Rapid detection method for defects of wave-absorbing coating based on incident angle modulation

Abstract

The invention provides a method for rapidly detecting defects of a wave-absorbing coating based on incident angle modulation, and belongs to the field of defect detection.The method comprises the steps that a detection system is initialized; carrying out multi-angle incidence frequency sweeping test on the surface of the defect-free coating; setting an incidence mode as grazing incidence, performing a sweep frequency test on the surface of the defective coating, and comparing and analyzing with reference data to judge whether surface defects exist or not; setting an incidence mode as a vertical incidence mode, carrying out a sweep frequency test on the surface of the defective coating, and judging whether a circular defect exists or not; when the existence of the circular defect is determined, executing multi-angle sweep frequency inversion to quantify the defect size, and correcting by using an ellipse correction model to obtain an actual physical size; and introducing a self-adaptive calibration module based on deep learning to obtain a final defect size. According to the invention, the requirements of invisible airplanes and other aerospace equipment can be met.

Description

Technical Field

[0001] The present invention belongs to the field of defect detection, and in particular relates to a method for quickly detecting defects in absorbing coatings based on incident angle modulation. Background Art

[0002] Stealth technology mainly includes two major areas: external stealth and material stealth. As an important branch of this, material stealth technology achieves efficient absorption and scattering modulation of radar waves by constructing a multifunctional absorbing coating system on the surface of the aircraft, thereby significantly improving the low detectability of the target. The absorbing coating not only achieves radar stealth, but also effectively suppresses signal crosstalk in complex electromagnetic environments, ensuring the reliable operation of electronic equipment. However, in practical applications, the absorbing coating faces the risk of failure under the action of thermal load shock and mechanical stress, which damages the structural integrity and thus affects the stealth effectiveness of the aircraft. Therefore, in order to ensure the stealth capability of the aircraft and reduce maintenance costs, it needs to be regularly inspected and maintained.

[0003] Current absorbing coating defect detection technologies are mainly divided into the following three categories:

[0004] (1) Traditional microwave detection technology: It is based on the interaction mechanism between electromagnetic waves and materials, and the reflection, scattering and transmission characteristics of microwaves at the interface of the medium are used to realize defect detection. The current mainstream reflective microwave detection technology establishes a quantitative mapping relationship between the defect area and the characteristic parameters by accurately measuring the phase offset of the reflection coefficient of the coating-substrate interface. Experimental data show that the use of waveguide probe plane scanning technology under Ka-band conditions can effectively identify circular hole-shaped debonding defects with a diameter of 1.6mm, with a spatial resolution of sub-millimeter level. The traditional microwave method has high test accuracy, but the waveguide probe needs to be scanned point by point during the test, which has low defect detection efficiency, and the microwave transceiver module design in the high-frequency band is complex.

[0005] (2) Ultrasonic testing technology: This method uses the propagation characteristics of ultrasonic waves in materials to detect defects. When the coating debonds, the propagation of ultrasonic waves inside the coating is affected, which in turn causes abnormal received signals. C-scan imaging technology is commonly used. This method has good penetration and high sensitivity and is suitable for defect detection in a variety of materials. However, it requires the use of a coupling agent and cannot detect curved surface structures. Contact measurement can easily damage the coating surface and is less efficient in large models and on-site testing.

[0006] (3) Infrared thermal imaging technology: Detection is achieved by analyzing temperature changes caused by defects. Discontinuities within the object being tested, such as defects or structural differences, can affect the object's thermal diffusion characteristics, resulting in abnormal surface temperature distribution. The temperature resolution reaches 0.05°C. Existing technologies have high detection efficiency for surface and near-surface defects, but are significantly affected by ambient temperature and have poor adaptability to complex curved surfaces.

[0007] Therefore, the existing rapid detection methods for absorbing coating defects have the following technical problems:

[0008] (1) Poor adaptability to different scenarios: Insufficient ability to detect curved surfaces. Ultrasonic testing relies on plane coupling, and infrared thermal imaging is affected by surface normal deviation, resulting in large positioning errors. (2) Poor flexibility: Traditional methods have a single detection mode and rely on fixed detection parameters, making them unable to adapt to different defect types. (3) High risk of damage: Contact detection methods (such as ultrasound) generate mechanical stress on the coating surface that can easily cause secondary damage. (4) Application targets: Traditional defect detection methods are mainly focused on coating thickness measurement. Summary of the Invention

