Damage identification and failure analysis method of composite material specimens based on wavelet packet decomposition

Abstract

The present invention relates to the research field of composite material acoustic emission detection, and aims to provide a composite material sample damage identification and failure analysis method based on wavelet packet decomposition. It includes: conducting an acoustic emission test on a composite material ply sample, extracting the acoustic emission signal and performing wavelet packet decomposition; establishing the correlation between the full-frequency domain response characteristics of the acoustic emission signal under different damage modes and the damage mechanism based on the wavelet packet energy ratio and the k-means++ algorithm; based on the similarity metric, realizing damage pattern recognition through clustering of high-dimensional vectors, and analyzing the failure process of the damage mode; verifying the damage pattern recognition and failure process analysis of the sample based on the accumulated wavelet packet energy. The present invention can obtain damage classification and failure process analysis of composite materials under a variety of environments, and is applicable to dynamic data sets, which greatly improves the damage identification operation efficiency; the present invention is applicable to damage classification and failure process analysis of composite materials under a variety of environments.

Description

Technical Field

[0001] The invention relates to the research field of composite material acoustic emission detection, and in particular to a composite material ply specimen damage pattern recognition and failure process analysis method based on wavelet packet decomposition. Background Art

[0002] With the continuous progress of society and the continuous development of science and technology, it is increasingly difficult for a single material to meet the stringent requirements of advanced technology for material strength, stiffness, density, durability and other aspects. Therefore, composite materials composed of two or more materials with different characteristics play an increasingly important role, among which fiber-reinforced polymer composite materials (FPR) are the most widely used. The layering process is a molding method for composite materials: prepreg materials (non-weft cloth, non-weft tape, woven fabric, etc.) are laid layer by layer in a mold in a predetermined direction and order until the required thickness (or number of layers), and the products are obtained by heating, pressurizing, curing, demoulding and trimming. The products formed by layering are relatively strong, and can be made into products with anisotropic strength according to the stress conditions in different directions. The harsh service environment and complex load conditions have brought great challenges to the long-term and stable service of composite structures. Therefore, it is necessary to develop non-destructive damage detection technology and structural health monitoring technology, among which acoustic emission technology has development potential in this regard.

[0003] When damage occurs inside the material, the damage source quickly releases energy to generate transient elastic waves, which is called acoustic emission. The elastic waves propagate in the material and cause small vibrations on the surface when they reach the surface. This weak vibration signal can usually be detected by a sensor and converted into an electrical signal, which is then recorded after being amplified, preprocessed and digitized by the acquisition system.

[0004] Acoustic emission signals contain information related to the initiation and evolution of damage. If we want to mine damage-related information, we must develop appropriate signal analysis and processing methods. According to certain definitions or mathematical models, various acoustic emission features can be extracted from the time domain waveform or spectrum of the impact signal, and features are usually an important basis for acoustic emission signal analysis and processing.

[0005] Acoustic emission feature selection is very important but often overlooked, especially in real-time monitoring. Acoustic emission signals are generated in the form of data streams, and their impact on feature selection has not yet been explored. Therefore, the reliability of damage pattern recognition based on all acoustic emission data remains to be verified; in addition to damage identification, other information related to damage characterization contained in acoustic emission features remains to be explored.

[0006] Conventional identification of acoustic emission response characteristics of composite material damage modes often relies on a feature set consisting of multiple screened conventional acoustic emission features, in which the amount of information contained in the original signal waveform is unknown, and the feature set often varies depending on instrument settings, material systems, load modes, structural forms and environments. The key is that the amount of information contained in the original waveform of such a feature set is difficult to measure, resulting in a lack of comparability in signal characterization under different conditions and a lack of comparability between different features under the same conditions, such as different dimensions of features, and the fluctuation range of signal values ​​in each feature dimension is usually not in the same order of magnitude. Therefore, the signal preprocessing process before damage identification is relatively cumbersome, and the signal characterization method is more suitable for damage identification based on static data sets (such as all acoustic emission data generated during the failure process of the specimen).

[0007] In view of the problems existing in existing damage identification models, such as the cumbersome signal pre-processing process and the difficulty in measuring feature coverage information, the present invention develops a feature selection method for acoustic emission data streams in composite materials to provide a simpler and more reliable acoustic emission signal characterization method, and thereby carries out damage pattern recognition and failure process analysis of composite ply specimens. Summary of the invention

[0008] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a method for damage pattern recognition and failure process analysis of composite material ply specimens, particularly one based on wavelet packet decomposition.

