CN114235957B - Method for detecting interface debonding defect of multi-layer bonding structure - Google Patents
Method for detecting interface debonding defect of multi-layer bonding structure Download PDFInfo
- Publication number
- CN114235957B CN114235957B CN202111543787.2A CN202111543787A CN114235957B CN 114235957 B CN114235957 B CN 114235957B CN 202111543787 A CN202111543787 A CN 202111543787A CN 114235957 B CN114235957 B CN 114235957B
- Authority
- CN
- China
- Prior art keywords
- energy
- defect
- signal
- bonding structure
- acoustic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000007547 defect Effects 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000001514 detection method Methods 0.000 claims abstract description 44
- 238000003064 k means clustering Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 51
- 239000011159 matrix material Substances 0.000 claims description 35
- 238000010606 normalization Methods 0.000 claims description 19
- 230000001186 cumulative effect Effects 0.000 claims description 17
- 238000012795 verification Methods 0.000 claims description 14
- 230000009467 reduction Effects 0.000 claims description 10
- 230000005236 sound signal Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 229910000831 Steel Inorganic materials 0.000 claims description 7
- 239000010959 steel Substances 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 229920001971 elastomer Polymers 0.000 claims description 5
- 238000000513 principal component analysis Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 239000003814 drug Substances 0.000 claims description 4
- 239000007787 solid Substances 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 22
- 239000000463 material Substances 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000001228 spectrum Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000010079 rubber tapping Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 2
- 238000012847 principal component analysis method Methods 0.000 description 2
- 229920000459 Nitrile rubber Polymers 0.000 description 1
- 239000000853 adhesive Substances 0.000 description 1
- 230000001070 adhesive effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004579 scanning voltage microscopy Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4454—Signal recognition, e.g. specific values or portions, signal events, signatures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4472—Mathematical theories or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2291/00—Indexing codes associated with group G01N29/00
- G01N2291/02—Indexing codes associated with the analysed material
- G01N2291/023—Solids
- G01N2291/0231—Composite or layered materials
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Pathology (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Mathematical Optimization (AREA)
- Probability & Statistics with Applications (AREA)
- Acoustics & Sound (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention discloses a method for detecting an interface debonding defect of a multi-layer bonding structure, which converts a plurality of linearly related energy characteristics of an acoustic signal time domain of the multi-layer bonding structure into a few linearly independent main components through main component analysis, clusters the acoustic signals through a K-means clustering algorithm, establishes a detection model of the interface debonding defect of the multi-layer bonding structure, realizes accurate judgment of the debonding defect, and solves the problems of low accuracy and difficult detection of judging defects such as knocking detection and ultrasonic detection.
Description
Technical Field
The invention belongs to the technical field of structural health monitoring and machine learning, and particularly relates to a method for detecting interface debonding defects of a multi-layer bonding structure.
Background
The multi-layer bonding structure has good mechanical property and thermodynamic property, and is widely applied to various fields such as aerospace, special equipment, industrial general technical equipment, military weapon equipment and the like. However, in the bonding process, the bonding interface often has debonding defects due to contamination of the material surface, insufficient curing of the adhesive, large differences in properties of the two materials, or other artifacts. The interfacial debonding defect of the multi-layer bonded structure may cause the entire structure to fail, resulting in major accidents and immeasurable economic loss. Therefore, the method has positive significance for accurately and efficiently detecting the interface bonding quality of the multi-layer bonding structure.
The determination of the debonding defect is usually achieved by analyzing the spectrum difference of the structural acoustic resonance signal, and detection methods including ultrasonic detection, manual knocking detection and the like are available in the prior art. However, when the acoustic impedance difference of the materials of the multi-layer bonding structure is large, ultrasonic waves enter the multi-layer bonding structure and then are scattered and reflected at interfaces formed by different media, and the acoustic energy is scattered and has large scattering attenuation, so that the ultrasonic waves are adopted to detect the bonding quality of the interfaces of the multi-layer bonding structure at the moment, so that the difficulty is high; however, when the debonding defect of the multi-layer bonding structure of the composite material is detected by adopting a manual knocking method, the debonding defect is identified by spectrum analysis and high error rate due to the influence of factors such as uneven human knocking force and environmental noise, and great challenges are brought to the identification of the debonding defect.
