Electromagnetic acoustic based rivet structure defect detection and identification method and system
By performing high-pass filtering and sym6 wavelet decomposition denoising on the electromagnetic ultrasonic signal, the problem of signal-noise separation in the detection of countersunk rivet structural defects was solved, and accurate detection and depth measurement of rivet structural defects were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGFEI AIRCRAFT EQUIP MFG (CHENGDU) CO LTD
- Filing Date
- 2026-06-12
- Publication Date
- 2026-07-14
AI Technical Summary
In existing technologies, when using electromagnetic ultrasonic testing to detect structural defects in countersunk rivets, the signal noise is strong and interference is significant, making it impossible to effectively separate the effective signal from the noise, resulting in poor testing performance.
After high-pass filtering preprocessing, 7-level discrete wavelet decomposition is performed using the sym6 wavelet basis. The denoising threshold is calculated using the min-max criterion, and the detail coefficients are denoised using a hard threshold function. Inverse wavelet transform is then performed to reconstruct the signal to extract the reflection coefficient and defect depth.
It effectively removes background noise, clearly identifies valid signals, accurately calculates the depth of defects in rivet structures, and significantly improves detection results.
Smart Images
Figure CN122385759A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of signal processing technology, and more specifically, to a method and system for detecting and identifying defects in rivet structures based on electromagnetic ultrasound. Background Technology
[0002] Countersunk rivets are widely used in critical parts of aerospace vehicles, air intakes, wings, and other equipment. Under long-term vibration, alternating loads, and complex environments, cracks are prone to form at the junction of the countersunk head and the shank, affecting the normal operation of the equipment.
[0003] Electromagnetic ultrasonic testing (EMAT) has advantages such as being non-contact, requiring no coupling agent, and having high detection efficiency, making it suitable for automated detection of structural defects in countersunk rivets. However, the following technical bottlenecks exist in actual testing:
[0004] First, the countersunk rivets are small in size, while the detection circuit that emits electromagnetic ultrasound has a high excitation voltage and a large receiving gain, resulting in strong noise and interference in the acquired raw signal. Second, electromagnetic ultrasound has an inherent detection blind zone, which causes the first defect echo of near-surface defects in the countersunk rivets to be submerged by the signal blind zone. The structural defects of the countersunk rivets can only be judged by the second defect echo. However, the amplitude of the second defect echo in the detection signal is weak and easily covered by noise. Conventional digital filtering cannot effectively separate the effective signal from the noise.
[0005] In summary, existing technologies suffer from poor noise reduction of countersunk rivet structural defect detection signals and an inability to effectively separate effective signals from noise. Summary of the Invention
[0006] The purpose of this application is to provide a method and system for detecting and identifying defects in rivet structures based on electromagnetic ultrasound, which solves the technical problems existing in the prior art, such as poor denoising effect on the detection signal of countersunk rivet structural defects and inability to effectively separate effective signals and noise.
[0007] To solve the above-mentioned technical problems, the solution adopted in this application is as follows:
[0008] A method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound includes:
[0009] S1: Collect the original time-domain signal obtained by electromagnetic ultrasonic wave to detect structural defects in the rivet structure, and perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal.
[0010] S2: The preprocessed original time-domain signal is decomposed into 7-level discrete wavelet based on the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer.
[0011] S3: Calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and use the hard thresholding function to denoise the detail coefficients of each wavelet decomposition layer to obtain the detail coefficients after hard thresholding denoising.
[0012] The approximation coefficients remain unchanged;
[0013] S4: Use the approximation coefficients and the detail coefficients after hard thresholding to perform inverse wavelet transform and reconstruct the denoised time domain signal;
[0014] S5: Locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo from the corresponding time positions in the original time domain signal, and calculate the reflection coefficient based on the signal amplitude.
[0015] S6: Calculate the depth of structural defects in the rivet structure based on the reflection coefficient.
[0016] Preferably, the specific implementation method of S1 includes:
[0017] Set the design parameters for the high-pass filter;
[0018] Design the transfer function of the high-pass filter based on the design parameters to obtain the high-pass filter;
[0019] The original time-domain signal is subjected to zero-phase filtering using a pre-designed high-pass filter to obtain the pre-processed original time-domain signal.
[0020] Preferably, the specific implementation method of S2 includes:
[0021] Select the sym6 wavelet basis and obtain the wavelet decomposition filter coefficients;
[0022] The preprocessed original time-domain signal is initialized to obtain the initial decomposition data. Specifically, the approximation coefficients of the initial wavelet decomposition layer are set to be equal to the preprocessed original time-domain signal.
[0023] Starting from the initial wavelet decomposition layer, the approximate coefficients of each wavelet decomposition layer are sequentially filtered, boundary processed, and halved downsampled to obtain the approximate coefficients of each wavelet decomposition layer and the detail coefficients of the highest layer, i.e., the set of wavelet coefficients.
[0024] The initial wavelet decomposition layer is layer 0, and the above processing is performed on the approximation coefficients of a total of 7 layers.
