A defect multi-parameter inversion method of capacitive imaging sensitivity field constraint

Abstract

The present application belongs to the technical field of nondestructive testing, and particularly relates to a defect multi-parameter inversion method with a capacitive imaging sensitivity field constraint. The inversion method combines finite element simulation to construct a standardized sensitivity field matrix and establish a dielectric constant-capacitance quantitative mapping relationship under the sensitivity field constraint. A closed-loop inversion algorithm combining forward modeling and reverse matching can quickly and stably invert the key parameters of defects, effectively improving the precision and reliability of quantitative detection of defects in insulating structures. The inversion method comprises: building a capacitive imaging detection system; constructing an electrostatic field model corresponding to the sensor to obtain a standardized sensitivity field matrix; deriving a quantitative discrete mapping relationship between defect parameters and charge signals; scanning the test piece to obtain a real charge map; presetting the initial defect parameters to generate a predicted charge map through forward modeling; and calculating the matching degree between the predicted charge map and the real charge map.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing technology, and in particular relates to a multi-parameter inversion method for defects constrained by capacitance imaging sensitivity field. Background Technology

[0002] In my country's industrial sector, non-conductive, insulating anti-corrosion coatings are widely used on pressure-bearing equipment such as pipelines and storage tanks. These coatings play a crucial role in protecting equipment in scenarios such as oil and gas transportation and chemical production, and are key to ensuring the stable operation of industrial systems. However, during long-term service, these coatings are prone to various defects due to corrosion, scratches, and aging. The presence of these defects can lead to coating failure, resulting in damage to the equipment itself, safety accidents, and economic losses. Therefore, accurate detection of non-conductive coatings has become a vital aspect of safety monitoring for pressure-bearing equipment. Among these methods, capacitive imaging detection technology, with its non-contact, high-sensitivity, and adaptability to non-conductive materials, has become the mainstream detection method in this field and is widely used in defect screening of non-conductive coatings.

[0003] In the application of capacitance imaging detection technology, the lack of theoretical research on defect inversion and the scarcity of data processing methods limit its ability to determine the presence and rough location of defects. Current conventional capacitance imaging detection procedures typically convert the acquired capacitance signal into a visualized charge map; then, technicians manually determine the presence of defects by observing the shape and characteristics of the charge map. There is no systematic method to support core aspects such as the charge response characteristics induced by defects and the prediction of charge distribution generation; these rely solely on the experience of technicians, resulting in poor reliability and consistency of the detection results. With the increasing demands for equipment safety monitoring in industrial production, simply screening defects is no longer sufficient to meet the practical needs of equipment simulation analysis, detection model verification, and signal feature prediction. Accurate inversion and prediction of charge generation based on defect information have become core requirements for the engineering application of capacitance imaging detection technology and represent an important direction for its development.

[0004] However, further research revealed that existing capacitance imaging defect inversion techniques still have the following significant shortcomings: (1) Lacking analytical models or explicit mapping relationships for multi-parameter collaborative inversion, existing technologies can usually only evaluate the key parameter size of the defect (such as length or width) one by one, while ignoring the coupling effect between parameters, and cannot achieve synchronous inversion of multiple key parameters; (2) Existing parameter inversion methods generally lack the use of the frequency correlation of "soft field" and cannot effectively decouple the coupling effect between the depth of the defect itself and the burial depth. (3) Pure data-driven algorithms lack the support of physical theory, require a large amount of training data, have poor generalization, and are difficult to meet the needs of in-situ detection in industrial sites. (4) Due to the limited signal-to-noise ratio of the measurement system and the fixed number of electrodes, the number of independent capacitance measurements that can be obtained is far less than the number of parameters to be inverted, resulting in an extremely underdetermined inverse problem. This easily leads to the phenomenon that multiple sets of parameter combinations simultaneously fit the measurement data, causing the inversion results to lose their uniqueness.