[0009] To address the above-mentioned technical problems, the present invention provides a rapid detection method for absorbing coating defects based on incident angle modulation. This method, based on the principle of reducing the radar cross section (RCS) of the target surface caused by the absorbing material, detects defects by observing the sudden RCS change caused by the presence of gaps or circular defects. This method enables continuous frequency sweeping of user-defined incident angles within a frequency range of 2 to 18 GHz with a user-defined step size, generating synchronized echo signal amplitude, phase, and polarization information. Frequency-domain S-parameters are then converted into time-domain signals through an inverse Fourier transform, achieving one-dimensional imaging and significantly improving defect location accuracy. Furthermore, defect size can be assessed based on multi-angle detection data. By comparing post-detection results with initial defect-free results, the present method is applicable to various complex shapes and curved structures, meeting the requirements of stealth aircraft and other aerospace equipment.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] A rapid detection method for absorbing coating defects based on incident angle modulation includes the following steps:

[0012] Step 1: Initialize the detection system, set the operating frequency range of the vector network analyzer, fix the sample to be tested on the 3D scanning platform, and install the antenna on an adjustable antenna bracket;

[0013] Step 2: Perform a multi-angle incident frequency sweep test on the defect-free coating surface to obtain echo signal data and perform inverse Fourier transform to obtain time domain data as the baseline data;

[0014] Step 3: Set the incidence mode to grazing incidence, perform a frequency sweep test on the defective coating surface, and determine whether there are surface defects by comparing and analyzing with the benchmark data;

[0015] Step 4: Set the incidence mode to vertical incidence, perform a frequency sweep test on the defective coating surface, and determine whether there is a circular defect by comparing and analyzing it with the benchmark data;

[0016] Step 5: Once the presence of a circular defect is confirmed, multi-angle frequency sweep inversion is performed to quantify the defect size. The projected size range of the circular defect on the incident plane is determined using time domain data at different incident angles, and the actual physical size of the circular defect is corrected using an ellipse correction model. The time domain data at different incident angles is referred to as multi-angle time domain data.

[0017] Step 6: Introduce a deep learning-based adaptive calibration module, build a defect feature-size mapping model, and use a convolutional neural network to jointly analyze multi-angle time domain data to obtain the final size of the circular defect.

[0018] Beneficial effects:

[0019] 1. The present invention compares initial non-destructive data with subsequent test data, and determines the presence of defects by comparing data differences, ignoring the influence of deviations caused by curved surfaces. It also uses multi-angle sweep frequency measurement to improve measurement accuracy. The antenna transmits and receives echo signals, eliminating the need for contact with the absorbing coating being tested, making it a non-destructive test.

[0020] 2. This invention uses multi-angle testing to identify defect types such as gaps and circular shapes. It also allows for customizable test frequencies based on a vector network analyzer, allowing for user-defined incident angles. Furthermore, this invention utilizes a handheld, miniaturized detection system, making it flexible and convenient for application scenarios.

[0021] 3. This invention focuses on identifying, locating, and sizing defects in absorbing coatings. By employing novel detection methods and data processing technologies, this handheld, vector network analyzer-based detection device can rapidly scan the absorbing coating surface from multiple angles to identify defect types and sizing. Traditional detection methods, on the other hand, can only qualitatively determine the presence of defects and are inadequate for detecting small gaps and circular defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a method for rapid detection of absorbing coating defects based on incident angle modulation according to the present invention;

[0023] Figure 2 Schematic diagram of the detection system. DETAILED DESCRIPTION

[0024] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0025] like Figure 1 As shown, the present invention provides a rapid detection method for absorbing coating defects based on incident angle modulation. The method uses multi-angle incident broadband scanning to obtain a series of time domain data by inverse Fourier transform of the echo signal intensity. The peak amplitude is used to quickly determine whether there are gaps or circular defects on the coating surface, and further deduce the defect size. The method specifically includes the following steps:

[0026] Step 1: Initialize the detection system and set the operating frequency range of the vector network analyzer according to the frequency band range of the antenna used.

[0027] like Figure 2 As shown, the detection system includes a vector network analyzer, horn antennas (transmitting and receiving antennas), an antenna bracket, a 3D scanning platform (for placing the sample under test or adjusting the antenna angle), a data transmission cable (such as a coaxial cable), and a calibration component. The first port of the vector network analyzer is connected to the transmitting antenna via a coaxial cable, and the second port is connected to the receiving antenna via a coaxial cable. The sample under test (i.e., the target to be tested) is fixed to the 3D scanning platform, and the antenna is mounted on an adjustable antenna bracket.

