Method for quantitative detection of defects in composite moulded materials by ultrasonic a-scan

Abstract

The application protects a detection method for quantitative detection of defects of composite die pressing materials through ultrasonic A scanning, relates to the technical field of die forming, specifically, flat-shaped parts are pressed, and artificial defect samples and to-be-tested samples are prepared; ultrasonic A scanning detection of artificial defects is carried out through an ultrasonic phased array instrument, and ultrasonic A scanning signals of the artificial defects are collected; the ultrasonic A scanning signals are decomposed into a linear combination of a group of intrinsic mode functions IMF through a WOA-VMD decomposition model optimized based on a whale optimization algorithm, feature vectors of the intrinsic mode function IMF components are selected, an optimized feature vector matrix is formed by using kernel principal component analysis, the artificial defects are taken as a training set, the to-be-tested samples are taken as a test set, IAO-SVM is used for classification and identification, quantitative detection of defects through ultrasonic A scanning is realized, a large amount of cost is saved, the detection precision is improved, the advantages of ultrasonic A scanning, ultrasonic C scanning and CT scanning technology are combined, and the method has important practical application value.

Description

Technical Field

[0001] This invention relates to the field of molding process technology, specifically to a detection method for quantitative detection of defects in composite molding materials using ultrasonic A-scan. Background Technology

[0002] Composite materials, due to their high specific modulus, high specific strength, and light weight, can effectively reduce energy consumption and are known as the next generation of green materials, widely used in industrial manufacturing. However, the complex composition and complicated manufacturing process of composite materials make them prone to internal damage and defects, affecting their performance. Therefore, non-destructive testing (NDT) of composite materials is particularly important. Existing NDT methods, such as the widely used ultrasonic A-scan, are simple, fast, widely applicable, and inexpensive, but they only collect ultrasonic signals. Defect identification relies on the experience of professionals, resulting in large errors and low accuracy. They often only provide qualitative detection of defects, failing to meet current industry requirements. While ultrasonic C-scan or CT scans can achieve high-precision defect identification, their high cost significantly limits their application. Therefore, improving the detection accuracy of ultrasonic A-scan has become a key focus of industry research. Currently, from the perspective of studying ultrasound A-scan signals, the research direction is to use mathematical models to quantify and visualize the collected ultrasound A-scan signal data of defects, so as to achieve high-precision detection without CT scans, thereby improving the effect of ultrasound A-scan detection and saving costs. This approach combines the advantages of ultrasound A-scan and CT scans.

[0003] Based on the above, and to address the difficulty in balancing detection accuracy and cost, this application constructs a mathematical model to process the ultrasonic A-scan signal of internal defects in composite material parts, thereby achieving quantitative detection and identification of defects. Summary of the Invention

[0004] I. Technical problems to be solved

[0005] This invention addresses the aforementioned deficiencies in existing technologies by proposing a method for quantitative detection of defects in composite molding materials using ultrasonic A-scanning. This method solves the problem that existing ultrasonic A-scanning methods can only perform qualitative detection of defects and damage in composite molding parts, resulting in low detection accuracy and the inability to perform quantitative detection.

[0006] II. Technical Solution

[0007] To address the aforementioned technical problems, this invention provides a method for quantitative detection of defects in composite molding materials using ultrasonic A-scan, comprising the following steps:

[0008] S1: Press flat plate-shaped parts and prepare artificial defect samples;

[0009] S2: Perform ultrasonic A-scan detection on artificial defects using an ultrasonic phased array instrument and collect ultrasonic A-scan signals of artificial defects;

[0010] S3: Decompose the ultrasonic A-scan signal. The defective ultrasonic A-scan signal is decomposed into a linear combination of intrinsic mode functions (IMFs) through the variational mode decomposition model WOA-VMD based on the whale algorithm optimization. The energy and energy ratio cumulative coefficient of each IMF component after decomposition are calculated. The IMF component corresponding to the energy ratio cumulative coefficient is selected when it exceeds 96%.

[0011] S4: Select seven parameters—slope, kurtosis, peak value index, clearing index, shape index, impulse index, and energy—of the intrinsic mode function (IMF) components as feature vectors to characterize the defective ultrasound A-scan signal. Use kernel principal component analysis to remove irrelevant and redundant noise information from the feature vector matrix composed of the feature vectors, forming an optimized feature vector matrix to prepare for the next step of classification and recognition.

[0012] S5: Press flat plate-shaped parts and prepare test samples;

[0013] S6: Perform ultrasonic A-scan detection on the test sample in S5, and execute the signal processing steps of S3 to S5 on the ultrasonic A-scan signal. Use the optimized feature vector matrix obtained from the artificial defect as the training set and the optimized feature vector matrix obtained from the test sample as the test set. Use the improved Tianying algorithm-optimized support vector machine model IAO-SVM to classify and identify the training set and the test set.

[0014] The specific steps for preparing the artificial defect sample in S1 are as follows:

[0015] Glass fiber reinforced composite prepreg is pressed into a flat plate shape using a compression molding machine. Five holes with a diameter of 0.5 mm, five holes with a diameter of 0.8 mm, and five holes with a diameter of 1.1 mm and a depth of 10 mm are drilled sequentially on the side of the flat plate shape. Five holes with a diameter of 2 mm, five holes with a diameter of 3 mm, and five holes with a diameter of 4 mm and a depth of 3 mm are drilled sequentially on the front of the flat plate shape, for a total of 30 holes.

[0016] Carbon fiber reinforced composite prepreg is pressed into a flat plate shape using a molding equipment. Small holes with a diameter of 0.5 mm, 0.8 mm, and 1.1 mm, and a depth of 10 mm are drilled in the thickness direction of the flat plate shape, for a total of 15 holes.

[0017] The specific steps for collecting the ultrasonic A-scan signal of artificial defects in S2 are as follows:

[0018] Each artificial defect was tested 10 times. Each group of artificial defects with the same aperture formed 50 sets of ultrasonic A-scan data samples. After analyzing and screening the data samples, 30 sets of data samples were selected as the best.

[0019] The specific steps for decomposing the ultrasonic A-scan signal of the defect in S3 using the variational mode decomposition model WOA-VMD based on the whale algorithm optimization are as follows:

[0020] First, the ultrasound A-scan signal is decomposed into a variational model of a linear combination of a set of intrinsic mode functions (IMFs) using variational mode decomposition (VMD).

[0021] Then, the Whale Optimization Algorithm (WOA) is used to optimize the parameter penalty term coefficient α and the number of parameter mode functions k in Variational Mode Decomposition (VMD), and the minimum value of the envelope entropy is used as the fitness function until the global optimal solution of the two parameters is obtained.

[0022] By substituting the global optimal solution of the two parameters into the variational model and determining the optimal center frequency and finite bandwidth of each mode group, the intrinsic mode function (IMF) components of the signal are obtained for effective decomposition.

