Electromagnetic regulation and control detection method and detection system for structural defects of offshore oil well control equipment

Abstract

The invention belongs to the technical field of nondestructive detection defect evaluation, and particularly relates to an electromagnetic regulation and control detection method and system for structural defects of offshore oil well control equipment. According to the electromagnetic regulation and control detection method and detection system for the structural defects of the offshore oil well control equipment, crack multi-feature signals are obtained through amplitude-frequency phase regulation and control of electromagnetic excitation signals, angle information of cracks is obtained by means of a residual network, and accurate identification of crack length and depth information is achieved; and the service life prediction network realizes effective prediction of the residual fatigue life. The detection system comprises a monitoring probe, an excitation signal generation unit and a monitoring signal processing and collecting unit, the monitoring probe consists of an exciting coil group, an array sensor and a front-end signal processing circuit; the excitation signal generation unit comprises a signal generator and a power amplifier; the monitoring signal processing and collecting unit comprises a multiplexing and amplifying filter circuit, a band-pass filter circuit, a phase-locked amplifying circuit and a signal collecting board card.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing and defect assessment technology, and particularly relates to an electromagnetic control detection method and system for structural defects in offshore oil well control equipment. Background Technology

[0002] Offshore oil well control equipment is highly susceptible to fatigue damage due to the complex loads caused by harsh operating conditions such as wind, waves, and currents. The resulting fatigue damage and cracks often propagate irregularly, posing significant challenges to defect detection and assessment. For example, current non-destructive testing (NDT) techniques for detecting structural defects in well control equipment typically require extensive removal of surface deposits and thorough cleaning of anti-corrosion coatings. This results in a cumbersome and inefficient defect detection process, along with high costs for surface cleaning and coating repair.

[0003] Alternating Current Field Measurement (ACFM), an emerging non-destructive testing (NDT) technique, is widely used for detecting surface cracks in conductive materials. This technique induces a uniform alternating current on the surface of a conductive specimen using a probe. When a defect is present, the path of the alternating current is disturbed, resulting in a distortion of the spatial magnetic field. By measuring this distorted magnetic field, cracks can be identified and quantitatively assessed. Due to its advantages, such as non-contact detection and quantitative analysis, ACFM has been widely applied in defect detection in various marine engineering structures.

[0004] However, further research revealed that existing ACFM detection technologies primarily target transverse cracks (i.e., cracks perpendicular to the induced current direction). Specifically, crack depth is assessed using the extreme amplitude of the characteristic signal Bz, and crack length is estimated using the extreme value spacing. However, for longitudinal cracks (i.e., cracks parallel to the induced current direction) and inclined cracks, effective assessment is difficult due to a significant reduction in detection sensitivity. Furthermore, existing ACFM detection technologies suffer from drawbacks such as inconsistent current field intensity and large differences in defect response signal modulation, all of which severely limit the accuracy of structural defect assessment. Summary of the Invention

[0005] This invention provides an electromagnetic control detection method and system for structural defects in offshore oil well control equipment. It acquires multi-feature signals of cracks through amplitude-frequency phase modulation of electromagnetic excitation signals, obtains crack angle information using a residual network, and achieves accurate identification of crack length and depth information using amplitude-frequency phase modulation and a physical heuristic neural network. A service life prediction network enables effective prediction of remaining fatigue life, and a spatiotemporal prediction neural network enables effective assessment of crack propagation, providing accurate data support for the health assessment of underwater structures of offshore oil well control equipment.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The electromagnetic control detection method for structural defects in offshore oil well control equipment includes the following steps: Step S101: Adjust the excitation induced current to obtain the spatial magnetic field distortion of the offshore oil well control equipment under test; calculate the 8×8 magnetic field signal matrix Z. Wherein, the magnetic field signal Bz matrix Z satisfies: Among them, Bz0 ij Let Bz be the value of the magnetic field signal in the i-th row and j-th column of the Bz matrix; Step S102: Perform absolute value operation on the magnetic field signal Bz matrix Z and draw a grayscale image to obtain the Bz image; Step S103: Smooth the Bz image using grayscale interpolation; By overlaying crack images from different angles, complete information about the crack outline can be obtained. Step S104: Add real angle labels to all crack angle fusion images to form an angle evaluation database; The angle evaluation database is trained using a residual network; once trained, the angle evaluation database can be used to obtain angle prediction values ​​for unknown samples. Step S105: Adjust the amplitude and phase parameters to make the induced current perpendicular to the crack and obtain information about the crack endpoint; Step S106: When the induced current is perpendicular to the crack, adjust the frequency of the excitation signal to change the skin depth of the induced current, and obtain at least a first frequency image, a second frequency image, and a third frequency image; wherein the first frequency is less than the second frequency, and the second frequency is less than the third frequency. The first frequency image, the second frequency image, and the third frequency image are placed in the R, G, and B channels of the color image, respectively, and stacked to obtain a crack size assessment fused image; Step S107: Construct a cross-fusion network to extract the Bz peak value from the Bz image. and peak spacing A crack prediction system was established using the random forest algorithm. If the predicted value of the cross-fusion network is set to OUTn, the predicted value of the crack prediction system is set to OUTe, and the true value of the crack size is set to Label, then the constructed loss function satisfies: ; in, The weights involved in the training of the crack prediction system; Step S108: Initialize the cross-fusion network, embed the crack prediction system, and train it using training data; wherein, the trained crack prediction system can be used to obtain the size prediction value of cracks at any angle in the structural defects of offshore oil well control equipment.