[0028] Typically, antennas in the 12-18 GHz frequency band are used, and the intermediate frequency bandwidth is set to 1 kHz. This bandwidth value is an ideal starting point for general measurements and can effectively balance measurement accuracy (signal-to-noise ratio / dynamic range) and measurement speed. At the same time, the number of frequency points can be set to 801. The more frequency points there are, the larger the upper limit of the corresponding time domain measurement, but the scanning time will also increase accordingly. The distance between the target to be detected and the antenna can be used to determine the maximum peak position in subsequent tests. It should be noted that the intermediate frequency bandwidth is not a fixed value. In actual applications, the bandwidth can be flexibly selected based on the specific accuracy or speed requirements (the narrower the bandwidth, the higher the accuracy and the lower the efficiency). Generally, the number of frequency points N needs to meet the following requirements:

[0029] ;

[0030] in, is the frequency range, is the distance between the antenna and the target to be detected, is the speed of light, usually taken By setting the above parameters, the initialization of the detection system is completed.

[0031] Step 2: Perform a multi-angle incident frequency sweep test on the defect-free coating surface to obtain a series of echo signals, which are substituted into the inverse Fourier transform formula. , and obtain the corresponding time domain data , as the benchmark data for subsequent tests, The incident angle is The time domain data obtained when is the incident angle index.

[0032] The inverse Fourier transform formula is as follows:

[0033] ;

[0034] in, The incident angle is Time domain data, is the incident angle index; The incident angle is Frequency domain data at ; represents the inverse Fourier transform operator; represents an imaginary unit; represents a frequency variable; Represents a time variable.

[0035] For multi-angle incident angle selection, a uniform or non-uniform angle scanning strategy can be selected according to actual needs.

[0036] For the uniformity angle scanning strategy, the smaller the step angle, the more reference data can be obtained, and the longer the scanning time. Based on the typical defect scattering pattern, the angle interval can be initially selected as (i.e. select 、 、 、 、 、 、 ) is scanned uniformly. In order to improve the size inversion accuracy, a smaller interval, such as 5°, can be used, but the corresponding scanning time will also increase.

[0037] For the non-uniform angle scanning strategy, the angle-sensitive area of ​​the coating can be considered. Usually, when the incident angle is close to the critical angle determined by the refractive index of the coating medium, the surface wave effect will greatly enhance the response sensitivity of small defects. It has the best detection efficiency for delamination defects when incident, but has no good detection effect on gap defects, which will help to distinguish different defect types in the future. Under the non-uniformity strategy, six characteristic angles based on the fifth-order Legendre polynomial node distribution optimization (i.e. 、 、 、 、 、 This distribution, generated through zero-point mapping of orthogonal polynomials, covers key sensitive regions such as near-grazing incidence, the Brewster angle region, and vertical incidence. This reduces the workload of full scans while ensuring the completeness of defect response characteristics. In practice, to improve data reliability and minimize the instability introduced by vector network analyzers, multiple measurements are repeated for each incident angle, and the average value is used as a benchmark for each angle.

[0038] Step 3: Set the incident mode to grazing incidence and perform a frequency sweep test on the defective coating surface.

[0039] Adjust the antenna to a pre-selected characteristic grazing angle. The characteristic grazing angle can usually be selected based on the critical angle calculated based on the electromagnetic characteristics of the absorbing coating. The typical value is At this time, the electromagnetic wave forms a coupling effect with the coating surface, which increases the scattering field intensity of the micron-level surface opening defects by 10-15dB.

[0040] During the test, a continuous wave signal is transmitted through a vector network analyzer to obtain the frequency domain scattering parameters. , converted into time domain data through inverse Fourier transform By comparing with the benchmark data obtained in step 2 (i.e., the time domain data of step 2 in Take 6 characteristic angles 、 、 、 、 、 in ) is analyzed, and it is defined that if If a peak value of more than 6dB increases at ns, it is determined that there is a surface defect. is the time window corresponding to the defect reflection peak, and is related to the size of the sample to be tested in the incident direction. and the distance between the antenna and the sample to be tested Related, it can usually be assumed that it satisfies In addition, in order to improve the stability of the judgment, multiple measurements can be performed and the average value can be taken to suppress the occurrence of false positives.

[0041] Step 4: Set the incident mode to vertical incidence and perform a frequency sweep test on the defective coating surface.

