第二特征值组成特征向量。
可选地，所述第一特征值包括：长轴、短轴、长轴倾斜角以及峰值，所述根据所述漏磁增强估计信号得到第一特征值，包括：根据预设的二值化阈值，对所述漏磁增强估计信号进行二值化处理，得到二值化漏磁信号；计算所述二值化漏磁信号的边缘轮廓的长轴、短轴、长轴倾斜角以及所述漏磁增强估计信号的峰值，将得到的长轴、短轴、长轴倾斜角以及峰值作为所述第一特征值；所述动磁信号或涡流信号包括前后线圈差分信号和左右线圈差分信号，所述第二特征值包括：所述前后线圈差分信号的峰值和峰-峰值间距，以及所述左右线圈差分信号的峰值和峰值间距。
可选地，所述裂纹信号正交孪生模型为通过卷积神经网络实现的自编码器模型，所述自编码器模型包括编码器部分和解码器部分；所述裂纹尺度估计模型为全连接神经网络。
可选地，所述漏磁增强估计信号为仿真最优三轴漏磁信号模值SAo与仿真的正交孪生 三轴漏磁信号模值S⊥的函数f(αSAo,αS⊥)，所述仿真最优三轴漏磁信号模值SAo是使||αSA-MA||F最小的仿真三轴漏磁信号模值，其中MA为所述三轴漏磁测量信号模值，SA为将裂纹尺度代入磁偶极子模型得到的仿真三轴漏磁信号模值，‖·‖F为矩阵的Frobenius范数，α为调节因子；分别是仿真三轴漏磁信号的X轴分量、Y轴分量以及Z轴分量，所述X轴方向与所述单向直流励磁的激励方向相同，所述Y轴方向在所述油气管道平面内与所述X轴方向垂直，所述Z轴方向分别与所述X轴方向和Y轴方向垂直；分别是正交孪生三轴漏磁信号的X轴分量、Y轴分量以及Z轴分量； K≥1；调节因子α表示等于MA矩阵的最大元素除以初始矩阵的最大元素，所述初始矩阵为将裂纹实测尺度代入磁偶极子模型计算所得的仿真三轴漏磁信号模值组成的矩阵。
可选地，所述方法还包括：对于多个裂纹样本，使用在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的所述漏磁增强估计信号训练所述裂纹信号正交孪生模型，其中，每个所述裂纹样本对应的漏磁增强估计信号通过如下方法生成：以所述裂纹样本的实测尺度为初始的裂纹尺度；用MA矩阵的最大元素除以初始矩阵的最大元素得到调节因子α，MA为所述三轴漏磁测量信号模值；计算||αSA-MA||F，其中SA为将裂纹尺度代入磁偶极子模型得到的仿真三轴漏磁信号模值，‖·‖F为矩阵的Frobenius范数；反复调节所述裂纹尺度，并将所述裂纹尺度输入裂纹磁偶极子模型，直到得到使||αSA-MA||F最小的最优裂纹尺度，与所述最优裂纹尺度对应的SA即为仿真最优三轴漏磁信号模值SAo；将所述最优裂纹尺度输入裂纹正交磁偶极子模型，得到所述仿真的正交孪生三轴漏磁信号模值S⊥；将所述SAo与S⊥代入公式K≥1，从而生成所述漏磁增强估计信号。
可选地，对于所述多个裂纹样本，使用在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的所述漏磁增强估计信号训练所述裂纹信号正交孪生模型，包括：将每个裂纹样本在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的漏磁增强估计信号标记为一组正交孪生映射对；将N组正交孪生映射对按照比例k:(1-k)随机拆分为训练集和测试集，其中，0<k<1；使用k×N组所述正交孪生映射对训练所述裂纹信号正交孪生模型，使用(1-k)×N组所述正交孪生映射对测试所述裂纹信号正交孪生模型。
可选地，在训练所述裂纹信号正交孪生模型之后，使用所述特征向量训练所述裂纹尺 度估计模型，包括：将每个裂纹样本的第一特征值和第二特征值与对应的裂纹尺寸和倾斜角的真实值标记为一组裂纹尺度估计映射对；使用k×N组所述正交孪生映射对对应的裂纹尺度估计映射对进行所述裂纹尺度估计模型的训练，使用(1-k)×N组所述正交孪生映射对对应的裂纹尺度估计映射对进行所述裂纹尺度估计模型的测试。
本公开实施例还提供了基于正交孪生的油气管道裂纹量化装置，包括：单向直流励磁条件下的漏磁传感器探头，单向直流励磁以及与直流励磁正交的动态磁场激励条件下的动磁或涡流传感器探头，存储指令、算法、模型的存储器，执行所述的油气管道裂纹量化方法的处理器，以及连接各个单元的总线系统。
本公开实施例还提供了一种存储介质，其上存储有计算机程序，该程序被处理器执行时实现本公开任一实施例所述的基于正交孪生的油气管道裂纹量化方法。
本公开实施例的基于正交孪生的油气管道裂纹量化方法及装置、存储介质，通过依据多种检测原理的裂纹响应信号，确定裂纹的客观存在性。当裂纹信号的信噪比较低时，对裂纹信号实施正交孪生，得到裂纹对应的增强信号，然后基于正交孪生的增强裂纹信号，应用机器学习模型，实现对原始测量的微弱裂纹信号的高精度尺寸量化。
本公开实施例的油气管道裂纹量化方法，通过裂纹信号正交孪生模型，基于单向直流励磁条件，提出一种准确检测、辨识、量化微小裂纹，尤其是可以准确量化与励磁方向平行分布的裂纹的新方法新技术，对油气管道、输氢管道、高铁钢轨、石油储罐等重大基础设施的安全运营具有重要实际意义。
本公开的其它特征和优点将在随后的说明书中阐述，并且，部分地从说明书中变得显而易见，或者通过实施本公开而了解。本公开的其他优点可通过在说明书以及附图中所描述的方案来实现和获得。
附图说明
附图用来提供对本公开技术方案的理解，并且构成说明书的一部分，与本公开的实施例一起用于解释本公开的技术方案，并不构成对本公开技术方案的限制。
图1为本公开示例性实施例一种油气管道裂纹椭球轮廓及正交磁场示意图；
图2为本公开示例性实施例一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质的流程示意图；
图3A为本公开示例性实施例一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质在模型训练阶段的框架结构图；
图3B为本公开示例性实施例一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质在模型使用阶段的框架结构图；
图4A为本公开示例性实施例一种“裂纹信号正交孪生模型”(MLOT)的损失函数迭代过程；
图4B为本公开示例性实施例一种“裂纹信号正交孪生模型”(MLOT)的确定系数迭代过程；
图5A为本公开示例性实施例一种“裂纹尺度估计模型”(MLsize)的损失函数迭代过程；
图5B为本公开示例性实施例一种“裂纹尺度估计模型”(MLsize)的确定系数迭代过程；
图6A为本公开示例性实施例对0°倾斜角裂纹的信号增强3-D效果图；
图6B为本公开示例性实施例对0°倾斜角裂纹的信号增强俯视效果图；
图7A为本公开示例性实施例对3°倾斜角裂纹的信号增强3-D效果图；
图7B为本公开示例性实施例对3°倾斜角裂纹的信号增强俯视效果图；
图8A为本公开示例性实施例对18°倾斜角裂纹的信号增强3-D效果图；
图8B为本公开示例性实施例对18°倾斜角裂纹的信号增强俯视效果图；
图9为本公开示例性实施例一种基于正交孪生的油气管道裂纹量化装置的结构示意图。
具体实施方式
为使本公开的目的、技术方案和优点更加清楚明白，下文中将结合附图对本公开的实施例进行详细说明。需要说明的是，在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互任意组合。
除非另外定义，本公开实施例公开使用的技术术语或者科学术语应当为本公开所属领域内具有一般技能的人士所理解的通常意义。本公开实施例中使用的“第一”、“第二”以及类似的词语并不表示任何顺序、数量或者重要性，而只是用来区分不同的组成部分。“包括”或者“包含”等类似的词语意指出该词前面的元件或物件涵盖出现在该词后面列举的元件或者物件及其等同，而不排除其他元件或者物件。
本公开提出“正交孪生”的概念，如图1所示，旨在仅有单一方向直流励磁HA的条件下，虚拟孪生出与直流激励磁场同平面、大小相等、方向垂直的激励磁场H⊥产生的漏磁响应信号。然后通过联合处理原始响应信号与孪生响应信号从而得到漏磁增强估计信号，然后通过利用漏磁增强估计信号开展裂纹长、宽、深、倾斜角的估计，最终实现对裂纹尤其是小倾斜角裂纹的高精度量化。如图1所示，裂纹的倾斜角定义为裂纹椭球轮廓的长轴方向与原始励磁方向HA之间的夹角，范围为(-90°,90°]。本公开的裂纹量化方法适用于小倾斜角裂纹但不限于小倾斜角裂纹，这里的“小倾斜角裂纹”是指以励磁方向为基准，正负偏差不超过20°倾斜角的裂纹。本公开实施例中，单向直流励磁是指具有单一方向的直流励磁方式，包括但不限于现有的轴向直流励磁方式、环向直流励磁方式、螺旋直流励磁方式。
如附图1所示，在油气管道内壁上建立空间直角坐标系。以裂纹椭球轮廓的几何中心为坐标原点，定义励磁方向(HA方向)为X轴方向；在管道壁平面内与X轴方向垂直的方向为Y轴方向；与管道壁垂直并指向管道内部的方向为Z轴方向；Y轴的反方向为H⊥方向；在管道壁平面内，裂纹椭球轮廓的长轴方向为X′轴方向，相对X轴方向的倾斜角为θ；在管道壁平面内与X′轴垂直的方向为Y′轴方向，它相对Y轴方向的倾斜角为θ；与Z轴重合的方向为Z′轴方向；管道内油、气的流动方向为轴向；与轴向垂直并沿管道壁周向的方向为环向；与Z轴重合的方向为径向。
如图2所示，本公开实施例提供了一种基于正交孪生的油气管道裂纹量化方法及装置、存储介质，包括如下步骤：
201、建立(训练)基于正交孪生的油气管道裂纹量化模型；
202、使用(测试)基于正交孪生的油气管道裂纹量化模型。
本公开实施例的裂纹量化方法以油气管道为例进行说明，在其它铁磁性金属材料中的实施步骤可参考此方法进行，本公开不再赘述。
在一些示例性实施方式中，参考图3A，已知裂纹尺度及其响应信号，步骤201可以包括如下步骤：
1)依据不同检测原理获得单向直流励磁条件下尺度为(L,W,D,θ)的裂纹响应微弱矩阵信号。依据不同检测原理采用3轴漏磁和2轴动磁或涡流传感器超高分辨集成探头获得单向直流励磁条件下尺度为(L,W,D,θ)的裂纹响应微弱矩阵信号，如获得在A方向直流励磁场HA条件下的三轴漏磁测量信号模值以及 在A方向直流励磁场HA条件下的动磁或涡流信号其中(L,W,D,θ)分别为裂纹实际测量的长、宽、深、倾斜角；分别表示三轴漏磁测量信号的X轴分量，Y轴分量以及Z轴分量；分别表示3轴漏磁和2轴动磁或涡流传感器超高分辨集成探头电路板前后线圈差分(以下简称“前后差分”)的动磁或涡流信号以及左右线圈差分(以下简称“左右差分”)的动磁或涡流信号。以上MA和DA都是矩阵信号。至少包括裂纹存在性的指示信号，如和信号。裂纹存在性指示信号包括裂纹的边界信息。通过信号可以区分裂纹是内壁裂纹(ID)还是外壁裂纹(OD)。
2)在相同的A方向直流励磁场HA条件下，应用内壁裂纹磁偶极子模型或者外壁裂纹磁偶极子模型，生成尺度(Ls,Ws,Ds,θs)虚拟裂纹的仿真三轴漏磁信号模值 其中，SA为矩阵信号，(Ls,Ws,Ds,θs)分别为虚拟裂纹的长、宽、深、倾斜角，属于中间变量，其初始值可为对应已知的裂纹实测尺度(L,W,D,θ)；分别是仿真三轴漏磁信号的X轴分量，Y轴分量以及Z轴分量。如果是内壁裂纹，基于“内壁裂纹磁偶极子模型”，由尺度(Ls,Ws,Ds,θs)计算的公式为(1)式至(2)式；如果是外壁裂纹，基于“外壁裂纹磁偶极子模型”，由尺度(Ls,Ws,Ds,θs)计算的公式为(3)式至(5)式。其中，其中，fxX′(·)，fyX′(·)，fzX′(·)分别为在HA的X′轴分量激励下的磁偶极子3轴(X′轴、Y′轴、Z′轴)积分算子；fxY′(·)，fyY′(·)，fzY′(·)分别为在HA的Y′轴分量激励下的磁偶极子3轴(X′轴、Y′轴、Z′轴)积分算子；μ0＝4π×10-7(H/m)为真空磁导率，μr为饱和条件下的管壁相对磁导率；σx′为磁偶极子X′轴方向磁荷面密度，σy′为磁偶极子Y′轴方向磁荷面密度；如果是内壁裂纹，h为传感器集成探头的提离值，如果是外壁裂纹，h为裂纹处的管壁剩余壁厚与传感器集成探头的提离值之和。