[0007] A more preferred approach also includes the following steps: Step S201: Obtain the control data and use the trained cross-fusion network to extract the crack length information features of the offshore oil well control equipment. ; In addition, acquiring stress data from offshore oil well control equipment. ; Step S202: Obtain the remaining fatigue life by cyclically loading until the offshore oil well control equipment workpiece completely fractures. ; Step S203: Construct a multi-layer fully connected neural network topology; Crack length information characteristics of offshore oil well control equipment Stress data of offshore oil well control equipment Remaining fatigue life As three nodes in the input layer, a pyramid-shaped structure is used as the hidden layer, and the fatigue life prediction value is used as a single node in the output layer to realize multi-dimensional physical constraints on fatigue life prediction. Step S204: Train the lifetime prediction network; The regression evaluation index MSE is used to calculate the deviation between predicted and measured values. And construct features including crack length information of offshore oil well control equipment. Stress data of offshore oil well control equipment Remaining fatigue life Physical consistency loss of three types of characteristic residuals ; Among them, the deviation between the predicted value and the measured value ,satisfy: ; These are actual observed values. The value is the model prediction, and n is the number of samples. Designing the weighted loss function, we obtain: ;in, Determined through cross-validation; ,satisfy: ;in, , , Let be the weights, y be the model prediction, B be the Basquin model output, and P be the Paris model output. This represents a non-physical value penalty term, used to ensure that y is non-negative, thereby avoiding unreasonable results from the model output; Step S205: In the fatigue life prediction process, a fixed time step is set, the load is set to a fixed value, the actual crack length is obtained, and the crack propagation image is monitored simultaneously. High-resolution, multi-frequency fusion magnetic field images of vertical cracks induced by current were obtained using SRCNN, and a database was built. Step S206: Construct the PredRNN++ spatiotemporal prediction neural network; The PredRNN++ spatiotemporal prediction neural network takes a known crack magnetic field image as its input and a predicted crack magnetic field image as its output. Step S207: Train the PredRNN++ spatiotemporal prediction neural network; The root mean square error (RMSE) and structural similarity index (SSIM) are used as metrics to evaluate the accuracy of the PredRNN++ spatiotemporal prediction neural network prediction results; whereby the root mean square error (RMSE) satisfies: ; These are actual observed values. The value is the model prediction, and n is the number of samples. The structural similarity index SSIM satisfies: ;in, is the predicted value, For the actual true value, , The mean, , For variance, For covariance, , It is a constant; The trained PredRNN++ spatiotemporal prediction neural network can be used to obtain crack propagation prediction images; Step S208: Add true length labels to all predicted crack propagation images to form a length evaluation database; train the length evaluation database using a cross-fusion network; the trained length evaluation database can then be used to obtain the predicted crack propagation length of the predicted crack propagation images in the structural defects of offshore oil well control equipment.

[0008] On the other hand, the present invention also provides an electromagnetic control detection system for structural defects in marine oil well control equipment, comprising: a monitoring probe, an excitation signal generation unit, and a monitoring signal processing and acquisition unit; The monitoring probe consists of an excitation coil group, an array sensor, and a front-end signal processing circuit. The excitation coil group includes an upper excitation coil group and a lower excitation coil group that are orthogonally distributed. Each of the upper and lower excitation coil groups consists of two semicircular coils. Opposite sinusoidal alternating currents are passed through the two semicircular coils of the same layer in both the upper and lower excitation coil groups to generate a uniform induced current field. The array sensor is positioned above the uniform induced current field generated by the excitation coil group and is used to collect the induced magnetic field signal perpendicular to the surface of the offshore oil well control equipment under inspection. The front-end signal processing circuit is connected to the excitation coil group and is used to adjust the resonant frequency of the monitoring probe according to the external signal. The excitation signal generation unit includes a signal generator and a power amplifier; wherein, the signal generator is used to provide a sinusoidal AC excitation signal to the excitation coil group; and the power amplifier is used to amplify the sinusoidal AC excitation signal. The monitoring signal processing and acquisition unit includes a multiplexing and amplification filtering circuit, a bandpass filtering circuit, a phase-locked amplifier circuit, and a signal acquisition board.

[0009] Preferably, the upper excitation coil group and the lower excitation coil group in the excitation coil group use the same frequency parameters, so as to induce X-direction induced current Jx and Y-direction induced current Jy on the surface of the offshore oil well control equipment to be inspected. Among them, the induced current Jx in the X direction and the induced current Jy in the Y direction satisfy the following: ; ; , For amplitude, Angular frequency, , This is the initial phase; The angle between the vector sum of the X-direction induced current Jx and the Y-direction induced current Jy and the superimposed current Js in the X-direction is... ,satisfy: and ; The magnitude of the superimposed current Js obtained by vector addition of the X-direction induced current Jx and the Y-direction induced current Jy ,satisfy: ; Adjusting the skin depth of the induced current ,satisfy: ;in, Angular frequency, For magnetic permeability, is the electrical conductivity.

[0010] Preferably, the array sensor is composed of differentially arranged rectangular coils.

[0011] This invention provides a method and system for detecting electromagnetic control of structural defects in offshore oil well control equipment. The method includes the following steps: Step S101: Adjusting the excitation induced current to obtain the spatial magnetic field distortion of the offshore oil well control equipment under inspection; Step S102: Performing absolute value calculation on the magnetic field signal Bz matrix Z and plotting a grayscale image to obtain a Bz image; Step S103: Smoothing the Bz image using grayscale interpolation; fusing crack images from different angles using overlay to obtain complete crack contour information; Step S104: Adding real angle labels to all fused crack angle images to form an angle evaluation database; training the angle evaluation database using a residual network; and training the angle evaluation database to complete the angle evaluation... The database can be used to obtain the angle prediction value of unknown samples; Step S105: Adjust the amplitude and phase parameters to make the induced current perpendicular to the crack and obtain the crack endpoint information; Step S106: When the induced current is perpendicular to the crack, adjust the excitation signal frequency to change the skin depth of the induced current and obtain at least a first frequency image, a second frequency image, and a third frequency image; wherein, the first frequency is less than the second frequency, and the second frequency is less than the third frequency; place the first frequency image, the second frequency image, and the third frequency image in the R, G, and B channels of the color image respectively, and stack them to obtain a crack size evaluation fusion image; Step S107: Build a cross-fusion network and extract the Bz peak value in the Bz image. and peak spacing Using the random forest algorithm, a crack prediction system is established; Step S108: Initialize the cross-fusion network, embed the crack prediction system, and train it using training data; The trained crack prediction system can then be used to obtain the size prediction value of cracks at any angle in the structural defects of offshore oil well control equipment.