[0025] Preferably, the step of sequentially filtering, boundary processing, and half-sampling the approximation coefficients of each wavelet decomposition layer is specifically implemented by the following steps:
[0026] For the current wavelet decomposition layer, the approximation coefficients of the previous wavelet decomposition layer are convolved with the low-pass filter coefficients and the high-pass filter coefficients respectively to obtain the original convolution result.
[0027] The original convolution result is subjected to boundary processing using periodic extension to obtain the boundary-processed convolution result.
[0028] The convolution result after boundary processing is downsampled by half, and the value at the even index position is retained to obtain the approximation coefficient and detail coefficient of the current wavelet decomposition layer.
[0029] Use the approximation coefficients of the current wavelet decomposition layer as the input to the next wavelet decomposition layer.
[0030] Preferably, the specific implementation method of S3 includes:
[0031] Using the min-max criterion, the denoising threshold of the detail coefficients of each wavelet decomposition layer is calculated based on the length of the detail coefficients of each wavelet decomposition layer.
[0032] A hard thresholding function is used to denoise the detail coefficients of each wavelet decomposition layer. Detail coefficients with amplitudes less than the denoising threshold are set to zero, while detail coefficients with amplitudes greater than or equal to the denoising threshold are retained, resulting in the detail coefficients after hard thresholding denoising.
[0033] The approximation coefficients of the highest wavelet decomposition layer are left unprocessed, resulting in the processed set of wavelet coefficients.
[0034] Preferably, the specific implementation method of S4 includes:
[0035] Set the reconstruction approximation coefficient of the highest wavelet decomposition layer to be equal to the approximation coefficient of the highest wavelet decomposition layer, and set the reconstruction detail coefficient of the highest wavelet decomposition layer to be equal to the detail coefficient of the highest wavelet decomposition layer after hard threshold denoising.
[0036] Starting from the highest wavelet decomposition layer, upsampling, filtering, and summing are performed layer by layer to recover the denoised time-domain signal.
[0037] Preferably, the specific implementation method of the inverse wavelet transform includes:
[0038] The reconstruction approximation coefficients and reconstruction detail coefficients of the current wavelet decomposition layer are upsampled by a factor of two.
[0039] The upsampled reconstruction approximation coefficients are convolved with the reconstruction low-pass filter, and the upsampled reconstruction detail coefficients are convolved with the reconstruction high-pass filter to obtain the low-frequency and high-frequency components of the denoised time-domain signal, respectively.
[0040] By adding the low-frequency and high-frequency components point by point, the reconstruction approximation coefficients of the next wavelet decomposition layer are obtained.
[0041] Preferably, the specific implementation method of S5 includes:
[0042] Using the first end echo time of the defect-free rivet as the reference time position, in the denoised time domain signal, with the reference time position as the center, search for the peak point with the largest absolute value within a preset time window, and mark it as the time position of the first end echo.
[0043] The difference signal is obtained by subtracting the denoised time domain signal from the reference denoised time domain signal. The peak point with the largest absolute value of the signal is searched in the difference signal using the sliding window search method and marked as the time position of the second defect echo.
[0044] Based on the time position of the first end echo, the amplitude of the first end echo is extracted from the original time domain signal;
[0045] Based on the time location of the second defect echo, the amplitude of the second defect echo is extracted from the original time-domain signal.
[0046] Calculate the ratio of the second defect echo amplitude to the first end echo amplitude, and use it as the reflection coefficient.
[0047] Preferably, the specific implementation method of S6 includes:
[0048] Establish a linear calibration model for the reflection coefficient and defect depth;
[0049] The reflection coefficient is input into the linear calibration model to calculate the defect depth.
[0050] The electromagnetic ultrasound-based rivet structure defect detection and identification system is applicable to the aforementioned electromagnetic ultrasound-based rivet structure defect detection and identification method, including:
[0051] The preprocessing module is used to perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal;
[0052] The wavelet decomposition module, connected to the preprocessing module, is used to perform 7-level discrete wavelet decomposition on the preprocessed original time-domain signal using the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer.
[0053] The thresholding module, connected to the wavelet decomposition module, is used to calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and to denoise the detail coefficients of each wavelet decomposition layer using a hard thresholding function to obtain the detail coefficients after hard threshold denoising.
[0054] The signal reconstruction module is connected to the threshold processing module and is used to reconstruct the denoised time domain signal by performing inverse wavelet transform using the approximation coefficients and the detail coefficients after hard threshold denoising.
[0055] The echo localization and amplitude extraction module, connected to the signal reconstruction module, is used to locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo at the corresponding time positions in the original time domain signal, and calculate the reflection coefficient based on the signal amplitude.
[0056] The defect depth calculation module is connected to the echo positioning and amplitude extraction module to calculate the structural defect depth of the rivet structure based on the reflection coefficient.