3)为使裂纹仿真产生的效果与裂纹实际测量的效果一致，建立目标函数和约束条件：
min||αSA-MA||F          (6)
其中，‖·‖F为矩阵的弗罗贝尼乌斯(Frobenius)范数，Os(Ls,Ws,Ds,θs)＝1为拟合裂纹轮廓的椭球方程；调节因子：α＝max(MA)/max(SA_initial)，等于MA矩阵的最大元素除以初始矩阵的最大元素，而初始矩阵(SA_initial)即为将裂纹实测尺度(L,W,D,θ)代入磁偶极子模型计算所得的仿真三轴漏磁信号模值。通过调节虚拟裂纹尺寸(Ls,Ws,Ds)，使上述目标函数(6)得以满足。使(6)得到满足的最优虚拟裂纹尺度记为对应的仿真三轴漏磁信号模值记为本文中，裂纹尺度包括裂纹尺寸和倾斜角。
复杂的裂纹形状可以分解为简单条形轮廓的组合，而条形轮廓可以抽象为长方体、椭球体、椭圆柱等简单的几何形状。
本公开实施例的裂纹量化方法中，将裂纹轮廓近似为椭球体，在另一些示例性实施方式中，裂纹轮廓还可以近似为长方体、椭圆柱等形状，本公开实施例对此不作限制。
4)应用内壁裂纹正交磁偶极子模型或者外壁裂纹正交磁偶极子模型计算正交孪生三轴漏磁信号模值。在与HA同平面、大小相等、方向垂直的孪生磁场H⊥激励的条件下(|H⊥|＝|HA|)，针对尺度为的虚拟裂纹，应用磁偶极子模型，生成对应的漏磁响应信号模值，即仿真的正交孪生三轴漏磁信号模值其中分别是仿真的正交孪生三轴漏磁信号的X轴分量，Y轴分量以及Z轴分量。如果是内壁裂纹，基于“内壁裂纹正交磁偶极子模型”，由尺度计算 的公式如(8)式至(9)式所示；如果是外壁裂纹，基于“外壁裂纹正交磁偶极子模型”，由尺度计算的公式如(10)式至(12)式所示。其中，σxOT′为磁偶极子X′轴方向正交磁荷面密度，σyOT′为磁偶极子Y′轴方向正交磁荷面密度。




