'Peak entropy'-based CFRP interface weak bonding defect dynamic detection method

Abstract

The invention discloses a CFRP interface weak bonding defect dynamic detection method based on peak entropy. The CFRP interface weak bonding defect dynamic detection method specifically comprises the following steps that 1, particle speed signals on the back face of a nondestructive test piece under laser shock are collected through a photon Doppler speed measurement system; step 2, preprocessing the particle speed signal; step 3, carrying out S transformation time-frequency analysis and extreme value analysis on the preprocessed particle speed signal, and calculating an amplitude difference between extreme value points; 4, calculating the Shannon entropy of the amplitude difference sequence between the extreme points, and generating a peak entropy feature; and 5, extracting peak entropy characteristics, and judging the damage state of the nondestructive test piece. According to the CFRP interface weak bonding defect dynamic detection method based on the peak entropy, dynamic response signals are collected through single-time laser shock, and single-time shock detection is achieved through S transformation time-frequency analysis and Shannon entropy feature extraction.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-destructive testing methods for composite materials, and particularly relates to a dynamic detection method for CFRP interface weak bonding defects based on "peak entropy". Background Art

[0002] Carbon fiber reinforced composite materials (CFRP) are widely used in the aerospace field due to their lightweight and high specific strength characteristics. However, their adhesive interfaces are prone to "weak bonding defects" due to manufacturing or service problems. Traditional non-destructive testing techniques detect defects inside or on the surface of materials through physical or chemical methods, such as ultrasonic and infrared thermography techniques, which mainly rely on changes in the geometric morphology of the material or differences in acoustic impedance to detect defects. However, for CFRP interface weak bonding defects, such as when the bonding strength is less than 20% of the normal value, traditional testing techniques usually do not cause obvious changes in geometric morphology, such as cracks, holes, etc., nor significant differences in acoustic impedance when detecting CFRP interface weak bonding defects. Therefore, traditional techniques such as ultrasonic and infrared thermography are difficult to detect such defects. Existing entropy methods, including permutation entropy and fuzzy entropy, etc., although they can characterize the disorder degree of signals, have the following deficiencies: they cannot distinguish the entropy change differences caused by material non-uniformity noise and local damage; they rely on signal analysis under quasi-static loads and are difficult to capture dynamic impact transient responses; they require a large amount of labeled data and have insufficient diagnostic accuracy under small sample conditions. The laser shock wave interface bonding strength detection technique (LBI) activates defects through the stress wave reflection and tensile effect, but the existing methods still have problems such as low efficiency and high cost.

[0003] The LSPT company in the United States uses a "low-high-low" three-time impact detection method (the first low-energy calibration reference signal, the second high-energy activation of defects, and the third low-energy verification of damage). This detection method's equipment relies on a fixed pulse width laser, requires multiple impacts to cover interfaces at different depths, and depends on complex signal comparison, resulting in a long detection cycle and large equipment loss. There is also the two-time impact detection method of "low-high" proposed by the Air Force Engineering University (the first low-energy calibration reference signal, the second high-energy activation of potential weak bonding defects, and by comparing the stress wave signals of the two impacts, determining whether there is a delamination phenomenon at the bonding interface), which requires high-precision extraction of dynamic response characteristics and is sensitive to noise. There are still deficiencies in the analysis of the signal coupling law of multi-interface complex structures. Summary of the Invention

[0004] The purpose of the present invention is to provide a dynamic detection method for CFRP interface weak bonding defects based on "peak entropy", which collects dynamic response signals through a single laser shock, and uses S-transform time-frequency analysis and Shannon entropy feature extraction to achieve single-impact detection.

[0005] The technical solution adopted by the present invention is a dynamic detection method for weak bonding defects at the CFRP interface based on "peak entropy", which specifically includes the following steps: Step 1: Collect the particle velocity signal on the back surface of the non-destructive specimen under laser shock through a photon Doppler velocimetry system; Step 2: Preprocess the particle velocity signal; Step 3: Perform S-transform time-frequency analysis and extreme value analysis on the preprocessed particle velocity signal, and calculate the amplitude difference between extreme points; Step 4: Calculate the Shannon entropy of the amplitude difference sequence between extreme points to generate peak entropy features; Step 5: Extract peak entropy features and judge the damage state of the non-destructive specimen. The characteristics of the present invention also lie in that, In Step 1, the non-destructive specimen is made of carbon fiber reinforced composite material.

[0006] Step 2 is specifically as follows: First, perform downsampling to reduce the sampling rate of the original signal from 20 GS / s to 1 GS / s, then perform baseline drift correction, extract the low-frequency trend term by wavelet decomposition, and eliminate the signal baseline offset through algebraic subtraction; finally, use a low-pass digital filter with a cut-off frequency of 500 MHz to eliminate environmental noise and high-frequency interference of the instrument.