[0057] The technical solution of this application has at least the following advantages and beneficial effects:
[0058] 1. In this invention, the detection signal of rivet structural defects is decomposed into a 7-level discrete wavelet based on the sym6 wavelet basis to obtain a set of wavelet coefficients. Hard thresholding denoising is then performed on the wavelet coefficient set in the wavelet transform domain to remove background noise from the detection signal. Based on the wavelet coefficients after hard thresholding denoising, the signal is reconstructed to restore the denoised detection signal. It can be seen from the restored denoised detection signal that the signal processing method in this invention is more effective at removing background noise than conventional filtering methods, and can clearly identify effective signals such as the first end echo signal and the second defect echo signal. Based on the clear identification of effective signals, the signal amplitudes of the first end echo signal and the second defect echo signal are extracted from the corresponding positions in the original detection signal, and the reflection coefficients are calculated. Finally, the depth of the structural defect is calculated based on the reflection coefficients. Through the above method, noise is significantly suppressed while the waveform characteristics of the effective signals are preserved relatively completely. Based on these waveform characteristics, the detection and depth measurement of rivet structural defects are achieved, resulting in good detection performance and solving the technical problems mentioned in the background art. Attached Figure Description
[0059] Figure 1 The waveform diagram for the sym6 wavelet basis;
[0060] Figure 2 Plot of sym6 wavelet basis scaling functions;
[0061] Figure 3 These are the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set obtained by decomposition.
[0062] Figure 4 Energy distribution of wavelet coefficients before hard thresholding denoising;
[0063] Figure 5The energy distribution diagram of the wavelet coefficient set after hard thresholding denoising;
[0064] Figure 6 The waveform of the reconstructed, denoised time-domain signal corresponding to a defect-free rivet;
[0065] Figure 7 The waveform of the reconstructed denoised time-domain signal corresponding to a rivet with a defect depth of 0.5 mm;
[0066] Figure 8 The reconstructed denoised time-domain signal waveform is shown for a rivet with a defect depth of 1 mm.
[0067] Figure 9 The reconstructed denoised time-domain signal waveform corresponding to a rivet with a defect depth of 1.5mm;
[0068] Figure 10 The reconstructed denoised time-domain signal waveform is shown for a rivet with a defect depth of 2mm.
[0069] Figure 11 The waveform of the detected signal is obtained by denoising the signal using a Butterworth high-pass filter.
[0070] Figure 12 The waveform of the detected signal is obtained by denoising the signal using a Butterworth bandpass filter.
[0071] Figure 13 This is a graph showing the relationship between the depth of structural defects and the reflection coefficient.
[0072] Figure 14 This is a flowchart of the method of the present invention. Detailed Implementation
[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0074] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. The terms "center," "upper," "lower," "inner," and "outer," indicating orientation or positional relationships based on the orientation or positional relationships shown in the figures, or the orientation or positional relationships commonly used when the product is in use, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or operated in a specific orientation, and therefore should not be construed as a limitation on this application. It should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two elements. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0075] See Figure 14 This invention discloses a method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound, comprising the following steps:
[0076] S1: Collect the original time-domain signal obtained by electromagnetic ultrasonic wave to detect structural defects in the rivet structure, and perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal.
[0077] S2: The preprocessed original time-domain signal is decomposed into 7-level discrete wavelet based on the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer.
[0078] S3: Calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and use the hard thresholding function to denoise the detail coefficients of each wavelet decomposition layer to obtain the detail coefficients after hard thresholding denoising.
[0079] The approximation coefficients remain unchanged;
[0080] S4: Use the approximation coefficients and the detail coefficients after hard thresholding to perform inverse wavelet transform and reconstruct the denoised time domain signal;
[0081] S5: Locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo from the corresponding time positions in the original time domain signal, and calculate the reflection coefficient based on the signal amplitude.
[0082] S6: Calculate the depth of structural defects in the rivet structure based on the reflection coefficient.
[0083] In this embodiment, in S1, the original time-domain signal undergoes high-pass filtering preprocessing, and the specific implementation method includes the following steps:
[0084] S1.1: Set the design parameters for the high-pass filter;
[0085] S1.2: Design the transfer function of the high-pass filter based on the design parameters to obtain the high-pass filter. ;
[0086] S1.3: The original time-domain signal is subjected to zero-phase filtering using a pre-designed high-pass filter to obtain the pre-processed original time-domain signal, specifically as follows:
[0087] Will Forward through high-pass filter The process is performed to obtain the first filtered result sequence. ;
[0088] The sequence of the first filtering results Time reversal, obtain ;
[0089] in, This is the sequence of the first filtered results after time reversal;
[0090] Will Passing through the high-pass filter again The process is performed to obtain the second filtered result sequence. ;
[0091] The sequence of the second filtering results Reverse time to obtain the output sequence , recorded as .
[0092] in, ( ) represents the original time-domain signal sequence.
[0093] As an example, the above zero-phase filtering process can also employ a zero-phase function. For the original time-domain signal sequence The present invention does not limit the processing method.