5)针对内壁裂纹和外壁裂纹，分别获得N个不同尺寸、不同倾斜角的正交孪生映射对以及实测尺寸集合。上述步骤1)至4)中的MA，αSAo，αS⊥，和实测尺度(L,W,D,θ)构成一一对应关系，定义第i组正交孪生映射对表示为与之对应的实际测量尺度为(Li,Wi,Di,θi)；则改变(L,W,D,θ)数值，即更换新的实测裂纹，重复执行步骤1)至4)，直到针对内壁裂纹和外壁裂纹，分别获得N个不同尺寸、不同倾斜角的正交孪生映射对，即与之对应的实测尺度集合表示为{(Li,Wi,Di,θi),i＝1,2,…,N}。
6)针对内壁裂纹和外壁裂纹，分别按照比例k:(1-k)(0<k<1)将N组正交孪 生映射对和实测尺度集合随机拆分为训练集和测试集，它们的数量分别为k×N个和(1-k)×N个。另外，针对内壁裂纹和外壁裂纹，分别从k×N个训练集中随机抽取一部分作为模型训练时的验证集。
7)针对内壁裂纹和外壁裂纹，对于训练集，以为输入，以为输出，i＝1,2,…,k×N，搭建并训练机器学习模型MLOT，将其命名为“裂纹信号正交孪生模型”，包含“内壁裂纹信号正交孪生模型”以及“外壁裂纹信号正交孪生模型”。其中，表示对和进行逻辑运算，例如K≥1；显然，即经过正交孪生变换的输出信号得到增强。
8)针对内壁裂纹和外壁裂纹，对于训练集，在第7)步的基础上，当i＝1,2,…,k×N，再将输入到训练好的MLOT模型中，得到k×N组估计结果
9)针对内壁裂纹和外壁裂纹，对于训练集，以为阈值(c在0到1之间，示例性的，c＝0.5)，对进行0-1二值化处理，得到其中i＝1,2,…,k×N。
10)针对内壁裂纹和外壁裂纹，对于训练集，通过和以及计算k×N组特征向量。的边缘轮廓通常为椭圆，记边缘轮廓的长轴为ai，短轴为bi，长轴的倾斜角为的峰值为pi，的峰-峰值为的峰-峰值间距为的峰值为的峰值间距为当i＝1,2,…,k×N时，共组成k×N组特征向量
11)针对内壁裂纹和外壁裂纹，对于训练集，以为输入，以(Li,Wi,Di,θi)为输出，i＝1,2,…,k×N，搭建并训练机器学习模型MLsize，将其命名为“裂纹尺度估计模型”，包含“内壁裂纹尺度估计模型”以及“外壁裂纹尺度估计模型”。
到此，训练好的MLOT模型和MLsize模型共同组成了“裂纹量化模型”。
在一些示例性实施方式中，在第2阶段，“裂纹量化模型”的使用或者测试主要适用于没有经过第1阶段模型训练的测试集或者真实油气管道上的新裂纹，参考图3B，步骤202可以包括如下步骤：
1)依据不同检测原理获得单向直流励磁条件下裂纹响应微弱矩阵信号。如获得在A方向直流励磁场HA条件下的三轴漏磁测量信号模值 以及在A方向直流励磁场HA条件下的动磁或涡流信号至少包括裂纹存在性的指示信号，如和信号。裂纹存在性指示信号包括裂纹的边界信息。通过信号可以区分裂纹是内壁裂纹(ID)还是外壁裂纹(OD)。
2)针对内壁裂纹或外壁裂纹，将MA信号输入到第1阶段训练好的MLOT(“内壁裂纹信号正交孪生模型”或“外壁裂纹信号正交孪生模型”)中，输出漏磁增强估计信号
3)以为阈值，对进行0-1二值化处理，得到c在0到1之间，示例性的，c＝0.5。
4)通过和以及计算特征向量记边缘轮廓的长轴为a，短轴为b，长轴的倾斜角为的峰值为p，的峰-峰值为的峰-峰值间距为的峰值为的峰值间距为由此组特征向量
5)将特征向量输入到第1阶段训练好的MLsize(“内壁裂纹尺度估计模型”或“外壁裂纹尺度估计模型”)中，输出内壁或外壁裂纹尺度的估计值
6)重复执行步骤1)至5)直到将第1阶段的测试集或者真实油气管道上的新裂纹全部量化完毕为止。
为了更好地理解本公开提供的基于正交孪生的油气管道裂纹量化方法及装置、存储介质，下面结合(X80钢材，15.30mm壁厚)油气管道裂纹量化这一示例性实施例对本公开的技术方案做进一步说明。
本实施例使用基于3轴漏磁和2轴动磁或涡流传感器超高分辨集成探头采集裂纹弱信号。3轴漏磁是指包含漏磁场的轴向分量、径向分量以及环向分量；2轴动磁或涡流是指探头电路板可以同时采集前后差分的动磁或涡流信号以及左右差分的动磁或涡流信号。前后差分的动磁或涡流信号可以对360°全向裂纹进行检测和内外识别，但是随着裂纹倾斜角模值的减小，前后差分的动磁或涡流信号幅值逐渐降低；而左右差分的动磁或涡流信号幅值却随着裂纹倾斜角模值的减小而增大；两者形成互补、弥补对方的不足。因此利用3轴漏磁和2轴动磁或涡流传感器超高分辨集成探头可以确定裂纹是否存在，判别内裂纹与外裂纹，以及确定裂纹的可疑区域。
整个实施步骤共分为2个阶段，第1阶段为“裂纹量化模型”的建立阶段(模型训练阶段)；第2阶段为“裂纹量化模型”的使用阶段(模型测试阶段)。
在第1阶段，已知裂纹尺度及其响应信号，“裂纹量化模型”的建立(模型训练)步骤包括：
按照步骤1)，利用轴向励磁方式的3轴漏磁和2轴动磁或涡流检测器开展对(X80钢材，15.30mm壁厚)人工裂纹油气管道的牵拉实验，由此获得人工裂纹的牵拉数据。采用轴向励磁意味着此时励磁方向(X轴方向)与轴向重合，建立的坐标系正如附图1所示。首先在牵拉管道上加工2160个不同尺寸和不同倾斜角的人工裂纹作为模型训练的样本，其中包含1080个内壁裂纹(ID)以及1080个外壁裂纹(OD)；针对内壁裂纹和外壁裂纹，分别随机分配其中75％的样本(810个)为训练集(用train标识)，另外25％的样本(270个)为测试集(用test标识)，而训练集中随机抽取25％的样本(203个)被定义为验证集(用val标识)，它们的尺度分布区间如表1所示。将上述牵拉数据导入“裂纹原始样本标定软件”，在加工裂纹的已知区域，框选并导出裂纹原始样本数据文件，将其以txt格式保存，每个数据文件以裂纹实测的“长(L)-宽(W)-深(D)-倾斜角(θ)-内或外”命名，由此获得轴向励磁条件下不同尺度(Li,Wi,Di,θi)的裂纹响应弱信号以及i＝1,2,…,2160。其中，每个裂纹原始样本数据文件由5种矩阵数据组成，分别为漏磁轴向数据矩阵漏磁径向数据矩阵漏磁环向数据矩阵动磁或涡流前后差分数据矩阵和动磁或涡流左右差分数据矩阵而且， 针对内壁裂纹或外壁裂纹，训练集中810个样本为第1阶段训练“裂纹量化模型”做好准备，测试集中270个样本为第2阶段测试“裂纹量化模型”做好准备。
表1模型训练的样本分布
按照步骤2)，提前测得牵拉实验的轴向励磁场HA的大小，然后在相同的轴向励磁场条件下，针对内壁裂纹和外壁裂纹，分别对1080每一个不同尺寸和不同倾斜角的人工裂纹样本，以它们文件名记录的实测尺度为初值，应用磁偶极子模型，生成虚拟裂纹的仿真三轴漏磁信号模值i＝1,2,…,1080；如果是内壁裂纹，基于“内壁裂纹磁偶极子模型”，由尺度计算的公式为(13)式至(14)式；如果是外壁裂纹，基于“外壁裂纹磁偶极子模型”，由尺度计算的公式为(15)式至(17)式。其中，fxX′(·)，fyX′(·)，fzX′(·)分别为在HA的X′轴分量激励下的磁偶极子3轴(X′轴、Y′轴、Z′轴)积分算子；fxY′(·)，fyY′(·)，fzY′(·)分别为在HA的Y′轴分量激励下的磁偶极子3轴(X′轴、Y′轴、Z′轴)积分算子；μ0＝4π×10-7(H/m)，为真空磁导率，μr为饱和条件下的管壁相对磁导率为与第i个人工裂纹样本对应的磁偶极子X′轴方向磁荷面密度，为与第i个人工裂纹样本对应的磁偶极子Y′轴方向磁荷面密度；如果是内壁裂纹，hi为第i个人工裂纹样本处传感器集成探头的提离值，如果是外壁裂纹，hi为第i个人工裂纹样本处的管壁剩余壁厚与传感器集成探头的提离值之和。