[0009] To solve the technical problem, the solution of the present invention is:

[0010] A method for damage pattern recognition and failure process analysis of composite material ply specimens based on wavelet packet decomposition is provided, comprising the following steps:

[0011] (1) Conduct acoustic emission tests on composite laminate specimens, extract acoustic emission signals and perform wavelet packet decomposition;

[0012] (2) Based on the wavelet packet energy ratio and k-means++ algorithm, the correlation between the full-frequency response characteristics of the acoustic emission signal under different damage modes and the damage mechanism is established; based on the similarity measurement, the damage pattern recognition is realized by clustering the high-dimensional vectors, and the failure process of the damage mode is analyzed;

[0013] (3) The damage pattern recognition and failure process analysis of the specimen are verified based on the accumulated wavelet packet energy.

[0014] As a preferred embodiment of the present invention, the step (1) specifically comprises:

[0015] (1.1) Extract the original waveform signal of each acoustic emission impact signal, and derive the frequency domain details of the input waveform by multiple iterations of discrete wavelet transform;

[0016] (1.2) The optimal wavelet packet basis is selected according to the minimum Shannon entropy, and the wavelet packet decomposition is used to decompose the wavelet subspace in multi-resolution analysis to achieve focusing on any frequency band of the signal, obtain the energy ratio of the reconstructed signal at different levels and frequency ranges, and characterize the acoustic emission waveform signal as a high-dimensional vector.

[0017] As a preferred embodiment of the present invention, the step (2) specifically comprises:

[0018] (2.1) The acoustic emission signal is regarded as a vector in a high-dimensional space, and the cosine similarity is used to measure the similarity of different signals;

[0019] (2.2) The cosine similarity between vectors represented by acoustic emission impact signals is used as the clustering basis for high-dimensional vectors represented by wavelet packet decomposition, and the k-means++ algorithm is used to realize damage pattern recognition of acoustic emission data; using the cumulative number of acoustic emission impact signal pairs, a time-related damage category transformation matrix is ​​constructed to characterize the transformation relationship between various damage modes and realize the analysis of failure process;

[0020] As a preferred embodiment of the present invention, there are four damage modes, namely: matrix cracking, fiber / matrix interface debonding, interlayer delamination, and fiber breakage; among them, the peak frequency of fiber breakage is the highest, followed by interlayer delamination and fiber / matrix interface debonding, and the least is matrix cracking.

[0021] As a preferred embodiment of the present invention, the step (3) specifically includes: utilizing the differences in the values ​​of the accumulated wavelet packet energy of each damage mode in different frequency bands and the differences in the rise time, as well as the changing trends of the accumulated wavelet packet energy in each frequency band under the same damage mode, to verify the damage mode recognition and failure process analysis results of the sample obtained in step (2).

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. The present invention extracts two sets of features, namely wavelet packet energy ratio and wavelet packet energy, and determines the response characteristics of four damage mechanisms of composite materials in the full frequency domain based on energy ratio and cluster analysis; completes the failure process analysis of FRP one-way plates based on the cumulative statistics of wavelet packet energy; can obtain damage classification and failure process analysis of composite materials under various environments, and is suitable for dynamic data sets.

[0024] 2. The present invention reduces the steps of noise removal, feature selection, feature normalization, etc. in damage identification based on acoustic emission response characteristics by decomposing FRP acoustic emission signals based on wavelet packets, thereby greatly improving the efficiency of damage identification operations.

[0025] 3. The identification of acoustic emission response characteristics of composite material damage modes in the prior art often relies on a feature set consisting of multiple screened conventional acoustic emission features. The amount of information contained in the original waveform of such a feature set is difficult to measure, resulting in a lack of comparability in signal representation under different conditions and a lack of comparability between different features under the same conditions. In addition, the cumbersome signal preprocessing steps in the early stage make it impossible to apply it to the dynamic data streams generated in real-time health monitoring.

[0026] Compared with the prior art, the present invention provides a more concise and reliable method for characterizing acoustic emission signals of composite material acoustic emission data streams, and performs damage pattern recognition and failure process analysis on composite material ply specimens.

[0027] 4. As the damage process continues, the energy ratio of each damage mode in the main frequency band fluctuates very little and can be ignored, indicating that damage identification based on wavelet packet energy ratio is robust to the damage process. Therefore, the present invention is suitable for damage classification and failure process analysis of composite materials under various environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of a method for characterizing acoustic emission signals of composite material ply specimens based on wavelet packet decomposition;

[0029] Figure 2 It is the geometric morphology of the composite material UDL specimen;

[0030] Figure 3 is the clustering quality evaluation of the acoustic emission signals of the UDL specimens under different clustering numbers: (a) specimen UDL1, (b) specimen UDL2, (c) specimen UDL3, (d) specimen UDL4;

[0031] Figure 4 is the energy distribution of the four types of central signals of the UDL specimen in each frequency band: (a) type 1, (b) type 2, (c) type 3, and (d) type 4;

[0032] Figure 5 It is a schematic diagram of intra-class enhancement and inter-class transformation of damage categories. DETAILED DESCRIPTION

[0033] The present invention will be described in detail below with reference to the accompanying drawings and embodiments, and the purpose and effects of the present invention will become more apparent.