[0094] Specifically, the high-pass filter has an order of 2 to 4 and a cutoff frequency of 2 MHz. The transfer function of the high-pass filter can be the transfer function of an existing high-pass filter, which can be implemented by those skilled in the art through existing high-pass filters, so it will not be elaborated here.
[0095] Through the above processing, the original time-domain signal is preprocessed with high-pass filtering, which suppresses the low-frequency circuit oscillation and baseline drift interference inherent in electromagnetic ultrasonic testing. At the same time, zero-phase filtering makes the time position of the echo peak more accurate. Through the above processing, a cleaner and higher quality signal is provided for the detection and identification of rivet structure defects.
[0096] In this embodiment, in S2, the preprocessed original time-domain signal is decomposed into a 7-level discrete wavelet decomposition using the sym6 wavelet basis to obtain a set of wavelet coefficients. The specific implementation method includes the following steps:
[0097] S2.1: Select the sym6 wavelet basis function and obtain the wavelet decomposition filter coefficients;
[0098] See Figure 1 and Figure 2 , Figure 1 The waveform diagram for the sym6 wavelet basis; Figure 2 The scaling function diagram of the sym6 wavelet basis is shown. In this invention, the sym6 wavelet basis is selected as the wavelet basis for decomposition. The sym6 wavelet has the following characteristics: support length 12, 6th order vanishing moment, and approximate symmetry.
[0099] The decomposition low-pass filter coefficients corresponding to the sym6 wavelet basis and high-pass filter coefficients The sequence is a predetermined constant with a length of 12. Its specific value can be obtained by querying the MATLAB wavelet toolbox.
[0100] In this invention, a three-cycle sinusoidal ultrasonic signal is used as the ultrasonic excitation signal for defect detection of rivet structures. The waveform shape of the sym6 wavelet is very similar to that of the ultrasonic excitation signal. Therefore, the sym6 wavelet is the best choice for wavelet decomposition of the preprocessed original time-domain signal.
[0101] S2.2: Initialize the preprocessed original time-domain signal to obtain the initialized decomposition data, specifically by setting the 0th level approximation coefficients... Equal to the preprocessed original time-domain signal,
[0102] ;
[0103] in, The number of sampling points. Set the total number of wavelet decomposition levels. ; For the combination of the approximate coefficients of the 0th layer, ; This is the preprocessed original time-domain signal;
[0104] S2.3: Perform filtering, boundary processing, and downsampling on the initial decomposed data to obtain a set of wavelet coefficients.
[0105] In this embodiment, in S2.3, for arrive The process involves sequentially performing filtering, boundary processing, and downsampling. The specific implementation includes the following steps:
[0106] S2.3.1: Approximate the coefficients of the previous wavelet decomposition layer respectively with the low-pass filter coefficients and high-pass filter coefficients Perform convolution to obtain the original convolution result. and ;
[0107] ;
[0108] ;
[0109] in, This represents the current wavelet decomposition level. For the first The approximate coefficients of the wavelet decomposition layer have a length of [missing information]. ; This is a sequence of low-pass filter coefficients. This is an index of the low-pass filter coefficient sequence. ,in =12 is the filter length; This is a sequence of high-pass filter coefficients. This is an index into the high-pass filter coefficient sequence. ,in =12 is the filter length; The index of the original convolution result; This is the alignment parameter, used to adjust the center alignment of the filter. It typically takes a value of 0 or 1; in this embodiment, it is set to 1. ;
[0110] S2.3.2: The original convolution result and Boundary treatment is performed using periodic continuation;
[0111] Specifically: when calculating convolution, when the index Exceeding the approximation coefficient of the previous wavelet decomposition layer Valid index range When, from the approximation coefficient The corresponding values are taken cyclically, and the corresponding formula is as follows:
[0112] ;
[0113] ;
[0114] To exceed the approximation coefficients of the previous wavelet decomposition layer Valid index range The offset.
[0115] After periodic extension, the boundary-corrected convolution result is obtained, which is still denoted as... and .
[0116] S2.3.3: The convolution result after boundary processing is downsampled by half, retaining the values at even-numbered index positions, to obtain the approximate coefficients of the current wavelet decomposition layer. and detail coefficient Approximation coefficients As the first Input to layer wavelet decomposition;
[0117] ;
[0118] ;
[0119] in, This is the coefficient index for downsampling; An even index value indicates that a value is taken every other point; Indicates the floor function;
[0120] See Figure 3 , Figure 3 This shows the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set obtained by decomposition. The wavelet coefficient set is... Including the approximation coefficients of the highest wavelet decomposition layer and the detail coefficients of each wavelet decomposition layer See Table 1, which shows the length of the coefficients in each layer of the wavelet coefficient set.
[0121] Table 1. Length of coefficients at each level in the wavelet coefficient set.