[0012] The electromagnetic control detection method and system for structural defects in offshore oil well control equipment with the above-mentioned characteristics have at least the following technical advantages compared with existing technologies: (1) The electromagnetic control detection system for structural defects in marine oil well control equipment provided by this invention uses a circular structure composed of semi-circular coils for both the upper and lower excitation coil groups. Compared with existing rectangular coils, this effectively avoids the local concentration of induced current at corners, making the induced current distribution more uniform and concentrated at the center of the array sensor position, thereby significantly enhancing the sensitivity of the monitoring probe and the crack detection rate. The front-end signal processing circuit connected to the excitation coil group is used to adjust the resonant frequency of the monitoring probe according to the external signal. When the excitation coil group is in series resonance, the current driven by the voltage source is maximized, thereby enhancing the alternating magnetic field to increase the detection eddy current density. When the detection coils are in parallel resonance, the impedance is maximized, resulting in the highest amplitude of the induced voltage, thereby improving the sensitivity of capturing the eddy current disturbance signal caused by defects. In addition, the array sensor is composed of differentially arranged rectangular coils, which improves the area utilization rate while effectively reducing missed detections, and can effectively reduce coil mutual inductance and signal noise. (2) The electromagnetic control and detection system for structural defects in marine oil well control equipment provided by this invention can control the direction of the induced current field by adjusting the sinusoidal signal parameters in the upper and lower excitation coil groups, changing the amplitude and phase at the same frequency, and obtaining a uniform current field with uniform amplitude. This solves the problem of the inability to separate the leakage magnetic signal and the AC electromagnetic field signal in time in the prior art, and eliminates the adverse effects of uneven current amplitude on the detection results. In addition, when the frequency parameters are changed, the skin depth of the induced current changes accordingly, thereby obtaining a variety of depth feature information. (3) The electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention obtains more information about cracks using a high-resolution network; by constructing a cross-fusion network, the single forward channel of the neural network is divided into two channels, which are used to extract the length and depth features of the crack, respectively. The feature fusion is achieved through the cross-fusion of the two channels. Then, the physical features in the crack Bz image are extracted, and a crack prediction system is established using the random forest algorithm; and the parameters are used to detect the crack. By adjusting the weights involved in the training of the crack prediction system, the training direction of the residual network was corrected. Finally, while enhancing the interpretability of the method, the prediction accuracy of crack length and depth was increased, avoiding misjudgments caused by single-dimensional feature analysis compared to the single detection mode of existing technologies. (4) The electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention transforms the three core fatigue laws—Paris equation, damage law, and SN curve—into calculable constraint terms through a life prediction network embedded with physical information. Through the synergistic effect of these three, multi-dimensional physical constraints for fatigue life prediction are realized, improving the generalization ability of the model and avoiding prediction distortion caused by data bias. Subsequently, the spatiotemporal prediction neural network captures spatiotemporal dependencies, especially capturing the temporal correlation of "intensified stress concentration caused by increased crack depth, which accelerates crack propagation." In addition, it can also capture local details such as crack edges, thereby integrating global distribution information and achieving the technical objective of more accurately predicting crack propagation length. Attached Figure Description

[0013] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the following drawings: Figure 1 A schematic diagram of the electromagnetic control detection system for structural defects in marine oil well control equipment provided by the present invention; Figure 2 This is a schematic diagram of the excitation coil group and array sensor in the monitoring probe; Figure 3 A schematic diagram of the array sensor structure in the monitoring probe; Figure 4a This is a schematic diagram of the induced current density of a rectangular coil; Figure 4b This is a schematic diagram of the induced current density of a circular coil; Figure 5a This is a schematic diagram showing the direction of the induced current at an angle of 0°. Figure 5b This is a schematic diagram showing the direction of the induced current at a 45° angle. Figure 5c This is a schematic diagram showing the direction of the induced current at an angle of 60°. Figure 5d This is a schematic diagram showing the direction of the induced current at a 90° angle. Figure 5e This is a schematic diagram of the induced current flow at an angle of 150°. Figure 6a A schematic diagram of the skin depth of an induced current at a frequency of 100Hz. Figure 6b A schematic diagram of the skin depth of an induced current at a frequency of 1 kHz. Figure 6c A schematic diagram of the skin depth of an induced current at a frequency of 10 kHz. Figure 7aA schematic diagram of the amplitude of the induced current after adjustment; Figure 7b A schematic diagram illustrating how to control the induced current to be perpendicular to the horizontal crack; Figure 7c A schematic diagram illustrating how to control the induced current to be perpendicular to the vertical crack; Figure 8 A schematic diagram of the equivalent resonant circuit for the monitoring probe; Figure 9a A schematic diagram of the series resonance of the excitation coil group (without resonance); Figure 9b A schematic diagram for detecting parallel resonance of coils (when resonance is present); Figure 10 A flowchart illustrating the electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention. Figure 11 This is a schematic diagram of the SRCNN (Super-Resolution Network) structure. Figure 12 A schematic diagram showing the complete information of the crack outline obtained by fusing and superimposing crack images from different angles; Figure 13 This is a schematic diagram of a crack size assessment fused image obtained by stacking 100Hz, 1kHz, and 10kHz images; Figure 14 A schematic diagram of the structure of the constructed cross-fusion network; Figure 15 A schematic diagram showing the result of crack propagation and complete fracture of the test specimen. Figure 16 A schematic diagram of the constructed multilayer fully connected neural network topology; Figure 17a This is a schematic diagram of the crack length propagation experiment results; Figure 17b A schematic diagram showing the predicted crack length propagation results; Figure 18 This is a schematic diagram of the structure of the PredRNN++ spatiotemporal prediction neural network. Detailed Implementation

[0014] This invention provides an electromagnetic control detection method and system for structural defects in offshore oil well control equipment. It acquires multi-feature signals of cracks through amplitude-frequency phase modulation of electromagnetic excitation signals, obtains crack angle information using a residual network, and achieves accurate identification of crack length and depth information using amplitude-frequency phase modulation and a physical heuristic neural network. A service life prediction network enables effective prediction of remaining fatigue life, and a spatiotemporal prediction neural network enables effective assessment of crack propagation, providing accurate data support for the health assessment of underwater structures of offshore oil well control equipment.

[0015] To facilitate understanding of the present invention by those skilled in the art, the electromagnetic control and detection system for structural defects in offshore oil well control equipment provided by the present invention will first be explained and described below. For example... Figure 1 As shown, the present invention provides an electromagnetic control detection system for structural defects in marine oil well control equipment, comprising: a monitoring probe, an excitation signal generation unit, and a monitoring signal processing and acquisition unit.

[0016] The monitoring probe further consists of an excitation coil assembly, an array sensor, and a front-end signal processing circuit. Specifically, such as... Figure 2 As shown, the excitation coil group includes an upper excitation coil group and a lower excitation coil group that are orthogonally distributed. Each of the upper and lower excitation coil groups consists of two semicircular coils. Opposite sinusoidal alternating currents are passed through the two semicircular coils of the same layer in both the upper and lower excitation coil groups to generate a uniform induced current field.

[0017] like Figure 3 As shown, the array sensor (preferably composed of differentially arranged rectangular coils) is positioned above the uniform induced current field generated by the excitation coil group to acquire the induced magnetic field signal perpendicular to the surface of the offshore oil well control equipment under inspection. It should be noted that the array sensor uses a differential arrangement of rectangular coils because the geometry of the rectangular coils better meets the regularity requirements of the array layout. Adjacent coils can be closely arranged, with smaller edge gaps. Taking the same coverage area as an example, the fill factor (effective coil area / total area) of rectangular coils is about 20%-30% higher than that of circular coils, increasing area utilization and reducing the likelihood of missed crack detection. Figure 4a , Figure 4b As shown ( Figure 4a This is a schematic diagram of the induced current density of a rectangular coil. Figure 4b (This is a schematic diagram of the induced current density of a circular coil). Furthermore, the differential arrangement reduces the mutual inductance between coils, thus reducing noise in the signal; the right-angled structure of the rectangular coil allows the magnetic field distribution to be more concentrated inside the coil, with faster attenuation of the edge magnetic field, further reducing the magnetic field coupling strength between adjacent coils. Additionally, the front-end signal processing circuit is connected to the excitation coil group to adjust the resonant frequency of the monitoring probe according to the external signal.

[0018] The excitation signal generation unit includes a signal generator and a power amplifier. The signal generator provides a sinusoidal AC excitation signal to the excitation coil group; the power amplifier amplifies the sinusoidal AC excitation signal. The monitoring signal processing and acquisition unit includes multiplexing and amplification filtering circuits, a bandpass filter circuit, a phase-locked amplifier circuit, and a signal acquisition board.

[0019] Furthermore, in a preferred embodiment of the present invention, the upper excitation coil group and the lower excitation coil group in the excitation coil group use the same frequency parameters, thereby inducing an X-direction induced current Jx and a Y-direction induced current Jy on the surface of the offshore oil well control equipment to be inspected.

[0020] Among them, the induced current Jx in the X direction and the induced current Jy in the Y direction satisfy the following: Equation (1); Equation (2); The angle between the vector sum of the X-direction induced current Jx and the Y-direction induced current Jy and the superimposed current Js in the X-direction is... ,satisfy: and Equation (3); The magnitude of the superimposed current Js obtained by vector addition of the X-direction induced current Jx and the Y-direction induced current Jy ,satisfy: Equation (4); Adjusting the skin depth of the induced current ,satisfy: Equation (5).

[0021] For specific details, please refer to the following: Figures 5a-5e As shown, select respectively , ; , ; , ; , ; , This allows for the determination of the induced current direction at angles of 0°, 45°, 60°, 90°, and 150°. By controlling the direction of the induced current, information about the crack at any angle can be obtained. Compared to existing rotating AC electromagnetic fields (where the current is more concentrated in a single direction), this effectively solves the problem of temporal overlap and inseparability between leakage magnetic field signals and AC electromagnetic field signals.

[0022] Furthermore, , The frequencies were chosen to be 100Hz, 1kHz, and 10kHz, respectively, to obtain schematic diagrams of current skin depth at different frequencies, as shown in the figure. Figure 6a , Figure 6b , Figure 6c As shown, it can be observed that by using multi-frequency excitation to make the induced current approach different skin depths, various depth feature information can be obtained.

[0023] It should be noted that amplitude-frequency phase modulation can make the current direction more concentrated and uniform, as shown in the example below. Figure 7a As shown. Among them, Figure 7a This involves adjusting the amplitude of the induced current in each direction at a frequency of 1 kHz, using superimposed currents of equal magnitude. The figure shows that this adjustment method highlights the influence of the current direction on the magnetic field distortion at the crack tip and eliminates the adverse effects of uneven current amplitude on the detection results. Furthermore, as... Figure 7b , Figure 7c As shown, it can be observed that adjusting the induced current to be perpendicular to the horizontal crack (reference) Figure 7b ), and adjust the induced current to be perpendicular to the vertical crack (refer to Figure 7c The amplitudes of the obtained magnetic field images are basically consistent.

[0024] In one possible embodiment of the present invention, the inductances of the excitation coil group and the array sensor were measured to be 170.9 μH and 27.7 μH, respectively, using an impedance analyzer. During this process, an adjustable capacitor (with an automatically adjustable capacitance value) is connected externally to the front-end signal processing circuit, ensuring that the resonant frequency of the monitoring probe is the excitation frequency. The equivalent resonant circuit of the monitoring probe can then be referenced as follows: Figure 8 As shown. When the excitation coil group resonates in series, the current is at its maximum under voltage source drive, as follows. Figure 9a As shown, the alternating magnetic field can be significantly enhanced, thereby increasing the eddy current density of the test piece. When the detection coils resonate in parallel, the impedance of the detection coils is at its maximum, and its induced voltage amplitude is at its highest. The minute eddy current disturbances caused by defects will significantly alter the voltage signal of the detection coils, such as... Figure 9b As shown, under the 1kHz excitation self-adjusting resonance, the sensitivity is enhanced by about 30%.