按照步骤3)，为使裂纹仿真产生的效果与裂纹实际测量的效果一致，针对内壁裂纹和外壁裂纹，分别对1080个不同尺寸和不同倾斜角的人工裂纹样本，逐一建立目标函数和约束条件：

其中，当i＝1,2,…,1080时，为与第i个人工裂纹样本对应的拟合裂纹轮廓的椭球方程；调节因子：表示矩阵的最大元素除以第i个初始矩阵的最大元素，而第i个初始矩阵即为将裂纹实测尺度(Li,Wi,Di,θi)代入磁偶极子模型计算所得的仿真三轴漏磁信号模值。以实测尺度(Li,Wi,Di,θi)为初值通过调节虚拟裂纹尺寸使上述目标函数(18)得以满足。使(18)得到满足的最优虚拟裂纹尺度记为对应的仿真三轴漏磁信号记为
按照步骤4)，应用内壁裂纹正交磁偶极子模型和外壁裂纹正交磁偶极子模型分别计算2160个不同尺寸和不同倾斜角的人工裂纹样本的正交孪生三轴漏磁信号模值。在与HA同平面、大小相等、方向垂直的孪生磁场H⊥激励的条件下(|H⊥|＝|HA|)，针对内壁裂纹或外壁裂纹，当i＝1,2,…,1080时，针对尺度为的虚拟裂纹，应用磁偶极子模型，生成对应的漏磁响应信号模值，即仿真的正交孪生三轴漏磁信号模值 如果是内壁裂纹，基于“内壁裂纹正交磁偶极子模型”，由尺度计算的公式如(20)式至(21)式所示；如果是外壁裂纹，基于“外壁裂纹正交磁偶极子模型”，由尺度计算的公式如(22)式至(24)式所示。其中，为与第i个人工裂纹样本对应的磁偶极子X′轴方向正交磁荷面密度，为与第i个人工裂纹样本对应的磁偶极子Y′轴轴方向正交磁荷面密度。




按照步骤5)，针对内壁裂纹和外壁裂纹，分别获得1080个不同尺寸、不同倾斜角的正交孪生映射对以及实测尺寸集合。在上述步骤1)至4)中，针对内壁裂纹或外壁裂纹，为1080个不同尺寸、不同倾斜角的正交孪生映射对，与之对应的实测尺度集合表示为{(Li,Wi,Di,θi),i＝1,2,…,1080}。
按照步骤6)，针对内壁裂纹和外壁裂纹，本实施例拆分训练集和测试集的比例系数被设定的k＝0.75，即训练集样本数占样本总数的75％；而训练集中验证集的占比为25％。
按照步骤7)，针对内壁裂纹和外壁裂纹，对于训练集样本(810个)，以为输入，以为输出，i＝1,2,…,810，搭建并训练机器学习模型MLOT，即“裂纹信号正交孪生模型”，包含“内壁裂纹信号正交孪生模型”以及“外壁裂纹信号正交孪生模型”。本实施例具体采用卷积神经网络自编码器(CNN Autoencoder)算法搭建了裂纹信号正交孪生模型。模型要求当i＝1,2,…,810时，输入为200×200的矩阵， 输出同样也是200×200的矩阵，两点之间的物理步进为0.5mm。如果物理步进超过0.5mm，可以进行三次样条插值。如果矩阵维度不满足要求，则以裂纹中心为矩阵中心，对超出200维度的边缘部分进行删减，对少于200维度的边缘部分进行补零。
本实施例设计的卷积神经网络自编码器共有37层结构，其中编码部分包含18层结构，解码部分包含19层结构。使用Python语言搭建的网络架构如下：