[0034] like Figure 1As shown, the present invention first extracts the original waveform of each acoustic emission impact signal, and then decomposes it into several sub-components in the frequency domain by wavelet packet decomposition, that is, an acoustic emission impact signal is represented as a vector composed of energy ratios of multiple continuous frequency bands, and two groups of simple features are extracted respectively by calculating the wavelet packet energy ratio and wavelet packet energy of each component, so that the acoustic emission waveform signal is characterized as a high-dimensional vector, and finally the two groups of extracted vectors are used to realize the damage identification and failure process analysis of the composite material sample respectively.

[0035] The method of the present invention specifically comprises the following steps:

[0036] (1) Acoustic emission tests were carried out on composite laminate specimens, and the acoustic emission signals were extracted and decomposed by wavelet packets.

[0037] The test object is a composite unidirectional laminate (UDL) specimen prepared according to the international standard ASTM D3039. During the tensile process of the composite unidirectional laminate. To prevent the occurrence of clamping damage, wedge-shaped reinforcement sheets are bonded at the clamping positions at both ends of the specimen. The reinforcement sheet material is glass fiber composite material. The thickness of the adhesive film J116B used for bonding is 0.1mm. The width of the specimen and the reinforcement sheet is 12mm. Before the test, the two specimens of UDL3 and UDL4 were placed in 70℃ water for 25 days for wet heat aging pretreatment. The correctness of this method for damage identification and failure process analysis of composite specimens was verified by comparing the two types of specimens. The acoustic emission test was carried out according to the set process, and the acoustic emission impact signal was recorded, that is, any signal that exceeded the threshold and caused a system channel to collect data. In order to exclude the air between the piezoelectric ceramic sensor and the UDL specimen, firstly, vacuum grease was evenly applied to the position of the specimen to be installed with the sensor, and then the two sensors were symmetrically fixed on the UDL specimen with adhesive tape. The distance D between the sensors was controlled within the range of 60-100mm. As the load increases, signals are continuously generated and first detected by two sensors with a sampling frequency of 1Mhz, then pre-processed by two corresponding 2 / 4 / 6AST preamplifiers with an amplification gain of 40dB to resist interference and improve the signal-to-noise ratio, and finally recorded in the PAC Samos-48 acoustic emission instrument. The positioning mode is set to linear positioning. Other important parameters include threshold, peak definition time, impact definition time, impact lockout time and maximum duration, which are set to 40dB, 50μs, 200μs, 300μs and 1000ms respectively. Before the test, multiple lead break tests were performed between the sensors, and the response amplitude of each time was above 95dB, and the linear positioning of the lead break position was accurate. See the attached geometric morphology of the composite UDL specimen. Figure 2 , the geometric dimensions and ply angles of the composite UDL specimens are shown in the following table.

[0038]

[0039] First, the original waveform signal x(t) of each acoustic emission impact signal is extracted. Its frequency domain information can be obtained by processing a series of filters with different center frequencies but the same bandwidth. That is, the frequency domain details of the input waveform can be derived by multiple iterations of discrete wavelet transform (DWT). Discrete wavelet transform comes from continuous wavelet transform (CWT), as shown in formula (1).

[0040]

[0041] Where a, b, ψ and * represent the expansion coefficient in the frequency domain, the translation coefficient in the time domain, the wavelet function and the complex conjugate operator respectively, and t is the time.

[0042] The reconstruction formula of the original signal restored by CWT is not unique, that is, there is redundancy in information representation in CWT, which increases the difficulty of analyzing and interpreting CWT results. Information redundancy can be eliminated by DWT, that is, discretizing a and b into 1 / 2 respectively. j and k / 2 j , where j and k represent the decomposition level and the factor in the time domain respectively, and the wavelet function ψ a，b (t) can also be written as formula (2).

[0043]

[0044] However, DWT only focuses on the approximate part of the low-frequency region in each layer of decomposition, while ignoring the details of the high-frequency region. Therefore, it is necessary to introduce wavelet packet decomposition (WPD) to decompose the wavelet subspace in multi-resolution analysis, so as to achieve the focus on any frequency band of the signal. In this way, WPD can avoid information redundancy and prevent information omission. If φ(t) and ψ(t) represent the scaling function and wavelet function respectively, the wavelet packet can be defined as a set of functions that satisfy the following recursive relationship {w n (t)}, as shown in formula (3).