[0023] In S4, the kernel function used in the kernel principal component analysis method is the Gaussian kernel function.

[0024] In S4, the kernel principal component analysis method is used to reduce the dimensionality of the eigenvector matrix and select the principal components corresponding to the cumulative contribution rate of more than 95% to form an optimized eigenvector matrix.

[0025] The specific classification steps of the IAO-SVM support vector machine model optimized using the improved Skyhawk algorithm in S6 are as follows:

[0026] The Support Vector Machine (SVM) model uses a radial basis function kernel and optimizes the radial basis function kernel parameter penalty factor C and kernel parameter σ using the improved Skyhawk Algorithm (IAO).

[0027] The improved Skyhawk Algorithm IAO introduces chaotic Tent mapping for population initialization based on the Skyhawk Algorithm AO. At the same time, it applies an adaptive weight factor ω to balance the search range and search capability. In the early stage of the AO algorithm iteration, it accelerates the iteration speed and expands the search range to determine the location of the target value.

[0028] The specific steps for preparing the test sample in S5 are as follows:

[0029] Glass fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 5 pieces. Three samples with a size of 3mm×10mm×4mm are taken out at intervals along the length of each flat plate-shaped part.

[0030] Carbon fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 3 pieces. Three samples with a size of 3mm×10mm×4mm are taken out from the length of each flat plate-shaped part at intervals.

[0031] The detection methods also include:

[0032] S7: Perform a CT scan on the sample to be tested in S5, and compare the CT scan results with the classification and identification results of the ultrasound A-scan signals.

[0033] The specific steps for performing a CT scan on the test sample in S7 are as follows:

[0034] The first stage involves cross-sectional scanning along the 3mm length of each sample, with each cross-section spaced 0.2mm apart, for a total of 15 cross-sections;

[0035] The second stage involves cross-sectional scanning along the 4mm thickness direction of each sample, with each cross-section spaced 0.1mm apart, for a total of 40 cross-sections;

[0036] All cross-sectional tomographic reconstructions were used to create CT scan 3D images of each sample segment.

[0037] III. Beneficial Effects

[0038] Compared with existing technologies, this invention establishes a mathematical model through WOA-VMD decomposition, kernel principal component analysis, and IAO-SVM classification and recognition algorithms to analyze the ultrasonic A-scan signal of defects in composite molding materials, thereby realizing quantitative detection of defects. This not only saves a lot of costs but also improves the accuracy of detection. It combines the advantages of ultrasonic A-scan with ultrasonic C-scan and CT scanning technologies, and has important practical application value. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating how WOA optimizes VMD parameters in this invention;

[0040] Figure 2 This is a WOA-VMD decomposition result of a φ0.5mm hole defect in a glass fiber reinforced composite molded part.

[0041] Figure 3 This is a WOA-VMD decomposition result of a φ0.8mm hole defect in a glass fiber reinforced composite molded part;

[0042] Figure 4 This is a WOA-VMD decomposition result of a φ1.1mm hole defect in a glass fiber reinforced composite molded part;

[0043] Figure 5 This is a WOA-VMD decomposition result of a φ2mm hole defect in a glass fiber reinforced composite molded part;

[0044] Figure 6 This is a WOA-VMD decomposition result of a φ3mm hole defect in a glass fiber reinforced composite molded part;

[0045] Figure 7 This is a WOA-VMD decomposition result of a φ4mm hole defect in a glass fiber reinforced composite molded part;

[0046] Figure 8 This is a WOA-VMD decomposition result of a φ0.5mm hole defect in a carbon fiber reinforced composite molded part.

[0047] Figure 9 This is a WOA-VMD decomposition result of a φ0.8mm hole defect in a carbon fiber reinforced composite molded part.

[0048] Figure 10 This is a WOA-VMD decomposition result of a φ1.1mm hole defect in a carbon fiber reinforced composite molded part.

[0049] Figure 11 This is a flowchart of the process of optimizing an SVM model using the IAO algorithm;

[0050] Figure 12 This is a schematic diagram of a CT scan of a carbon fiber reinforced composite material;

[0051] Figure 13 This is a schematic diagram of a CT scan of a glass fiber reinforced composite material;

[0052] Figure 14 This is a waveform diagram of an ultrasonic phased array A-scan of the defect and damage region of a carbon fiber reinforced composite material.

[0053] Figure 15 yes Figure 14 The corresponding WOA-VMD decomposition result diagram of the defect;

[0054] Figure 16 This is a waveform diagram of an ultrasonic phased array A-scan of the defect and damage region of a glass fiber reinforced composite material.

[0055] Figure 17 yes Figure 16 The corresponding WOA-VMD decomposition result diagram of the defect;

[0056] Figure 18 This is a comparison chart of categories for the classification and identification of defects in carbon fiber composite material specimens under test using IAO-SVM.

[0057] Figure 19 This is a comparison chart of categories for the classification and identification of defects in glass fiber composite material samples under test using IAO-SVM. Detailed Implementation

[0058] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0059] In this embodiment, the detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan includes the following steps:

[0060] S1: Press flat plate-shaped parts and prepare artificial defect samples;

[0061] The specific steps for preparing artificial defect samples in S1 are as follows:

[0062] Glass fiber reinforced composite prepreg is pressed into a flat plate shape using a compression molding machine. Five holes with a diameter of 0.5 mm, five holes with a diameter of 0.8 mm, and five holes with a diameter of 1.1 mm and a depth of 10 mm are drilled sequentially on the side of the flat plate shape. Five holes with a diameter of 2 mm, five holes with a diameter of 3 mm, and five holes with a diameter of 4 mm and a depth of 3 mm are drilled sequentially on the front of the flat plate shape, for a total of 30 holes.

[0063] Carbon fiber reinforced composite prepreg is pressed into a flat plate shape using a molding equipment. Small holes with a diameter of 0.5 mm, 0.8 mm, and 1.1 mm, and a depth of 10 mm are drilled in the thickness direction of the flat plate shape, for a total of 15 holes.

[0064] The specific manufacturing process of the flat plate part is as follows:

[0065] (1) Turn on the mold temperature controller and heat the heat transfer oil to the set temperature. The heat transfer oil flows between the mold and the mold temperature controller, and heats the mold to the required temperature through heat transfer.

[0066] (2) Cut the prepreg to the set size. The total weight of the prepreg must meet the requirement that the thickness of the composite plate after molding reaches 4mm.

[0067] (3) Lay the cut prepreg in sequence in the middle of the lower mold cavity.

[0068] (4) Start the press. The mold closes under the action of the press. The prepreg softens in the high temperature mold and flows to form the shape.

[0069] (5) While the molding process is underway, the vacuum device is activated to expel the gas from the mold cavity.

[0070] (6) Set the segmented pressure through the press control panel to fully pressurize the molten prepreg so that the prepreg fills the entire cavity as much as possible, and the resin matrix and reinforcing fiber are fully fused.