[0007] Step 3 is specifically as follows: Perform S-transform time-domain analysis on the preprocessed signal to locate the transient energy peak region caused by stress wave reflection-unloading coupling; then identify the wave peaks and wave valleys of the particle velocity curve by the second-order difference method, and eliminate the pseudo-peaks with an adjacent extreme value interval less than 0.8 μs; starting from the first main wave peak, select the time interval when the velocity decays to 20%-30% of the peak value as the feature analysis window, and statistically analyze the probability distribution of the amplitude difference between consecutive extreme points within this time interval.

[0008] In Step 3, identifying the wave peaks and wave valleys of the particle velocity curve by the second-order difference method is specifically as follows: First, calculate the first-order difference of the denoised velocity curve to obtain the change rate of the particle velocity signal; then perform a difference operation on the first-order difference data again to obtain the second-order difference, so as to reflect the curvature change of the particle velocity signal, as shown in formula (1): (1) In the formula, is the value of the particle velocity signal at point ; h is the step size.

[0009] In Step 3, calculating the amplitude difference between extreme points is specifically as follows: (2) In the formula, and respectively represent adjacent extreme points represents the amplitude difference .

[0010] Adopt the adaptive binning method to statistically analyze the probability distribution of the amplitude differences between consecutive extreme points within the time interval of step 3.

[0011] Step 4 is specifically as follows: Use the Shannon entropy formula to calculate the peak entropy value, which reflects the degree of chaos in the extreme value changes in the particle velocity signal, and is specifically expressed as follows: (3) In the formula, represents the probability of the

[0012] Step 5 is specifically as follows: When the peak entropy has a step change and the peak entropy corresponding to the undamaged specimen is "0", the signal disorder degree is relatively low, and the corresponding specimen is in an undamaged state; when the peak entropy is greater than "0", the degree of signal chaos increases, and the corresponding specimen is in a damaged state. Compared with the prior art, the beneficial effects of the present invention are: (1) The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" provided by the present invention realizes the quantitative description of the signal chaos degree by statistically analyzing the probability distribution of the amplitude differences between consecutive extreme values in the signal. The detection can be completed with a single impact, avoiding the cumbersome processes of "low-high-low" three impacts or "low-high" two impacts; the peak entropy feature is sensitive to interface damage, and the diagnostic accuracy rate is high; it reduces the loss of the laser and the experimental time consumption, and the cost is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a schematic flow chart of the dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" of the present invention; Figure 2 is a schematic curve diagram of the back particle velocity with a specimen thickness of 1.5 mm in Example 7 of the present invention; Figure 3 is a schematic curve diagram of the back particle velocity with a specimen thickness of 3 mm in Example 7 of the present invention; Figure 4 is a schematic diagram of the influence trend of peak entropy in the undamaged and damaged states of the specimen under different impact parameters in Example 7 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. The described embodiments are only some of the embodiments of the present invention, rather than all of the embodiments.

[0015] Example 1 The present invention provides a dynamic detection method for weak bonding defects at the CFRP interface based on "peak entropy", as Figure 1 shown, specifically: Step 1: Collect the particle velocity signal on the back of a carbon fiber reinforced polymer (CFRP) specimen under laser shock through a photon Doppler velocimetry (PDV) system; Step 2: Preprocess the particle velocity signal: Step 3: Perform S-transform time-frequency analysis and extreme value analysis on the preprocessed particle velocity signal, and calculate the amplitude difference between extreme points; Step 4: Calculate the Shannon entropy of the amplitude difference sequence between extreme points to generate peak entropy features; Step 5: Extract the peak entropy features and judge the damage state of the non-damaged specimen.

[0016] Example 2 On the basis of Example 1, Step 2 is specifically: Reduce the sampling rate from 20 GS / s to 1 GS / s by downsampling, but it must satisfy the Nyquist sampling theorem. As long as the sampling rate is not lower than twice the highest frequency in the signal, the true shape of the signal can still be restored after reducing the sampling rate without distortion.

[0017] Among them, the data volume is reduced to 1 / 20 of the original, and subsequent processing is faster, while retaining the key features of the signal.

[0018] Then perform baseline drift correction, select the db4 wavelet basis for 5-layer decomposition, split the signal into 5 different thickness levels, extract the low-frequency trend term of the bottom layer, that is, the drift trend of the overall signal. Eliminate the signal baseline offset through algebraic subtraction; finally, use a low-pass digital filter with a cut-off frequency of 500 MHz to eliminate environmental noise and high-frequency interference of the instrument.