[0025] The specific implementation process of this electromagnetic control and detection system for structural defects in offshore oil well control equipment can be described as follows: Adjust the required signal generator parameters to emit a dual-channel sinusoidal AC signal; amplify the amplitude of the excitation signal through a power amplifier, so that it reaches the excitation coil group of the monitoring probe and induces a current on the structural surface. , At this point, structural defects (especially cracks) in the offshore oil well control equipment cause induced current disturbances, leading to distortion of the spatial magnetic field, which is then received by the array sensor. For example, the 64 differential signals emitted by an 8×8 array sensor are simplified into 4 differential signals after multiplexing and amplification filtering circuits. After further processing by a bandpass filter circuit, high-frequency noise and coupling interference signals are removed. The lock-in amplifier circuit amplifies the AC signal by 2 times, converting it into DC amplitude and DC phase signals, which are then finally acquired by the signal acquisition board and sent to the PC.

[0026] On the other hand, the present invention also provides a method for detecting electromagnetic control defects in the structure of offshore oil well control equipment, such as... Figure 10 As shown, it includes the following steps: Step S101: Adjust the excitation induced current to obtain the spatial magnetic field distortion of the offshore oil well control equipment under test. Calculate the 8×8 magnetic field signal matrix Z.

[0027] Wherein, the magnetic field signal Bz matrix Z satisfies: .

[0028] To facilitate understanding of the testing method by those skilled in the art, a specimen with the following dimensional parameters (a crack with a length of 20 mm, a depth of 5 mm, and an angle of 60° on the surface) is provided here, and the testing process of the specimen is described as an example.

[0029] Specifically, by changing the amplitude and phase parameters of the quadrature excitation coil group, the purpose of regulating the excitation induced current is achieved, thereby obtaining the spatial magnetic field distortion caused by the disturbance of the induced current; then, by subtracting the calibrated background magnetic field, the magnetic field signal Bz matrix Z with a size of 8×8 can be obtained.

[0030] Step S102: Perform absolute value operation on the magnetic field signal Bz matrix Z and draw a grayscale image to obtain the Bz image.

[0031] Step S103: Smooth the Bz image using grayscale interpolation; By overlaying crack images from different angles, complete information about the crack contour is obtained.

[0032] Based on step S101, the absolute value of the magnetic field signal Bz matrix Z is further calculated and a grayscale image is plotted to obtain the Bz image; then, the Bz image is smoothed. For details, refer to... Figure 11 As shown, a pre-trained SRCNN network can first be used to process the low-resolution Bz image; for example, a three-layer convolutional structure can be used to sequentially perform feature extraction, nonlinear mapping, and high-resolution image reconstruction, thereby outputting a high-resolution image. The purpose of this is to obtain more crack information using limited sensor data. Then, by overlaying crack images from different angles, complete crack contour information can be obtained. The complete crack contour information obtained after overlaying can be referenced as shown in the figure. Figure 12 As shown.

[0033] Step S104: Add real angle labels to all crack angle fusion images to form an angle evaluation database.

[0034] The angle evaluation database is trained using a residual network; once trained, the angle evaluation database can be used to obtain the angle prediction values ​​for unknown samples.

[0035] Based on steps S102 and S103, an angle evaluation database is further constructed. Specifically, a residual network (preferably a ResNet-18 residual network) is used to train the angle evaluation database, and the predicted angle of the crack is 59.78°.

[0036] Step S105: Adjust the amplitude and phase parameters to make the induced current perpendicular to the crack and obtain information about the crack endpoint.

[0037] Specifically, the amplitude and phase parameters adjusted in step S105 can be referenced as follows: , The purpose is to make the induced current perpendicular to the crack, so as to obtain information about the crack endpoint.

[0038] Step S106: When the induced current is perpendicular to the crack, adjust the frequency of the excitation signal to change the skin depth of the induced current, and obtain at least a first frequency image, a second frequency image, and a third frequency image; wherein the first frequency is less than the second frequency, and the second frequency is less than the third frequency.

[0039] The first frequency image, the second frequency image, and the third frequency image are placed in the R, G, and B channels of the color image, respectively, and stacked to obtain a crack size assessment fused image.

[0040] Based on step S105, a crack size assessment fused image is further obtained through stacking. In a preferred embodiment of the invention, the excitation signal frequency is adjusted to obtain 100Hz, 1kHz, and 10kHz images. Different frequency images correspond to different induced current skin depths, reflecting the characteristics of the crack at different depth levels.

[0041] Furthermore, these three images are placed into the R, G, and B channels of the color image respectively, and a stacked defense is used to construct RGB-like 3D samples to generate a crack size assessment fused image. For details, please refer to [reference needed]. Figure 13 As shown, this fusion method can integrate information on the skin depth of different induced currents into the same image; compared with images of a single frequency, it can present crack depth information more comprehensively and intuitively, effectively reducing the depth quantization error caused by single-frequency detection.

[0042] Step S107: Construct a cross-fusion network to extract the Bz peak value from the Bz image. and peak spacing A crack prediction system was established using the random forest algorithm.

[0043] If the predicted value of the cross-fusion network is set to OUTn, the predicted value of the crack prediction system is set to OUTe, and the true value of the crack size is set to Label, then the constructed loss function satisfies: ; in, The weights involved in the training of the crack prediction system.

[0044] Based on step S106, a crack prediction system is further constructed. It is worth noting that, with reference to... Figure 14 As shown, a cross-fusion network is first constructed. This network employs a dual-path structure to specifically extract crack length and depth features, reducing mutual quantization interference and strengthening feature correlation through cross-fusion between channels. Then, by adjusting the weight parameters, the weights involved in the crack prediction system during training are adjusted to correct the training direction of the residual network.

[0045] Step S108: Initialize the cross-fusion network, embed the crack prediction system, and train it using training data; wherein, the trained crack prediction system can be used to obtain the size prediction value of cracks at any angle in the structural defects of offshore oil well control equipment.

[0046] Specifically, the aforementioned crack size assessment fused image is loaded into the trained cross-fusion network, and the final output shows a crack length of 20.32 mm and a crack depth of 5.09 mm. It can be seen that, compared to traditional single detection methods, the electromagnetic control detection method for structural defects in marine oil well control equipment provided by this invention can simultaneously consider multi-dimensional information on crack length and depth, avoiding size misjudgments caused by single feature analysis.

[0047] In addition, the electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention further includes the following steps: Step S201: Obtain the control data and use the trained cross-fusion network to extract the crack length information features of the offshore oil well control equipment. .

[0048] In addition, acquiring stress data from offshore oil well control equipment. .

[0049] To facilitate understanding of the present invention by those skilled in the art, the test specimen is further explained as follows. Specifically, a workpiece containing a 10mm pre-set crack is selected and subjected to cyclic loading. The stress on the workpiece, as obtained by the stress sensor, is 4e. 7 Pa, using the above detection method, the predicted length of the pre-crack is approximately 10.03 mm.

[0050] Step S202: Obtain the remaining fatigue life by cyclically loading until the offshore oil well control equipment workpiece completely fractures. .

[0051] Based on completing step S201, the remaining fatigue life is obtained through further cycles. The test specimen was subjected to cyclic loading until the crack propagated and complete fracture occurred, yielding a remaining fatigue life of approximately 5.2 e. 4 For details, please refer to the following: Figure 15 As shown.

[0052] Step S203: Construct a multi-layer fully connected neural network topology.

[0053] Crack length information characteristics of offshore oil well control equipment Stress data of offshore oil well control equipment Remaining fatigue life As the three nodes of the input layer, a pyramid-shaped structure is used as the hidden layer, and the fatigue life prediction value is used as the single node of the output layer to realize the multi-dimensional physical constraints of fatigue life prediction.

[0054] Building upon step S202, a multi-layer fully connected neural network topology is further constructed. For details, please refer to... Figure 16 As shown, the input layer of this multi-layer fully connected neural network topology uses crack length information features from offshore oil well control equipment. Stress data of offshore oil well control equipment Remaining fatigue life Three types of features are used as input nodes; and a pyramid structure is used as the hidden layer, with the number of neurons in the hidden layer decreasing by 2 / 3 of the previous layer (preferably, the ReLU activation function is used to introduce nonlinear mapping capability, thereby enhancing the model's expressive energy for complex fatigue life relationships); the output layer outputs the fatigue life prediction value through a single node.

[0055] Step S204: Train the lifetime prediction network.

[0056] The regression evaluation index MSE is used to calculate the deviation between predicted and measured values. And construct features including crack length information of offshore oil well control equipment. Stress data of offshore oil well control equipment Remaining fatigue life Physical consistency loss of three types of characteristic residuals .

[0057] Among them, the deviation between the predicted value and the measured value ,satisfy: ; Designing the weighted loss function, we obtain: ; in, This was determined through cross-validation.

[0058] Building upon step S203, the life prediction network is further trained. Specifically, the workpiece stress 4e... 7 Pa and a crack length of 10.03 mm were applied to the trained PINN network; after inverse normalization, the predicted remaining fatigue life of the workpiece was approximately 5.18 e. 4 It is worth noting that the prediction results embedded with physical information not only fit the experimental data but also satisfy the theoretical relationship between stress intensity factor and crack propagation rate, reducing errors while being suitable for small datasets.

[0059] Step S205: In the fatigue life prediction process, a fixed time step is set, the load is set to a fixed value, the actual crack length is obtained, and the crack propagation image is monitored simultaneously.

[0060] High-resolution, multi-frequency fusion images of vertical cracks caused by induced current were obtained using SRCNN, and a database was built.

[0061] Building upon step S204, the actual crack length is further obtained, and crack propagation images are monitored simultaneously. Specifically, during crack propagation, a static load of 3kN is applied after every 4000 loading cycles, and the actual crack length is obtained using calipers. Then, the propagated crack image is obtained using the method described in the example above. Figure 17a , Figure 17b As shown; where, as Figure 17a A schematic diagram of the crack length propagation experiment results. Figure 17b This is a schematic diagram of the crack length propagation prediction results. Then, the initial crack images obtained above are compared by arranging the first 5 frames (historical crack images) in chronological order as known images and the last 5 frames (historical crack images) in chronological order as predicted images, forming the input sequence. The purpose of this is to improve the prediction accuracy of crack propagation through multi-frequency fusion.

[0062] Step S206: Construct the PredRNN++ spatiotemporal prediction neural network.

[0063] The PredRNN++ spatiotemporal prediction neural network takes a known crack magnetic field image as input and outputs a predicted crack magnetic field image as output.

[0064] Building upon step S205, a PredRNN++ spatiotemporal prediction neural network is further constructed. (See reference...) Figure 18As shown, spatial features and temporal dependencies are first extracted simultaneously through five Causal LSTM convolutional layers. Each Causal LSTM layer contains 3×3 convolutions and a gating mechanism, with residual connections between layers to mitigate the degradation problem of deep networks. Then, a Gradient Highway Unit (GHU) is embedded to directly transmit long-distance gradients through gating signals, capturing the long-term dynamic patterns of crack propagation. Finally, the output layer generates a single-channel predicted crack magnetic field image through 1×1 convolutions and upsampling.

[0065] Step S207: Train the PredRNN++ spatiotemporal prediction neural network.

[0066] The root mean square error (RMSE) and structural similarity index (SSIM) are used as metrics to evaluate the accuracy of the PredRNN++ spatiotemporal prediction neural network. Specifically, the root mean square error (RMSE) satisfies: The structural similarity index SSIM satisfies: .

[0067] The trained PredRNN++ spatiotemporal prediction neural network can be used to obtain crack propagation prediction images.

[0068] Building upon step S206, the PredRNN++ spatiotemporal prediction neural network is further trained. The output layer of the PredRNN++ spatiotemporal prediction neural network outputs the predicted crack magnetic field image of the workpiece, as shown in the example below. Figure 17b As shown, the RMSE of the predicted image is on the order of 10e. -2 This indicates that the model's prediction error is at a low level. Meanwhile, an SSIM greater than 0.94 signifies a high similarity between the predicted and actual values. Therefore, simultaneously extracting spatial features and temporal dependencies can significantly enhance the accuracy of the prediction results.

[0069] Step S208: Add true length labels to all predicted crack propagation images to form a length evaluation database; train the length evaluation database using a cross-fusion network; the trained length evaluation database can then be used to obtain the predicted crack propagation length of the predicted crack propagation images in the structural defects of offshore oil well control equipment.

[0070] It is worth noting that by loading the aforementioned crack propagation prediction images into the trained length evaluation database and the cross-fusion network, the predicted crack propagation lengths are obtained: 18.66mm, 20.98mm, 23.34mm, 27.48mm, and 32.30mm; corresponding to actual crack lengths of 19.3mm, 21.0mm, 23.3mm, 26.0mm, and 30.5mm. The comparison confirms that the average error is approximately 0.532mm.