[0045]

[0046] Where: n is the frequency parameter, k is the time domain factor, h(k) and g(k) are the coefficients of the low-pass filter and high-pass filter in multi-resolution analysis, respectively. These two coefficients satisfy the orthogonal relationship, that is, g(k) = (-1) k h(1-k).

[0047] If n is the subdivided octave parameter, that is, n = 2 l +m, the wavelet packet can be simplified as a function as shown in formula (4).

[0048] ψ j，k，n (t) = 2-j / 2 ψ n (t·2 -j -k) (4)

[0049] Where: The integers m and l range from 0 to 2. l -1 and 1 to j.

[0050] Compared with DWT, WPD introduces a frequency parameter n to improve the defect of poor frequency resolution of DWT. Here the frequency parameter represents ψ n The number of zero crossings of (t) is the number of oscillations of the waveform. The wavelet packet coefficient c of the input waveform x(t) is j，k，n It can be calculated according to formula (5).

[0051]

[0052] This set of coefficients is also the coefficients of the fitting polynomial used to describe the waveform. Therefore, each component of the signal can be reconstructed according to equation (6).

[0053]

[0054] The original signal can be obtained by accumulating each reconstructed component, as shown in formula (7).

[0055]

[0056] Different combinations of orthogonal wavelet packet bases can achieve waveform decomposition, as long as the selected base (component) can cover the entire space. This will lead to different feature representations between different signals, that is, different signals will lack comparability. Therefore, the criterion for selecting the best wavelet packet basis is crucial for feature extraction. The most widely used wavelet packet basis selection method is the minimum Shannon entropy, which is defined as shown in formula (8).

[0057]

[0058] Where: c n is the wavelet packet coefficient. However, the optimal wavelet packet basis corresponding to different acoustic emission signals may still be different, resulting in the local frequency domain of the reconstructed form being not completely the same.

[0059] In order to obtain a consistent signal representation form, each acoustic emission waveform is completely decomposed at the same level. In this example, the Daubechies wavelet basis function with a vanishing moment of 5 (denoted as db5) is used to decompose the acoustic signal to the third level, and 8 components can be obtained. At a sampling frequency of 1 MHz, the frequency domain from 0 to 500 kHz is divided into 8 frequency bands, denoted as F1-F8. The energy ratio of the reconstructed signal in the nth frequency range of the jth level can be calculated according to formula (9).

[0060]

[0061] Where: E j，n (t) represents the wavelet packet energy of the signal in the nth frequency range of the jth level, x j，n (t) represents the reconstructed signal in the nth frequency range of the jth level. Since the wavelet packet coefficients are coefficients that describe the similarity between the scale function and the wavelet function and the original signal, the wavelet packet energy can reflect the amplitude of the original signal, that is, the intensity of the damage source, to a certain extent. Therefore, it is used as the first set of features after wavelet packet decomposition.

[0062] The AE signal is decomposed into the third layer using db5 wavelet, and the range of each frequency band is about 50kHz, which is enough to distinguish various damage sources. Therefore, an AE impact signal can be represented as a vector composed of the energy ratio of eight frequency bands, and then the clustering analysis of all signals in the failure process of the specimen can be completed by calculating the cosine distance between the vectors and using the clustering algorithm k-means++.

[0063] 2. Based on the wavelet packet energy ratio and k-means++ algorithm, the correlation between the full-frequency response characteristics of the acoustic emission signal and the damage mechanism under different environments is established. Based on the similarity measurement, the damage pattern recognition is realized through the clustering of high-dimensional vectors, and the failure process of the damage pattern is analyzed.

[0064] Different damage sources have different acoustic emission response characteristics in the time domain and frequency domain. If the signals from different damage modes (such as fiber breakage and matrix cracking) are completely decomposed in the same limited complementary frequency band, the signal can be characterized by the energy ratio of the decomposed limited frequency band. The energy ratio avoids the huge difference in the absolute value of the wavelet packet energy of different signals, so that different signals are comparable. At the same time, the energy ratio of different frequency bands of the same signal is also comparable. Specifically, the frequency bands corresponding to each node are regarded as different feature dimensions, and the energy ratio of each frequency band represents the response value of the signal on this feature dimension. Therefore, the second set of features of the signal after wavelet packet decomposition is the combination of the energy ratios of each frequency band, that is, a signal can be represented as a vector in a high-dimensional space composed of all complementary frequency bands, or as a point in a hyperplane. By extracting these two sets of naive features, the acoustic emission waveform signal is characterized as a high-dimensional vector.