[0005] Therefore, there is an urgent need for those skilled in the art to provide a capacitance imaging detection method that can accurately invert non-conductive coating defects and predict charge generation, so as to solve the technical problems of low inversion accuracy, lack of physical theoretical support, and poor accuracy of predicted charge generation in the existing technology. Summary of the Invention

[0006] This invention provides a multi-parameter defect inversion method constrained by capacitance imaging sensitivity field. This method combines finite element simulation to construct a standardized sensitivity field matrix and establishes a quantitative mapping relationship between dielectric constant and capacitance under sensitivity field constraints. After preprocessing the measured charge signal, a closed-loop inversion algorithm combining forward modeling and reverse matching is employed, along with neural network optimization, to achieve accurate optimization of multiple defect parameters. This method can quickly and stably invert key parameters such as defect length, width, depth, and burial depth, effectively improving the accuracy and reliability of quantitative detection of internal defects in insulation structures.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A multi-parameter inversion method for defects constrained by capacitance imaging sensitivity field includes the following steps: Step S101: Build the capacitance imaging detection system required for the defect multi-parameter inversion method and perform empty field calibration on it; wherein, the capacitance imaging detection system includes a single pair of coplanar electrode capacitance imaging sensors, a triaxial scanning unit, a signal conditioning unit, a data acquisition unit and a control unit; Step S102: Construct the electrostatic field model corresponding to a single pair of coplanar electrode capacitive imaging sensors; calculate and normalize to obtain the standardized sensitivity field matrix; Step S103: Establish a dielectric constant-capacitance mapping model; derive the quantitative discrete mapping relationship between defect parameters and charge signals; Step S104: Scan the test piece to acquire the charge signal and record its spatial coordinates; preprocess the original charge map to obtain the true charge map; Step S105: Preset initialization defect parameters; By combining the standardized sensitivity field matrix and the initialized defect parameters, the dielectric constant matrix is ​​generated. Based on the dielectric constant-capacitance mapping model established in step S103, a predicted charge map is generated through forward modeling. Step S106: Calculate the matching degree between the predicted charge map and the actual charge map; With the goal of maximizing the matching degree between the predicted charge map and the actual charge map, the defect parameters are iteratively updated until the matching degree reaches a preset threshold, and the current defect parameters are output as the optimal solution for defect inversion. Among them, the defect parameters include at least the length, width, depth and burial depth of the defect.

[0008] Preferably, the qualification criteria for the empty field calibration of the capacitance imaging detection system in step S101 are: the empty field background charge signal curve is stable, the amplitude fluctuation is ≤ ±0.5mV, and there is no clutter or baseline drift.

[0009] Preferably, the process of calculating and normalizing the standardized sensitivity field matrix in step S102 is specifically described as follows: A two-dimensional electrostatic field model was constructed using finite element simulation software based on the actual geometric parameters of a single pair of coplanar electrode capacitive imaging sensors. Dielectric perturbations are sequentially applied to each micro-element within the detection region of the two-dimensional electrostatic field model; The sensitivity coefficients of each infinitesimal element are derived using the point perturbation method; where, the sensitivity coefficients of each infinitesimal element... ,satisfy: ; In the formula, The vacuum permittivity, For excitation voltage, The two electrostatic field interfaces are located at the following positions. The electric field strength obtained by solving at point; The sensitivity coefficients of each infinitesimal element are linearly normalized to the [0,1] interval to obtain the standardized sensitivity field matrix.

[0010] Preferably, the process of establishing the dielectric constant-capacitance mapping model in step S103 is specifically described as follows: The defective specimen is equivalent to a defect-free uniform substrate superimposed with a defect negative polarization region; The detection area is discretized into several non-overlapping micro-elements; Based on the physical definition of the sensitivity field, a discrete mapping formula for the total charge perturbation is established, satisfying: ; In the formula, Q represents the total charge disturbance detected by a single pair of coplanar electrode capacitance imaging sensors, and N represents the total number of discrete infinitesimal elements in the detection region. For the first The sensitivity coefficient at each infinitesimal position. For the first The volume of a microelement; For the first The dielectric constant perturbation of each infinitesimal element satisfies: ; in, For the first The actual dielectric constant of a microelement is the uniform dielectric constant of a defect-free substrate.

[0011] Preferably, the process of preprocessing the original charge map to obtain the true charge map in step S104 is specifically described as follows: The charge signals in the original charge map are linearly normalized to the [0,1] interval; The normalized signal is filtered using a Gaussian kernel. Using the background signal acquired during field calibration as a reference, the filtered signal is subjected to background correction to obtain the true charge map.