[0071] (7) Under the simultaneous action of high temperature and high pressure, the resin matrix undergoes a cross-linking and curing reaction, solidifying into a flat part. The mold can be opened and the part removed after the molding process is completed.

[0072] S2: Perform ultrasonic A-scan detection on artificial defects using an ultrasonic phased array instrument and collect ultrasonic A-scan signals of artificial defects;

[0073] The specific steps for collecting ultrasonic A-scan signals of artificial defects in S2 are as follows:

[0074] Each artificial defect was tested 10 times. For each group of artificial defects with the same aperture, 50 sets of ultrasonic A-scan data samples were generated. After analysis and screening, 30 sets were selected as the best. During the actual testing, because the flat plate used was relatively thin and the defect was located on or near the surface, a dual-crystal probe was used for the ultrasonic phased array instrument. To address the blind zone issue in near-surface testing, a wedge was added between the ultrasonic probe and the sample. The wedge was made of plexiglass, 20mm high, and glycerol was used as the coupling agent. Glycerol has appropriate viscosity, is non-corrosive to the workpiece, and harmless to the human body. The coupling agent also acts as a lubricant, effectively protecting the probe and extending its service life.

[0075] Table 1 shows the types of defects and the collected ultrasonic test data for glass fiber reinforced composite flat plate parts, and Table 2 shows the types of defects and the collected ultrasonic test data for carbon fiber reinforced composite flat plate parts.

[0076] Table 1:

[0077]

[0078]

[0079] Table 2:

[0080]

[0081] S3: Decompose the ultrasonic A-scan signal. The defective ultrasonic A-scan signal is decomposed into a linear combination of intrinsic mode functions (IMFs) through the variational mode decomposition model WOA-VMD based on the whale algorithm optimization. The energy and energy ratio cumulative coefficient of each IMF component after decomposition are calculated. The IMF component corresponding to the energy ratio cumulative coefficient is selected when it exceeds 96%.

[0082] Specifically, the steps for decomposing the ultrasonic A-scan signal of the defect in S3 using the variational mode decomposition model WOA-VMD optimized based on the whale algorithm are as follows:

[0083] First, the ultrasound A-scan signal is decomposed into a variational model of a linear combination of intrinsic mode functions (IMFs) using variational mode decomposition (VMD). VMD decomposes the signal into a linear combination of IMFs. The optimal center frequency and finite bandwidth of each mode are determined by iteratively searching for the optimal solution of the variational model, thus obtaining the effective decomposition components of the signal and the optimal solution.

[0084] (1) Construction of variational model

[0085] In the variational model, the intrinsic mode function (IMF) is calculated as shown in equation (1.1):

[0086]

[0087] in It is phase, A k (t) is the instantaneous amplitude, and the formula for calculating the instantaneous frequency is shown in equation (1.2):

[0088] ω k (t)=φ′ k (t) (1.2)

[0089] Variational modeling (VMD) decomposes a signal into a linear combination of a set of intrinsically significant factors (IMFs). Each IMF must possess sufficient sparsity to characterize the original signal. The construction of the variational model involves the following steps:

[0090] (a) For each u k (t) Perform Hilbert transform to obtain the corresponding analytic signal and obtain the one-sided spectrum. The formula for calculating the one-sided spectrum is shown in equation (1.3):

[0091]

[0092] (b) Demodulate the spectrum of each mode onto the baseband to obtain the center frequency ω of each mode. k (t), the calculation formula is shown in equation (1.4):

[0093]

[0094] (c) Each mode must meet the requirements of a narrowband signal and remain smooth in the time domain.

[0095] In summary, the variational binding model is constructed, and the calculation formula is shown in equation (1.5):

[0096]

[0097] (2) Solving the variational model

[0098] By introducing a quadratic penalty term and a Lagrange factor into the formula, the constrained variational model is transformed into an unconstrained variational model. The calculation formula for the unconstrained variational model is shown in equation (1.6):

[0099]

[0100] The Alternating Direction Method (ADMM) of multiplicative operators was then used to solve the problem.

[0101] However, if only the ultrasound A-scan signal is decomposed using VMD, with the quadratic penalty term coefficient α set to 100 and the number of modal functions k set to 5, both of which are empirical values, a significant error can occur. Therefore, the Whale Algorithm (WOA) is further employed to optimize the two parameters of VMD, using the minimum envelope entropy as the fitness function. Envelope entropy represents the sparsity of the original signal; when the IMF component contains less feature information, the envelope entropy value is relatively large, and vice versa. Assume the received signal is S... i (i = 1, 2, ... N), then S i Envelope entropy E p The calculation formula (1.7) is as follows:

[0102]

[0103] Where α(i) is the envelope signal obtained after the original signal is decomposed by VMD and then transformed by Hilbert, ε(i) is the probability distribution sequence obtained after normalizing α(i), and N is the number of sampling points for ultrasound detection.

[0104] The Whale Optimization Algorithm (WOA) is an optimization algorithm proposed in 2016 by Australian scholar Mirjalili. Compared to other algorithms, it boasts advantages such as fast computation and strong global convergence. It simulates the hunting mechanism of humpback whales, comprising three stages: encirclement, bubble net predation, and search. In the Whale Optimization Algorithm, each whale represents a potential optimal solution to an extreme value optimization problem. The whales continuously update their positions during the iteration process until the global optimum is obtained. The specific computational steps are as follows:

[0105] (1) Encirclement and capture

[0106] First, it is necessary to determine the location surrounding the prey. According to the WOA algorithm, it is assumed that the target prey is the current potential optimal solution. After defining the best search agent, other search agents will explore and update positions towards the best search agent. Mirjalili proposed the following mathematical model to describe this behavior, and the calculation formula is shown in equation (2.1):

[0107]

[0108] In the formula, A and C are coefficients, t is the current iteration number, and X is... * Let X(t) be the current optimal position and X(t) be the current position. If a better solution exists, X... * (t) needs to be continuously updated iteratively. Therefore, the formulas (2.2) for calculating A and C are as follows:

[0109]

[0110] In the formula, r1 and r2 are arbitrary numbers taken from (0,1), and a ranges from 2 to 0, taking values ​​linearly downwards. t is the current iteration number, T max It is the maximum number of iterations.

[0111] (2) Start the hunt

[0112] Humpback whales employ two hunting mechanisms: encirclement and spiral hunting. When a humpback whale swims towards its prey in a spiral motion, the mathematical model is shown in equation (2.3):

[0113]

[0114] D p Let be the distance between the whale and its prey, b be a constant defining the spiral, and l be an arbitrary number taken from (-1, 1). Since humpback whales have two hunting mechanisms, it is assumed that the whale has P... i The probability of choosing to narrow the encirclement is 1-P i If the probability of choosing a spiral is 1, then the modified mathematical model is shown in equation (2.4):

[0115]

[0116] As the number of iterations t increases, both parameters A and a gradually decrease. When |A| < 1, the system is in a local optimization state.