上述Python语言代码中，Conv2D表示二维卷积层，SpatialDropout2D表示二维空间随机置零层，BatchNormalization表示数据批量标准化层，ReLU表示激活函数层，MaxPooling2D表示二维池化层，UpSampling2D表示二维上采样层，Activation表示激活层。每一行Python语言表示一层结构，每层结构的输出作为下一层结构的输入，下一层结构的输出作为下下一层结构的输入。
按照步骤8)，针对内壁裂纹和外壁裂纹，对于训练集样本(810个)，在第7)步的基础上，当i＝1,2,…,810，再将输入到训练好的MLOT模型中，得到810组估计结果
按照步骤9)，针对内壁裂纹和外壁裂纹，对于训练集样本(810个)，当i＝1,2,…,810，以为阈值，对进行0-1二值化处理，得到
按照步骤10)，针对内壁裂纹和外壁裂纹，对于训练集样本(810个)，当i＝1,2,…,810，通过和以及计算810组特征向量。的边缘轮廓通常为椭圆，记边缘轮廓的长轴为ai，短轴为bi，长轴的倾斜角为的峰值为pi，的峰-峰值为的峰-峰值间距为的峰值为的峰值间距为当i＝1,2,…,810时，共组成810组特征向量
按照步骤11)，针对内壁裂纹和外壁裂纹，对于训练集样本(810个)，当i＝1,2,…,810，以为输入，以(Li,Wi,Di,θi)为输出，搭建并训练机器学习模型MLsize，即“裂纹尺度估计模型”，包含“内壁裂纹尺度估计模型”以及“外壁裂纹尺度估计模型”。本实施例具体采用全连接神经网络(DNN)算法搭建了裂纹尺度估计模型。模型要求当i＝1,2,…,810时，输入为8×1的向量，输出(Li,Wi,Di,θi)是4×1的向量。
本实施例设计的全连接神经网络共有5层结构，分别是1个输入层、3个隐藏层、1个输出层。使用Python语言搭建的网络架构如下：
上述Python语言代码中，Dense用于实现全连接层，中间层有10个神经元，输出层有4个神经元，kernel_initializer为指定初始化方法的关键字，activation为指定激活函数的关键字。
以“内壁裂纹信号正交孪生模型”为例，设置模型训练的迭代步数为500，模型训练的迭代过程如图4A、4B、5A、5B所示。图4A是“裂纹信号正交孪生模型”(MLOT)的损失函数(loss function)迭代过程，图4B是“裂纹信号正交孪生模型”(MLOT)的确定系数(coefficient of determination,即R2)迭代过程，迭代结束“裂纹信号正交孪生模型”(MLOT)的评价指标如表2所示。可见MLOT模型经过收敛，训练集和验证集的损失函数(loss)最终都减小到0.0002，说明MLOT模型最终的均方误差(MSE)非常小，且没有出现过拟合。另一方面，MLOT模型训练集的确定系数最终达到0.9371，验证集的确定系数最终达到0.9281，说明MLOT模型最终的拟合效果较好，基本满足正交孪生变换的要求。
表2迭代结束“裂纹信号正交孪生模型”(MLOT)的评价指标
以“内壁裂纹信号正交孪生模型”为例，图5A是“裂纹尺度估计模型”(MLsize)的损失函数(loss function)迭代过程，图5B是“裂纹尺度估计模型”(MLsize)的确定系数(coefficient of determination,即R2)迭代过程，迭代结束“裂纹尺度估计模型”(MLsize)的评价指标如表3所示。可见MLsize模型经过收敛，训练集的损失函数(loss)最终减小到0.0044，验证集的损失函数(loss)最终减小到0.0027，说明MLsize模型最终的均方误差(MSE)也非常小，且没有出现过拟合。另一方面，MLsize模型训练集的确定系数最终达到0.9907，验证集的确定系数最终达到0.9940，说明MLsize模型最终的拟合效果非常好，对裂纹长、宽、深、倾斜角的估计准确性非常高。
表3迭代结束“裂纹尺度估计模型”(MLsize)的评价指标
在第2阶段，针对第1阶段未被训练的270个测试集样本开展模型测试。“裂纹量化 模型”的使用(模型测试)步骤为：
按照步骤1)，依据不同检测原理获得单向直流励磁条件下裂纹响应微弱矩阵信号。在执行完第1阶段的步骤1)之后，已经在轴向励磁条件下，针对内壁裂纹或外壁裂纹，获得270个测试集样本的裂纹响应弱信号以及i＝1,2,…,270。其中，每个裂纹原始样本数据文件由5种矩阵数据组成，分别为漏磁轴向数据矩阵漏磁径向数据矩阵漏磁环向数据矩阵动磁或涡流前后差分数据矩阵和动磁或涡流左右差分数据矩阵至少包括裂纹存在性的指示信号，如和信号。裂纹存在性指示信号包括裂纹的边界信息。通过信号可以区分裂纹是内壁裂纹(ID)还是外壁裂纹(OD)。
按照步骤2)，针对内壁裂纹或外壁裂纹，对于测试集样本(270个)，当i＝1,2,…,270，将信号输入到MLOT(“内壁裂纹信号正交孪生模型”或“外壁裂纹信号正交孪生模型”)中，输出漏磁增强估计信号
按照步骤3)，针对内壁裂纹或外壁裂纹，对于测试集样本(270个)，当i＝1,2,…,270，以为阈值，对进行0-1二值化处理，得到
按照步骤4)，针对内壁裂纹或外壁裂纹，对于测试集样本(270个)，当i＝1,2,…,270，通过和以及计算270组特征向量。的边缘轮廓通常为椭圆，记边缘轮廓的长轴为ai，短轴为bi，长轴的倾斜角为的峰值为pi，的峰-峰值为的峰-峰值间距为的峰值为的峰值间距为当i＝1,2,…,270时，共组成270组特征向量
按照步骤5)，针对内壁裂纹或外壁裂纹，对于测试集样本(270个)，当i＝1,2,…,270，将特征向量输入到MLsize(“内壁裂纹尺度估计模型”或“外壁裂纹尺度估计模型”)中，输出内壁或外壁裂纹尺度的估计值
按照步骤6)，针对内壁裂纹或外壁裂纹，确保第1阶段测试集的270个裂纹全部量化完毕为止。
图6A、图6B、图7A、图7B、图8A、图8B分别是测试集样本中3组不同尺度内壁裂纹信号的正交孪生效果图，从左往右三幅图分别是原始输入的漏磁测量信号模值MA，图片标题为“输入信号”，经过MLOT输出的漏磁增强估计信号图片标题为 “预测输出信号”；以及MLOT输出的理想信号f(αSAo,αS⊥)，图片标题为“真实输出信号”。
图6A和图6B展示的是测试集中一个L＝30mm，W＝0.3mm，D＝1.4mm，θ＝0°的内壁裂纹，经过MLOT(“内壁裂纹信号正交孪生模型”)后信号增强效果图。可见，原始输入信号MA的漏磁场峰值仅有10Gs，经过正交孪生变换之后，MLOT输出信号的漏磁场峰值达到120Gs，信号被增强12倍；此外，由于该裂纹样本的倾斜角θ＝0°，属于轴向裂纹，图6B表明，原始输入信号MA呈现双峰特征，甚至无法分辨它是一个轴向裂纹，然而经过正交孪生变换之后，MLOT输出信号清楚反映了它的轴向裂纹特征，因而信号对该轴向裂纹的尺度量化提供了有利条件。