[0012] Preferably, the formula for calculating the matching degree between the predicted charge map and the actual charge map in step S106 satisfies: ; in, To predict the charge value of the charge map, The charge values ​​are those of the actual charge map.

[0013] This invention provides a multi-parameter defect inversion method constrained by a capacitance imaging sensitivity field. The method includes the following steps: Step S101: Constructing the capacitance imaging detection system required for the multi-parameter defect inversion method and performing empty-field calibration; Step S102: Constructing an electrostatic field model corresponding to a single pair of coplanar electrode capacitance imaging sensors; calculating and normalizing to obtain a standardized sensitivity field matrix; Step S103: Establishing a dielectric constant-capacitance mapping model; deriving the quantitative discrete mapping relationship between defect parameters and charge signals; Step S104: Scanning the test piece, acquiring charge signals and recording their spatial coordinates; processing the original charge map... Preprocessing is performed to obtain the true charge map; Step S105: Preset initialization defect parameters; Combine the standardized sensitivity field matrix with the initialization defect parameters to generate the dielectric constant matrix; Based on the dielectric constant-capacitance mapping model established in step S103, the predicted charge map is generated through forward modeling; Step S106: Calculate the matching degree between the predicted charge map and the true charge map; With the goal of maximizing the matching degree between the predicted charge map and the true charge map, the defect parameters are iteratively updated until the matching degree reaches the preset threshold, and the current defect parameters are output as the optimal solution for defect inversion.

[0014] The multi-parameter inversion method for capacitance imaging sensitivity field constraint with the above-mentioned step characteristics has at least the following advantages compared with the prior art; (1) The multi-parameter inversion method for defects with capacitance imaging sensitivity field constraint provided by the present invention significantly reduces the ill-posedness of defect quantification and improves the identification accuracy of buried defects and micro-defects by introducing the physical constraint of sensitivity field into the inversion process. (2) The multi-parameter inversion method for defects with capacitance imaging sensitivity field constraint provided by the present invention realizes the synchronous inversion of multiple parameters such as defect length, width, depth and burial depth, without the need for multiple detections or complex modeling, and the detection efficiency is higher. (3) The multi-parameter inversion method for the capacitance imaging sensitivity field constraint provided by the present invention combines forward modeling and reverse matching closed-loop iteration with neural network optimization solution, thus taking into account both physical reliability and algorithm efficiency, and has a stronger industrial adaptability. Attached Figure Description

[0015] 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 flowchart illustrating a multi-parameter inversion method for capacitance imaging sensitivity field constraint provided by the present invention; Figure 2 This is a schematic diagram of the capacitive imaging detection system used in this invention; Figure 3 for Figure 2 The diagram shows the structure of a single pair of coplanar electrode capacitance imaging sensors in a capacitance imaging detection system. Figure 4 A reference schematic diagram of the standardized sensitivity field matrix obtained in step S102 of the defect multi-parameter inversion method for capacitance imaging sensitivity field constraint provided by the present invention. Figure 5 This is a schematic diagram of the scanning method for the test piece; Figure 6 This is a schematic diagram comparing the simulation results, experimental results, and predicted charge distribution images obtained in this invention.

[0016] Figure label: 1. Single pair of coplanar electrode capacitive imaging sensor; 2. Triaxial scanning unit; 3. Signal conditioning unit; 4. Data acquisition unit; 5. Control unit; 6. Signal generator; 7. Lock-in amplifier; 8. Test piece. Detailed Implementation

[0017] This invention provides a multi-parameter defect inversion method constrained by capacitance imaging sensitivity field. This method combines finite element simulation to construct a standardized sensitivity field matrix and establishes a quantitative mapping relationship between dielectric constant and capacitance under sensitivity field constraints. After preprocessing the measured charge signal, a closed-loop inversion algorithm combining forward modeling and reverse matching is employed, along with neural network optimization, to achieve accurate optimization of multiple defect parameters. This method can quickly and stably invert key parameters such as defect length, width, depth, and burial depth, effectively improving the accuracy and reliability of quantitative detection of internal defects in insulation structures.

[0018] like Figure 1 As shown, the present invention provides a multi-parameter inversion method for capacitance imaging sensitivity field constraint defects, comprising the following steps: Step S101: Build the capacitance imaging detection system required for the multi-parameter defect inversion method and perform empty field calibration on it.