[0122]
[0123] In this step, the preprocessed original time-domain signal is decomposed into 7-level wavelet coefficients using the sym6 wavelet basis. Since the ultrasonic excitation signal for defect detection of the rivet structure is a three-period sinusoidal signal, and the sym6 wavelet is symmetrical and similar to the ultrasonic excitation signal, the sym6 wavelet is selected to process the preprocessed original time-domain signal in this invention. This maps the preprocessed original time-domain signal to multiple frequency sub-bands, effectively separating the effective signal from the noise in the preprocessed original time-domain signal. Periodic extension and other processing techniques are then used to process both ends of the original time-domain signal, ensuring the boundary fidelity of the original time-domain signal and providing wavelet coefficients with concentrated energy and high signal-to-noise ratio for subsequent steps.
[0124] In this embodiment, in S3, the denoising threshold of each level of detail coefficients in the wavelet coefficient set is calculated using the min-max criterion, and a hard thresholding function is used to denoise each level of detail coefficients to obtain the detail coefficients after hard thresholding denoising. The specific implementation method includes:
[0125] S3.1: Calculate the denoising threshold for the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion. The specific formula for calculating the threshold using the min-max criterion is as follows:
[0126] ;
[0127] in, for Length of layer detail factor;
[0128] S3.2: Use a hard thresholding function to evaluate the detail coefficients c of each wavelet decomposition layer. Denoising is performed to obtain the detail coefficients after hard thresholding. The expression for the hard threshold function is:
[0129] ;
[0130] S3.3: For approximation coefficients No action was taken. The processed wavelet coefficient set is obtained. .
[0131] In this step, the hard thresholding function does not change the amplitude of the detail coefficients. It is used to process the second defect echo in the obtained rivet structure defect detection signal, thus better preserving the signal's authenticity. See also Figure 4 and Figure 5 , Figure 4 This is the energy distribution map of the wavelet coefficient set before hard thresholding denoising. Figure 5The image shows the energy distribution of the wavelet coefficient set after hard threshold denoising. It can be seen that before denoising, the energy distribution of the wavelet coefficients was mixed with the energy distribution of the interference signal. After denoising, the energy distribution of the wavelet coefficients is concentrated, and the energy of the interference signal is greatly reduced. Thus, hard threshold denoising effectively removes the noise interference in the original wavelet coefficient set.
[0132] In this embodiment, in step S4, the detail coefficients and approximation coefficients after hard threshold denoising are used to perform inverse wavelet transform to reconstruct the denoised time-domain signal. The specific implementation method includes:
[0133] S4.1: Set the approximation coefficients for the 7th layer reconstruction and reconstruction detail coefficient ;
[0134] ;
[0135] ;
[0136] S4.2: For Up to step 1, perform upsampling, filtering, and summing processes layer by layer to recover the denoised time-domain signal. ;
[0137] Furthermore, the specific implementation methods of the inverse wavelet transform in S4.2 include:
[0138] S4.2.1: Approximation coefficients for reconstruction and reconstruction detail coefficient Perform upsampling by 2 times each;
[0139] S4.2.2: Reconstruct the approximation coefficients after upsampling. With the reconstructed low-pass filter coefficients Perform convolution to reconstruct the detail coefficients after upsampling. With reconstructed high-pass filter coefficients Convolution is performed to obtain the low-frequency and high-frequency components of the denoised time-domain signal, respectively.
[0140] Among them, the reconstructed low-pass filter coefficients Reconstructing the high-pass filter coefficients The low-pass filter coefficients used in wavelet transform with S2 High-pass filter coefficients They satisfy an orthogonal mirror relationship and have the same length;
[0141] In addition, the same periodic extension method as in S2 can be used to process the signal boundaries during convolution to ensure the continuity of both ends of the signal.
[0142] S4.2.3: Add the low-frequency and high-frequency components of the obtained denoised time-domain signal point by point to obtain the first... Reconstruction approximation coefficients of the layer .
[0143] See Figure 6-10 , Figure 6-10 The reconstructed denoised time-domain signal is shown. Figure 6 The waveform of the reconstructed, denoised time-domain signal corresponding to a defect-free rivet; Figure 7 The waveform of the reconstructed denoised time-domain signal corresponding to a rivet with a defect depth of 0.5 mm; Figure 8 The reconstructed denoised time-domain signal waveform is shown for a rivet with a defect depth of 1 mm. Figure 9 The reconstructed denoised time-domain signal waveform corresponding to a rivet with a defect depth of 1.5mm; Figure 10 The image shows the reconstructed denoised time-domain signal waveform for a rivet with a defect depth of 2mm. In this step, the signal in the wavelet transform domain is subjected to inverse wavelet transform to restore the denoised time-domain signal. Figure 6-10 In the graph, the horizontal axis represents distance (unit: mm), and the vertical axis represents signal voltage (unit: V). 1 indicates the signal dead zone, 2 indicates the initial pulse of the detection signal, 3 indicates the first end echo, 4 indicates the second end echo, and 5 indicates the second defect echo. As can be seen from the graph, the first defect echo is submerged in the signal dead zone after the initial pulse and cannot be identified or processed. The second defect echo, following the first end echo, is not submerged in interference signals and can be effectively identified and processed through noise reduction.