[0042] Adjust the antenna to In the incident state, the slot-shaped defect has little effect on the echo signal at this angle, while the circular defect will produce a significant backscattering enhancement. During the test, the same sweep parameters as the benchmark are maintained to obtain the frequency domain scattering parameters of the incident direction. , converted into time domain data through inverse Fourier transform By comparing with the benchmark data (i.e., the time domain data of step 2 in Take 6 characteristic angles 、 、 、 、 、 in )and Comparative analysis, definition if at the corresponding time If a peak value of more than 6dB increases at ns, it is determined that a circular defect exists, otherwise the defect does not exist. is the time window corresponding to the defect reflection peak, and is related to the size of the sample to be tested in the incident direction. and the distance between the antenna and the sample to be tested Related, it can usually be assumed that it satisfies In order to improve the stability of the judgment, 3-10 measurements can be taken and the average value can be taken to suppress the occurrence of false positives.

[0043] By comparing with time domain data Comparative analysis, if the original time position ns The peak increase phenomenon occurs in If there is no response or a response less than 2dB at the same time position, it indicates that the defect has directionally selective scattering characteristics, consistent with a slot-type defect. To improve the stability of the judgment, 3-10 measurements can be made to reduce the occurrence of false positives.

[0044] Step 5: After the circular defect is determined to exist in step 4, a multi-angle frequency sweep inversion is performed to quantify the defect size.

[0045] You can set the incident angle to perform multi-angle sweep test. The incident angle value is recommended to follow the symmetrical scanning principle. 、 、 、 、 、 、 Seven characteristic angles, the sequence covers the main lobe area of ​​the defect scattering field ( ) and side lobe sensitive area ( ), when more accurate detection results are required, the adjacent angle intervals can be reduced accordingly. The continuous wave signal is transmitted by the vector network analyzer to obtain the frequency domain scattering parameters, which are converted into a series of time domain data through the inverse Fourier transform. (i.e. multi-angle time domain data), for each incident angle, the average value can be re-measured 3-10 times to improve the stability and reliability of the detection. After comparison, the peak value of more than 6dB increases the time position (defined as the angle of incidence When the peak value of more than 6dB increases compared to the baseline data, the time node) may be different, that is, different reflection peaks are formed in the time domain. Therefore, in step 4, the corresponding time position ns, for each incident angle, find the specific time point in the time domain data where the peak value increases by more than 6dB compared to the baseline data at the same incident angle. At all the time points obtained In this case, the minimum echo time and corresponding angle are and , the maximum echo time and the corresponding angle are and , the approximate projection size range of the circular defect on the incident plane can be determined :

[0046] ;

[0047] ;

[0048] in, Indicates the maximum echo time, Indicates the minimum echo time, represents the incident angle corresponding to the minimum echo time, represents the incident angle corresponding to the maximum echo time, represents the average angle of incidence.

[0049] This size is the equivalent diameter of the defect in the detection plane, the actual physical size Need to be corrected by ellipse correction model:

[0050] ;

[0051] in, Indicates the approximate projection size range of the circular defect on the incident plane, represents the incident angle corresponding to the minimum echo time, Indicates the incident angle corresponding to the maximum echo time. Indicates the difference between the incident angle corresponding to the maximum echo time and the incident angle corresponding to the minimum echo time.

[0052] Step 6: Introduce a deep learning-based adaptive calibration module to achieve higher-precision defect quantification by building a defect feature-size mapping model.

[0053] After obtaining the preliminary defect size range, a pre-trained convolutional neural network is used to analyze the multi-angle time domain data. Conduct conjoint analysis:

[0054] First, the time domain signal amplitude sequences corresponding to different incident angles are arranged according to the angle dimension, that is, the multi-angle time domain data With benchmark data Perform normalized difference processing to form a two-dimensional time-angle feature matrix :

[0055] ;

[0056] ;

[0057] in, is the incident angle index; is the time domain sampling point, T represents the number of sampling points; Indicates the incident angle The maximum absolute value of the lower differential signal, Indicates all time sampling points at this angle; Represents the value corresponding to the i-th row and j-th column of the two-dimensional time-angle characteristic matrix; Indicates the angle of incidence Defective coating below Time domain data at the moment; Indicates the angle of incidence Defect-free coating under Time domain data at the moment; Indicates the angle of incidence exist The differential signal value at the moment.

[0058] Maxwell's equations are solved by the finite-difference time-domain method (FDTD), and the corresponding frequency bands and different dielectric constants ( ), defect shape (circular / crack / complex defect), incident angle ( , step size 0.2°) in the frequency domain , a time-domain training set is generated through Fourier transform. Data enhancement strategies can be introduced, such as adding Gaussian white noise with a signal-to-noise ratio of 30-50dB to simulate actual measurement environment interference.