In fact, the debonding defect judgment is carried out by adopting the multi-signal characteristic to replace the single-signal characteristic, so that the information contained in the signal can be more comprehensively utilized, the judgment precision and reliability are improved, but the use of the multi-signal characteristic can lead to a large increase in data quantity, and the efficiency of subsequent characteristic analysis and identification is reduced; meanwhile, due to the correlation among the features, the discrimination precision cannot be infinitely improved by too many features, and when the information is rich enough, the efficiency ratio can be reduced by adding too much information, so that the dimension reduction processing of the multi-signal features is considered. In addition, in the research of analyzing and identifying the features, the currently commonly used machine learning methods mainly include BP neural network, SVM, K-means clustering method and the like. The K-means clustering algorithm is a common unsupervised learning algorithm. The algorithm does not need a large number of samples to train model parameters, can find internal connection of data, and has good effect on fault diagnosis.
Disclosure of Invention
In view of this, the present invention provides a method for detecting the interfacial debonding defect of a multi-layer bonding structure, which can realize high discrimination accuracy and high calculation efficiency for the interfacial debonding defect of the multi-layer bonding structure while more comprehensively utilizing the signal information of the multi-layer bonding structure.
The invention provides a method for detecting interface debonding defects of a multi-layer bonding structure, which comprises the following steps:
step S1: collecting an acoustic signal of a multilayer bonding structure to be detected;
step S2: extracting the same p energy characteristics from each acoustic signal, and constructing an energy characteristic matrix;
step S3: and (3) firstly analyzing the energy feature matrix by using a principal component through a detection model, reducing the dimension, and then carrying out K-means clustering with the K being 2 on acoustic signals corresponding to the energy feature matrix after the dimension reduction, so as to detect the defect area and the defect-free area of the multi-layer bonding structure.
Further, the detection model is obtained by:
step 1: establishing an artificial defect calibration sample of the multilayer bonding structure to be detected, and collecting n acoustic signals of a defect area and a defect-free area of the artificial defect calibration sample;
step 2: firstly discarding a total of sound signals at the initial acquisition stage and the final acquisition stage, and removing sound signals containing coarse errors in the residual signals; then the rest n 1 The personal sound signals are used as training sets;
step 3: extracting the same p energy features from each acoustic signal of the training set, and constructing an energy feature matrix of the training set;
step 4: performing dimension reduction treatment on the training set energy feature matrix by using principal component analysis, and then primarily selecting m principal components;
step 5: training set n using m principal components 1 K=2K means clustering is carried out on the individual sound signals;
step 6: judging whether the accuracy of the training set acoustic signal clustering result meets the requirement or not, if so, obtaining a detection model, and outputting a clustering center T; otherwise, returning to the step 4 to adjust the number of the main components.
Further, the method further comprises the operation of verifying the accuracy of the detection model, wherein the verification steps are as follows:
a: collecting the total n of the defect area and the defect-free area of the artificial defect calibration sample 2 A personal acoustic signal;
b: b acoustic signals at the initial stage and the final stage of acquisition are abandoned, the acoustic signals containing coarse errors in the residual signals are removed, and then the residual n 3 The individual acoustic signals are used as verification sets;
c: extracting the same p energy characteristics from each acoustic signal of the verification set, and constructing an energy characteristic matrix of the verification set;
d: firstly, reducing the dimension of the verification set energy feature matrix through the detection model;
e: taking the clustering center T output in the step 6 as an initial clustering center of the verification sound collecting signal, and carrying out K-means clustering of K=2 on the verification sound collecting signal so as to verify the accuracy of the training model;
f: judging whether the accuracy of the gathering result meets the requirement or not, and if so, adopting a training model; otherwise, returning to the step 4 to adjust the number of the main components.
Further, the step 4 includes:
step 401, performing a Z normalization process on the data of the training set energy feature matrix according to formula (1):
wherein,
i＝1,2,…,n 1 ；
j＝1,2,…,p；
for the average value of the j-th feature before normalization, var (x j ) Is the standard deviation, x of the j-th feature before normalization ij For the value of the j-th characteristic of the i-th signal before normalization, +.>The value of the j-th feature of the i-th signal after normalization;
step 402, calculating a correlation coefficient matrix of the normalized training set energy feature matrix, and solving a plurality of feature values lambda of the correlation coefficient matrix j A plurality of corresponding feature vectors Z j ＝(Z j1 ,Z j2 ,…,Z jp ) Where j=1, 2, …, p;
step 403, calculating the contribution rate q of the jth principal component according to formula (2) j ：
Setting an initial cumulative contribution rate to 90%, and determining the number m of required main components according to a cumulative contribution rate calculation formula (3);
calculating m principal components from formula (4):
F r ＝a r1 x 1 * +a r2 x 2 * +…+a rp x p * (4)
wherein r=1, 2, …, m; a, a r1 、a r2 、…、a rp The value in the feature vector corresponding to the principal component r;
x 1 * 、x 2 * 、…、x p * the p energy characteristic parameters after the normalization treatment.