[0019] To facilitate understanding of this solution by those skilled in the art, a set of capacitance imaging detection systems is further provided below as an example, and the structure and parameters of each component of the capacitance imaging detection system are described in detail below.

[0020] like Figure 2 As shown, the capacitance imaging detection system includes a single pair of coplanar electrode capacitance imaging sensors 1, a triaxial scanning unit 2, a signal conditioning unit 3, a data acquisition unit 4, and a control unit 5. In addition, depending on actual experimental requirements, a signal generator 6 and a lock-in amplifier 7 can be further added to the capacitance imaging detection system.

[0021] Specifically, the single-pair coplanar electrode capacitance imaging sensor 1, as the core component of the system, consists of two parts: an excitation electrode and a measurement electrode, as shown in the figure below. Figure 3 As shown in the figure. Both the excitation electrode and the measurement electrode are isosceles triangular copper electrodes, and the substrate is a polyimide sheet. The base of the excitation electrode and the measurement electrode are b=10mm, the height is c=10mm, the electrode spacing is a=5mm, and the electrode surface is gold-plated to reduce contact impedance and signal loss.

[0022] First, a defect-free flat plate test block with the same structural material as the test piece is selected as the calibration test piece. Then, it is fixed on the triaxial scanning unit 2, the lift-off height of the single pair of coplanar electrode capacitance imaging sensor 1 is adjusted to 1mm, and it is controlled to complete the 100mm×100mm area scanning in 1mm steps to collect the empty field charge signal and complete the entire empty field calibration process.

[0023] As a preferred embodiment of the present invention, the qualification criteria for the empty field calibration of the capacitance imaging detection system in step S101 are: the empty field background charge signal curve is stable, the amplitude fluctuation is ≤ ±0.5mV, and there is no clutter or baseline drift.

[0024] Step S102: Construct an electrostatic field model corresponding to a single pair of coplanar electrode capacitive imaging sensors; calculate and normalize to obtain a standardized sensitivity field matrix.

[0025] Based on completing step S101, step S102 is further implemented. In a preferred embodiment of the present invention, the process of calculating and normalizing the standardized sensitivity field matrix in step S102 is specifically described as follows: A two-dimensional electrostatic field model was constructed using finite element simulation software based on the actual geometric parameters of a single pair of coplanar electrode capacitive imaging sensors.

[0026] Specifically, a two-dimensional electrostatic field model was constructed at a 1:1 scale using finite element simulation software to fully reproduce the shape, size, and substrate structure of the excitation and measurement electrodes in a single-pair coplanar electrode capacitance imaging sensor. After constructing the two-dimensional electrostatic field model, the mesh around the excitation and measurement electrodes and the effective detection area was further refined, and an infinite element domain was set to simulate the open boundary to eliminate the influence of boundary effects. Specifically, the mesh element size for the electrode gap and electrode edge regions can be set to 0.1 mm, while the mesh element size for the effective detection area can be set to 0.5 mm, thus balancing solution accuracy and computational efficiency.

[0027] Dielectric perturbations (e.g., 5V excitation voltage) are sequentially applied to each micro-element within the detection region of the two-dimensional electrostatic field model; the sensitivity coefficient of each micro-element is derived using the point perturbation method.

[0028] Among them, the sensitivity coefficient of each infinitesimal element ,satisfy: ; In the formula, The vacuum permittivity, For excitation voltage, The two electrostatic field interfaces are located at the following positions. The electric field strength obtained by solving the problem (specifically, the two electrostatic fields obtained when voltages are applied to the excitation electrode and the measuring electrode respectively).

[0029] Then, the sensitivity coefficients of each infinitesimal element are linearly normalized to the [0,1] interval to obtain the standardized sensitivity field matrix. A reference diagram of the standardized sensitivity field matrix is ​​shown below. Figure 4 As shown.

[0030] Step S103: Establish a dielectric constant-capacitance mapping model; derive the quantitative discrete mapping relationship between defect parameters and charge signals.

[0031] Based on completing step S102, step S103 is further implemented. As a preferred embodiment of the present invention, the process of establishing the dielectric constant-capacitance mapping model in step S103 is specifically described as follows: The defective specimen is equivalent to a defect-free uniform substrate superimposed with a defect-negative polarization region; the detection area is discretized into several non-overlapping micro-elements.