[0019] Example 3 On the basis of Example 2, Step 3 is specifically: Perform S-transform time-domain analysis on the preprocessed signal, with the Gaussian window time-width coefficient of 0.8; locate the transient energy peak region caused by stress wave reflection-unloading coupling; then identify the wave peaks and wave valleys of the particle velocity curve through the second-order difference method, and eliminate the pseudo-peaks with the adjacent extreme value interval less than 0.8 μs; take the first main wave peak as the starting point, select the time interval when the velocity decays to 20%-30% of the peak value as the feature analysis window, and statistically analyze the probability distribution of the amplitude difference between consecutive extreme points within this time interval.

[0020] Among them, identifying the wave peaks and wave valleys of the particle velocity curve through the second-order difference method is specifically: First, calculate the first-order difference of the denoised velocity curve to obtain the change rate of the particle velocity signal; then, perform a difference operation on the first-order difference data again to obtain the second-order difference, thereby reflecting the curvature change of the particle velocity signal, as shown in formula (1): (1) In the formula, is the value of the particle velocity signal at point ; h is the step size.

[0021] By calculating the second-order difference, the concavity and convexity of the signal can be judged, thereby identifying the wave peaks and wave valleys. According to the sign change of the second-order difference: when it changes from a positive value to a negative value, it indicates that the signal changes from rising to falling, corresponding to a local maximum (wave peak); on the contrary, when the second-order difference changes from a negative value to a positive value, it indicates that the signal changes from falling to rising, corresponding to a local minimum (wave valley).

[0022] Secondly, after identifying the wave peaks and wave valleys, it is necessary to delimit the time interval that can reflect the key dynamic response characteristics. Take the first wave peak in the signal as the starting moment, and the time range from 0.05 to 10 times the time it takes for the velocity in the curve signal to reach the peak value of the first wave peak is the effective analysis interval. This analysis interval is crucial for accurately evaluating spall damage. For specimens with different thicknesses under different impact parameters, there are significant differences in the particle velocity on the back surface. If a single set of data is analyzed separately, the calculation amount will be greatly increased, and the constructed discrimination method lacks generality. From the feasibility, it can be seen that there are significant differences in the descending edge of the first main wave peak of the particle velocity curve on the back surface of the specimen in the undamaged and damaged states. Therefore, the present invention takes the first wave peak in the signal as the starting moment. In addition, the time for the stress wave to experience a complete propagation coupling in specimens with different thicknesses is inconsistent, and different cut-off times need to be set for different specimen thicknesses, which seriously affects the calculation efficiency and the generality of the discrimination method. Therefore, the present invention takes the moment when the velocity in the curve signal reaches a specific proportion of the peak value of the first wave peak as the cut-off moment.

[0023] Within this time interval, use the second-order difference method to identify all the wave peaks and wave valleys in the particle velocity curve. Calculate the amplitude difference between adjacent extreme points, that is (2) In the formula, and respectively represent adjacent extreme points, represents the amplitude difference, .

[0024] Among them, the probability distribution of the amplitude difference between consecutive extreme points within the time interval is statistically analyzed using the adaptive binning method.

[0025] Based on the adaptive binning method, the probability distribution of the amplitude differences between consecutive extreme points within a selected time interval is statistically analyzed. The adaptive binning method dynamically adjusts the width of each bin according to the distribution characteristics of the data, making the number of data points in each bin more uniform. Compared with the fixed-width binning, this method can more accurately reflect the actual distribution of the data, especially when the data distribution is uneven or there are outliers.

[0026] Example 4 Based on Example 3, step 4 is specifically as follows: The peak entropy value is calculated using the Shannon entropy formula to reflect the degree of chaos in the extreme value changes in the particle velocity signal, as specifically represented below: (3) In the formula, represents the probability of the xi-th amplitude difference, and the "peak entropy" is calculated and used as a quantitative characteristic index for spall damage. The higher the peak entropy value, the more irregular the extreme value changes in the signal and the more severe the damage degree.

[0027] Example 5 Based on Example 4, the peak entropy is the core feature, and other features include time series statistics such as extreme point density and amplitude difference kurtosis; after single impact, signal processing is completed and the damage probability score is output. When the peak entropy is "0", it represents that the signal is relatively smooth within the selected time interval, and the corresponding specimen is in a non-damaged state; when the peak entropy is greater than "0", it represents that the signal is relatively chaotic within the selected time interval, and the corresponding specimen is in a damaged state.