[0117] (3) Searching for prey

[0118] The mathematical model for this process is shown in equation (2.5):

[0119]

[0120] In the formula X rand (t) represents the random whale position, and D is the distance between the current individual and the random individual. When A≥1, a random search agent is selected, the current prey is abandoned, a new position is randomly updated, and a more suitable prey is selected.

[0121] The steps for optimizing VMD parameters using WOA are as follows: Figure 1 As shown. First, initialize the position [k, α] of the humpback whale pod, and take the envelope entropy E as the fitness function. p The fitness of each individual whale is calculated. An iterative formula is selected based on the convergence factor to iterate until the optimal VMD parameters are output. The initial population size is set to 20, and the maximum number of iterations is set to 50.

[0122] In this embodiment, the ultrasonic testing signal of a φ0.5mm hole defect in a glass fiber reinforced composite molded part was used as the object. WOA-VMD was applied to obtain the optimal parameters k = 9 and α = 1225, with the optimization process taking 21 minutes. The final result after WOA-VMD decomposition is as follows: Figure 2 As shown.

[0123] Using the ultrasonic testing signal of a φ0.8mm pore defect in a glass fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 7 and α = 112, with the optimization process taking 18 minutes. The final results after WOA-VMD decomposition are as follows. Figure 3 As shown.

[0124] Using the ultrasonic testing signal of a φ1.1mm pore defect in a glass fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 8 and α = 139, with the optimization process taking 18 minutes. The final results after WOA-VMD decomposition are as follows. Figure 4 As shown.

[0125] Using the ultrasonic testing signal of a φ2mm pore defect in a glass fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 5 and α = 1145, with the optimization process taking 20 minutes. The final results after WOA-VMD decomposition are as follows. Figure 5 As shown.

[0126] Using the ultrasonic testing signal of a φ3mm pore defect in a glass fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 5 and α = 116, with the optimization process taking 19 minutes. The final results after WOA-VMD decomposition are as follows: Figure 6 As shown.

[0127] Using the ultrasonic testing signal of a φ4mm pore defect in a glass fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 3 and α = 100, with the optimization process taking 19 minutes. The final results after WOA-VMD decomposition are as follows: Figure 7 As shown.

[0128] Using the ultrasonic testing signal of a φ0.5mm pore defect in a carbon fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 8 and α = 336, with the optimization process taking 20 minutes. The final results after WOA-VMD decomposition are as follows. Figure 8 As shown.

[0129] Using the ultrasonic testing signal of a φ0.8mm pore defect in a carbon fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 10 and α = 741, with the optimization process taking 19 minutes. The final results after WOA-VMD decomposition are as follows. Figure 9 As shown.

[0130] Using the ultrasonic testing signal of a φ1.1mm pore defect in a carbon fiber reinforced composite molded part as the object, WOA-VMD was applied to obtain the optimal parameters k = 10 and α = 1793, with the optimization process taking 20 minutes. The final results after WOA-VMD decomposition are as follows. Figure 10 As shown.

[0131] exist Figures 2 to 10 In the table, table a shows the convergence curve of WOA, table b shows the time-domain plot of the mode function obtained by WOA-VMD decomposition, and table c shows the Fourier transform spectrum (frequency domain plot).

[0132] The WOA-VMD method decomposes the signal into k effective components and one residual component. The decomposition effect is ideal in both the mid-to-high frequency band and the low frequency band, and there is no obvious mode function overlap.

[0133] The above process involves using the Whale Optimization Algorithm (WOA) to optimize the coefficients α of the parameter penalty term and the number k of the parameter mode functions in Variational Mode Decomposition (VMD), using the minimum value of the envelope entropy as the fitness function, until the global optimal solution for the two parameters is obtained.

[0134] By substituting the global optimal solution of the two parameters into the variational model and determining the optimal center frequency and finite bandwidth of each mode group, the intrinsic mode function (IMF) components of the signal are obtained for effective decomposition.

[0135] Observing the frequency domain diagrams of the decomposed IMF components reveals that as the number of IMF decomposition layers increases, the frequency of the decomposed signal gradually decreases, approaches a flattening level, and contains less and less effective information. Therefore, adding features of these IMF components containing very little defect information to the final feature set would not only increase computational complexity but also affect the accuracy of final defect identification. Thus, these IMF components with relatively small contributions must be removed. By calculating the energy contained in the IMF components, the IMF components containing most of the energy of the original signal are selected.

[0136] Specifically, the formula for calculating signal energy is shown in equation (3.1):

[0137]

[0138] Where h(t) is the time-domain sequence of the modal components. If we denote the frequency domain function of the IMF component as F(jω), then the formula for calculating the energy density function is shown in equation (3.2):

[0139] G(w)=|F(jω)| 2 (3.2)

[0140] According to Passevar's theorem, as shown in equation (3.3):

[0141]

[0142] Here, we introduce the cumulative energy ratio factor (ERA), which is calculated using the formula shown in equation (3.4):

[0143]

[0144] Where k is the number of selected IMF components, and n is the total number of IMF components after decomposition. Currently, ERA is generally considered... i When the accuracy is ≥96%, selecting the corresponding k IMF components is more reasonable.

[0145] Taking the φ0.5mm hole defect in the aforementioned glass fiber reinforced composite molded part as an example, the energy ratio and ERA of the IMF component after WOA-VMD decomposition are calculated, and the specific results are shown in Table 3 below.

[0146] Table 3:

[0147]

[0148] Table 3 records the energy distribution of each order of IMF components. The data in the table shows that the ERA of the first four order IMF components has reached 96.1%, exceeding the specified 96%. Therefore, the first four order IMF components were selected for feature extraction.

[0149] Similarly, the same method was used to decompose the φ0.8mm, φ1.1mm, φ2mm, φ3mm, and φ4mm hole defects in glass fiber reinforced composite molded parts and the φ0.5mm, φ0.8mm, and φ1.1mm hole defects in carbon fiber reinforced composite molded parts.

[0150] Table 4 records the number and energy percentage of IMF components selected for the six types of defects in glass fiber reinforced composite molded parts.

[0151] Table 5 records the number and energy percentage of the three types of defects in the final selected IMF components of the carbon fiber reinforced composite molded parts.

[0152] The WOA-VMD method was applied to decompose defect signals. For glass fiber reinforced composite molded parts, the ERA of the first four IMF components of the φ0.5mm hole defect signal reached 96.1%, therefore, the first four IMF components were selected for feature extraction. For the φ0.8mm hole defect signal, the first three IMF components were selected; for the φ1.1mm hole defect signal, the first five IMF components were selected; for the φ2mm hole defect signal, the first four IMF components were selected; for the φ3mm hole defect signal, the first four IMF components were selected; and for the φ4mm hole defect signal, the first three IMF components were selected. For carbon fiber reinforced composite molded parts, the first eight IMF components of the φ0.5mm hole defect signal, the first five IMF components of the φ0.8mm hole defect signal, and the first six IMF components of the φ1.1mm hole defect signal were selected.