图7A和图7B展示的是测试集中一个L＝40mm，W＝0.3mm，D＝2.5mm，θ＝3°的内壁裂纹，经过MLOT(“内壁裂纹信号正交孪生模型”)后信号增强效果图。可见，原始输入信号MA的漏磁场峰值仅有12Gs，经过正交孪生变换之后，MLOT输出信号的漏磁场峰值达到150Gs，信号被增强12.5倍；同样，由于该裂纹样本的倾斜角θ＝3°，属于小倾斜角裂纹，图7B表明，该裂纹经过正交孪生变换之后，MLOT输出信号增强了裂纹信号特征，为对其进行尺度的高精度量化提供了有利条件。
图8A和图8B展示的是测试集中一个L＝60mm，W＝0.4mm，D＝2.9mm，θ＝18°的内壁裂纹，经过MLOT(“内壁裂纹信号正交孪生模型”)后信号增强效果图。可见，原始输入信号MA的漏磁场峰值仅有60Gs，经过正交孪生变换之后，MLOT输出信号的漏磁场峰值达到200Gs，信号被增强3.3倍，因而MLOT输出信号也有利于该裂纹的高精度量化。
通过对测试集270个内壁裂纹的估计误差进行区间估计，在90％的置信度下，计算结果如表4所示。可见，内壁裂纹的长度估计误差的置信区间为(-1.046mm,1.026mm)，宽度估计误差的置信区间为(-0.006mm,0.005mm)，深度估计误差的置信区间为(-0.214mm,0.193mm)，倾斜角估计误差的置信区间为(-2.282°,2.452°)。综上所述，本公开的基于正交孪生的油气管道裂纹量化方法及装置、存储介质具有非常高的尺度量化精度。
表4裂纹量化模型估计误差的置信区间
本公开实施例还提供了基于正交孪生的油气管道裂纹量化装置，包括：1)3轴漏磁 和2轴动磁/涡流传感器超高分辨集成探头，可实现采集前后差分动磁/涡流信号以及左右差分动磁/涡流信号；2)存储指令、算法、模型的存储器；3)执行基于正交孪生的油气管道裂纹量化方法的处理器；4)显示量化结果的显示器；5)连接各个单元的总线系统。
在一个示例中，如图9所示，基于正交孪生的油气管道裂纹量化装置可包括：3轴漏磁和2轴动磁/涡流传感器超高分辨集成探头910、存储器920、处理器930、显示器940和总线系统950，其中，3轴漏磁和2轴动磁/涡流传感器超高分辨集成探头910、存储器920、处理器930、显示器940通过该总线系统950相连；3轴漏磁和2轴动磁/涡流传感器超高分辨集成探头910用于检测并获得单向直流励磁条件下尺度为(L,W,D,θ)的裂纹响应微弱信号，包括3轴漏磁信号和前后差分动磁/涡流信号和左右差分动磁/涡流信号；存储器920用于存储指令及“裂纹信号正交孪生模型”和“裂纹尺度估计模型”等，处理器930用于执行存储器920存储的指令，以通过基于正交孪生的油气管道裂纹量化方法进行裂纹量化。具体地，处理器930可以分别训练“裂纹信号正交孪生模型”和“裂纹尺度估计模型”；可对不同倾斜角的裂纹弱信号进行增强并进行尺度估计；最后通过显示器940将量化结果显示出来。
应理解，存储器920可以包括只读存储器和随机存取存储器，并向处理器930提供指令和数据，包括所述的“裂纹信号正交孪生模型”和“裂纹尺度估计模型”等。存储器920的一部分还可以包括非易失性随机存取存储器。例如，存储器920还可以存储设备类型的信息。
处理器930可以是中央处理单元(Central Processing Unit，CPU)，处理器930还可以是其他通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现场可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器930也可以是任何常规的处理器等。
显示器940除了显示裂纹的量化结果之外，还可以通过裂纹分析软件显示存储器920中的裂纹检测数据。
总线系统950除包括数据总线之外，还可以包括电源总线、控制总线和状态信号总线等。
在实现过程中，基于正交孪生的油气管道裂纹量化装置所执行的处理可以通过处理器930中的硬件的集成逻辑电路或者软件形式的指令完成。即本公开实施例的基于正交孪生的油气管道裂纹量化装置步骤可以由硬件处理器执行完成，或者用处理器930中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等存储介质中。该存储介质位于存储器920，处理器930读取存储器920中的信息，结合其硬件完成上述方法的步骤。为避免重复，这里不再详细描述。
本公开实施例还提供了一种存储介质，该存储介质存储有可执行指令，该可执行指令被处理器执行时可以实现本公开上述任一实施例提供的基于正交孪生的油气管道裂纹量化方法；此外，存储介质还可以存储裂纹量化模型，包括“裂纹信号正交孪生模型”和“裂纹尺度估计模型”；对所述的小倾斜角裂纹，分别使用训练好的“裂纹信号正交孪生模型”和“裂纹尺度估计模型”进行量化。
本领域普通技术人员可以理解，上文中所公开方法中的全部或某些步骤、系统、装置中的功能模块/单元可以被实施为软件、固件、硬件及其适当的组合。在硬件实施方式中，在以上描述中提及的功能模块/单元之间的划分不一定对应于物理组件的划分；例如，一个物理组件可以具有多个功能，或者一个功能或步骤可以由若干物理组件合作执行。某些组件或所有组件可以被实施为由处理器，如数字信号处理器或微处理器执行的软件，或者被实施为硬件，或者被实施为集成电路，如专用集成电路。这样的软件可以分布在计算机可读介质上，计算机可读介质可以包括计算机存储介质(或非暂时性介质)和通信介质(或暂时性介质)。如本领域普通技术人员公知的，术语计算机存储介质包括在用于存储信息(诸如计算机可读指令、数据结构、程序模块或其他数据)的任何方法或技术中实施的易失性和非易失性、可移除和不可移除介质。计算机存储介质包括但不限于RAM、ROM、EEPROM、闪存或其他存储器技术、CD-ROM、数字多功能盘(DVD)或其他光盘存储、磁盒、磁带、磁盘存储或其他磁存储装置、或者可以用于存储期望的信息并且可以被计算机访问的任何其他的介质。此外，本领域普通技术人员公知的是，通信介质通常包含计算机可读指令、数据结构、程序模块或者诸如载波或其他传输机制之类的调制数据信号中的其他数据，并且可包括任何信息递送介质。
虽然本公开所揭露的实施方式如上，但所述的内容仅为便于理解本公开而采用的实施方式，并非用以限定本公开。任何本公开所属领域内的技术人员，在不脱离本公开所揭露的精神和范围的前提下，可以在实施的形式及细节上进行任何的修改与变化，但本公开的保护范围，仍须以所附的权利要求书所界定的范围为准。