[0071] This invention provides a method and system for detecting electromagnetic control of structural defects in offshore oil well control equipment. The method includes the following steps: Step S101: Adjusting the excitation induced current to obtain the spatial magnetic field distortion of the offshore oil well control equipment under inspection; Step S102: Performing absolute value calculation on the magnetic field signal Bz matrix Z and plotting a grayscale image to obtain a Bz image; Step S103: Smoothing the Bz image using grayscale interpolation; fusing crack images from different angles using overlay to obtain complete crack contour information; Step S104: Adding real angle labels to all fused crack angle images to form an angle evaluation database; training the angle evaluation database using a residual network; and training the angle evaluation database to complete the angle evaluation... The database can be used to obtain the angle prediction value of unknown samples; Step S105: Adjust the amplitude and phase parameters to make the induced current perpendicular to the crack and obtain the crack endpoint information; Step S106: When the induced current is perpendicular to the crack, adjust the excitation signal frequency to change the skin depth of the induced current and obtain at least a first frequency image, a second frequency image, and a third frequency image; wherein, the first frequency is less than the second frequency, and the second frequency is less than the third frequency; place the first frequency image, the second frequency image, and the third frequency image in the R, G, and B channels of the color image respectively, and stack them to obtain a crack size evaluation fusion image; Step S107: Build a cross-fusion network and extract the Bz peak value in the Bz image. and peak spacing Using the random forest algorithm, a crack prediction system is established; Step S108: Initialize the cross-fusion network, embed the crack prediction system, and train it using training data; The trained crack prediction system can then be used to obtain the size prediction value of cracks at any angle in the structural defects of offshore oil well control equipment.

[0072] The electromagnetic control detection method and system for structural defects in offshore oil well control equipment with the above-mentioned characteristics have at least the following technical advantages compared with existing technologies: (1) The electromagnetic control detection system for structural defects in marine oil well control equipment provided by this invention uses a circular structure composed of semi-circular coils for both the upper and lower excitation coil groups. Compared with existing rectangular coils, this effectively avoids the local concentration of induced current at corners, making the induced current distribution more uniform and concentrated at the center of the array sensor position, thereby significantly enhancing the sensitivity of the monitoring probe and the crack detection rate. The front-end signal processing circuit connected to the excitation coil group is used to adjust the resonant frequency of the monitoring probe according to the external signal. When the excitation coil group is in series resonance, the current driven by the voltage source is maximized, thereby enhancing the alternating magnetic field to increase the detection eddy current density. When the detection coils are in parallel resonance, the impedance is maximized, resulting in the highest amplitude of the induced voltage, thereby improving the sensitivity of capturing the eddy current disturbance signal caused by defects. In addition, the array sensor is composed of differentially arranged rectangular coils, which improves the area utilization rate while effectively reducing missed detections, and can effectively reduce coil mutual inductance and signal noise. (2) The electromagnetic control and detection system for structural defects in marine oil well control equipment provided by this invention can control the direction of the induced current field by adjusting the sinusoidal signal parameters in the upper and lower excitation coil groups, changing the amplitude and phase at the same frequency, and obtaining a uniform current field with uniform amplitude. This solves the problem of the inability to separate the leakage magnetic signal and the AC electromagnetic field signal in time in the prior art, and eliminates the adverse effects of uneven current amplitude on the detection results. In addition, when the frequency parameters are changed, the skin depth of the induced current changes accordingly, thereby obtaining a variety of depth feature information. (3) The electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention obtains more information about cracks using a high-resolution network; by constructing a cross-fusion network, the single forward channel of the neural network is divided into two channels, which are used to extract the length and depth features of the crack, respectively. The feature fusion is achieved through the cross-fusion of the two channels. Then, the physical features in the crack Bz image are extracted, and a crack prediction system is established using the random forest algorithm; and the parameters are used to detect the crack. By adjusting the weights involved in the training of the crack prediction system, the training direction of the residual network was corrected. Finally, while enhancing the interpretability of the method, the prediction accuracy of crack length and depth was increased, avoiding misjudgments caused by single-dimensional feature analysis compared to the single detection mode of existing technologies. (4) The electromagnetic control detection method for structural defects in offshore oil well control equipment provided by this invention transforms the three core fatigue laws—Paris equation, damage law, and SN curve—into calculable constraint terms through a life prediction network embedded with physical information. Through the synergistic effect of these three, multi-dimensional physical constraints for fatigue life prediction are realized, improving the generalization ability of the model and avoiding prediction distortion caused by data bias. Subsequently, the spatiotemporal prediction neural network captures spatiotemporal dependencies, especially capturing the temporal correlation of "intensified stress concentration caused by increased crack depth, which accelerates crack propagation." In addition, it can also capture local details such as crack edges, thereby integrating global distribution information and achieving the technical objective of more accurately predicting crack propagation length.

[0073] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting structural defects in offshore oil well control equipment using electromagnetic control, characterized in that, The steps include the following: Step S101: Adjust the excitation induced current to obtain the spatial magnetic field distortion of the offshore oil well control equipment under test; The magnetic field signal Bz matrix Z of size 8×8 was calculated; Wherein, the magnetic field signal Bz matrix Z satisfies: ;in, Let Bz be the value of the magnetic field signal in the i-th row and j-th column of the Bz matrix; Step S102: Perform absolute value operation on the magnetic field signal Bz matrix Z and draw a grayscale image to obtain the Bz image; Step S103: Smooth the Bz image using grayscale interpolation; By overlaying crack images from different angles, complete information about the crack outline can be obtained. Step S104: Add real angle labels to all crack angle fusion images to form an angle evaluation database; The angle evaluation database is trained using a residual network; once trained, the angle evaluation database can be used to obtain angle prediction values ​​for unknown samples. Step S105: Adjust the amplitude and phase parameters to make the induced current perpendicular to the crack and obtain information about the crack endpoint; Step S106: When the induced current is perpendicular to the crack, adjust the frequency of the excitation signal to change the skin depth of the induced current, and obtain at least a first frequency image, a second frequency image, and a third frequency image; wherein the first frequency is less than the second frequency, and the second frequency is less than the third frequency. The first frequency image, the second frequency image, and the third frequency image are placed in the R, G, and B channels of the color image, respectively, and stacked to obtain a crack size assessment fused image; Step S107: Construct a cross-fusion network to extract the Bz peak value from the Bz image. and peak spacing A crack prediction system was established using the random forest algorithm. If the predicted value of the cross-fusion network is set to OUTn, the predicted value of the crack prediction system is set to OUTe, and the true value of the crack size is set to Label, then the constructed loss function satisfies: ; in, The weights involved in the training of the crack prediction system; Step S108: Initialize the cross-fusion network, embed the crack prediction system, and train it using training data; wherein, the trained crack prediction system can be used to obtain the size prediction value of cracks at any angle in the structural defects of offshore oil well control equipment.

2. The electromagnetic control detection method for structural defects in offshore oil well control equipment according to claim 1, characterized in that, It also includes the following steps: Step S201: Obtain the control data and use the trained cross-fusion network to extract the crack length information features of the offshore oil well control equipment. ; In addition, acquiring stress data from offshore oil well control equipment. ; Step S202: Obtain the remaining fatigue life by cyclically loading until the offshore oil well control equipment workpiece completely fractures. ; Step S203: Construct a multi-layer fully connected neural network topology; Crack length information characteristics of offshore oil well control equipment Stress data of offshore oil well control equipment Remaining fatigue life As three nodes in the input layer, a pyramid-shaped structure is used as the hidden layer, and the fatigue life prediction value is used as a single node in the output layer to realize multi-dimensional physical constraints on fatigue life prediction. Step S204: Train the lifetime prediction network; The regression evaluation index MSE is used to calculate the deviation between predicted and measured values. And construct features including crack length information of offshore oil well control equipment. Stress data of offshore oil well control equipment Remaining fatigue life Physical consistency loss of three types of characteristic residuals ; Among them, the deviation between the predicted value and the measured value ,satisfy: ; These are actual observed values. The value is the model prediction, and n is the number of samples. Designing the weighted loss function, we obtain: ;in, Determined through cross-validation; ,satisfy: ;in, , , Let be the weights, y be the model prediction, B be the Basquin model output, and P be the Paris model output. This represents a non-physical value penalty term, used to ensure that y is non-negative, thereby avoiding unreasonable results from the model output; Step S205: In the fatigue life prediction process, a fixed time step is set, the load is set to a fixed value, the actual crack length is obtained, and the crack propagation image is monitored simultaneously. High-resolution, multi-frequency fusion magnetic field images of vertical cracks induced by current were obtained using SRCNN, and a database was built. Step S206: Construct the PredRNN++ spatiotemporal prediction neural network; The PredRNN++ spatiotemporal prediction neural network takes a known crack magnetic field image as its input and a predicted crack magnetic field image as its output. Step S207: Train the PredRNN++ spatiotemporal prediction neural network; The root mean square error (RMSE) and structural similarity index (SSIM) are used as metrics to evaluate the accuracy of the PredRNN++ spatiotemporal prediction neural network prediction results; whereby the root mean square error (RMSE) satisfies: ; These are actual observed values. The value is the model prediction, and n is the number of samples. The structural similarity index SSIM satisfies: Where x is the predicted value, y is the actual value, and μ is the predicted value. x μ y The mean, , Let σ be the variance. xy For covariance, C1 and C2 are constants; The trained PredRNN++ spatiotemporal prediction neural network can be used to obtain crack propagation prediction images; Step S208: Add true length labels to all predicted crack propagation images to form a length evaluation database; train the length evaluation database using a cross-fusion network; the trained length evaluation database can then be used to obtain the predicted crack propagation length of the predicted crack propagation images in the structural defects of offshore oil well control equipment.

3. An electromagnetic control and detection system for structural defects in offshore oil well control equipment, characterized in that, It includes: a monitoring probe, an excitation signal generation unit, and a monitoring signal processing and acquisition unit; The monitoring probe consists of an excitation coil group, an array sensor, and a front-end signal processing circuit. The excitation coil group includes an upper excitation coil group and a lower excitation coil group that are orthogonally distributed. Each of the upper and lower excitation coil groups consists of two semicircular coils. Opposite sinusoidal alternating currents are passed through the two semicircular coils of the same layer in both the upper and lower excitation coil groups to generate a uniform induced current field. The array sensor is positioned above the uniform induced current field generated by the excitation coil group and is used to collect the induced magnetic field signal perpendicular to the surface of the offshore oil well control equipment under inspection. The front-end signal processing circuit is connected to the excitation coil group and is used to adjust the resonant frequency of the monitoring probe according to the external signal. The excitation signal generation unit includes a signal generator and a power amplifier; wherein, the signal generator is used to provide a sinusoidal AC excitation signal to the excitation coil group; and the power amplifier is used to amplify the sinusoidal AC excitation signal. The monitoring signal processing and acquisition unit includes a multiplexing and amplification filtering circuit, a bandpass filtering circuit, a phase-locked amplifier circuit, and a signal acquisition board.

4. The electromagnetic control and detection system for structural defects in offshore oil well control equipment according to claim 3, characterized in that, The upper and lower excitation coil groups in the excitation coil group use the same frequency parameters, thereby inducing X-direction induced current Jx and Y-direction induced current Jy on the surface of the offshore oil well control equipment to be inspected. Among them, the induced current Jx in the X direction and the induced current Jy in the Y direction satisfy the following: ； ； , For amplitude, Angular frequency, , This is the initial phase; The angle between the vector sum of the X-direction induced current Jx and the Y-direction induced current Jy and the superimposed current Js in the X-direction is... ,satisfy: and ； The magnitude of the superimposed current Js obtained by vector addition of the X-direction induced current Jx and the Y-direction induced current Jy ,satisfy: ; Adjusting the skin depth of the induced current ,satisfy: ;in, Angular frequency, For magnetic permeability, is the electrical conductivity.

5. The electromagnetic control and detection system for structural defects in offshore oil well control equipment according to claim 3, characterized in that, The array sensor is composed of differentially arranged rectangular coils.

Images