[0032] It should be noted that the above methods limit the detection process to the range of low-frequency electrostatic field, and a three-dimensional rectangular coordinate system is established with the defect center as the origin, forming several non-overlapping micro-element detection areas (preferably, the detection area is discretized into cubic micro-elements of 1mm×1mm×0.1mm).

[0033] Further, based on the physical definition of the sensitivity field, a discrete mapping formula for the total charge perturbation is established, satisfying: ; In the formula, Q represents the total charge disturbance detected by a single pair of coplanar electrode capacitance imaging sensors, and N represents the total number of discrete infinitesimal elements in the detection region. For the first The sensitivity coefficient at each infinitesimal position. For the first The volume of a microelement; For the first The dielectric constant perturbation of each infinitesimal element satisfies: ; in, For the first The actual dielectric constant of a microelement is the uniform dielectric constant of a defect-free substrate.

[0034] Step S104: Scan the test piece to acquire the charge signal and record its spatial coordinates; preprocess the original charge map to obtain the real charge map.

[0035] Based on the completion of step S103, step S104 is further implemented. Specifically, the scanning process of the test piece can be described as follows: The test piece 8 containing artificial standard defects is stably fixed on the triaxial scanning unit 2; with the assistance of an industrial camera, the defect area of ​​the test piece 8 is aligned with the effective detection center of the single pair of coplanar electrode capacitance imaging sensor 1; then, the single pair of coplanar electrode capacitance imaging sensor 1 is controlled to perform surface scanning of the test piece 8 along the XY axis in 1mm steps, and the data acquisition unit 4 synchronously acquires charge signals at a frequency of 1kHz and records the corresponding spatial coordinates, finally generating a 33×33 original charge map, whose charge signals cover the defect area and the defect-free background area. The scanning method of the test piece can be referred to as follows. Figure 5 As shown.

[0036] The scanning method for the test piece can be described as follows: Figure 5 The X and Y axes are the motion coordinate axes of the three-axis scanning unit. Continuous arrow lines represent the complete reciprocating scanning path of the sensor. Specifically, a surface scanning mode of continuous scanning in the X-axis direction and interval stepping in the Y-axis direction is adopted. Several square artificial defects with gradient sizes of 4mm-15mm side length and 1mm step size are evenly distributed on the plane of the test piece. The dashed boxes are the positioning marks of each defect. The upper cross-sectional structure distinguishes the defect depth d and the burial depth bd. By superimposing glass fiber sheets of different thicknesses, multiple sets of burial conditions of 0mm, 0.5mm, 1mm, 1.5mm, and 2mm can be detected. The scanning trajectory completely covers the entire defect area and the surrounding defect-free substrate. After sampling at each coordinate point, the current spatial coordinates and the collected charge value are stored simultaneously. After all scans are completed, a regular grid data is formed, and a 33×33 original charge distribution map is output. In addition, to reduce random measurement noise, the same test piece is scanned three times, and the average value of multiple charge maps is calculated.

[0037] Then, the original charge map is preprocessed. In a preferred embodiment of the present invention, the process of preprocessing the original charge map in step S104 to obtain the true charge map is specifically described as follows: The charge signals in the original charge map are linearly normalized to the [0,1] interval; The normalized signal is filtered using a Gaussian kernel. Using the background signal acquired during field calibration as a reference, the filtered signal is subjected to background correction to obtain the true charge map.

[0038] Specifically, normalization can effectively eliminate amplitude fluctuations caused by excitation voltage fluctuations and uneven scanning speeds; filtering with a 5×5 Gaussian kernel can suppress random high-frequency noise and retain effective features in the charge signal; finally, background correction is performed using the background signal acquired by the empty field calibration as a reference to eliminate baseline interference in defect-free areas, ultimately obtaining a true charge map with high signal-to-noise ratio and clear features.

[0039] It is worth noting that, in this process, the calibration specimens are made of the same batch of sheet metal as the test specimens. The dimensions of the test specimens can be referenced as 300mm×200mm×5mm, and the test specimens are machined with several standard artificial defects to cover surface defects and buried defects. The defect depths include 0.6mm, 1mm, 1.4mm and 1.8mm. With different length, width and burial depth parameters, different types of defects commonly found in industrial sites can be covered.