Claims (10)

  1. 一种基于正交孪生的油气管道裂纹量化方法，包括：
    获取在单向直流励磁条件下内壁裂纹或者外壁裂纹的三轴漏磁测量信号模值，获取在与直流励磁正交的动态磁场激励条件下的动磁信号或涡流信号，所述单向直流励磁指具有单一方向的直流励磁方式；
    将所述三轴漏磁测量信号模值输入裂纹信号正交孪生模型，得到漏磁增强估计信号，所述漏磁增强估计信号为仿真的正交孪生三轴漏磁信号模值的函数，所述正交孪生三轴漏磁信号模值为在虚拟正交孪生直流励磁条件下的三轴漏磁响应信号模值，所述虚拟正交孪生直流励磁与所述单向直流励磁同平面、大小相等且方向垂直；
    从所述漏磁增强估计信号以及所述动磁信号或涡流信号中提取特征向量，将所述特征向量输入裂纹尺度估计模型，得到裂纹的尺寸和倾斜角；
    所述正交孪生被定义为，在已知单向直流励磁的原始响应信号的条件下，得到与单向直流励磁同平面、大小相等、方向垂直的孪生磁场对应的响应信号，得到的响应信号被定义为孪生响应信号，所述孪生响应信号与所述原始响应信号构成孪生关系，将所述原始响应信号变换为所述孪生响应信号的过程定义为正交孪生变换。
  2. 根据权利要求1所述的方法，其中，从所述漏磁增强估计信号以及所述动磁信号或涡流信号中提取特征向量，包括：
    根据所述漏磁增强估计信号得到至少一个第一特征值；
    根据所述动磁信号或涡流信号得到至少一个第二特征值；
    将所述第一特征值和第二特征值组成特征向量。
  3. 根据权利要求2所述的方法，其中，所述第一特征值包括：长轴、短轴、长轴倾斜角以及峰值，所述根据所述漏磁增强估计信号得到第一特征值，包括：
    根据预设的二值化阈值，对所述漏磁增强估计信号进行二值化处理，得到二值化漏磁信号；
    计算所述二值化漏磁信号的边缘轮廓的长轴、短轴、长轴倾斜角以及所述漏磁增强估计信号的峰值，将得到的长轴、短轴、长轴倾斜角以及峰值作为所述第一特征值；
    所述动磁信号或涡流信号包括前后线圈差分信号和左右线圈差分信号，所述第二特征 值包括：所述前后线圈差分信号的峰值和峰-峰值间距，以及所述左右线圈差分信号的峰值和峰值间距。
  4. 根据权利要求2所述的方法，其中，所述裂纹信号正交孪生模型为通过卷积神经网络实现的自编码器模型，所述自编码器模型包括编码器部分和解码器部分；所述裂纹尺度估计模型为全连接神经网络。
  5. 根据权利要求1所述的方法，其中，所述漏磁增强估计信号为仿真最优三轴漏磁信号模值SAo与仿真的正交孪生三轴漏磁信号模值S⊥的函数f(αSAo,αS⊥)，所述仿真最优三轴漏磁信号模值SAo是使||αSA-MA||F最小的仿真三轴漏磁信号模值，其中MA为所述三轴漏磁测量信号模值，SA为将裂纹尺度代入磁偶极子模型得到的仿真三轴漏磁信号模值，‖·‖F为矩阵的Frobenius范数，α为调节因子；
    分别是仿真三轴漏磁信号的X轴分量、Y轴分量以及Z轴分量，所述X轴方向与所述单向直流励磁的激励方向相同，所述Y轴方向在所述油气管道平面内与所述X轴方向垂直，所述Z轴方向分别与所述X轴方向和Y轴方向垂直；
    分别是正交孪生三轴漏磁信号的X轴分量、Y轴分量以及Z轴分量；
    调节因子α表示等于MA矩阵的最大元素除以初始矩阵的最大元素，所述初始矩阵为将裂纹实测尺度代入磁偶极子模型计算所得的仿真三轴漏磁信号模值组成的矩阵。
  6. 根据权利要求1所述的方法，其中，所述方法还包括：
    对于多个裂纹样本，使用在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的所述漏磁增强估计信号训练所述裂纹信号正交孪生模型，其中，每个所述裂纹样本对应的漏磁增强估计信号通过如下方法生成：
    以所述裂纹样本的实测尺度为初始的裂纹尺度；
    用MA矩阵的最大元素除以初始矩阵的最大元素得到调节因子α，MA为所述三轴漏磁测量信号模值；
    计算||αSA-MA||F，其中SA为将裂纹尺度代入磁偶极子模型得到的仿真三轴漏磁信 号模值，‖·‖F为矩阵的Frobenius范数；
    反复调节所述裂纹尺度，并将所述裂纹尺度输入裂纹磁偶极子模型，直到得到使||αSA-MA||F最小的最优裂纹尺度，与所述最优裂纹尺度对应的SA即为仿真最优三轴漏磁信号模值SAo；
    将所述最优裂纹尺度输入裂纹正交磁偶极子模型，得到所述仿真的正交孪生三轴漏磁信号模值S⊥；
    将所述SAo与S⊥代入公式K≥1，从而生成所述漏磁增强估计信号。
  7. 根据权利要求6所述的方法，其中，对于所述多个裂纹样本，使用在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的所述漏磁增强估计信号训练所述裂纹信号正交孪生模型，包括：
    将每个裂纹样本在单向直流励磁条件下的三轴漏磁测量信号模值以及对应的漏磁增强估计信号标记为一组正交孪生映射对；
    将N组正交孪生映射对按照比例k:(1-k)随机拆分为训练集和测试集，其中，0<k<1；
    使用k×N组所述正交孪生映射对训练所述裂纹信号正交孪生模型，使用(1-k)×N组所述正交孪生映射对测试所述裂纹信号正交孪生模型。
  8. 根据权利要求7所述的方法，其中，在训练所述裂纹信号正交孪生模型之后，使用所述特征向量训练所述裂纹尺度估计模型，包括：
    将每个裂纹样本的第一特征值和第二特征值与对应的裂纹尺寸和倾斜角的真实值标记为一组裂纹尺度估计映射对；
    使用k×N组所述正交孪生映射对对应的裂纹尺度估计映射对进行所述裂纹尺度估计模型的训练，使用(1-k)×N组所述正交孪生映射对对应的裂纹尺度估计映射对进行所述裂纹尺度估计模型的测试。
  9. 一种基于正交孪生的油气管道裂纹量化装置，包括：单向直流励磁条件下的漏磁传感器探头，单向直流励磁以及与直流励磁正交的动态磁场激励条件下的动磁或涡流传感器探头，存储指令、算法、模型的存储器，执行如权利要求1至8中任一项所述的油气管 道裂纹量化方法的处理器，以及连接各个单元的总线系统。
  10. 一种存储介质，其上存储有基于正交孪生的油气管道裂纹量化方法的程序，该程序被处理器执行时实现如权利要求1至8中任一项所述的油气管道裂纹量化方法。
PCT/CN2023/116749 2023-08-16 2023-09-04 基于正交孪生的油气管道裂纹量化方法及装置、存储介质 Pending WO2025035512A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202311034794.9 2023-08-16
CN202311034794.9A CN116754632B (zh) 2023-08-16 2023-08-16 基于正交孪生的油气管道裂纹量化方法及装置、存储介质

Publications (1)

Publication Number Publication Date
WO2025035512A1 true WO2025035512A1 (zh) 2025-02-20

Family

ID=87961236

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/116749 Pending WO2025035512A1 (zh) 2023-08-16 2023-09-04 基于正交孪生的油气管道裂纹量化方法及装置、存储介质

Country Status (2)

Country Link
CN (1) CN116754632B (zh)
WO (1) WO2025035512A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120627860A (zh) * 2025-08-15 2025-09-12 中国石油大学(华东) 一种油气管道全壁厚裂纹长度量化方法与裂纹检测探头
CN120847227A (zh) * 2025-09-19 2025-10-28 中国石油大学(华东) 海洋石油井控装备结构缺陷电磁调控检测方法及检测系统
CN121740295A (zh) * 2026-02-27 2026-03-27 国机传感科技有限公司 一种油气输送管道应力量化方法及系统

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN121165100A (zh) * 2025-08-01 2025-12-19 茂名新金明石油有限公司 一种原油泄漏检测与处理方法、系统、设备及介质