[0153] Table 4:

[0154]

[0155] Table 5:

[0156]

[0157] S4: Seven parameters—slope, kurtosis, peak value index, clearing index, shape index, impulse index, and energy—of the intrinsic mode function (IMF) components are selected as feature vectors characterizing the defective ultrasound A-scan signal. Kernel principal component analysis (KPCA) is used to remove irrelevant and redundant noise information from the feature vector matrix formed by the feature vectors, forming an optimized feature vector matrix to prepare for the next step of classification and recognition. In S4, KPCA is used to reduce the dimensionality of the feature vector matrix, and the principal components corresponding to the cumulative contribution rate exceeding 95% are selected to form the optimized feature vector matrix.

[0158] Regarding the seven parameters, assuming the acquired signal is S, the mean... It is the average value of the signal, and is the first moment. The calculation formula is shown in equation (4.1).

[0159]

[0160] The standard deviation σ of signal S is the variance σ 2 The square root of the signal energy is the variance, which is the average of the squares of the differences between each signal value and the mean of all signal values. It represents the dynamic component of the signal energy and is calculated as shown in equation (4.2).

[0161]

[0162] The slope is the average of the cubes of the differences between each signal value and the average of all signal values, divided by the cube of the standard deviation. The calculation formula is shown in Equation (4.3).

[0163]

[0164] Kurtosis is the average of the fourth power of the difference between each signal value and the mean of all signal values, divided by the fourth power of the standard deviation. The calculation formula is shown in Equation (4.4).

[0165]

[0166] The formula for calculating the peak index is shown in equation (4.5).

[0167]

[0168] The calculation formula for the clearing index is shown in equation (4.6).

[0169]

[0170] The formula for calculating the shape index is shown in equation (4.7).

[0171]

[0172] The impulse index is the maximum value and the mean value of a signal. The ratio is calculated using the formula shown in equation (4.8).

[0173] The formula for calculating signal energy is shown in equation (4.9).

[0174]

[0175] Taking the φ0.5mm hole defect in a glass fiber reinforced composite molded part as an example, the defect signal was decomposed using WOA-VOM and the effective IMF components were selected. Then, the seven eigenvalues ​​of each IMF component were calculated. Table 6 shows the eigenvalues ​​of the first four IMF components after WOA-VMD decomposition.

[0176] Table 6:

[0177]

[0178] Thus, the ultrasonic signal sample of each φ0.5mm hole defect is decomposed using WOA-VMD, and effective IMF components are selected. Features of each order of IMF components are extracted, resulting in 7×4 feature values. These 28 feature values ​​can be used to characterize the ultrasonic detection signal of the defect. In this embodiment, a total of 30 ultrasonic detection signal samples of φ0.5mm hole defects were collected, ultimately yielding a 30×28 matrix vector.

[0179] Similarly, the effective IMF components after WOA-VMD decomposition for different defects are obtained. For the selected IMF components, the eigenvalues ​​of each IMF component are calculated, and the eigenvector matrices of defects in glass fiber reinforced composite molded parts are finally obtained as shown in Table 7, and the eigenvector matrices of defects in carbon fiber reinforced composite molded parts are shown in Table 8.

[0180] Table 7:

[0181]

[0182] Table 8:

[0183]

[0184] In the eigenvector matrices of IMF components at each order, although each feature contains some defect information, the amount of information contained in each feature is not the same, and therefore the importance of these features also varies. Furthermore, these features often have some correlation, so the effective information is often obscured by a lot of irrelevant and redundant noise information, greatly increasing the computational complexity and reducing the accuracy of classification. Therefore, kernel principal component analysis is used to resynthesize low-dimensional features carrying the maximum amount of feature information from numerous feature parameters. The kernel function used in kernel principal component analysis is the Gaussian kernel function.

[0185] Specifically, assuming the original low-dimensional matrix X is composed of the feature vectors of the defect ultrasonic detection signal, matrix K = {K} is calculated using a kernel function. ij} n×n K ij =k(x i x j ),x i x j ∈X, k is the kernel function.

[0186] Calculate the eigenvalues ​​of K, arrange them in descending order, and find the corresponding eigenvectors α. l After normalizing the eigenvectors, the formula for calculating the l-th nonlinear principal component of the original sample is shown in equation (5.1):

[0187]

[0188] In the above formula, x i Let X be the i-th sample. Selecting the top m principal components with a cumulative contribution rate greater than 95% can characterize the original sample X.

[0189] The formula for calculating the Gaussian kernel function is shown in equation (5.2):

[0190]

[0191] Ultrasonic detection signals of φ0.5mm pore defects in glass fiber reinforced composite molded parts were selected and decomposed using WOA-VMD to obtain a 30*28 feature matrix. KPCA was then used to optimize this feature matrix. As the number of principal components increases, the contribution rate of each principal component decreases, meaning the defect information contained in each principal component decreases. When the first 7 principal components were selected, the cumulative contribution rate was already greater than 95%. To reduce information redundancy and simplify calculations, the first 7 principal components were selected to construct the optimized feature matrix. The original 30*28 feature matrix was optimized to a 30*7 feature matrix. Kernel principal component analysis was applied to the remaining feature matrices in the same way. The final optimized feature matrix sizes for glass fiber reinforced composite molded parts are shown in Table 9, and the optimized feature matrix sizes for carbon fiber reinforced composite molded parts are shown in Table 10.

[0192] Table 9:

[0193]

[0194]

[0195] Table 10:

[0196]

[0197] S5: Press flat plate-shaped parts and prepare test samples;

[0198] The specific steps for preparing the test sample in S5 are as follows:

[0199] Glass fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 5 pieces. Three samples with a size of 3mm×10mm×4mm are taken out at intervals along the length of each flat plate-shaped part.

[0200] Carbon fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 3 pieces. Three samples with a size of 3mm×10mm×4mm are taken out from the length of each flat plate-shaped part at intervals.

[0201] S6: Perform ultrasonic A-scan detection on the test sample from S5, and execute the signal processing steps from S3 to S5 on the ultrasonic A-scan signal. Use the optimized feature vector matrix obtained from artificial defects as the training set, and the optimized feature vector matrix obtained from the test sample as the test set. Use the improved Tianying algorithm-optimized support vector machine model IAO-SVM to classify and identify the training set and the test set. Based on the consistency of the classification and identification results, the possible defects in the test sample can be determined.