[0065] Obviously, there is no overlap between any two frequency bands, that is, the signal representation method based on the wavelet packet energy ratio is orthogonal, which can avoid the data redundancy existing in the conventional feature combination representation method, and the combination of all frequency bands can just cover the entire frequency domain space, and can also reduce information omission. Therefore, here, the signal preprocessing steps such as feature normalization and feature selection used in traditional technologies can be omitted, which makes this set of features not only be used for the clustering of static data, but also for the clustering of dynamic acoustic emission data streams.

[0066] Experiments have shown that the energy distribution of the same damage pattern in each frequency band should be similar, that is, the vectors corresponding to the same type of damage signals should appear in similar directions. Since the acoustic emission signal can be regarded as a vector in a high-dimensional space, the similarity of different signals can be measured by cosine similarity, and the distance between any two high-dimensional vectors x and y can be calculated according to formula (10).

[0067] D(x,y)=1-similarity(x,y)=1-cos<x,y> (10)

[0068] Where: Represents the cosine of the angle between vectors.

[0069] Considering that the sum of the energy proportions of the signal in each frequency band is always equal to 1, the limitation of the absolute value of cosine similarity can be avoided. Based on the similarity metric, damage pattern recognition can be achieved by clustering high-dimensional vectors, so that the damage evolution law of composite materials can be explored from two aspects: on the one hand, for the same type of damage, the change in the energy proportion distribution of this type of signal in each frequency band can reflect the change of the main characteristics of this type and the degree of its influence by other damages to a certain extent. On the other hand, for all damage categories, the self-reinforcement within the class and the mutual transformation between classes can reflect the mutual correlation between the damage patterns and the damage evolution law to a certain extent. Therefore, the combination of multiple wavelet packet decomposition features can determine the category of heterogeneous signals.

[0070] Based on the above processing, this application further uses the k-means++ algorithm to realize composite material damage pattern recognition of acoustic emission data, which mainly improves the problem that the clustering results of the k-means algorithm are easily affected by the initial value. In this method, the total number of clusters needs to be determined in advance. In order to select an appropriate total number of clusters, two evaluation indicators are introduced to measure the clustering quality under a certain number of clusters, namely the silhouette coefficient (SI) and the Davidson-Bodin index (DBI).

[0071] SI is an indicator for measuring clustering quality based on compactness and separation, that is, the closer the data points within a category are and the more distant the data points between categories are, the higher the quality of the clustering result is. i , its silhouette coefficient can be defined according to formula (11).

[0072]

[0073] Where: represents the average length of all lines from xi to all other data points in the same category, Indicates that from x i The minimum average length of all links to all data points in other categories. The smaller the size, the tighter the clustering within the class. The larger the value, the higher the degree of separation between clusters. When the quality of clustering improves, the average SI of all data points will increase. DBI is an internal metric used to evaluate clustering quality and is defined as shown in formula (12).

[0074]

[0075] Where: d jk Represents the distance between the j-th class center and the k-th class center. SD j and SD k They represent the discreteness of the acoustic emission data points in the jth and kth categories respectively. The discreteness of the jth category can be defined according to formula (13).

[0076]

[0077] In the formula, |C j | represents the total number of data points in the jth class, and the constant q controls how the distance metric is calculated. j It represents the average distance between each data point in the jth class and the center of the class. The lower the DBI, the better the clustering quality.

[0078] The clustering quality of the k-means++ algorithm with different cluster numbers was evaluated by SI and DBI. Considering the possible damage modes of composite materials, the number of clusters to be evaluated ranged from 2 to 10. The evaluation results of the experiment are shown in the attached Figure 3 When the number of clusters is 4, the SI value is relatively high and the DBI value is relatively low, so 4 is a more appropriate total number of clusters. Afterwards, the cosine similarity between the vectors represented by the acoustic emission impact signal is used as the clustering basis for the high-dimensional vector represented by the wavelet packet decomposition. The k-means++ algorithm is used to divide the data set composed of all acoustic emission impact signals into 4 categories. The energy distribution of the four types of central signals of the sample in each frequency band is shown in the attached figure. Figure 4 shown.

[0079] Referring to the peak frequency characteristics of existing studies, the four common damage modes in carbon fiber reinforced resin-based composites include matrix cracking, fiber / matrix interface debonding, interlayer delamination and fiber breakage. Since moisture-heat aging will significantly affect the energy distribution ratio of the signal clustering center in each frequency band, the peak frequency range of each damage mode is not fixed for composite materials of different periods, but can be distinguished by the peak frequency size. Among them, the peak frequency size of fiber breakage is the highest, followed by interlayer delamination and fiber / matrix interface debonding, and the least is matrix cracking. Therefore, the composite material damage mode recognition of acoustic emission data can be preliminarily realized. Figure 4 In the figure, classes 1, 2, 3, and 4 correspond to matrix cracking, fiber / matrix interface debonding, interlaminar delamination, and fiber breakage, respectively, and the correctness of the identification results can be further confirmed in the subsequent damage evolution process.

[0080] For the unaged carbon fiber reinforced resin matrix composite specimen, further damage pattern recognition and evolution analysis are as follows:

[0081] The main frequency bands of matrix cracking, fiber / matrix interface debonding, interlaminar delamination and fiber breakage are F1 (0-62.5kHz), F4 (187.5-250kHz), F7 (375-437.5kHz) and F8 (437.5-500kHz), respectively.

[0082] Most of the impact signals corresponding to matrix cracking have a higher energy proportion in the frequency band F1 (0-62.5kHz) and a lower energy proportion in other frequency bands.

[0083] The fiber / matrix interface debonding is very active in the main frequency band F4 (187.5-250kHz), and is intermittently active in the minor frequency bands F1 (0-62.5kHz) and F7 (375-437.5kHz), especially in F1.

[0084] The main frequency band of interlayer delamination is F7 (375-437.5kHz), which is relatively stable in the early stage and significantly increases with the increase of load in the later stage. Another minor frequency band of interlayer delamination, F1, which is the main frequency band of matrix cracking, is activated at regular intervals at the beginning of the test.

[0085] The damage evolution of fiber breakage is the most chaotic because it can be regarded as a combination of the other three damage modes, that is, fiber breakage is difficult to occur alone without affecting the matrix, fiber / matrix interface, and interlayer interface. In the last 25% of the load time, fiber breakage signals frequently broke out, accompanied by active delamination signals (F7 and F6) and fiber / matrix interface debonding signals (F4 and F3). Before this, the activity of fiber breakage was not strong.

[0086] In addition to the main frequency bands, each damage mode still has other secondary frequency bands. For matrix cracking, F2, F4 and F7 are secondary frequency bands. Compared with the other three damage modes, the energy dispersion of matrix cracking signals in the secondary frequency bands is not obvious. For the other three damages, quite a few impact signals have an energy proportion of more than 20% in certain frequency bands, especially F1, F2 and F7 corresponding to fiber / matrix interface debonding, F1, F3 and F4 corresponding to interlayer delamination, and F1, F3, F4, F6 and F7 corresponding to fiber breakage.

[0087] After wet-heat aging, the energy ratio distribution of the acoustic emission signals of the four damage modes in each frequency band has changed. The most important of these is the change in the main frequency bands of each damage mode after aging. The main frequency band of fiber / matrix interface debonding is reduced from F4 (187.5-250kHz) to F1 (0-62.5kHz); the main frequency band of interlayer delamination is reduced from F7 (375-437.5kHz) to F4 (187.5-250kHz); the main frequency band of fiber breakage is reduced from F8 (437.5-500kHz) to the adjacent frequency band F7; for matrix cracking, the energy of the signal is more concentrated in the frequency band F1 (0-62.5kHz), so that more signals have a relatively high energy ratio on F1. Through the distribution of the main and secondary frequency bands before and after aging, the type of damage mode can be further determined, and the identification results based on the peak frequency size of each frequency band after clustering can be further verified.

[0088] From the above, it can be inferred that the acoustic emission signal of fiber / matrix interface debonding is mixed with the characteristics of matrix cracking and interlayer delamination. The occurrence of fiber breakage is accompanied by matrix cracking, fiber / matrix interface debonding and interlayer delamination, and the occurrence of interlayer delamination is accompanied by fiber / matrix interface debonding and matrix cracking. It can be seen that the various damage modes are interrelated and influence each other.

[0089] Based on the above damage identification results, the energy ratio distribution changes of the acoustic emission signals of the four damage modes at different times in each frequency band are recorded to further study the changes in the acoustic emission response characteristics of each damage mechanism during the load-bearing process. In order to simply represent the transformation relationship between the damage modes, the cumulative number of acoustic emission impact signal pairs is considered and set as follows: For the acoustic emission impact signal x generated at any time t (t) , and its corresponding damage category can be expressed as a cluster label where n c is the total number of damage categories, then the cluster label of the acoustic emission signal generated at the next moment is like and If they are equal, it means that the damage has undergone self-reinforcement within the class, and the cumulative number of collision signal pairs representing the reinforcement within the class of the damage needs to be increased by one. Otherwise, it means that the first The damage is transformed into To identify the type of damage, the cumulative number of collision signal pairs representing the corresponding type of transformation needs to be increased by one. Obviously, the enhancement of the activity of a certain damage pattern may come from its own direct enhancement or from the indirect enhancement of a non-same type of damage pattern. Therefore, a time-dependent n c ×n c The damage category transformation matrix describes the damage evolution process, as shown in formula (7). The schematic diagram of the damage category intra-category enhancement and inter-category transformation is shown in the attached figure. Figure 5 shown.

[0090]

[0091] Where: cum(j, k) (t) It represents the cumulative number of acoustic emission impact signal pairs that transform the j-th damage into the k-th damage by time t. Figure 2 As shown in the figure, the path of intra-class enhancement and inter-class conversion is: Class 1 damage is first converted to Class 3, then Class 3 undergoes intra-class enhancement, then converts to Class 2, and then converts to Class 3. After a series of conversion processes, Class 2 is finally converted to Class 1. At this point, the damage category conversion matrix is ​​a third-order square matrix with all elements being 1. The process of forming this matrix can be considered as the damage evolution process of the composite material based on the acoustic emission signal.

[0092] Among the four damage modes, the main frequency band of matrix cracking is the most stable during the damage evolution process, and the secondary frequency band is the least significant, which means that the initiation and extension of matrix cracking have the least impact on other damage modes. Therefore, matrix cracking is the basic damage mode of composite laminates. In the first half of the load cycle, the matrix cracking activity is intermittently enhanced. In the later stage of the load cycle, the matrix cracking activity is not significantly enhanced, that is, the main damage modes before fracture are the other three.

[0093] The main frequency bands of matrix cracking, fiber / matrix interface debonding and interlayer delamination are very stable and do not change significantly during the damage process, indicating that the signal characterization model based on wavelet packets is robust to the damage process. The clustering results based on this signal characterization model mean that the relative energy ratio of the same damage mode in different frequency bands remains dynamically stable, and the relative energy ratio of different damage modes in the same frequency band does not change due to the damage process. However, the secondary frequency bands of each damage mode fluctuate with the damage process, which means that the same damage mode in different time periods is not affected by other damage modes in the same way, or its influence on other damage modes is different. The damage modes influence and correlate with each other and are competitive with each other. This phenomenon does not disappear due to hygrothermal aging.

[0094] The fourth-order damage category conversion matrix of the UDL specimen is calculated to represent the change of the acoustic emission impact signal over time for each damage mode undergoing intra-class reinforcement or inter-class transformation. Taking specimen UDL1 as an example, the probability of each damage mode transforming to matrix cracking in the early stage of load bearing is the highest, followed by fiber / matrix interface debonding, then interlaminar delamination, and finally fiber fracture. As the load gradually increases, the activity of interlaminar delamination and fiber fracture gradually increases. As the load approaches the failure load, the probability of fiber fracture transforming to interlaminar delamination exceeds the probability of its transformation to fiber / matrix interface debonding and matrix cracking, and at the same time, the probability of interlaminar delamination transforming to fiber fracture also increases significantly. Subsequently, the probability of intra-class reinforcement of interlaminar delamination and fiber fracture exceeds the probability of its transformation to fiber / matrix interface debonding and matrix cracking, and this state will continue until the specimen fails completely.

[0095] 3. The damage pattern recognition and failure process analysis of the specimen are verified based on the accumulated wavelet packet energy.

[0096] The result of wavelet packet decomposition not only includes a set of wavelet packet energy ratios suitable for damage pattern recognition, but also a set of wavelet packet energies that may act similarly to acoustic emission energy. The cumulative acoustic emission energy refers to the integral of the square of the signal voltage from 0 to the present time divided by the resistance, and the cumulative wavelet packet energy refers to the integral of the square of the signal voltage from 0 to the present time in each wavelet packet decomposition frequency band divided by the resistance.

[0097] The cumulative wavelet packet energy of each damage mode in the main frequency band is higher than that in other frequency bands, that is, matrix cracking, fiber / matrix interface debonding, interlayer delamination and fiber fracture show the highest cumulative wavelet packet energy in frequency bands F1 (0-62.5kHz), F4 (187.5-250kHz), F7 (375-437.5kHz) and F8 (437.5-500kHz), respectively. In addition to the main frequency band, the cumulative wavelet packet energy of each damage mode in some secondary frequency bands is also high, such as F2 and F4 for matrix cracking, F1 and F2 for fiber / matrix interface debonding, F1, F3 and F4 for interlayer delamination and F3, F6 and F7 for fiber fracture. For the same damage mode, the cumulative wavelet packet energy of each frequency band shows a trend of increasing in stages, that is, a step-by-step increase in the early stage of loading, an increasing platform period in the middle stage of loading and a rapid increase in the late stage of loading, which is basically consistent with the accumulated acoustic emission energy. Therefore, wavelet packet energy can also play a role similar to acoustic emission energy to a certain extent, that is, qualitatively characterize the degree of damage to the specimen. The accumulated wavelet packet energy of the four damage modes in the main features has an obvious high-low order, with matrix cracking being the highest, followed by fiber / matrix interface debonding, then interlayer delamination, and fiber breakage being the lowest, which further verifies the damage mode recognition results based on the peak frequency size of each frequency band after clustering.

[0098] The frequency bands with the highest accumulated wavelet packet energy of each damage mode in the UDL specimen after hygrothermal aging treatment also change with the change of the main frequency band, that is, matrix cracking, fiber / matrix interface debonding, interlayer delamination and fiber breakage accumulate the highest wavelet packet energy in the frequency bands F1 (0-62.5kHz), F1 (0-62.5kHz), F4 (187.5-250kHz) and F7 (375-437.5kHz), respectively. In addition to the main frequency band, the accumulated wavelet packet energy of each damage mode in some secondary frequency bands is also high, that is, F2 and F4 for matrix cracking, F4 and F7 for fiber / matrix interface debonding, F1 and F7 for interlayer delamination, and F1, F3 and F4 for fiber breakage. Compared with the unaged specimens, the two accumulated energies of the same damage mode in the aged specimens have a similar step-by-step upward trend and are more significant. For different damage modes, the time of the accumulated wavelet packet energy jump is not the same, which can be used to analyze the evolution of each damage mode during the failure process.

[0099] For different damage modes, the time of cumulative wavelet packet energy jump is not consistent. In the early stage of bearing, the cumulative wavelet packet energy of matrix cracking in specimen UDL1 first jumps at about 13s, while the time of the first jump of the cumulative energy of the other three damage modes is 19s. In addition, the two cumulative energies of matrix cracking are numerically the highest among the four damage modes, which shows that matrix cracking is the most basic damage mode in composite materials. Before the failure of specimen UDL1, the cumulative energy jumps of fiber breakage and interlayer delamination are slightly earlier than those of matrix cracking and fiber / matrix interface debonding, which is consistent with the failure process analysis based on wavelet packet energy ratio.

Claims

1. A composite material specimen damage identification and failure analysis method based on wavelet packet decomposition, characterized in that: The following steps are involved: (1) Conduct acoustic emission tests on composite laminate specimens, extract acoustic emission signals and perform wavelet packet decomposition; (2) Based on the wavelet packet energy ratio and k-means++ algorithm, the correlation between the full-frequency response characteristics of the acoustic emission signal under different damage modes and the damage mechanism is established; based on the similarity measurement, the damage mode recognition is realized by clustering the high-dimensional vectors, and the failure process of the damage mode is analyzed; specifically, the following are included: (2.1) The acoustic emission signal is regarded as a vector in a high-dimensional space, and the cosine similarity is used to measure the similarity between different signals; (2.2) The cosine similarity between vectors represented by acoustic emission impact signals is used as the clustering basis for high-dimensional vectors represented by wavelet packet decomposition, and the k-means++ algorithm is used to realize damage pattern recognition of acoustic emission data; using the cumulative number of acoustic emission impact signal pairs, a time-related damage category transformation matrix is ​​constructed to characterize the transformation relationship between various damage modes and realize the analysis of failure process; (3) The damage pattern recognition and failure process analysis of the specimen are verified based on the accumulated wavelet packet energy.

2. The method according to claim 1, characterized in that The step (1) specifically includes: (1.1) Extract the original waveform signal of each acoustic emission impact signal, and derive the frequency domain details of the input waveform by multiple iterations of discrete wavelet transform; (1.2) The optimal wavelet packet basis is selected according to the minimum Shannon entropy, and the wavelet packet decomposition is used to decompose the wavelet subspace in multi-resolution analysis to achieve focusing on any frequency band of the signal, obtain the energy ratio of the reconstructed signal at different levels and frequency ranges, and characterize the acoustic emission waveform signal as a high-dimensional vector.

3. The method according to claim 1, characterized in that There are four types of damage modes, namely: matrix cracking, fiber / matrix interface debonding, interlayer delamination, and fiber breakage; among them, the peak frequency of fiber breakage is the highest, followed by interlayer delamination and fiber / matrix interface debonding, and the least is matrix cracking.

4. The method according to claim 1, characterized in that The step (3) specifically includes: The damage pattern recognition and failure process analysis results of the sample obtained in step (2) are verified by using the differences in the accumulated wavelet packet energy values ​​and the jump time differences of each damage mode in different frequency bands, as well as the changing trends of the accumulated wavelet packet energy in each frequency band under the same damage mode.

Images