[0028] Example 6 The dynamic detection method for weak bonding defects at the CFRP interface based on "peak entropy" in this example is specifically as follows: S1, The laser shock uses a LABER-H50 type tunable pulse width laser, with an output wavelength of 1053 nm, a pulse width adjustment range of 10 - 300 ns, an energy output range of 1 - 50 J, and an adjustable spot diameter of 2 - 15 mm; the dynamic signal acquisition uses a photon Doppler velocimetry (PDV) system, including a 1550 nm continuous laser light source, a 4 GHz bandwidth oscilloscope (sampling rate 20 GS / s), and the probe is 2 - 5 mm away from the back of the specimen; the constraint layer system: deionized water constraint layer (thickness 1 - 2 mm), and the absorption layer is black PVC tape; the three-dimensional positioning platform has a repeat positioning accuracy of ±0.01 mm and a load-bearing capacity of 15 kg; the CT verification uses a Diondo D2 type high-resolution CT system, and the scanning parameters are X-ray source voltage 90 kV, current 90 μA, and spatial resolution 0.02 mm.

[0029] S2, PDV signal acquisition and downsampling, single laser shock parameter setting. According to the specimen thickness (1.5 mm), match the laser energy (1 - 4 J), pulse width (20 - 100 ns), and spot diameter of 5 mm; for the dynamic response signal downsampling, reduce the original sampling rate of 20 GS / s to 1 GS / s to ensure that the signal bandwidth ≤ 500 MHz (meeting the Nyquist criterion); then for the baseline drift correction, use the db4 wavelet basis for 5 - layer decomposition, extract the 5th - layer low - frequency approximation coefficient as the trend term, and eliminate the baseline offset through algebraic subtraction; finally, for the high - frequency noise suppression, design a FIR low - pass filter (cut - off frequency 500 MHz, transition bandwidth 100 MHz) to eliminate high - frequency noise interference.

[0030] S3, perform S - transform on the denoised signal (Gaussian window time - width coefficient 0.8), extract the time - domain energy matrix, and locate the transient energy peak region of stress - wave reflection - unloading coupling (energy threshold set to 98% of the global peak); screen the extreme points, identify the wave peaks and wave valleys based on the second - order difference method, and eliminate the pseudo - peaks with adjacent extreme value intervals < 0.8 μs; delimit the characteristic window starting from the first main wave peak, and intercept the time interval when the velocity decays to 20% - 30% of the peak value.

[0031] S4, for the screened extreme - point sequence , calculate the amplitude difference between adjacent extreme points according to the formula: (2) In the formula, and respectively represent adjacent extreme points, represents the amplitude difference, .

[0032] And dynamically divide 10 - 15 intervals according to the amplitude - difference distribution to ensure uniform data volume in each bin; use kernel density estimation (bandwidth 0.05) to calculate the normalized probability density distribution.

[0033] S5, calculate the Shannon entropy of the amplitude - difference sequence between extreme points to generate the peak - entropy feature, specifically: (3) In the formula, represents the peak - entropy feature, represents the probability of the th amplitude difference. For normal specimens ≈0 (single - peak attenuation, concentrated amplitude - difference distribution); for defective specimens >0 (multi - peak oscillation, random amplitude - difference distribution).

[0034] Then input the characteristic peak - entropy , extreme point density (number of extreme points per unit time), amplitude difference kurtosis; after single impact, signal processing is completed, and the damage probability score (0 - 1) is output.

[0035] Under different characteristic window ratios, the stability of peak entropy is excellent, and its sensitivity to spall damage remains consistent within a reasonable range of laser parameters. For specimens with different thicknesses, when the same threshold condition is used, the misjudgment rate does not increase significantly.

[0036] Example 7 The dynamic detection method for weak bonding defects at the CFRP interface based on "peak entropy" in this example uses the parameters provided in Example 6, specifically: Collect the PDV dynamic response signals on the back surface of the laser - shocked composite material, and extract the free particle velocity on the back surface of the specimen under different impact conditions.

[0037] As Figure 2 and Figure 3 shown, they are schematic diagrams of the specimen with a thickness of 1.5 mm and a pulse width of 50 ns and the specimen with a thickness of 3 mm and a pulse width of 30 ns respectively. From Figure 2 and Figure 3 it can be seen that there are significant differences in the occurrence times of the signal characteristics of the back - surface particle velocity under different specimen thicknesses or impact conditions. To ensure good comparability of each group of experimental data and capture the signal characteristics closely related to the damage mechanism, the effective analysis interval is set as 0.05 - 10 times the time taken for the velocity in the curve signal to reach the peak value of the first wave peak. The first wave peak in the signal is determined as the starting moment, and at the same time, the moment corresponding to the particle velocity in the signal decaying to a specific ratio (0.1 - 0.5) of the peak value of the first wave peak is used as the cut - off moment.

[0038] Taking the laser pulse width of 50 ns, specimen thickness of 1.5 mm and laser pulse width of 30 ns, specimen thickness of 3 mm as examples, calculate the influence law of peak entropy under different laser energies, and verify the detailed value range of the ratio combined with CT data. The influence of the ratio on peak entropy is as Figure 4 shown, Figure 4 showing the influence trends of peak entropy in the non - damaged and damaged states of the specimen under different impact parameters. As the laser energy increases, the peak entropy undergoes a step change. When the peak entropy of the non - damaged specimen is "0", the signal disorder degree is low, and the corresponding specimen is in a non - damaged state; when the peak entropy is greater than "0", the signal chaos degree increases, and the corresponding specimen is in a damaged state.

[0039] The dynamic detection method for weak bonding defects of CFRP interfaces based on "peak entropy" provided by the present invention collects dynamic response signals through single laser shock, and uses S-transform time-frequency analysis and Shannon entropy feature extraction, with the detection efficiency increased to 5 seconds per piece. Compared with the "low-high-low" three-shock and "low-high" two-shock methods, the present invention realizes single-shock detection through optimization, with improved efficiency and accuracy, and is applicable to the rapid quality inspection and monitoring of aviation CFRP components.

Claims

1. A dynamic detection method for weak bonding defects at the CFRP interface based on "peak entropy", characterized in that Specifically, it includes the following steps: Step 1: Collect the particle velocity signals on the back surface of the non-destructive specimen under laser shock through a photon Doppler velocimetry system; Step 2: Preprocess the particle velocity signals; Step 3: Perform S-transform time-frequency analysis and extreme value analysis on the preprocessed particle velocity signals, and calculate the amplitude difference between extreme points; Step 4: Calculate the Shannon entropy of the amplitude difference sequence between extreme points to generate peak entropy features; Step 5: Extract the peak entropy features to judge the damage state of the non-destructive specimen.

2. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 1, characterized in that, In Step 1, the non-destructive specimen is made of carbon fiber reinforced composite material.

3. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 1, characterized in that The specific content of Step 2 is as follows: First, perform downsampling to reduce the sampling rate of the original signal from 20 GS / s to 1 GS / s, then perform baseline drift correction, extract the low-frequency trend term by wavelet decomposition, and eliminate the signal baseline offset through algebraic subtraction; finally, use a low-pass digital filter with a cut-off frequency of 500 MHz to eliminate environmental noise and high-frequency interference of the instrument.

4. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 1, characterized in that, The specific content of Step 3 is as follows: Perform S-transform time-domain analysis on the preprocessed signal to locate the transient energy peak region caused by stress wave reflection-unloading coupling; then identify the peaks and valleys of the particle velocity curve by the second-order difference method, and eliminate the pseudo-peaks with an adjacent extreme value interval less than 0.8 μs; take the first main peak as the starting point, select the time interval when the velocity decays to 20%-30% of the peak value as the feature analysis window, and statistically analyze the probability distribution of the amplitude difference between consecutive extreme points within this time interval.

5. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 4, characterized in that The specific method of identifying the peaks and valleys of the particle velocity curve by the second-order difference method in Step 3 is as follows: First, calculate the first-order difference of the denoised velocity curve to obtain the change rate of the particle velocity signal; then perform a difference operation on the first-order difference data again to obtain the second-order difference, so as to reflect the curvature change of the particle velocity signal, as shown in formula (1): （1） In the formula, is the value of the particle velocity signal at the point ; h is the step size.

6. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 4, characterized in that The specific method of calculating the amplitude difference between extreme points in Step 3 is as follows: （2） Wherein, and respectively represent adjacent extreme points, represents the amplitude difference, .

7. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 4, characterized in that, Use the adaptive binning method to statistically analyze the probability distribution of the amplitude difference between consecutive extreme points within the time interval in Step 3.

8. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 1, characterized in that, The specific content of Step 4 is as follows: Calculate the peak entropy value using the Shannon entropy formula to reflect the degree of chaos of extreme value changes in the particle velocity signal, which is specifically expressed as follows: （3） In the formula, represents the probability of the nth amplitude difference.

9. The dynamic detection method for CFRP interface weak bonding defects based on "peak entropy" according to claim 1, characterized in that The specific content of Step 5 is as follows: When the peak entropy has a step change and the peak entropy of the non-destructive specimen is "0", the signal disorder is low and the corresponding specimen is in a non-destructive state; when the peak entropy is greater than "0", the signal chaos increases and the corresponding specimen is in a damaged state.

Images