[0144] See Figure 11-12 , Figure 11 The waveform of the detected signal is obtained by denoising the signal using a Butterworth high-pass filter. Figure 12 The image shows the signal waveform obtained by denoising the detection signal using a Butterworth bandpass filter. Figure 11-12 In the diagram, the horizontal axis represents distance (unit: mm), and the vertical axis represents signal voltage (unit: V). Figure 11 This diagram shows the signal waveform after denoising the detection signal using a 2MHz Butterworth high-pass filter. Figure 12 This represents the signal waveform after denoising the detected signal using a 6-12MHz Butterworth bandpass filter. Figure 11 , Figure 12In the two sub-figures, 1 indicates the signal dead zone, 2 indicates the initial pulse of the detection signal, 3 indicates the first end echo, and 4 indicates the second end echo. It can be seen from the figures that there is a significant signal dead zone after the initial pulse of the detection signal, followed by a noticeable oscillation signal. This is due to the resonance between the coil and the receiving circuit in the circuit emitting the ultrasonic excitation signal. At distances of 16mm and 20mm, the ultrasonic signals can be clearly identified, representing the first and second end echoes, respectively. Simultaneously, background noise exists between the first and second end echoes. Even with digital high-pass and band-pass filters for noise reduction, the background noise remains significant, and its frequency is very close to that of the ultrasonic signal, making the second defect echo difficult to identify and remove. Meanwhile, in comparison... Figure 6-10 The detection signal shown is obtained after processing by wavelet transform, threshold denoising, etc. It is obvious that the processing method of the present invention has effectively suppressed the background noise, and the background noise has been almost completely filtered out. As can be seen from the figure, the clutter around the second defect echo is small and can be easily identified and extracted.
[0145] In this embodiment, in step S5, the time positions of the second defect echo and the first end echo are located in the denoised time-domain signal, and the amplitudes of the second defect echo and the first end echo are extracted from the corresponding time positions in the original time-domain signal. The reflection coefficient is calculated based on the signal amplitude. The specific implementation method includes:
[0146] S5.1: Using the first end echo time of the defect-free rivet as the reference time position, in the actual denoised time domain signal... In the process, with the reference time position as the center, the peak point of the denoised time-domain signal with the largest absolute value is searched within a preset time window, and its time point is marked as [missing information]. ;
[0147] Specifically, the first end echo time of a defect-free rivet can be calculated based on the rivet's geometry and the propagation speed of electromagnetic ultrasonic waves. The reference time position is marked as follows. The preset time window is ;
[0148] Typically, the value is taken as 0.2µs-0.5µs. In this embodiment, Take 0.3us.
[0149] S5.2: Denoise the time-domain signal Compared with the reference denoised time domain signal Subtracting them yields the differential signal. and in differential signal The sliding window search method is used to find the peak point with the largest absolute value of the signal, and its time point is marked as [the peak point]. ;
[0150] Specifically, differential signals The location of the highest mid-peak value is the possible location of the second defect echo; reference denoised time-domain signal The denoised time-domain signal of a defect-free rivet of the same model;
[0151] The second defect echo appears after the first end echo, and its time delay is related to the defect depth.
[0152] If no valid [test] is detected This indicates that the tested rivet has no defects and no further processing is required.
[0153] S5.3: Based on time In the original time domain signal Extract the first end echo amplitude value ;
[0154] S5.4: Based on time In the original time domain signal Extracting the second defect echo amplitude ;
[0155] S5.5: Based on the first end echo amplitude value and the amplitude of the second defect echo Calculate the reflection coefficient ;
[0156] ;
[0157] In this step, the time point with the largest amplitude of the first end echo and the second defect echo is first found in the denoised time-domain signal. Based on this time point, the amplitudes of the first end echo and the second defect echo are extracted from the original time-domain signal, and the reflection coefficient is calculated based on these amplitudes. Finding the time point with the largest amplitude of the first end echo and the second defect echo in the time-domain signal is because noise removal makes it easier and more accurate to locate their positions. Extracting the amplitude from the original time-domain signal based on this location is because the original time-domain signal has not undergone wavelet transform, denoising, or other processing, and therefore has no nonlinear changes. Thus, the extracted amplitude data is more accurate, and the reliability of the reflection coefficient is higher.
[0158] In this embodiment, in step S6, the depth of the structural defect in the rivet structure is calculated based on the reflection coefficient.
[0159] S6.1: Establish a calibration model for the reflection coefficient and defect depth;
[0160] Specifically, several sample rivets of the same type, material, and surface condition as the rivet to be tested are used. Groove defects of varying depths are machined at the junction of the countersunk head and the shank. The depth of the defect is denoted as... For each sample rivet, the inspection is carried out according to steps S1-S5, and the reflection coefficient corresponding to its defect depth is calculated. ;
[0161] Defect depth as the dependent variable The reflection coefficient is the independent variable. The calibration model was fitted using the least squares method.