[0040] Step S105: Preset initialization defect parameters; By combining the standardized sensitivity field matrix and the initialized defect parameters, the dielectric constant matrix is ​​generated. Based on the dielectric constant-capacitance mapping model established in step S103, a predicted charge map is generated through forward modeling.

[0041] Based on step S104, step S105 is further implemented. Specifically, step S105 generates a predicted charge map through forward modeling. The forward modeling process can be described as follows: using a standardized sensitivity field matrix as a rigid physical constraint, combined with preset initial defect parameters such as defect length, width, depth, and burial depth, a dielectric constant matrix corresponding one-to-one with discrete micro-elements in the detection area is constructed, and the dielectric constant perturbation value of each micro-element is clearly obtained; then, based on the dielectric constant-capacitance mapping relationship constrained by the sensitivity field, the charge contribution is calculated point by point according to the single micro-element polarization charge increment formula, and the charge contribution of the micro-elements in the entire detection area is volume-weighted and superimposed, and solved through the forward problem of electromagnetic tomography, finally generating a predicted charge map with the same spatial dimension, coordinate system, and resolution as the real charge.

[0042] It should be noted that the construction rules of the dielectric constant matrix in this process are as follows: based on the preset defect parameters, the spatial position and range of the defect region in the detection grid are determined, and the corresponding dielectric constant perturbation value is assigned to the micro-elements in the defect region; among them, the dielectric constant perturbation value of the micro-elements in the defect-free region is 0, thereby ensuring that the dielectric constant matrix is ​​completely matched with the spatial distribution of the actual defect.

[0043] Furthermore, it is worth noting that the formula for calculating the polarization charge increment of a single infinitesimal element is: ;in, Let be the polarization charge increment of the i-th infinitesimal element. After calculating the charge contribution of each infinitesimal element using this formula, the charge contributions of all infinitesimal elements in the entire detection area are volume-weighted and superimposed to obtain the predicted charge value corresponding to each scanning position. Finally, by stitching them together, a predicted charge map that matches the actual charge can be formed.

[0044] Step S106: Calculate the matching degree between the predicted charge map and the actual charge map; With the goal of maximizing the matching degree between the predicted charge map and the actual charge map, the defect parameters are iteratively updated until the matching degree reaches a preset threshold, and the current defect parameters are output as the optimal solution for defect inversion. Among them, the defect parameters include at least the length, width, depth and burial depth of the defect.

[0045] Based on step S105, step S106 is further implemented. Specifically, before calculating the matching degree between the predicted charge map and the real charge map, the convolutional neural network used in the iterative update process of the defect parameters is explained as follows. A CNN model convolutional neural network is selected, using a 33×33 real charge map as input and outputting four parameters: defect length, width, depth, and burial depth. The specific structure of this convolutional neural network can be described as follows: three convolutional layers, two max-pooling layers, two fully connected layers, and one output layer are set sequentially. Specifically, the convolutional kernel sizes are 3×3, 3×3, and 2×2; the number of output channels are 32, 64, and 128 respectively; the activation function is ReLU; the pooling layer kernel size is 2×2; the number of neurons in the fully connected layer is 256 and 128 respectively, with a Dropout coefficient of 0.3 to prevent overfitting; the output layer has four neurons, corresponding to the defect length, width, depth, and burial depth parameters respectively; and the optimizer is Adam.

[0046] The training and validation process of the convolutional neural network is further described as follows: First, the sample dataset used in the training process is determined (this dataset should contain simulation and experimental samples of known standard defects). The total sample size is 2256 groups, including 1680 simulation samples and 768 experimental samples. The defect parameters in the samples cover lengths of 1-15mm, widths of 1-15mm, depths of 0.6-2mm, and burial depths of 0-2mm, covering common defect types in industrial settings. During training, the initial learning rate is set to 0.001, the batch size is 16, and the mean squared error (MSE) is less than 1×10⁻⁶. -4 To train the convergence conditions, a stable and convergent CNN model is obtained after training.

[0047] After training, the trained convolutional neural network is used to predict defect features based on the real charge map. After obtaining the corresponding predicted defect feature parameters, the predicted charge map can be generated through forward modeling, and the matching degree between the predicted charge map and the real charge map can be further calculated.