[0059] A deep residual network (ResNet-34) is trained end-to-end. An angle encoding layer is added to the input to map the incident angle into a 128-dimensional position encoding vector, which is then concatenated with the time-domain signal before being fed into a convolutional layer. The network architecture includes an initial convolutional layer (64 7×7 convolution kernels with a stride of 2 and 64 output channels), a max pooling layer (3×3 pooling window with a stride of 2), and four residual blocks (with increasing channel counts of 64, 128, 256, and 512). A Monte Carlo dropout layer is added at the end to achieve joint angle-waveform feature extraction and defect size regression.

[0060] In practical applications, the pre-trained model is adapted to the electromagnetic properties of the current coating through transfer learning, and based on the benchmark data in step 2 and the multi-angle time domain data in step 5 Construct the time-angle feature matrix and input it into the deep residual network. Add a probability output layer at the end of the network, use the randomness of dropout (random dropout), estimate the uncertainty of the model prediction through 100 Monte Carlo forward propagations, and calculate the prediction variance. , get the confidence , threshold For predictions with confidence levels lower than 90%, the angular sensitivity is calculated using the Gradient Class Activation Map (Grad-CAM):

[0061] ;

[0062] in, is the angle The global sensitivity score of , where the larger the value, the more important the angle is for defect detection; Represents the defect size prediction value output by the network; Represents the time-angle characteristic matrix at angle , time point Dynamically select the 3-5 optimal incident angles with the highest sensitivity to the current defect for additional scanning and update the two-dimensional time-angle feature matrix , until two consecutive prediction errors are within the error threshold Thus, a closed-loop optimization system is formed to control the final size error within level.

[0063] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A rapid detection method for absorbing coating defects based on incident angle modulation, characterized in that: The following steps are involved: Step 1: Initialize the detection system, set the operating frequency range of the vector network analyzer, fix the sample to be tested on the 3D scanning platform, and install the antenna on an adjustable antenna bracket; Step 2: Perform a multi-angle incident frequency sweep test on the defect-free coating surface to obtain echo signal data and perform inverse Fourier transform to obtain time domain data as the baseline data; Step 3: Set the incidence mode to grazing incidence, perform a frequency sweep test on the defective coating surface, and determine whether there are surface defects by comparing and analyzing with the benchmark data; Step 4: Set the incidence mode to vertical incidence, perform a frequency sweep test on the defective coating surface, and determine whether there is a circular defect by comparing and analyzing it with the benchmark data; Step 5: Once the presence of a circular defect is confirmed, multi-angle frequency sweep inversion is performed to quantify the defect size. The projected size range of the circular defect on the incident plane is determined using time domain data at different incident angles, and the actual physical size of the circular defect is corrected using an ellipse correction model. The time domain data at different incident angles is referred to as multi-angle time domain data. Step 6: Introduce a deep learning-based adaptive calibration module, build a defect feature-size mapping model, and use a convolutional neural network to jointly analyze multi-angle time domain data to obtain the final size of the circular defect.

2. The rapid detection method of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In the step 1, the first port of the vector network analyzer is connected to a transmitting antenna via a coaxial line, and the second port of the vector network analyzer is connected to a receiving antenna via a coaxial line.

3. The rapid detection method of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 2, the multi-angle incident includes incident at a uniform angle or a non-uniform angle.

4. The rapid detection method of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 3, the angle of grazing incidence is selected based on the critical angle calculated from the electromagnetic characteristic parameters of the absorbing coating.

5. The rapid detection method of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 4, the vertical incidence angle is 90°.

6. The rapid detection method of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 5, the value of the incident angle follows the symmetrical scanning principle.

7. The method for rapid detection of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 6, the convolutional neural network is a deep residual network.

8. The method for rapid detection of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 6, the deep learning-based adaptive calibration module solves Maxwell's equations through the time-domain finite difference method, obtains the frequency domain scattering field under the corresponding frequency band, different dielectric constants, defect shapes, and incident angles through simulation, and generates a time domain training set through Fourier transform.

9. The method for rapid detection of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 6, a data enhancement strategy is introduced into the adaptive calibration module based on deep learning.

10. The method for rapid detection of absorbing coating defects based on incident angle modulation according to claim 1, characterized in that: In step 6, the deep learning-based adaptive calibration module calculates the prediction variance and confidence through 100 Monte Carlo forward propagations. For prediction results with a confidence level lower than 90%, the gradient class activation map is used to calculate the angle sensitivity, and the 3-5 incident angles with the highest angle sensitivity are dynamically selected for supplementary scanning and updating until two consecutive prediction errors are within the error threshold.

Images