[0162] ;
[0163] in, To calibrate the slope of the model, The intercept is used to calibrate the model.
[0164] S6.2: Input the reflection coefficient obtained in S5 into the calibration model to calculate the defect depth.
[0165] In conclusion, Figure 6-10 The results of processing the detection signal of rivet structural defects using the method of the present invention are shown in the figure. Figure 6-10 In the diagram, the horizontal axis is converted to distance in millimeters based on the speed of electromagnetic ultrasound. It can be seen that there is a noticeable initial pulse at the beginning of the detection signal. This is due to the high-pass filtering of the original time-domain signal in the first step of signal processing. Figure 6 The detection signal of a countersunk rivet without structural defects is shown, and the signal is relatively pure after processing by S1-S6; from Figure 7-10 It can be observed that after noise reduction, a clear second defect echo can be identified. The difference between the horizontal coordinates of the second defect echo and the first end echo is about 1.5 mm, which is the same as the actual defect location. Figure 13 The paper demonstrates the relationship between the reflection coefficient calculated using experimental signals and the defect depth, and normalizes the calculation results. It can be seen that the reflection coefficient is linearly positively correlated with the defect depth, which indicates that the method of calculating the reflection coefficient using the second defect echo and the first end echo in this invention can effectively characterize the defect depth.
[0166] Another aspect of the present invention discloses a rivet structure defect detection and identification system based on electromagnetic ultrasound, applicable to the aforementioned rivet structure defect detection and identification method based on electromagnetic ultrasound, comprising:
[0167] The preprocessing module is used to perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal;
[0168] The wavelet decomposition module, connected to the preprocessing module, is used to perform 7-level discrete wavelet decomposition on the preprocessed original time-domain signal using the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer.
[0169] The thresholding module, connected to the wavelet decomposition module, is used to calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and to denoise the detail coefficients of each wavelet decomposition layer using a hard thresholding function to obtain the detail coefficients after hard threshold denoising.
[0170] The signal reconstruction module is connected to the threshold processing module and is used to reconstruct the denoised time domain signal by performing inverse wavelet transform using the approximation coefficients and the detail coefficients after hard threshold denoising.
[0171] The echo localization and amplitude extraction module, connected to the signal reconstruction module, is used to locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo at the corresponding time positions in the original time domain signal, and calculate the reflection coefficient based on the signal amplitude.
[0172] The defect depth calculation module is connected to the echo positioning and amplitude extraction module to calculate the structural defect depth of the rivet structure based on the reflection coefficient.
[0173] The various embodiments of the present invention have now been described in detail. To avoid obscuring the concept of the invention, some details known in the art have not been described. Those skilled in the art will fully understand how to implement the technical solutions of this invention based on the above description, and the scope of the invention is defined by the appended claims.
Claims
1. A method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound, characterized in that, include: S1: Collect the original time-domain signal obtained by electromagnetic ultrasonic wave to detect structural defects in the rivet structure, and perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal. S2: The preprocessed original time-domain signal is decomposed into 7-level discrete wavelet based on the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer. S3: Calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and then use a hard thresholding function to denoise the detail coefficients of each wavelet decomposition layer to obtain the detail coefficients after hard thresholding denoising. The approximation coefficients remain unchanged; S4: Use the approximation coefficients and the detail coefficients after hard thresholding to perform inverse wavelet transform and reconstruct the denoised time domain signal; S5: Locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo at the corresponding time positions in the original time domain signal, respectively. Calculate the reflection coefficient based on the extracted signal amplitudes. S6: Calculate the depth of structural defects in the rivet structure based on the reflection coefficient.
2. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 1, characterized in that, The specific implementation method of S1 includes: Set the design parameters for the high-pass filter; Design the transfer function of the high-pass filter based on the design parameters to obtain the high-pass filter; The original time-domain signal is subjected to zero-phase filtering using a pre-designed high-pass filter to obtain the pre-processed original time-domain signal.
3. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 2, characterized in that, The specific implementation method of S2 includes: Select the sym6 wavelet basis and obtain the wavelet decomposition filter coefficients; The preprocessed original time-domain signal is initialized to obtain the initial decomposition data. Specifically, the approximation coefficients of the initial wavelet decomposition layer are set to be equal to the preprocessed original time-domain signal. Starting from the initial wavelet decomposition layer, the approximate coefficients of each wavelet decomposition layer are sequentially filtered, boundary processed, and halved downsampled to obtain the approximate coefficients of each wavelet decomposition layer and the detail coefficients of the highest layer, i.e., the set of wavelet coefficients. The initial wavelet decomposition layer is layer 0, and a total of 7 layers of wavelet decomposition processing are performed.
4. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 3, characterized in that, The step of sequentially filtering, boundary processing, and half-downsampling the approximation coefficients of each wavelet decomposition layer includes the following steps: For the current wavelet decomposition layer, the approximation coefficients of the previous wavelet decomposition layer are convolved with the low-pass filter coefficients and the high-pass filter coefficients respectively to obtain the original convolution result. The original convolution result is subjected to boundary processing using periodic extension to obtain the boundary-processed convolution result. The convolution result after boundary processing is downsampled by half, and the value at the even index position is retained to obtain the approximation coefficient and detail coefficient of the current wavelet decomposition layer. Use the approximation coefficients of the current wavelet decomposition layer as the input to the next wavelet decomposition layer.
5. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 4, characterized in that, The specific implementation method of S3 includes: Using the min-max criterion, the denoising threshold of the detail coefficients of each wavelet decomposition layer is calculated based on the length of the detail coefficients of each wavelet decomposition layer. A hard thresholding function is used to denoise the detail coefficients of each wavelet decomposition layer. Detail coefficients with amplitudes less than the denoising threshold are set to zero, while detail coefficients with amplitudes greater than or equal to the denoising threshold are retained, resulting in the detail coefficients after hard thresholding denoising. The approximation coefficients of the highest wavelet decomposition layer are left unprocessed, resulting in the processed set of wavelet coefficients.
6. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 5, characterized in that, The specific implementation method of S4 includes: Set the reconstruction approximation coefficient of the highest wavelet decomposition layer to be equal to the approximation coefficient of the highest wavelet decomposition layer, and set the reconstruction detail coefficient of the highest wavelet decomposition layer to be equal to the detail coefficient of the highest wavelet decomposition layer after hard threshold denoising. Starting from the highest wavelet decomposition layer, upsampling, filtering, and summing are performed layer by layer to recover the denoised time-domain signal.
7. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 6, characterized in that, The specific implementation method of the inverse wavelet transform includes: The reconstruction approximation coefficients and reconstruction detail coefficients of the current wavelet decomposition layer are upsampled by a factor of two. The upsampled reconstruction approximation coefficients are convolved with the reconstruction low-pass filter, and the upsampled reconstruction detail coefficients are convolved with the reconstruction high-pass filter to obtain the low-frequency and high-frequency components of the denoised time-domain signal, respectively. The low-frequency and high-frequency components are added point by point to obtain the reconstruction approximation coefficients of the next wavelet decomposition layer.
8. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 7, characterized in that, The specific implementation method of S5 includes: Using the first end echo time of the defect-free rivet as the reference time position, in the denoised time domain signal, with the reference time position as the center, search for the peak point with the largest absolute value within a preset time window, and mark it as the time position of the first end echo. The difference signal is obtained by subtracting the denoised time domain signal from the reference denoised time domain signal. The peak point with the largest absolute value of the signal is searched in the difference signal using the sliding window search method and marked as the time position of the second defect echo. Based on the time position of the first end echo, the amplitude of the first end echo is extracted from the original time domain signal; Based on the time location of the second defect echo, the amplitude of the second defect echo is extracted from the original time-domain signal. Calculate the ratio of the second defect echo amplitude to the first end echo amplitude, and use it as the reflection coefficient.
9. The method for detecting and identifying defects in rivet structures based on electromagnetic ultrasound according to claim 8, characterized in that, The specific implementation method of S6 includes: Establish a linear calibration model for the reflection coefficient and defect depth; The reflection coefficient is input into the linear calibration model to calculate the defect depth.
10. A rivet structure defect detection and identification system based on electromagnetic ultrasound, applicable to the rivet structure defect detection and identification method based on electromagnetic ultrasound as described in any one of claims 1-9, characterized in that, include: The preprocessing module is used to perform high-pass filtering preprocessing on the original time-domain signal to obtain the preprocessed original time-domain signal; The wavelet decomposition module, connected to the preprocessing module, is used to perform 7-level discrete wavelet decomposition on the preprocessed original time-domain signal using the sym6 wavelet basis to obtain a set of wavelet coefficients. The set of wavelet coefficients includes the detail coefficients of each wavelet decomposition layer and the approximation coefficients of the highest wavelet decomposition layer. The thresholding module, connected to the wavelet decomposition module, is used to calculate the denoising threshold of the detail coefficients of each wavelet decomposition layer in the wavelet coefficient set using the min-max criterion, and to denoise the detail coefficients of each wavelet decomposition layer using a hard thresholding function to obtain the detail coefficients after hard threshold denoising. The signal reconstruction module is connected to the threshold processing module and is used to reconstruct the denoised time domain signal by performing inverse wavelet transform using the approximation coefficients and the detail coefficients after hard threshold denoising. The echo localization and amplitude extraction module, connected to the signal reconstruction module, is used to locate the time positions of the second defect echo and the first end echo in the denoised time domain signal, and extract the amplitude of the second defect echo and the amplitude of the first end echo at the corresponding time positions in the original time domain signal, and calculate the reflection coefficient based on the signal amplitude. The defect depth calculation module is connected to the echo positioning and amplitude extraction module to calculate the structural defect depth of the rivet structure based on the reflection coefficient.