[0048] In a preferred embodiment of the present invention, the formula for calculating the matching degree between the predicted charge map and the actual charge map satisfies: ; in, To predict the charge value of the charge map, The charge values ​​are those of the actual charge map.

[0049] It is worth noting that in this process, the matching degree between the predicted charge map and the real charge map is used as the optimization objective to iteratively optimize until the matching degree reaches the preset threshold requirement, and the current defect parameters are output as the optimal solution for defect inversion.

[0050] Further assuming a preset threshold for the matching degree between the predicted charge map and the true charge map is 0.95; if the matching degree does not reach the preset threshold, the current error is fed back to the convolutional neural network, the preset defect parameters are updated, forward modeling is re-executed to generate a new predicted charge map, and the matching degree is checked again, thus forming a closed-loop iterative process of "forward modeling - reverse matching - error feedback - parameter update" until the matching degree reaches the threshold, and the defect inversion result of the optimal solution is output. References are provided below. Figure 6 As shown, Figure 6 The diagram shows a comparison of the simulation results, experimental results, and predicted charge distribution images. The comparison reveals that the matching degree between the simulated and predicted charge images is 0.9904, and the matching degree between the experimental and predicted charge images is 0.9899. Both exceed the preset threshold of 0.95, indicating that the matching accuracy meets the standard. Further inversion calculations confirm that the geometric parameters of both the simulated and experimental defects under this condition match the preset defect parameters: length 10 mm, width 10 mm, depth 1 mm, and burial depth 0 mm.

[0051] Thus, the multi-parameter defect inversion method with capacitance imaging sensitivity field constraint provided by this invention achieves synchronous, rapid, and high-precision inversion of multiple defect parameters under the physical constraint of the sensitivity field. Its inversion results have strong anti-interference ability and stable and reliable results, and are suitable for non-destructive quantitative detection of defects in insulation structures in industrial fields.

[0052] This invention provides a multi-parameter defect inversion method constrained by a capacitance imaging sensitivity field. The method includes the following steps: Step S101: Constructing the capacitance imaging detection system required for the multi-parameter defect inversion method and performing empty-field calibration; Step S102: Constructing an electrostatic field model corresponding to a single pair of coplanar electrode capacitance imaging sensors; calculating and normalizing to obtain a standardized sensitivity field matrix; Step S103: Establishing a dielectric constant-capacitance mapping model; deriving the quantitative discrete mapping relationship between defect parameters and charge signals; Step S104: Scanning the test piece, acquiring charge signals and recording their spatial coordinates; processing the original charge map... Preprocessing is performed to obtain the true charge map; Step S105: Preset initialization defect parameters; Combine the standardized sensitivity field matrix with the initialization defect parameters to generate the dielectric constant matrix; Based on the dielectric constant-capacitance mapping model established in step S103, the predicted charge map is generated through forward modeling; Step S106: Calculate the matching degree between the predicted charge map and the true charge map; With the goal of maximizing the matching degree between the predicted charge map and the true charge map, the defect parameters are iteratively updated until the matching degree reaches the preset threshold, and the current defect parameters are output as the optimal solution for defect inversion.

[0053] The multi-parameter inversion method for capacitance imaging sensitivity field constraint with the above-mentioned step characteristics has at least the following advantages compared with the prior art; (1) The multi-parameter inversion method for defects with capacitance imaging sensitivity field constraint provided by the present invention significantly reduces the ill-posedness of defect quantification and improves the identification accuracy of buried defects and micro-defects by introducing the physical constraint of sensitivity field into the inversion process. (2) The multi-parameter inversion method for defects with capacitance imaging sensitivity field constraint provided by the present invention realizes the synchronous inversion of multiple parameters such as defect length, width, depth and burial depth, without the need for multiple detections or complex modeling, and the detection efficiency is higher. (3) The multi-parameter inversion method for the capacitance imaging sensitivity field constraint provided by the present invention combines forward modeling and reverse matching closed-loop iteration with neural network optimization solution, thus taking into account both physical reliability and algorithm efficiency, and has a stronger industrial adaptability.

[0054] 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 multi-parameter inversion method for defects in capacitance imaging sensitivity field constraint, characterized in that, The steps include the following: Step S101: Build the capacitance imaging detection system required for the defect multi-parameter inversion method and perform empty field calibration on it; wherein, the capacitance imaging detection system includes a single pair of coplanar electrode capacitance imaging sensors, a triaxial scanning unit, a signal conditioning unit, a data acquisition unit and a control unit; Step S102: Construct the electrostatic field model corresponding to a single pair of coplanar electrode capacitive imaging sensors; calculate and normalize to obtain the standardized sensitivity field matrix; Step S103: Establish a dielectric constant-capacitance mapping model; derive the quantitative discrete mapping relationship between defect parameters and charge signals; Step S104: Scan the test piece to acquire the charge signal and record its spatial coordinates; preprocess the original charge map to obtain the true charge map; Step S105: Preset initialization defect parameters; By combining the standardized sensitivity field matrix and the initialized defect parameters, the dielectric constant matrix is ​​generated. Based on the dielectric constant-capacitance mapping model established in step S103, a predicted charge map is generated through forward modeling. Step S106: Calculate the matching degree between the predicted charge map and the actual charge map; With the goal of maximizing the matching degree between the predicted charge map and the actual charge map, the defect parameters are iteratively updated until the matching degree reaches a preset threshold, and the current defect parameters are output as the optimal solution for defect inversion. Among them, the defect parameters include at least the length, width, depth and burial depth of the defect.

2. The defect multi-parameter inversion method for capacitance imaging sensitivity field constraint according to claim 1, characterized in that, The qualification criteria for the empty field calibration of the capacitance imaging detection system in step S101 are: the empty field background charge signal curve is stable, the amplitude fluctuation is ≤ ±0.5mV, and there is no clutter or baseline drift.

3. The defect multi-parameter inversion method for capacitance imaging sensitivity field constraint according to claim 1, characterized in that, The process of calculating and normalizing the standardized sensitivity field matrix in step S102 is specifically described as follows: A two-dimensional electrostatic field model was constructed using finite element simulation software based on the actual geometric parameters of a single pair of coplanar electrode capacitive imaging sensors. Dielectric perturbations are sequentially applied to each micro-element within the detection region of the two-dimensional electrostatic field model; The sensitivity coefficients of each infinitesimal element are derived using the point perturbation method; where, the sensitivity coefficients of each infinitesimal element... ,satisfy: ; In the formula, The vacuum permittivity, For excitation voltage, , The two electrostatic field interfaces are located at the following positions. The electric field strength obtained by solving at point; The sensitivity coefficients of each infinitesimal element are linearly normalized to the [0,1] interval to obtain the standardized sensitivity field matrix.

4. The multi-parameter inversion method for capacitance imaging sensitivity field constraint according to claim 1, characterized in that, The process of establishing the dielectric constant-capacitance mapping model in step S103 is specifically described as follows: The defective specimen is equivalent to a defect-free uniform substrate superimposed with a defect negative polarization region; The detection area is discretized into several non-overlapping micro-elements; Based on the physical definition of the sensitivity field, a discrete mapping formula for the total charge perturbation is established, satisfying: ; In the formula, Q represents the total charge disturbance detected by a single pair of coplanar electrode capacitance imaging sensors, and N represents the total number of discrete infinitesimal elements in the detection region. For the first The sensitivity coefficient at each infinitesimal position. For the first The volume of a microelement; For the first The dielectric constant perturbation of each infinitesimal element satisfies: ; in, For the first The actual dielectric constant of a microelement is the uniform dielectric constant of a defect-free substrate.

5. The defect multi-parameter inversion method for capacitance imaging sensitivity field constraint according to claim 1, characterized in that, The process of preprocessing the original charge map to obtain the real charge map in step S104 is specifically described as follows: The charge signals in the original charge map are linearly normalized to the [0,1] interval; The normalized signal is filtered using a Gaussian kernel. Using the background signal acquired during field calibration as a reference, the filtered signal is subjected to background correction to obtain the true charge map.

6. The defect multi-parameter inversion method for capacitance imaging sensitivity field constraint according to claim 1, characterized in that, The formula for calculating the matching degree between the predicted charge map and the actual charge map in step S106 satisfies: ; in, To predict the charge value of the charge map, The charge values ​​are those of the actual charge map.

Images