[0202] Specifically, the classification steps of the IAO-SVM support vector machine model optimized using the improved Skyhawk algorithm in S6 are as follows:

[0203] The Support Vector Machine (SVM) uses the radial basis function (RBF) kernel function, and optimizes the penalty factor C and kernel parameter σ using the improved Eagle Algorithm (IAO). SVM, proposed by Cortes and Vapnik in 1995, is a generalized linear classifier for binary classification. The RBF kernel function is currently the most widely used kernel function, suitable for classifying various distributions. The RBF kernel function has two main parameters: the penalty factor C and the kernel parameter σ. The penalty factor C primarily characterizes the generalization ability of the SVM; if the value is too small, underfitting may occur; however, if it is too large, the generalization ability will weaken. The kernel parameter σ mainly affects the dimensionality of the high-dimensional space mapped from the original samples; if it is too large, the fitting ability will decrease, and conversely, the generalization ability of the SVM model will deteriorate. Therefore, the selection of parameters is crucial for the construction of the SVM model.

[0204] The improved Skyhawk Algorithm IAO introduces chaotic Tent mapping for population initialization based on the Skyhawk Algorithm AO. At the same time, it applies an adaptive weight factor ω to balance the search range and search capability. In the early stage of the AO algorithm iteration, it accelerates the iteration speed and expands the search range to determine the location of the target value.

[0205] Specifically, the SVM decision function is shown in equation (6.1):

[0206]

[0207] SVM addresses the problem of nonlinear inseparability by introducing a kernel function. Assuming the introduced kernel function is k, the rewritten formula for the objective function is shown in equation (6.2):

[0208]

[0209] The rewritten calculation formula for the decision function is shown in equation (6.3):

[0210]

[0211] The performance of the SVM algorithm is highly dependent on the choice of kernel function. In this embodiment, a radial basis function (RBF) kernel is used. The expression for the RBF kernel function is shown in equation (6.4):

[0212]

[0213] Furthermore, this embodiment optimizes the parameters of the SVM using an improved Skyhawk algorithm, thereby improving the accuracy of pattern recognition.

[0214] (1) Principle of the Skyhawk Algorithm

[0215] The Eagle Optimization Algorithm (AO) is an optimization algorithm that simulates the predatory behavior of eagles. The predatory behavior of eagles consists of four stages. The first stage involves high-altitude flight, searching for prey, and selecting the optimal predatory area. The behavior in this stage is shown in equation (6.5):

[0216]

[0217] Where x best (t) represents the optimal solution before the t-th iteration, x M (t) represents the mean of the current solution at the t-th iteration, rand represents any value between 0 and 1, and t and T represent the current iteration number and the maximum iteration number, respectively.

[0218] The second stage involves narrowing the predation range, as shown in equation (6.6):

[0219]

[0220] Where D is the dimension, x R (t) represents a random solution between 1 and N in the Rth iteration, s equals 0.01, u and v are arbitrary numbers between 0 and 1, β is 1.5, and y and x represent the shape of the search region. The calculation formula is shown in equation (6.7):

[0221]

[0222] In the above formula, r1 takes a value between 1 and 20, u takes 0.00565, D1 takes an integer from 1 to the length of the search region, and ω takes 0.005.

[0223] The third stage is when the Eagle begins its descent and initiates its initial attack. This behavior is called a low-flying, slow-descent attack. The mathematical expression for this behavior is shown in equation (6.8):

[0224] x3=(x best (t)-x M (t))×α-ran d+((UB-LB)×UB+LB)×δ (6.8)

[0225] Where α and δ are constants of 0.1, LB is the lower bound, and UB is the upper bound.

[0226] The fourth stage is when the eagle approaches its prey and begins a random attack, as shown in equation (6.9):

[0227]

[0228] (2) Principle of the improved Skyhawk algorithm

[0229] While the conventional AO algorithm possesses excellent search capabilities, it suffers from the difficulty of balancing search range and search power. Therefore, in this experiment, an improved algorithm, IAO, is obtained by introducing chaotic Tent mapping and applying an adaptive weight factor strategy.

[0230] (a) Population initialization based on chaotic Tent mapping

[0231] Chaos possesses randomness, ergodicity, and sensitivity to initial conditions, thus accelerating convergence and improving accuracy. This experiment employs the chaotic Tent map to initialize the population, mitigating the uneven population distribution during optimization. The Tent chaotic map expression (6.10) is as follows:

[0232]

[0233] (b) Adaptive weighting factor

[0234] An adaptive weighting factor ω is introduced to balance the search range and search capability. By introducing ω, the iteration speed is accelerated in the early stages of the AO algorithm iteration, thereby expanding the search range and quickly determining the location of the target value; in the later stages of iteration, the iteration speed is reduced to enhance the search capability and avoid getting trapped in local optima. The mathematical expression of the adaptive weighting factor ω is shown in equation (6.11):

[0235]

[0236] In the above formula, t max To maximize the number of iterations, the improved mathematical expression is shown in equation (6.12):

[0237] x1=x best (t)×(1-ω) (6.12)

[0238] The IAO algorithm is applied to optimize the SVM model, and cross-validation is used to train the samples. The accuracy of cross-validation is used as the fitness value. The process is as follows: Figure 11 As shown, the specific steps are as follows:

[0239] (1) Select training and test sets and normalize them;

[0240] (2) Initialize the algorithm parameters, set the population size to 20, and the maximum number of iterations to 100;

[0241] (3) Train the samples, calculate the cross-validation accuracy and use it as the fitness value, and obtain the optimal fitness and position.

[0242] (4) Calculate the fitness value of the new position and compare it with the previous best fitness to obtain the latest best fitness and position.

[0243] (5) Determine whether the maximum number of iterations has been reached. If it has, obtain the optimal parameters and model. If not, repeat step 3 until the maximum number of iterations is reached, thereby obtaining the IAO-SVM model.

[0244] S7: Perform a CT scan on the sample to be tested in S5, and compare the CT scan results with the classification and identification results of the ultrasound A-scan signal.

[0245] The specific steps for performing a CT scan on the test sample in S7 are as follows:

[0246] The first stage involves cross-sectional scanning along the 3mm length of each sample, with each cross-section spaced 0.2mm apart, for a total of 15 cross-sections;

[0247] The second stage involves cross-sectional scanning along the 4mm thickness direction of each sample, with each cross-section spaced 0.1mm apart, for a total of 40 cross-sections;

[0248] All cross-sectional tomographic reconstructions were used to create CT scan 3D images of each sample segment.

[0249] The specific experimental procedure is as follows:

[0250] CT scans were performed on nine carbon fiber reinforced composite material samples. The CT scan results of eight samples were largely consistent, revealing small, irregularly shaped pores within the samples (these tiny pores have negligible impact on the mechanical properties of the composite parts). One sample's CT scan results are as follows: Figure 12 As shown in Figure a, a large hole defect was detected inside, with an average hole diameter of approximately 0.8 mm and a length of approximately 3 mm. Figure 12 Figure b shows the planar location of the defect.

[0251] CT scans were performed on 15 glass fiber reinforced composite material samples. The CT scan results of 14 of the samples were largely consistent, showing only small, irregularly shaped pores inside the samples. One sample's CT scan result was as follows: Figure 13 As shown in Figure a, a large hole defect was detected inside, with an average hole diameter of approximately 0.5 mm and a length of approximately 1.5 mm. Figure 13 Figure b shows the planar location of the defect.

[0252] The defect detection process for the test sample in S5 based on ultrasonic A-scan is as follows (data for test samples without defects is not shown):

[0253] Figure 14 The image shows the waveform of an ultrasonic phased array A-scan of the defect area of ​​a carbon fiber reinforced composite material. To ensure the accuracy of the ultrasonic testing, the defect area was repeatedly tested 15 times using an ultrasonic phased array instrument. After discarding unreasonable data, 8 sets of data were finally selected for subsequent analysis. It can be seen from the waveform that the surface echo, bottom echo, and defect echo are all quite obvious, but there is also a lot of noise. This is related to the fact that the carbon fiber composite material contains many reinforcing fiber bundles and there are a large number of tiny defects in this area.

[0254] The defect signal was then decomposed using the WOA-VMD model. The initial population size was set to 20, and the maximum number of iterations was set to 50. Figure 15 Table 1 shows the results of WOA-VMD decomposition, where table a is the optimized convergence curve of WOA, table b is the time-domain plot of the mode function obtained by WOA-VMD decomposition, and table c is the Fourier transform spectrum plot (frequency domain plot).

[0255] Depend on Figure 15 As shown in Table a, after 6 iterations, the optimal value of the envelope entropy, 3.31, was reached. At this point, the optimal parameter k was 7 and α was 620. Through WOA-VMD decomposition, 7 decomposition moduli were finally obtained. The energy proportion and ERA of each IMF component were calculated, and the specific results are shown in Table 11. The data in the table shows that the ERA of the first 6 IMF components has reached 99.5%, exceeding the specified 96%. Therefore, it is determined that the first 6 IMF components contain most of the feature information of the original defect, and the first 6 IMF components are selected for feature extraction.

[0256] Table 11:

[0257]

[0258] Process it in the same way as described above. Figure 16The image shows the waveform of an ultrasonic phased array A-scan of the defect and damage region in a glass fiber reinforced composite material. Figure 17 The result of WOA-VMD decomposition is... Figure 17 As shown in Table a, after 8 iterations, the optimal value of the envelope entropy, 3.42, was reached. At this point, the optimal parameter k was 7 and α was 258. Through WOA-VMD decomposition, 7 decomposition moduli were finally obtained. The energy proportion and ERA of each IMF component were calculated, and the specific results are shown in Table 12. The ERA of the first 5 IMF components reached 98.5%, exceeding the specified 96%. Therefore, it was determined that the first 5 IMF components contained most of the feature information of the original defect, and the first 5 IMF components were selected for feature extraction.

[0259] Table 12:

[0260]

[0261] After determining the order of the IMF components, the slope, kurtosis, peak value index, clearance index, shape index, impulse index, and energy were calculated as characteristic values ​​of the ultrasonic detection signal characterizing the defect. Table 13 shows the characteristic values ​​of the first 6 IMF components of carbon fiber reinforced composites after WOA-VMD decomposition, and Table 14 shows the characteristic values ​​of the first 5 IMF components of glass fiber reinforced composites after WOA-VMD decomposition.

[0262] Table 13:

[0263]

[0264] Table 14:

[0265]

[0266]

[0267] As shown in Tables 13 and 14, for the carbon fiber reinforced composite specimen, a matrix containing 42 eigenvalues ​​was obtained to characterize the defect. Similarly, the remaining 7 sets of ultrasonic test data for this defect were processed in the same way, resulting in an 8×42 eigenvector matrix. For the glass fiber reinforced composite specimen, a matrix containing 35 eigenvalues ​​was obtained to characterize the defect. The remaining 7 sets of ultrasonic test data for this defect were processed, resulting in an 8×35 eigenvector matrix.

[0268] To reduce computational complexity and remove redundant noise information, kernel principal component analysis (KPI) was used for dimensionality reduction, with a Gaussian kernel function selected. The contribution rates and cumulative contribution rates of each eigenvalue of the carbon fiber reinforced composite material specimens are shown in Table 15, and those of the glass fiber reinforced composite material specimens are shown in Table 16.

[0269] Table 15:

[0270]

[0271] Table 16:

[0272]

[0273] For the carbon fiber reinforced composite material sample, the cumulative contribution rate was greater than 95% when the first 8 principal components were selected. The optimized eigenvector matrix was constructed using the first 8 principal components, resulting in an 8×8 eigenvector matrix. For the glass fiber reinforced composite material sample, the cumulative contribution rate was greater than 95% when the first 5 principal components were selected. The optimized eigenvector matrix was constructed using the first 5 principal components, resulting in an 8×5 eigenvector matrix.

[0274] Defect identification based on IAO-SVM uses the feature vector matrix of artificial defects as the training set and the feature vector matrix of the sample defects in this embodiment as the test set. The specific process is as follows:

[0275] First, the carbon fiber reinforced composite material samples are classified and identified. The feature vector matrices of artificial defects (0.5mm, 0.8mm, 1.1mm, 2mm, 3mm, 4mm pores) in glass fiber reinforced composite molded parts and the feature vector matrices of artificial defects (0.5mm, 0.8mm, 1.1mm pores) in carbon fiber reinforced composite molded parts are labeled as nine categories: 1, 2, 3, 4, 5, 6, 7, 8, and 9, respectively, and used as the training set. Then, the defect feature vector matrices of the carbon fiber composite material samples to be tested in this embodiment are labeled nine times, sequentially as categories 1, 2, 3, 4, 5, 6, 7, 8, and 9, and used as the test set, corresponding one-to-one with artificial defects. The IAO-SVM category correspondence is as follows: Figure 18 As shown, Figure 18 Table a shows the fitness curves, derived from... Figure 18 As shown in Table a, after two iterations, the optimal cross-validation accuracy of 93.5% was achieved. At this point, the penalty factor C was 224.2, the optimized kernel parameter σ was 10, and the optimization time was 29 minutes. Figure 18 Table b shows the results for the test set. Figure 18Table c shows the confusion matrix of the test set. The classification categories of the test set are all category 8. This indicates that after classification and identification by IAO-SVM, the feature vector matrix of the defect is finally identified as matching the feature vector matrix of the φ0.8mm hole defect in the carbon fiber reinforced composite molded part. This is consistent with the experimental results of CT scan.

[0276] For glass fiber reinforced composite specimens, the same classification and identification method is used. For example... Figure 19 As shown, Figure 19 Table a shows the fitness curves, derived from... Figure 19 As shown in Table a, after two iterations, the optimal cross-validation accuracy of 93.7% was achieved. At this point, the penalty factor C was 243.4, the optimized kernel parameter σ was 9.5, and the optimization process took 30 minutes. Figure 19 Table b shows the results for the test set. Figure 19 Table c shows the confusion matrix of the test set. All categories in the test set are category 1. This proves that the IAO-SVM classification and recognition model ultimately identifies the feature vector matrix of the defect as being consistent with the feature vector matrix of the φ0.5mm hole defect in the glass fiber reinforced composite molded part, which is also consistent with the experimental results of CT scan.

[0277] Therefore, this method also proves that for defects of unknown size, whether glass fiber reinforced composites or carbon fiber reinforced composites, the ultrasonic A-scan signal of the defect can be acquired, the signal can be decomposed, the feature vector of the signal can be extracted and optimized, and then the IAO-SVM model can be used to classify and identify the defect and detect the size of the defect.

Claims

1. A method for quantitative detection of defects in composite molding materials using ultrasonic A-scan, characterized in that, The method for quantitative detection of defects in composite molding materials by ultrasonic A-scan includes the following steps: S1: Press flat plate-shaped parts and prepare artificial defect samples; S2: Perform ultrasonic A-scan detection on artificial defects using an ultrasonic phased array instrument and collect ultrasonic A-scan signals of artificial defects; S3: Decompose the ultrasonic A-scan signal. The defective ultrasonic A-scan signal is decomposed into a linear combination of intrinsic mode functions (IMFs) through the variational mode decomposition model WOA-VMD based on the whale algorithm optimization. The energy and energy ratio cumulative coefficient of each IMF component after decomposition are calculated. The IMF component corresponding to the energy ratio cumulative coefficient is selected when it exceeds 96%. S4: Select seven parameters—slope, kurtosis, peak value index, clearing index, shape index, impulse index, and energy—of the intrinsic mode function (IMF) components as feature vectors to characterize the defective ultrasound A-scan signal. Use kernel principal component analysis to remove irrelevant and redundant noise information from the feature vector matrix composed of the feature vectors, forming an optimized feature vector matrix to prepare for the next step of classification and recognition. S5: Press flat plate-shaped parts and prepare test samples; S6: Perform ultrasonic A-scan detection on the test sample in S5, and execute the signal processing steps of S3 to S5 on the ultrasonic A-scan signal. Use the optimized feature vector matrix obtained from the artificial defect as the training set and the optimized feature vector matrix obtained from the test sample as the test set. Use the improved Tianying algorithm-optimized support vector machine model IAO-SVM to classify and identify the training set and the test set.

2. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The specific steps for preparing artificial defect samples in S1 are as follows: Glass fiber reinforced composite prepreg is pressed into a flat plate shape using a compression molding machine. Five holes with a diameter of 0.5 mm, five holes with a diameter of 0.8 mm, and five holes with a diameter of 1.1 mm and a depth of 10 mm are drilled sequentially on the side of the flat plate shape. Five holes with a diameter of 2 mm, five holes with a diameter of 3 mm, and five holes with a diameter of 4 mm and a depth of 3 mm are drilled sequentially on the front of the flat plate shape, for a total of 30 holes. Carbon fiber reinforced composite prepreg is pressed into a flat plate shape using a molding equipment. In the thickness direction of the flat plate shape, five holes with a diameter of 0.5 mm, five holes with a diameter of 0.8 mm, and five holes with a diameter of 1.1 mm and a depth of 10 mm are drilled sequentially, for a total of 15 holes.

3. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The specific steps for collecting ultrasonic A-scan signals of artificial defects in S2 are as follows: Each artificial defect was tested 10 times. Each group of artificial defects with the same aperture formed 50 sets of ultrasonic A-scan data samples. After analyzing and screening the data samples, 30 sets of data samples were selected as the best.

4. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The specific steps for decomposing the ultrasonic A-scan signal of the defect in S3 using the variational mode decomposition model WOA-VMD optimized based on the whale algorithm are as follows: First, the ultrasound A-scan signal is decomposed into a variational model of a linear combination of a set of intrinsic mode functions (IMFs) using variational mode decomposition (VMD). Then, the Whale Optimization Algorithm (WOA) is used to optimize the parameter penalty term coefficient α and the number of parameter mode functions k in Variational Mode Decomposition (VMD), and the minimum value of the envelope entropy is used as the fitness function until the global optimal solution of the two parameters is obtained. By substituting the global optimal solution of the two parameters into the variational model and determining the optimal center frequency and finite bandwidth of each mode group, the intrinsic mode function (IMF) components of the signal are obtained for effective decomposition.

5. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, In S4, the kernel function used in kernel principal component analysis is the Gaussian kernel function.

6. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, In S4, the kernel principal component analysis method is used to reduce the dimensionality of the eigenvector matrix and select the principal components that have a cumulative contribution rate of more than 95% to form an optimized eigenvector matrix.

7. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The specific classification steps of the IAO-SVM support vector machine model optimized using the improved Skyhawk algorithm in S6 are as follows: The Support Vector Machine (SVM) model uses a radial basis function kernel and optimizes the radial basis function kernel parameter penalty factor C and kernel parameter σ using the improved Skyhawk Algorithm (IAO).

8. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 7, characterized in that, The improved Skyhawk Algorithm IAO introduces chaotic Tent mapping for population initialization based on the Skyhawk Algorithm AO. At the same time, it applies an adaptive weight factor ω to balance the search range and search capability. In the early stage of the AO algorithm iteration, it accelerates the iteration speed and expands the search range to determine the location of the target value.

9. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The specific steps for preparing the test sample in S5 are as follows: Glass fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 5 pieces. Three samples with a size of 3mm×10mm×4mm are taken out at intervals along the length of each flat plate-shaped part. Carbon fiber reinforced composite prepreg is pressed into flat plate-shaped parts using a molding equipment. The flat plate-shaped parts are 80mm×10mm×4mm in size, and there are 3 pieces. Three samples with a size of 3mm×10mm×4mm are taken out from the length of each flat plate-shaped part at intervals.

10. The detection method for quantitative detection of defects in composite molding materials by ultrasonic A-scan according to claim 1, characterized in that, The detection method further includes: S7: Perform a CT scan on the sample to be tested in S5, and compare the CT scan results with the classification and identification results of the ultrasound A-scan signal; The specific steps for performing a CT scan on the test sample in S7 are as follows: The first stage involves cross-sectional scanning along the 3mm length of each sample, with each cross-section spaced 0.2mm apart, for a total of 15 cross-sections; The second stage involves cross-sectional scanning along the 4mm thickness direction of each sample, with each cross-section spaced 0.1mm apart, for a total of 40 cross-sections; All cross-sectional tomographic reconstructions were used to create CT scan 3D images of each sample segment.

Images