Further, the adjusting the number of principal components means adjusting the number of principal components by adjusting the initial cumulative contribution rate.
Further, the p energy features include root mean square value, standard deviation, absolute mean, square root amplitude, peak-to-peak value, waveform factor, peak factor, pulse factor, margin factor, and kurtosis.
Further, before extracting the same p energy features from each acoustic signal, discarding unstable acoustic signals in the initial and final stages of acquisition, and removing acoustic signals containing coarse errors in the residual signals.
Further, the to-be-detected multi-layer bonding structure is a solid cylinder and comprises an outer layer steel structure, an intermediate layer rubber structure and an inner core medicament structure.
The beneficial effects are that:
(1) According to the invention, firstly, a plurality of linearly related energy characteristics of the acquired acoustic signal time domain of the multilayer bonding structure are converted into a few main components which can reflect most of information of original data and are not linear, so that dimension reduction of the acoustic signal data is realized, and then, the dimension-reduced acoustic signal data is clustered by using a K-means clustering algorithm (K=2), so that high discrimination precision and high calculation efficiency of debonding defects of the interface of the multilayer bonding structure are ensured;
meanwhile, after the training set acoustic signals are clustered through the K-means clustering algorithm, an initial clustering center containing actual working conditions can be obtained, and the clustering accuracy and the clustering speed are improved;
in addition, the invention detects the debonding defect of the interface of the multi-layer bonding structure by extracting the energy characteristic of the acoustic signal of the multi-layer bonding structure and taking the acoustic signal energy characteristic parameter as the basis, thereby effectively avoiding the problem of low defect detection accuracy caused by the frequency spectrum of the acoustic signal in the knocking detection; and the problem that ultrasonic detection is difficult when the acoustic impedance difference of materials of the multi-layer bonding structure is large is effectively avoided.
(2) According to the invention, before the interface debonding defect of the multi-layer bonding structure is detected, the acoustic signals at the initial stage and the final stage of acquisition are abandoned, the acoustic signals containing coarse errors in the residual signals are removed, and the influence of factors such as man-made knocking habit, knocking force non-uniformity and the like on the detection precision is reduced.
(3) According to the invention, the accuracy of the detection model is checked through the test set, so that the accuracy of the follow-up detection of the debonding defect of the multi-layer bonding structure in the actual working condition by using the detection model is ensured.
(4) According to the invention, the artificial defect calibration sample is firstly established for the multi-layer bonding structure in the spacecraft, and the detection model capable of detecting the interface debonding defect of the multi-layer bonding structure is obtained by training the artificial defect calibration sample, so that the interface debonding defect of the multi-layer bonding structure in the actual spacecraft can be rapidly and accurately detected.
Drawings
FIG. 1 is a flow chart of a method for detecting interfacial debonding defects in a multi-layer bonded structure according to the present invention;
FIG. 2 is a schematic diagram of a multi-layer bonding structure artificial defect calibration sample (a) and a two-dimensional model (b) thereof according to the method for detecting interfacial debonding defects of multi-layer bonding structures of the present invention;
FIG. 3 is a graph showing the frequency spectrum of the defect-free signal (c) and defect-free signal (d) and the debonding signal (e) and debonding signal (f) of the multi-layered bond structure artificial defect calibration sample (a) of FIG. 2;
FIG. 4 is a time domain waveform diagram of defect free signal (g) and defect free signal (h) and debonding signal (m) and debonding signal (n) of the multi-layer bonded structure artificial defect calibration sample (a) of FIG. 2;
FIG. 5 is a training set signal cluster scatter plot according to an embodiment of the present invention;
FIG. 6 is a scatter plot of test set signal clusters according to an embodiment of the present invention;
1-medicament structure, 2-rubber structure, 3-steel structure, 4-two interfaces, 5-one interface and 6-zero interface.
Detailed Description
The invention will now be described in detail by way of example with reference to the accompanying drawings.
The invention provides a method for detecting interface debonding defects of a multi-layer bonding structure, which comprises the following steps: collecting an acoustic signal of a multilayer bonding structure to be detected; extracting the same p energy characteristics from each acoustic signal, and constructing an energy characteristic matrix; and (3) firstly analyzing the energy feature matrix by using a principal component through a detection model, reducing the dimension, and then carrying out K-means clustering with the K being 2 on acoustic signals corresponding to the energy feature matrix after the dimension reduction, so as to detect the defect area and the defect-free area of the multi-layer bonding structure.