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020153883A1 (en) * 2001-04-23 2002-10-24 Korea Atomic Energy Research Institute Method of detecting through-wall crack and measuring through-wall length of crack in nuclear steam generator tube
US20030128887A1 (en) * 2002-01-10 2003-07-10 Nec Corporation Two-dimensional orthogonal transformation and quantization method and its device and program
CN101261246A (zh) * 2007-03-09 2008-09-10 清华大学 管道裂纹远场涡流检测方法
CN105092696A (zh) * 2015-08-10 2015-11-25 北京工业大学 一种用于铁磁性管道内壁裂纹检测的低频交流漏磁检测方法
CN106648541A (zh) * 2016-10-17 2017-05-10 山西大学 一种基于孪生光束产生真随机数的装置及方法
CN109100416A (zh) * 2018-09-21 2018-12-28 东北大学 基于正交多频电磁检测的铁磁性管道内壁缺陷检测装置
CN114636754A (zh) * 2022-03-09 2022-06-17 清华大学 一种基于漏磁空间积分的裂纹缺陷量化方法和装置
CN115308297A (zh) * 2022-08-05 2022-11-08 国家石油天然气管网集团有限公司 一种基于周向励磁装置的缺陷信号量化方法和系统
US20230003626A1 (en) * 2020-01-22 2023-01-05 Mitsubishi Electric Corporation Crack estimation device, crack estimation method, crack inspection method, and failure diagnosis method
CN115586245A (zh) * 2022-10-28 2023-01-10 南京航空航天大学 一种基于脉冲旋转电磁场铁磁性材料裂纹量化方法
US20230049804A1 (en) * 2020-01-31 2023-02-16 Mitsubishi Electric Corporation Crack estimation device, failure diagnosis device, crack estimation method, and failure diagnosis method for rotating machine
CN115758084A (zh) * 2022-11-21 2023-03-07 清华大学 一种深度神经网络裂纹量化方法及装置、存储介质
CN115809583A (zh) * 2022-12-22 2023-03-17 国网辽宁省电力有限公司电力科学研究院 一种基于pod的变压器振动场数字孪生模型降阶方法
CN115809595A (zh) * 2022-11-25 2023-03-17 沈阳建筑大学 一种反映滚动轴承缺陷扩展的数字孪生模型构建方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2591584C1 (ru) * 2015-05-19 2016-07-20 Открытое акционерное общество "Акционерная компания по транспорту нефти "Транснефть" (ОАО "АК "Транснефть") Способ оценки геометрических размеров дефектов стенки трубной секции и сварных швов по данным магнитного внутритрубного дефектоскопа с помощью универсальной нейросетевой модели, подходящей для дефектоскопов с различными диаметрами и магнитными системами
CN108037177B (zh) * 2017-10-21 2018-08-28 东北石油大学 漏磁法检测在役管道焊缝裂纹扩展的磁-结构耦合方法
CN112329590B (zh) * 2020-10-30 2024-05-10 中海石油(中国)有限公司 一种管道组件检测系统及检测方法
CN115047064A (zh) * 2022-06-07 2022-09-13 国家石油天然气管网集团有限公司 管道缺陷量化方法、处理器和管道缺陷量化装置

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020153883A1 (en) * 2001-04-23 2002-10-24 Korea Atomic Energy Research Institute Method of detecting through-wall crack and measuring through-wall length of crack in nuclear steam generator tube
US20030128887A1 (en) * 2002-01-10 2003-07-10 Nec Corporation Two-dimensional orthogonal transformation and quantization method and its device and program
CN101261246A (zh) * 2007-03-09 2008-09-10 清华大学 管道裂纹远场涡流检测方法
CN105092696A (zh) * 2015-08-10 2015-11-25 北京工业大学 一种用于铁磁性管道内壁裂纹检测的低频交流漏磁检测方法
CN106648541A (zh) * 2016-10-17 2017-05-10 山西大学 一种基于孪生光束产生真随机数的装置及方法
CN109100416A (zh) * 2018-09-21 2018-12-28 东北大学 基于正交多频电磁检测的铁磁性管道内壁缺陷检测装置
US20230003626A1 (en) * 2020-01-22 2023-01-05 Mitsubishi Electric Corporation Crack estimation device, crack estimation method, crack inspection method, and failure diagnosis method
US20230049804A1 (en) * 2020-01-31 2023-02-16 Mitsubishi Electric Corporation Crack estimation device, failure diagnosis device, crack estimation method, and failure diagnosis method for rotating machine
CN114636754A (zh) * 2022-03-09 2022-06-17 清华大学 一种基于漏磁空间积分的裂纹缺陷量化方法和装置
CN115308297A (zh) * 2022-08-05 2022-11-08 国家石油天然气管网集团有限公司 一种基于周向励磁装置的缺陷信号量化方法和系统
CN115586245A (zh) * 2022-10-28 2023-01-10 南京航空航天大学 一种基于脉冲旋转电磁场铁磁性材料裂纹量化方法
CN115758084A (zh) * 2022-11-21 2023-03-07 清华大学 一种深度神经网络裂纹量化方法及装置、存储介质
CN115809595A (zh) * 2022-11-25 2023-03-17 沈阳建筑大学 一种反映滚动轴承缺陷扩展的数字孪生模型构建方法
CN115809583A (zh) * 2022-12-22 2023-03-17 国网辽宁省电力有限公司电力科学研究院 一种基于pod的变压器振动场数字孪生模型降阶方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120627860A (zh) * 2025-08-15 2025-09-12 中国石油大学(华东) 一种油气管道全壁厚裂纹长度量化方法与裂纹检测探头
CN120847227A (zh) * 2025-09-19 2025-10-28 中国石油大学(华东) 海洋石油井控装备结构缺陷电磁调控检测方法及检测系统
CN121740295A (zh) * 2026-02-27 2026-03-27 国机传感科技有限公司 一种油气输送管道应力量化方法及系统

Also Published As

Publication number Publication date
CN116754632B (zh) 2023-11-21
CN116754632A (zh) 2023-09-15

Similar Documents

Publication Publication Date Title
Piao et al. Fast reconstruction of 3-D defect profile from MFL signals using key physics-based parameters and SVM
CN116754632B (zh) 基于正交孪生的油气管道裂纹量化方法及装置、存储介质
CN108198574B (zh) 变声检测方法及装置
WO2024108717A1 (zh) 一种深度神经网络裂纹量化方法及装置、存储介质
CN109325544B (zh) 一种基于中层特征的管道异常分类方法
Qiao et al. A crack identification method for concrete structures using improved U‐Net convolutional neural networks
CN109508709B (zh) 一种基于机器视觉的单指针仪表读数方法
Xu et al. Image recognition model of pipeline magnetic flux leakage detection based on deep learning
CN116182089A (zh) 长距离输送管道泄漏检测方法及其系统
He et al. Dominant Symmetry Plane Detection for Point‐Based 3D Models
WO2025035513A1 (zh) 一种油气管道裂纹信号的正交孪生方法及装置、存储介质
Li et al. A multidirectional magnetic flux leakage detection based crack 3-D profile reconstruction method
CN115047064A (zh) 管道缺陷量化方法、处理器和管道缺陷量化装置
CN118585919A (zh) 斜拉桥损伤识别方法、设备、存储介质及产品
CN108319954B (zh) 一种无接触测重方法
CN117471560A (zh) 一种基于CNN-LightGBM的地面瞬变电磁反演方法
CN115937116B (zh) 一种输电线路耐张线夹的检测方法
CN120876823A (zh) 一种供热管道缺陷检测方法、装置、设备、介质及产品
CN115471727A (zh) 一种基于迁移学习的复合材料缺陷检测方法
CN120543783A (zh) 地下管廊微裂缝检测与测量方法、装置、设备、介质及产品
CN110517300A (zh) 基于局部结构算子的弹性图像配准算法
CN107123085B (zh) 一种基于三角剖分的点云比例标准化方法
Chen et al. A global centralized magnetic flux leakage small defect detection network
CN118537486A (zh) 装配异常检测中的稀疏ct重建方法与系统
She et al. Simplified analytical model for pulsed eddy current testing of asymmetric cylinders and tubes with physics-guided multi-scale ResNet for property estimation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23948938

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE