System and method for detecting defects in CFRP based on phase change of superimposed magnetic field

Abstract

The application discloses a CFRP defect detection system and method based on superimposed magnetic field phase change, belongs to the technical field of nondestructive testing, and comprises a scanning terminal, an acquisition module and a defect identification module; the scanning terminal is used for point-by-point scanning of the CFRP of a detection area, and acquiring reflected polarized light after superimposed magnetic field modulation of the CFRP surface; the acquisition module is used for receiving the reflected polarized light and converting the reflected polarized light into a differential mode electric signal, simultaneously acquiring real-time working temperature, extracting an effective signal of the differential mode electric signal, generating a phase matrix and an amplitude matrix; the defect identification module is used for offline calibration of weight and defect mapping, constructing a magnetic field characteristic matrix through temperature drift correction and weighted fusion, adaptively performing preliminary detection on a defect area through the corrected phase matrix, identifying the defect area, performing multi-directional phase gradient fusion on the defect area, combining the magnetic field characteristic matrix, generating a defect imaging graph, inversely calculating defect depth in combination with offline calibration of the defect mapping, and generating a defect detection package.

Description

Technical Field

[0001] This invention relates to a CFRP defect detection system and method based on the phase change of superimposed magnetic fields, belonging to the field of nondestructive testing technology. Background Technology

[0002] Carbon fiber reinforced polymer (CFRP) composites, as a new type of composite material, have become the preferred material for structural components in aerospace, automotive manufacturing, and wind power generation due to their excellent properties such as high strength, high modulus, fatigue resistance, and corrosion resistance. Eddy current magneto-optical imaging (EMI) testing technology, based on the laws of electromagnetic induction and the Faraday magneto-optical effect, induces eddy currents on the surface of the material under test through an excitation device. The magneto-optical sensor converts the magnetic field distribution disturbance caused by defects into changes in light intensity, thereby realizing the visualization of defects. Eddy current MPI testing has become a reliable method for non-destructive testing of CFRP defects because the probe does not need to contact the specimen and has high sensitivity to surface and near-surface defects.

[0003] However, eddy current magneto-optical imaging technology has limitations in its suitability for detecting weakly conductive materials. Specifically, due to the anisotropy and overall weak conductivity of CFRP, the intensity of the induced eddy currents inside it under alternating excitation is much lower than that of metallic materials. The magnetic field amplitude disturbance caused by defects is extremely weak, and the light intensity signal output by the magneto-optical sensor is easily submerged by various noises. This not only requires a significant increase in the excitation current intensity to improve the magnetic field response, but also results in a large excitation coil and severe equipment overheating, making it difficult to achieve portable testing in industrial settings. Furthermore, it is difficult to obtain standard defect-free CFRP samples in actual engineering scenarios, and the reference signal is easily affected by environmental factors such as temperature drift and lift-off fluctuations, resulting in poor detection robustness. In addition, amplitude imaging is highly dependent on the defect direction and cannot simultaneously detect cracks, delamination, and impact defects with different orientations, posing a significant risk of missed detections. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a CFRP defect detection system and method based on superimposed magnetic field phase change. The system uses the phase signal as the core detection quantity, compensates for temperature drift and local phase relative change rate to offset lift-off fluctuation interference, eliminates defect orientation dependence by multi-directional gradient adaptive fusion, and achieves quantitative detection of standard-free defects by combining an offline mapping library.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The CFRP defect detection system based on superimposed magnetic field phase change includes: a scanning terminal, a data acquisition module, and a defect identification module;

[0007] The scanning terminal is used to scan the CFRP in the detection area point by point to obtain the reflected polarized light modulated by the superimposed magnetic field on the CFRP surface.

[0008] The acquisition module is used to receive the reflected polarized light and convert it into a differential mode electrical signal, while acquiring the real-time operating temperature. Through effective signal extraction of the differential mode electrical signal, a phase matrix and an amplitude matrix are generated.

[0009] The defect identification module is used to perform offline calibration of weights and defect mappings. By correcting the phase matrix for temperature drift and performing weighted fusion with the amplitude matrix, a magnetic field feature matrix is ​​constructed. By performing adaptive initial detection of defect regions on the corrected phase matrix, defect regions are identified. Multi-directional phase gradient fusion is performed on the defect regions. Combined with the magnetic field feature matrix, a defect imaging map is generated. Combined with the offline calibration of the defect mapping, the defect depth is inverted to generate a defect detection package.

[0010] Specifically, the defect identification module includes a feature extraction unit, a defect detection unit, and a quantitative inversion unit;

[0011] The feature extraction unit is used for the three-dimensional simulation model. By performing multi-condition simulation of the three-dimensional simulation model with control variables, a weight matrix combining phase and amplitude is constructed. By quantitatively fitting the phase change rate with the defect depth, a defect mapping library is constructed. Based on the standard temperature and the real-time working temperature, the phase matrix is ​​corrected for temperature drift point by point. Combined with the weight matrix and amplitude matrix, a magnetic field feature matrix is ​​constructed.

[0012] The defect detection unit is used to complete the edge of the phase matrix, calculate the phase change rate point by point with the sliding window, identify the defect region by configuring the anomaly judgment threshold and connected component filtering through local statistics, perform multi-directional phase gradient fusion on the defect region, and generate a defect imaging map by combining the magnetic field feature matrix.

[0013] The quantitative inversion unit is used to extract the average phase change rate of the defect region in the defect image, invert the defect depth by matching with the defect mapping library, and identify the defect type by combining the position and contour size of the defect in the defect image to generate a defect detection package.

[0014] Specifically, generating the phase matrix and magnitude matrix includes:

[0015] The differential-mode electrical signal is subjected to bandpass filtering to generate the signal to be tested;

[0016] Configure a reference signal for the lock-in amplifier demodulation circuit, and separate the in-phase reference signal and the quadrature reference signal;

[0017] The signal under test is multiplied with the in-phase reference signal and the quadrature reference signal respectively, and then low-pass filtered to generate in-phase components and quadrature components. The original amplitude and original phase of the corresponding scanning point are then calculated.

[0018] Read the spatial coordinates of each scanning point and the real-time operating temperature of the magneto-optical sensor in the scanning terminal, combine them with the original phase and original amplitude, and add a unified timestamp to form a single-point data packet;

[0019] Based on the scanning points in the scanning area, a grid coordinate system is constructed, and according to the single scanning cycle, point data packets are filled in at the corresponding grid positions.

[0020] Based on the row and column order of the grid, the original phase and original amplitude of all scan points are assembled separately to form the phase matrix and amplitude matrix under the same scan period.

[0021] Specifically, constructing the joint phase and magnitude weight matrix includes:

[0022] Using the controlled variable method, the lift-off degree of the magneto-optical sensor is used as the first-level controlled variable, and the defect parameter group, including defect type, defect direction, defect depth, and defect size, is used as the second-level controlled variable to divide the simulation conditions.

[0023] The three-dimensional simulation model is solved to obtain the magnetic field phase and magnetic field amplitude corresponding to all scanning points in the detection area under each simulation condition;

[0024] For the distribution of magnetic field phase and magnetic field amplitude under the same simulation conditions, calculate the phase sensitivity and amplitude sensitivity of each scanning point;

[0025] The phase sensitivity and amplitude sensitivity of each scan point are normalized, and the phase weight and amplitude weight of each scan point are calculated.

[0026] Based on the row and column arrangement of the grid coordinate system, a weight matrix combining phase and amplitude is constructed for each simulation condition.

[0027] Specifically, the steps for building a defect mapping library include:

[0028] The magnetic field phase and magnetic field amplitude of the full simulation working condition are obtained, classified and collected according to the defect type, and the phase distribution characteristics of the corresponding detection area are extracted. The clustering algorithm is used to screen typical features, and the Fisher discriminant method is used to perform cross-type verification of the typical characteristics to generate feature templates for each defect type.

[0029] Based on the defective and non-defective regions divided by the three-dimensional simulation model, the average phase of the defective region and the average phase of the non-defective region are obtained, and the phase change rate under the corresponding working conditions is calculated.

[0030] Summarize the phase change rate under the same defect type, and construct mapping data pairs based on the phase change rate and defect depth under each working condition;

[0031] Using the nonlinear least squares method, with the phase change rate as the independent variable and the defect depth as the dependent variable, a quantitative mapping function is fitted for the corresponding defect type.

[0032] Defect types, corresponding feature templates, and quantitative mapping functions are associated and stored to construct a defect mapping library.

[0033] Specifically, the steps for constructing the magnetic field feature matrix include:

[0034] Obtain the offline pre-calibrated calibration curve containing the temperature of the magneto-optical material and the Faraday rotation angle, and combine it with the configured standard temperature to obtain the reference rotation angle;

[0035] During online testing, the real-time operating temperature of each scanning point is substituted into the calibration curve to read the real-time rotation angle;

[0036] The rotation offset is obtained by the difference between the real-time rotation angle and the reference rotation angle. Combined with the original phase of the corresponding scan point, temperature drift compensation is performed to obtain the phase of each scan point after temperature compensation.

[0037] Based on joint weights, the magnitude matrix and the corrected phase matrix are summed point by point to construct the magnetic field characteristic matrix.

[0038] Specifically, the adaptive initial detection of the defect area includes:

[0039] Set up a square sliding window, with a single scan point as the sliding step size, and translate position by position along the row and column directions of the grid coordinate system. Set the scan point at the center of the window as the point to be detected.

[0040] The edges of the phase matrix are extended and completed using the mirror filling method to generate an extended phase matrix.

[0041] For any point to be detected, the central detection area and the local background area are divided. Based on the local background area, a statistical sample is set, and the anomaly judgment threshold is set using the 3σ criterion.

[0042] Calculate the mean background phase of the local background region, and combine it with the phase of the point to be detected to calculate the phase change rate of the point to be detected;

[0043] If the phase change rate of the point to be detected is greater than the anomaly determination threshold, it is marked as a defect candidate point; otherwise, it is a background point.

[0044] Traverse the scan points in the phase matrix to construct an initial defect set, and construct candidate connected regions using the 8-neighborhood connectivity rule, with the number of scan points in the candidate connected regions being the number of pixels;

[0045] Set a connectivity threshold and combine it with the number of pixels to filter out defective areas.

[0046] Specifically, the multi-directional phase gradient fusion includes:

[0047] Configure the gradient direction and use the Sobel gradient operator to calculate the phase gradient magnitude of each gradient direction point by point;

[0048] The mean gradient of the region is calculated by the mean of the phase gradient magnitude in a single gradient direction, and the fusion weight of each gradient direction is calculated based on the inverse weighting and normalization process.

[0049] For each scan point within the defect region, gradient-weighted fusion is performed using the phase gradient magnitude in each gradient direction and the fusion weight to obtain the fused gradient features of the scan point.

[0050] The fusion gradient features of all defect regions are summarized, and the optimal segmentation threshold is calculated using Otsu's method.

[0051] The fused gradient features of each scan point are compared with the optimal segmentation threshold to divide defect pixels and background pixels, generate an initial binary mask, and then combine morphological closing and opening operations to obtain the defect binary mask.

[0052] Using the magnetic field feature matrix as a grayscale base, the defect binary mask is superimposed and mapped onto the grid coordinate system to generate a defect image.

[0053] Specifically, the steps for generating a defect detection package include:

[0054] Extract the boundary coordinates of the defect region and calculate the center coordinates of the defect as the coordinate position of the defect;

[0055] Extract the contour boundary of the defect and calculate its shape and size parameters;

[0056] Using a single defect region as the target region, a background reference region is generated by expanding outward from the target region, and the phase change of the target region is calculated.

[0057] Extract the phase distribution features of the target region, call the feature templates of various defects in the defect feature mapping library, use cosine similarity to calculate the similarity between the defects in the target region and each feature template, and take the defect type with the highest similarity as the recognition result.

[0058] Based on the identified defect type, the corresponding quantitative mapping function is called from the defect feature mapping library. The phase change rate of the target area is substituted into the quantitative mapping function to solve for the defect depth of the target area.

[0059] For each individual defect, the coordinate position, defect type, defect depth, shape and size parameters, and contour boundary are bound together to generate a single defect detection package.

[0060] CFRP defect detection methods based on superimposed magnetic field phase changes include:

[0061] A three-dimensional simulation model is established, and a weight matrix is ​​constructed by combining multi-condition simulation of control variables.

[0062] A quantitative correspondence between the phase change rate and the defect depth was fitted to construct a defect mapping library;

[0063] To obtain the reflected polarized light from the CFRP surface and the real-time operating temperature of the magneto-optical sensor;

[0064] The reflected polarized light is converted into a differential mode electrical signal. The phase matrix and amplitude matrix are constructed using the effective signal at the target frequency. The magnetic field feature matrix is ​​constructed by temperature drift correction and weighted fusion.

[0065] The phase matrix after temperature drift correction is used to perform adaptive initial detection of the defect region, identify the defect region, and perform multi-directional phase gradient fusion on the defect region. Combined with the magnetic field feature matrix, a defect image is generated.

[0066] Based on the matching of defect imaging images and defect mapping libraries, defect depth is inverted, and defect category identification is combined to generate a defect detection package.

[0067] The beneficial effects of this invention are:

[0068] Using the phase signal, which is more sensitive to changes in eddy current depth, as the core detection quantity, and combining it with the amplitude signal through point-by-point weighted fusion to construct a magnetic field feature matrix, this method retains the high defect sensitivity of the phase signal while enhancing imaging contrast with amplitude information. Since it does not rely on significantly increasing the excitation current to strengthen the magnetic field response, it avoids the problems of increased coil size and equipment overheating caused by large excitation, supporting the engineering implementation of portable testing equipment and adapting to the mobile testing needs of industrial sites. Simultaneously, it completes point-by-point temperature drift compensation of the phase signal through offline pre-calibrated temperature calibration curves, and uses the relative phase change rate of local background statistics for defect judgment, using the local background area around the defect as a reference instead of the global standard defect-free sample. In principle... It offsets global signal shifts caused by temperature drift and lift-off fluctuations, eliminating dependence on standard samples and improving the stability and robustness of detection results under complex working conditions. Addressing the defect orientation dependence and missed detection risk caused by CFRP anisotropy, it uses multi-directional phase gradient calculation and reverse adaptive weighted fusion to smooth out differences in detection response intensity for defects with different orientations, eliminating directional sensitivity bias caused by fiber orientation. This allows for the uniform detection of various defects such as cracks, delamination, and impact damage with different orientations, reducing the probability of missed detection. Combined with an offline pre-built defect feature mapping library, it enables defect type identification and depth quantitative inversion without on-site calibration, comprehensively improving the engineering applicability and detection reliability of CFRP magneto-optical imaging detection. Attached Figure Description

[0069] Figure 1 This is a structural diagram of a CFRP defect detection system based on the phase change of a superimposed magnetic field;

[0070] Figure 2 This is a flowchart of constructing the weight matrix in this invention;

[0071] Figure 3 This is a flowchart of multi-directional phase gradient fusion in this invention;

[0072] Figure 4 This is a flowchart of a CFRP defect detection method based on the phase change of superimposed magnetic fields. Detailed Implementation

[0073] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0074] Example 1

[0075] refer to Figures 1 to 3 As shown, this embodiment introduces a CFRP defect detection system based on the phase change of a superimposed magnetic field, including:

[0076] Scanning terminal, data acquisition module, and defect identification module;

[0077] The scanning terminal includes a low-current excitation component and a magneto-optical sensor. It scans the CFRP within the detection area point by point through a two-dimensional displacement platform to acquire the reflected polarized light modulated by the superimposed magnetic field on the CFRP surface. The low-current excitation component has a built-in 50mA constant current excitation source. It applies a stable alternating excitation magnetic field to the CFRP through a hollow cylindrical excitation coil, which induces eddy currents inside the CFRP and forms a superimposed magnetic field. The superimposed magnetic field is a vector superposition field in space between the alternating excitation magnetic field and the induced magnetic field generated by the induced eddy currents inside the CFRP. The magneto-optical sensor then collects the reflected polarized light modulated by the superimposed magnetic field on the CFRP surface based on the Faraday magneto-optical effect.

[0078] The acquisition module is used to acquire reflected polarized light from the CFRP surface. It uses a polarization beam splitter to split the light and a differential balanced photodetector to suppress common-mode noise and ambient light interference in the optical path. The reflected polarized light is converted into a differential-mode electrical signal, and the real-time operating temperature of the magneto-optical sensor is acquired. The temperature is then timestamped with the differential-mode electrical signal. Based on the lock-in amplifier demodulation circuit, the effective signal of the target frequency is extracted from the differential-mode electrical signal, which is the time-domain phase difference relative to the alternating excitation current reference signal. The time-domain phase difference carries the time-domain phase information of the superimposed magnetic field, thereby reflecting the phase distribution characteristics of the eddy current field inside the CFRP and generating the corresponding phase matrix and amplitude matrix of the CFRP.

[0079] Furthermore, generating the phase matrix and magnitude matrix corresponding to the CFRP includes:

[0080] The differential-mode electrical signal, which is collected point by point by the scanning terminal and output by the beam splitter and differential balanced photodetector, is sent to a bandpass filter circuit with the center frequency matching the frequency of the alternating excitation current. The filter operation removes the DC component, out-of-band high-frequency noise, and low-frequency ambient light drift noise, resulting in the filtered differential-mode electrical signal, which is used as the demodulated signal to be measured. The current flowing through the hollow cylindrical excitation coil is the alternating excitation current.

[0081] A synchronous reference signal for the excitation source is synchronously extracted from a 50mA constant current excitation source. This signal has the same frequency and initial phase as the alternating excitation current supplied to the hollow cylindrical excitation coil and serves as the reference signal for the lock-in amplifier demodulation circuit.

[0082] The reference signal is divided into two paths. One path is directly used as the in-phase reference signal, which is in phase with the alternating excitation current. The other path is processed by a 90° phase shifting circuit and used as the quadrature reference signal. The two reference signals have the same frequency and a fixed phase difference of 90°.

[0083] The signal under test is multiplied by the in-phase reference signal and the quadrature reference signal, respectively. Then, the high-frequency harmonic components are filtered out by a low-pass filter circuit to obtain the in-phase component and the quadrature component. The original amplitude and original phase of the corresponding scanning point are then calculated. The expressions are as follows:

[0084]

[0085]

[0086] In the formula, For in-phase components, For orthogonal components, , These are the original amplitude and the original phase, respectively. The original amplitude is the vector magnitude of the in-phase component and the quadrature component, and the original phase is the time-domain phase difference between the signal under test and the alternating excitation current reference signal.

[0087] Using the zero-crossing point of the alternating excitation current as the global synchronous trigger clock, at the same trigger moment when the amplitude and phase calculations are completed at each scanning point, the spatial coordinates of the displacement encoder of the scanning terminal and the real-time operating temperature of the magneto-optical sensor are read synchronously. The original phase, original amplitude, spatial coordinates, and real-time operating temperature of the corresponding scanning point are bound together and a unified timestamp is added to form a single-point data packet.

[0088] When the scanning terminal completes traversing all scanning points within the scanning area, a single scanning cycle is completed. A two-dimensional rectangular grid coordinate system is established along the scanning row and column directions with the upper left corner vertex of the detection area as the origin. The row and column spacing of the grid is equal to the scanning step length. The total number of grid rows and columns is obtained by dividing the length and width of the detection area by the scanning step length and then rounding up. The scanning step length is set according to the minimum detectable defect size, taking 1 / 5 to 1 / 3 of the minimum detectable defect size to ensure that the defect covers at least 3 to 5 grid points, avoiding defect omissions due to insufficient sampling density. For example, if the preset detection area size is 100mm × 100mm, the minimum detectable defect size is 1mm, and the scanning step length is 0.2mm (1 / 5 of the minimum defect size), then the total number of grid rows and columns is 500, corresponding to a total of 250,000 scanning points.

[0089] Single-point data packets are mapped one by one to the corresponding grid positions according to spatial coordinates. Due to mechanical motion errors, rounding up edge grids, and occasional sampling loss, some grid positions have no measured sampling data, which produces grid gaps. For all missing grids, bilinear interpolation is used to calculate and fill the gap value using the values ​​of the surrounding adjacent measured points.

[0090] Based on the row and column order of the grid, the original phase and original amplitude of all scan points are assembled separately to form the phase matrix and amplitude matrix under the same scan period.

[0091] The defect identification module generates a weight matrix and a defect mapping library through offline pre-construction. It constructs a magnetic field feature matrix by performing temperature drift correction and weighted fusion using the phase matrix, amplitude matrix, and real-time operating temperature. It identifies defect areas through adaptive initial detection of CFRP defect areas, performs multi-directional phase gradient fusion on the defect areas, and generates a defect image by combining the magnetic field feature matrix. It also inverts the defect depth by combining the offline calibration of the defect mapping and generates a defect detection package.

[0092] The defect identification module includes a feature extraction unit, a defect detection unit, and a quantitative inversion unit;

[0093] The feature extraction unit is used to establish a three-dimensional simulation model including the CFRP laminate structure and the configuration of the scanning terminal. By performing multi-condition simulation with controlled variables on the three-dimensional simulation model, it simulates the phase and amplitude changes of the superimposed field of all scanning points under different lift-off degrees and different defect parameter groups. It calculates the sensitivity of phase and amplitude under the same conditions to construct a weight matrix. At the same time, it fits the quantitative correspondence between the phase change rate and the defect depth to construct a defect mapping library. Based on the standard temperature and the real-time working temperature, it performs point-by-point temperature drift correction on the phase matrix. Combined with the weight matrix, it performs weighted fusion of the corrected phase and amplitude to amplify the defect features under weak magnetic fields and construct a magnetic field feature matrix.

[0094] The defect detection unit performs adaptive initial detection of the defect region, completes the edge of the phase matrix after temperature drift correction to eliminate boundary traversal blind spots, calculates the phase change rate point by point using a sliding window to eliminate lift-off fluctuation interference, identifies the defect region by configuring anomaly judgment threshold and connected component filtering through local statistics, calculates the gradient magnitude through the configured gradient direction, performs multi-directional phase gradient fusion on the defect region to eliminate the defect orientation dependence caused by CFRP anisotropy, and generates a defect imaging map by combining the magnetic field feature matrix.

[0095] The quantitative inversion unit is used to extract the average phase change rate of the defect region in the defect image. By matching it with the defect mapping library, the defect depth is inverted. Combined with the location and contour size of the defect in the defect image, the defect type is identified to generate a defect detection package.

[0096] Furthermore, the steps for constructing the weight matrix include:

[0097] Based on the laminated structure of alternating fiber layers and resin matrix layers in CFRP, virtual components are constructed. According to the material manual of the corresponding CFRP, the orthogonal anisotropic conductivity tensor and single-layer thickness of the fiber layer, the bulk conductivity and single-layer thickness of the resin matrix layer are set respectively. Multi-layer stacking is completed according to the layup angle to completely restore the layered conductive structure of the material.

[0098] Using a three-dimensional electromagnetic field finite element simulation tool, such as COMSOL simulation software, a virtual component is imported and configured with excitation parameters consistent with the scanning terminal, including the geometric dimensions, number of turns, amplitude and frequency of the 50mA alternating excitation current of the hollow cylindrical excitation coil. The vertical distance between the lower surface of the magneto-optical sensor's sensitive film and the upper surface of the CFRP component is used as the lift-off degree. The detection plane position is set to be aligned with the actual detection surface of the magneto-optical sensor. The air domain boundary conditions and time-harmonic solver configuration are completed to form a parameterizable three-dimensional simulation model.

[0099] Using the controlled variable method, the lift-off of the magneto-optical sensor is used as the primary controlled variable, and a set of defect parameters (defect type, defect direction, defect depth, and defect size) are used as secondary controlled variables to define the simulation conditions. Specifically, based on the mechanical mounting clearance of the scanning terminal and the surface flatness tolerance range of the CFRP, the lift-off fluctuation range is set. Within this range, several discrete values ​​are taken at equal intervals. At each fixed lift-off, the defect type covers cracks, delamination, and impact damage. The defect direction is taken at equal angular intervals within the 0°~90° range to adapt to the detection direction sensitivity caused by the anisotropy of CFRP. The defect depth is taken according to the CFRP layup position, from the surface layer to the inner interlayer, according to the layup intervals to match the defect distribution characteristics of the laminated structure. The defect size is taken at equal intervals according to the equivalent length (crack) or equivalent diameter (delamination, impact damage). The parameter range covers common defect specifications of CFRP in various fields such as aerospace and wind power.

[0100] For example, the lift-off fluctuation range is set to 0.5mm~1.5mm, and values ​​are taken at 0.5mm intervals to obtain a total of 3 discrete lift-off values. The defect types are selected as crack, delamination and impact damage. The defect orientation is selected as 0°, 45° and 90° to cover the core scenarios of parallel, oblique and perpendicular to the fiber direction. The defect depth is selected as two positions: surface and half plate thickness. The defect size is selected as two typical specifications: small equivalent and conventional equivalent. Each combination of parameters of lift-off, defect type, defect orientation, defect depth and defect size corresponds to one independent simulation condition.

[0101] For each simulation condition, the corresponding lift-off degree and defect parameter set are imported into the three-dimensional simulation model. Defects conforming to geometric parameters are implanted at the corresponding positions of the virtual components. The solver is driven to perform time-harmonic electromagnetic field calculation. After the solution is completed, the magnetic field phase and magnetic field amplitude corresponding to all scanning points in the detection area under each simulation condition are obtained.

[0102] For the distribution of magnetic field phase and magnetic field amplitude under the same simulation conditions, the phase sensitivity and amplitude sensitivity of each scanning point are calculated, including: based on the defect-free reference simulation under the same lift-off degree, the reference phase, reference amplitude, and background noise standard deviation of each scan under the defect-free state are obtained; based on the defect geometric parameters implanted in the three-dimensional simulation model, the defect attribution of each scanning point is obtained, and defect-free and defect-free regions are divided; the ratio of the phase difference between the defective and defect-free scanning points to the background noise standard deviation of the phase signal under the defect-free state is used as the phase sensitivity of the corresponding scanning point; the ratio of the amplitude difference between the defective and defect-free scanning points to the background noise standard deviation of the amplitude signal under the defect-free state is used as the amplitude sensitivity of the corresponding scanning point.

[0103] For phase sensitivity and amplitude sensitivity under the same simulation conditions, the phase weight and amplitude weight of each scanning point are calculated through normalization processing, so that the sum of the weights of each scanning point is 1. According to the row and column arrangement of the grid coordinate system, the phase weight and amplitude weight of all scanning points are filled into the grid position, and a joint phase and amplitude weight matrix is ​​constructed for each simulation condition to facilitate the weighted fusion of phase and amplitude in online detection.

[0104] Furthermore, the steps for building a defect mapping library include:

[0105] The magnetic field phase and magnetic field amplitude of the full simulation working condition are obtained to construct a simulation dataset. The simulation dataset is classified and grouped according to the defect type, and the phase distribution features of the corresponding detection area are extracted, that is, the phase set of all scanning points in the detection area, thereby generating a phase distribution feature set classified by defect type.

[0106] Based on the three-dimensional simulation model, the defect area and the defect-free area are divided into defect area and defect-free area. That is, the projection area of ​​the defect geometric parameters on the detection area is the defect area and the remaining area is the defect-free area. The arithmetic mean of the phase of all scanning points in the defect area and the phase of all scanning points in the defect-free area are respectively obtained to obtain the defect average phase and the defect-free average phase. The ratio of the difference between the defect average phase and the defect-free average phase to the defect-free average phase is used as the phase change rate under the corresponding working condition.

[0107] The phase change rate under the same defect type is summarized. Based on the phase change rate and defect depth under various working conditions, mapping data pairs are constructed. Using the nonlinear least squares method, with the phase change rate as the independent variable and the defect depth as the dependent variable, a quantitative mapping function for the corresponding defect type is fitted. Combining the physical law that the eddy current field decays exponentially with depth, the fitting function adopts an exponential form:

[0108]

[0109] In the formula, For defect depth, The phase change rate, , , These are the undetermined coefficients obtained from the fitting process;

[0110] For the phase distribution characteristics of the same defect type, clustering algorithms, such as K-means clustering, are used to group the data into several clusters. The center sample of each cluster is selected as the typical feature of the corresponding cluster. Fisher's discriminant method is used to perform cross-type verification on the initially selected typical features. Using all typical features of the three types of defects as the sample set, intra-class scatter matrix (representing the degree of clustering of similar features) and inter-class scatter matrix (representing the degree of difference of features of different classes) are constructed respectively. The criterion value is calculated based on the ratio of the trace of the inter-class scatter matrix to the trace of the intra-class scatter matrix. After removing any typical feature in turn, the criterion value is recalculated. When the criterion value increases after removal, it indicates that this typical feature has reduced the overall discriminative power and is removed. Finally, feature templates for each defect type are generated.

[0111] Defect types, corresponding feature templates, and quantitative mapping functions are associated and stored to construct a defect mapping library.

[0112] Furthermore, the steps for constructing the magnetic field characteristic matrix include:

[0113] Obtaining offline pre-calibrated calibration curves containing the temperature of magneto-optical materials and the Faraday rotation angle includes: placing the magneto-optical sensor in a constant-temperature environment without an external magnetic field; setting multiple discrete temperature points at equal intervals (e.g., every 5°C) within the engineering-allowed operating temperature range (e.g., 0°C to 50°C, covering the annual ambient temperature variation range of a typical industrial site); measuring the Faraday rotation angle value corresponding to the magneto-optical material at each temperature point; and obtaining a continuous temperature and rotation angle calibration curve by polynomial fitting, such as a quadratic polynomial, with the Faraday rotation angle as the vertical axis and temperature as the horizontal axis.

[0114] During online testing, the real-time operating temperature of the magneto-optical sensor at each scanning point is substituted into the calibration curve, and the corresponding Faraday rotation angle on the calibration curve is read as the real-time rotation angle.

[0115] Based on the room temperature of the detection area, a standard temperature is configured, such as 25° room temperature. The standard temperature is substituted into the calibration curve to obtain the reference rotation angle, which remains unchanged throughout the detection process to measure the drift of the real-time temperature relative to the standard state.

[0116] The rotation offset is obtained by the difference between the real-time rotation angle and the reference rotation angle. Combined with the original phase of the corresponding scanning point, temperature drift compensation is performed. That is, the rotation offset is subtracted from the original phase to obtain the phase of each scanning point after temperature compensation, thus eliminating the interference of Faraday rotation angle drift caused by temperature fluctuation on the phase measurement results.

[0117] Based on the weight matrix, the magnitude matrix and the corrected phase matrix are summed point by point to obtain the fused magnetic field characteristics. After the weighted calculation of all scanning points is completed, they are arranged and assembled according to the row and column order of the grid coordinate system to construct the magnetic field characteristic matrix.

[0118] Furthermore, the steps of adaptive initial inspection of defect areas include:

[0119] A square sliding window is set up, with a single scan point as the sliding step size, and it is translated position by position along the row and column directions of the grid coordinate system. At the same time, the scan point at the center of each sliding window is set as the point to be detected. The side length of the window is set to twice the number of scan points corresponding to the minimum detectable defect size, so as to ensure that the potential defect area and sufficient surrounding background scan points can be covered simultaneously within the window, and to ensure the stability of the background statistical results.

[0120] Due to the limitation of the phase matrix boundary, when the sliding window moves to the edge of the matrix, the effective range of the window will exceed the matrix boundary, and it will be impossible to obtain complete background scan point samples. Therefore, the scan points at the edge of the phase matrix cannot be directly used as the detection points. The edge of the phase matrix needs to be completed first. The mirror filling method is used to expand and complete the edge of the phase matrix. That is, the scan points of the outermost row or column of the phase matrix are used as the mirror symmetry axis, and half the side length of the window is used as the expansion width. The scan point values ​​on the side of the symmetry axis closer to the inside of the matrix are symmetrically copied to the expansion position outside the matrix along the symmetry axis. The background scan points required for traversing the edge of the window are completed, and the expanded phase matrix is ​​generated. Based on the expanded phase matrix, any scan point in the phase matrix is ​​used as the detection point, and the corresponding sliding window is constructed.

[0121] For example, the phase matrix is ​​3 rows and 2 columns, that is... If the sliding window is 2×2, then the phase matrix needs to be expanded by 1 row and 1 column.

[0122] Using the outermost row of the phase matrix as the mirror axis of symmetry, fill the 0th row with the following elements: Fill the 4th row with After row filling, the phase matrix is: ;

[0123] Similarly, column filling and phase matrix are performed. ;

[0124] For any point to be detected, combined with the corresponding sliding window, the point to be detected is taken as the geometric center. In order to cover the core abnormal area of ​​a single defect, a square area with a side length of 1 / 3 of the window side length is taken as the central area to be detected. The remaining scanning points in the sliding window, except for the central area to be detected, constitute the local background area.

[0125] The arithmetic mean of the phases of all scan points in the local background region is taken to obtain the mean background phase. The difference between the phase of the point to be detected and the mean background phase is then divided by the mean background phase to obtain the phase change rate of the point to be detected.

[0126] Using the phase change rate of all scan points in the local background region corresponding to the detection point as a statistical sample, the sample mean and sample standard deviation of the phase change rate in the statistical sample are calculated using the 3σ criterion. The sum of the sample mean and 3 times the sample standard deviation is used as the anomaly judgment threshold of the detection point.

[0127] If the phase change rate of the point to be detected is greater than the anomaly judgment threshold, the point to be detected is marked as a defect candidate point; if the phase change rate of the point to be detected is not greater than the anomaly judgment threshold, the point to be detected is marked as a background point.

[0128] After all the scan points in the phase matrix have been traversed, candidate defect points are counted to construct an initial defect set.

[0129] Using the 8-neighborhood connectivity rule, all candidate defect points are divided into connected components to obtain several candidate connected components. The number of scan points in each candidate connected component is the number of pixels.

[0130] Based on half of the scan points corresponding to the smallest detectable defect, a connectivity threshold is set, and candidate connected regions with fewer than the connectivity threshold are eliminated. The remaining candidate connected regions are the defect regions.

[0131] Furthermore, multi-directional phase gradient fusion includes:

[0132] Because the orthogonal anisotropy of CFRP makes the eddy current distribution directional, the detection sensitivity of defects with different orientations varies significantly under a single-directional gradient. Based on the horizontal, vertical, and oblique directions, gradient directions are configured at 0°, 45°, 90°, and 135° respectively to cover the main orientation range of defects in the project.

[0133] For each scan point within the defect region, the phase gradient magnitude of each gradient direction is calculated point by point using the Sobel gradient operator. This includes: performing convolution operations between the Sobel convolution kernel of each gradient direction and the phase of the neighborhood. For example, if each gradient direction corresponds to a 3×3 convolution kernel, then with the scan point as the geometric center, the phase values ​​of a total of 9 scan points in the surrounding 3×3 spatial neighborhood are taken as operation samples to obtain the phase gradient components of the scan point in each gradient direction. The absolute value of the phase gradient component is taken as the phase gradient magnitude of the corresponding gradient direction.

[0134] The phase gradient magnitudes of all scanning points within the defect area in each gradient direction are integrated. The arithmetic mean of the phase gradient magnitudes in a single gradient direction is used to calculate the regional gradient mean in each gradient direction, so as to characterize the overall detection intensity in each direction within the defect area.

[0135] Based on inverse weighting, the reciprocal of the mean of the regional gradient is used as the initial weight of the corresponding gradient direction. The initial weights of all gradient directions are normalized to obtain the fused weights of each gradient direction, so that the sum of the weights is 1.

[0136] For each scan point within the defect region, gradient-weighted fusion is performed. The phase gradient magnitude of each gradient direction is multiplied by the corresponding fusion weight and then summed to obtain the fusion gradient feature of the corresponding scan point.

[0137] The fusion gradient features of all defect regions are summarized as statistical samples. Using Otsu's method, the optimal segmentation threshold is calculated to maximize the inter-class variance between the defect class and the background class.

[0138] Scan points whose fused gradient features are higher than the optimal segmentation threshold are marked as defect pixels and assigned a value of 1; scan points whose fused gradient features are not higher than the optimal segmentation threshold are marked as background pixels and assigned a value of 0, thus generating an initial binary mask.

[0139] Morphological closing and opening operations are performed on the initial binary mask to fill the internal holes of the defect and remove the edge noise, thus obtaining the defect binary mask.

[0140] The numerical values ​​in the magnetic field feature matrix are linearly mapped to grayscale values, which serve as the background grayscale base for the detection area. The binary mask of the defect is superimposed and mapped to the corresponding position in the grid coordinate system in the form of a highlighted mark, generating a defect image with spatial coordinates.

[0141] Furthermore, the steps for generating the defect detection package include:

[0142] Based on the binary mask of defects and the row and column index of the grid coordinate system, each defect region is spatially located, the boundary coordinates of the defect region are extracted, and the center coordinates of the defect are calculated by the arithmetic mean of the boundary coordinates, which is used as the coordinate position of the defect.

[0143] Extract the continuous contour boundary of the defect and calculate its shape and size parameters, including the length and width of the minimum bounding rectangle, the diameter of the equivalent circle, the aspect ratio, and the roundness.

[0144] Taking a single defect area as the target area, the target area is expanded outward, with the expansion width being 1 / 3 of the diameter of the equivalent circle of the target area. An annular area is set as the background reference area. Based on the difference between the average phase values ​​of the target area and the background reference area, the average phase value of the background reference area is removed, and the phase change rate of the target area is calculated. The equivalent circle of the target area is a circle with the same area as the target area.

[0145] Extract the phase distribution features of the target region, call the feature templates of various defects in the defect feature mapping library, use cosine similarity to calculate the similarity between the defects in the target region and each feature template, and take the defect type with the highest similarity as the recognition result.

[0146] Based on the identified defect type, the corresponding quantitative mapping function is called from the defect feature mapping library. The phase change rate of the target area is substituted into the quantitative mapping function to solve for the defect depth of the target area.

[0147] For each individual defect, the coordinate position, defect type, defect depth, shape and size parameters, and contour boundary are bound together to generate a single defect detection package.

[0148] Example 2

[0149] Please see Figure 4 Another embodiment of the present invention provides a CFRP defect detection method based on superimposed magnetic field phase change, comprising the following steps:

[0150] A three-dimensional simulation model was established, and the phase and amplitude changes of the superimposed field of all scanning points under different lift-off degrees and different defect parameter groups were simulated through multi-condition simulation with controlled variables, and a weight matrix was constructed.

[0151] A quantitative correspondence between the phase change rate and the defect depth was fitted to construct a defect mapping library;

[0152] By using a low-current excitation component and a magneto-optical sensor, the reflected polarized light from the CFRP surface and the real-time operating temperature of the magneto-optical sensor are obtained.

[0153] The reflected polarized light is converted into a differential mode electrical signal. The phase matrix and amplitude matrix are constructed using the effective signal at the target frequency. The magnetic field feature matrix is ​​constructed by correcting the phase temperature drift and weighted fusion of the phase and amplitude.

[0154] Adaptive preliminary inspection of defect regions is performed on the phase matrix after temperature drift correction to identify defect regions;

[0155] By configuring the gradient direction, multi-directional phase gradient fusion is performed on the defect region, and combined with the magnetic field feature matrix, a defect imaging map is generated.

[0156] Based on the matching of defect imaging images and defect mapping libraries, defect depth is inverted, and defect category identification is combined to generate a defect detection package.

[0157] Working principle and effects:

[0158] This method utilizes eddy current magneto-optical imaging to detect defects in CFRP (Computer-Assisted Reinforced Plastic) materials. It prioritizes phase signals, which are more sensitive to changes in eddy current depth in weakly conductive materials, as the core detection basis. A magnetic field feature matrix is ​​constructed by weighting and fusing amplitude signals point-by-point. This enhances defect signal contrast without increasing excitation current intensity, avoiding the increased equipment size and heat generation associated with large excitations. Offline calibrated temperature curves are used to compensate for phase signals point-by-point, eliminating the interference of temperature drift on measurement results. Initial defect screening is performed using the relative phase change rate statistically derived from local background statistics via a sliding window. The local area surrounding the defect serves as a reference benchmark, eliminating reliance on standard defect-free samples and offsetting global signal shifts caused by lift-off fluctuations, thus improving robustness under complex operating conditions. Multi-directional phase gradient reverse weighted fusion flattens the response intensity of defects with different orientations, eliminating defect orientation dependence and reducing the risk of missed detection of various defects such as cracks, delamination, and impact damage. Based on an offline defect feature mapping library, defect type identification and depth quantitative inversion are completed, achieving accurate, full-dimensional defect detection without on-site calibration.

[0159] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A CFRP defect detection system based on superimposed magnetic field phase change, characterized in that, include: Scanning terminal, data acquisition module, and defect identification module; The scanning terminal is used to scan the CFRP in the detection area point by point to obtain the reflected polarized light modulated by the superimposed magnetic field on the CFRP surface. The acquisition module is used to receive the reflected polarized light and convert it into a differential mode electrical signal, while acquiring the real-time operating temperature. Through effective signal extraction of the differential mode electrical signal, a phase matrix and an amplitude matrix are generated. The defect identification module is used to perform offline calibration of weights and defect mappings. By correcting the phase matrix for temperature drift and performing weighted fusion with the amplitude matrix, a magnetic field feature matrix is ​​constructed. By performing adaptive initial detection of defect regions on the corrected phase matrix, defect regions are identified. Multi-directional phase gradient fusion is performed on the defect regions. Combined with the magnetic field feature matrix, a defect imaging map is generated. Combined with the offline calibration of the defect mapping, the defect depth is inverted to generate a defect detection package.

2. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 1, characterized in that: The defect identification module includes a feature extraction unit, a defect detection unit, and a quantitative inversion unit; The feature extraction unit is used for the three-dimensional simulation model. By performing multi-condition simulation of the three-dimensional simulation model with control variables, a weight matrix combining phase and amplitude is constructed. By quantitatively fitting the phase change rate with the defect depth, a defect mapping library is constructed. Based on the standard temperature and the real-time working temperature, the phase matrix is ​​corrected for temperature drift point by point. Combined with the weight matrix and amplitude matrix, a magnetic field feature matrix is ​​constructed. The defect detection unit is used to complete the edge of the phase matrix, calculate the phase change rate point by point with the sliding window, identify the defect region by configuring the anomaly judgment threshold and connected component filtering through local statistics, perform multi-directional phase gradient fusion on the defect region, and generate a defect imaging map by combining the magnetic field feature matrix. The quantitative inversion unit is used to extract the average phase change rate of the defect region in the defect image, invert the defect depth by matching with the defect mapping library, and identify the defect type by combining the position and contour size of the defect in the defect image to generate a defect detection package.

3. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 1, characterized in that, The generation of the phase matrix and magnitude matrix includes: The differential-mode electrical signal is subjected to bandpass filtering to generate the signal to be tested; Configure a reference signal for the lock-in amplifier demodulation circuit, and separate the in-phase reference signal and the quadrature reference signal; The signal under test is multiplied with the in-phase reference signal and the quadrature reference signal respectively, and then low-pass filtered to generate in-phase components and quadrature components. The original amplitude and original phase of the corresponding scanning point are then calculated. Read the spatial coordinates of each scanning point and the real-time operating temperature of the magneto-optical sensor in the scanning terminal, combine them with the original phase and original amplitude, and add a unified timestamp to form a single-point data packet; Based on the scanning points in the scanning area, a grid coordinate system is constructed, and according to the single scanning cycle, point data packets are filled in at the corresponding grid positions. Based on the row and column order of the grid, the original phase and original amplitude of all scan points are assembled separately to form the phase matrix and amplitude matrix under the same scan period.

4. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 2, characterized in that, Constructing the joint phase and magnitude weight matrix includes: Using the controlled variable method, the lift-off degree of the magneto-optical sensor is used as the first-level controlled variable, and the defect parameter group, including defect type, defect direction, defect depth, and defect size, is used as the second-level controlled variable to divide the simulation conditions. The three-dimensional simulation model is solved to obtain the magnetic field phase and magnetic field amplitude corresponding to all scanning points in the detection area under each simulation condition; For the distribution of magnetic field phase and magnetic field amplitude under the same simulation conditions, calculate the phase sensitivity and amplitude sensitivity of each scanning point; The phase sensitivity and amplitude sensitivity of each scan point are normalized, and the phase weight and amplitude weight of each scan point are calculated. Based on the row and column arrangement of the grid coordinate system, a weight matrix combining phase and amplitude is constructed for each simulation condition.

5. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 4, characterized in that, The steps to build a defect mapping library include: The magnetic field phase and magnetic field amplitude of the full simulation working condition are obtained, classified and collected according to the defect type, and the phase distribution characteristics of the corresponding detection area are extracted. Typical features are screened using clustering algorithm, and cross-type verification of the typical characteristics is performed using Fisher's discriminant method to generate feature templates for each defect type. Based on the defective and non-defective regions divided by the three-dimensional simulation model, the average phase of the defective region and the average phase of the non-defective region are obtained, and the phase change rate under the corresponding working conditions is calculated. Summarize the phase change rate under the same defect type, and construct mapping data pairs based on the phase change rate and defect depth under each working condition; Using the nonlinear least squares method, with the phase change rate as the independent variable and the defect depth as the dependent variable, a quantitative mapping function is fitted for the corresponding defect type. Defect types, corresponding feature templates, and quantitative mapping functions are associated and stored to construct a defect mapping library.

6. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 2, characterized in that, The steps for constructing the magnetic field characteristic matrix include: Obtain the offline pre-calibrated calibration curve containing the temperature of the magneto-optical material and the Faraday rotation angle, and combine it with the configured standard temperature to obtain the reference rotation angle; During online testing, the real-time operating temperature of each scanning point is substituted into the calibration curve to read the real-time rotation angle; The rotation offset is obtained by the difference between the real-time rotation angle and the reference rotation angle. Combined with the original phase of the corresponding scan point, temperature drift compensation is performed to obtain the phase of each scan point after temperature compensation. Based on joint weights, the magnitude matrix and the corrected phase matrix are summed point by point to construct the magnetic field characteristic matrix.

7. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 1, characterized in that, The adaptive initial detection of the defect area includes: Set up a square sliding window, with a single scan point as the sliding step size, and translate position by position along the row and column directions of the grid coordinate system. Set the scan point at the center of the window as the point to be detected. The edges of the phase matrix are extended and completed using the mirror filling method to generate an extended phase matrix. For any point to be detected, the central detection area and the local background area are divided. Based on the local background area, a statistical sample is set, and the anomaly judgment threshold is set using the 3σ criterion. Calculate the mean background phase of the local background region, and combine it with the phase of the point to be detected to calculate the phase change rate of the point to be detected; If the phase change rate of the point to be detected is greater than the anomaly determination threshold, it is marked as a defect candidate point; otherwise, it is a background point. Traverse the scan points in the phase matrix to construct an initial defect set, and construct candidate connected regions using the 8-neighborhood connectivity rule, with the number of scan points in the candidate connected regions being the number of pixels; Set a connectivity threshold and combine it with the number of pixels to filter out defective areas.

8. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 7, characterized in that, The multi-directional phase gradient fusion includes: Configure the gradient direction and use the Sobel gradient operator to calculate the phase gradient magnitude of each gradient direction point by point; The mean gradient of the region is calculated by the mean of the phase gradient magnitude in a single gradient direction, and the fusion weight of each gradient direction is calculated based on the inverse weighting and normalization process. For each scan point within the defect region, gradient-weighted fusion is performed using the phase gradient magnitude in each gradient direction and the fusion weight to obtain the fused gradient features of the scan point. The fusion gradient features of all defect regions are summarized, and the optimal segmentation threshold is calculated using Otsu's method. The fused gradient features of each scan point are compared with the optimal segmentation threshold to divide defect pixels and background pixels, generate an initial binary mask, and then combine morphological closing and opening operations to obtain the defect binary mask. Using the magnetic field feature matrix as a grayscale base, the defect binary mask is superimposed and mapped onto the grid coordinate system to generate a defect image.

9. The CFRP defect detection system based on superimposed magnetic field phase change according to claim 8, characterized in that, The steps to generate a defect detection package include: Extract the boundary coordinates of the defect region and calculate the center coordinates of the defect as the coordinate position of the defect; Extract the contour boundary of the defect and calculate its shape and size parameters; Using a single defect region as the target region, a background reference region is generated by expanding outward from the target region, and the phase change of the target region is calculated. Extract the phase distribution features of the target region, call the feature templates of various defects in the defect feature mapping library, use cosine similarity to calculate the similarity between the defects in the target region and each feature template, and take the defect type with the highest similarity as the recognition result. Based on the identified defect type, the corresponding quantitative mapping function is called from the defect feature mapping library. The phase change rate of the target area is substituted into the quantitative mapping function to solve for the defect depth of the target area. For each individual defect, the coordinate position, defect type, defect depth, shape and size parameters, and contour boundary are bound together to generate a single defect detection package.

10. A CFRP defect detection method based on superimposed magnetic field phase change, used to implement the CFRP defect detection system based on superimposed magnetic field phase change as described in any one of claims 1-9, characterized in that, include: A three-dimensional simulation model is established, and a weight matrix is ​​constructed by combining multi-condition simulation of control variables. A quantitative correspondence between the phase change rate and the defect depth was fitted to construct a defect mapping library; To obtain the reflected polarized light from the CFRP surface and the real-time operating temperature of the magneto-optical sensor; The reflected polarized light is converted into a differential mode electrical signal. The phase matrix and amplitude matrix are constructed using the effective signal at the target frequency. The magnetic field feature matrix is ​​constructed by temperature drift correction and weighted fusion. The phase matrix after temperature drift correction is used to perform adaptive initial detection of the defect region, identify the defect region, and perform multi-directional phase gradient fusion on the defect region. Combined with the magnetic field feature matrix, a defect image is generated. Based on the matching of defect imaging images and defect mapping libraries, defect depth is inverted, and defect category identification is combined to generate a defect detection package.

Images