Therefore, the method reduces the dimension of the energy characteristics of the acoustic signals of the multi-layer bonding structure through the principal component analysis method, so that a plurality of characteristics are converted into a few principal components which can reflect most of information of the original data and are irrelevant in linearity, signal classification based on clustering can be realized, and high discrimination precision and high calculation efficiency of the interface debonding defects of the multi-layer bonding structure are ensured. Meanwhile, the invention is characterized by the energy characteristic of the acoustic signal of the multi-layer bonding structure, and the debonding defect is carried out on the multi-layer bonding structure based on the energy characteristic parameter of the acoustic signal, so that the problem that the defect detection accuracy is low by taking the frequency spectrum of the acoustic signal as the basis for knocking detection is effectively avoided, and the problem that the ultrasonic detection has larger difficulty when the acoustic impedance difference of the materials of the multi-layer bonding structure is larger.
Fig. 1 is a flowchart of a method for detecting an interfacial debonding defect of a multi-layer bonding structure, as shown in fig. 1, wherein the method for detecting an interfacial debonding defect of a multi-layer bonding structure to be detected comprises the following steps:
step 1: and establishing an artificial defect calibration sample of the multilayer bonding structure to be detected, and collecting n acoustic signals of a defect area and a defect-free area of the artificial defect calibration sample.
In this embodiment, the multi-layer bonding structure to be detected is a structure on a certain spacecraft, so the above-mentioned artificial defect calibration sample is shown in fig. 2, and the model is a solid cylinder formed by three layers of different materials, including an outer layer steel structure 3, a middle layer rubber structure 2 and an inner core medicament structure 1. Wherein the outer layer steel structure 3 can be made of high-strength steel 30CrMnSiA with the outer diameter of 400mm and the wall thickness of 3 mm; the middle layer rubber structure 2 is made of heat insulating layer material, and in the embodiment, nitrile rubber with the thickness of 1-10 mm is selected. The contact surface of the outer layer steel structure 3 and a structure such as air outside the multi-layer bonding structure model is a zero interface 6, the contact surface of the outer layer and the middle layer is an interface 5, the contact surface of the middle layer and the inner core is a two interface 4, the acoustic impedance difference of materials at two sides of the two interfaces 4 is large, and the method can detect the debonding defect of the two interfaces 4.
Step 2: firstly discarding a plurality of acoustic signals at the initial acquisition stage and the final acquisition stage, and removing acoustic signals containing coarse errors in the residual signals; the residual acoustic signal is then split into a training set and a test set.
In this embodiment, a force hammer is used to excite the debonded area and the nondefective area of the multi-layer bonded structure, and a high-precision microphone is fixedly placed near the excitation area to collect resonance acoustic signals, so as to obtain 84 acoustic signals in total.
In this embodiment, the test set is extracted simultaneously with the extraction of the training set. In practice, it is also possible to separate the process to extract a acoustic signals, remove a acoustic signals and coarse errors by means of step 2 to obtain n 1 A training set of individual acoustic signals; in addition extract n 2 The b sound signals are removed and coarse errors are removed in the mode of step 2, and n is obtained 3 Test set of individual acoustic signals.
In the above example, considering the instability of the tapping quality at the initial stage and the final stage of tapping, 10 acoustic signals at the initial stage and the final stage of tapping are removed, the signal characteristics of the remaining 74 acoustic signals are subjected to normal distribution test, 95% confidence intervals are obtained, and finally 68 acoustic signals are reserved as detection signal data spaces. 40 of the acoustic signals (including 23 acoustic signals with debonding defects and 17 acoustic signals with normal bonding) were randomly taken as training sets, and 28 as test sets (including 12 acoustic signals with debonding defects and 16 acoustic signals with normal bonding).
Step 3: and extracting the same p energy features from each acoustic signal of the training set, and constructing an energy feature matrix of the training set.
The method specifically comprises the following steps:
step 301: p energy features are extracted from each acoustic signal.
As shown in fig. 3, the signal with the debonding defect and the signal without defects are difficult to distinguish in frequency due to errors in the magnitude, angle and the like of the knocking force of the manual knocking, but the signal with the debonding defect and the signal without defects have obvious differences in amplitude as shown in fig. 4, so that signal identification can be performed by extracting the energy characteristics of the acoustic signals. In this embodiment, 10 time domain feature parameters related to signal energy are selected, including: root mean square value, standard deviation, absolute mean, square root amplitude, peak-to-peak value, waveform factor, peak factor, pulse factor, margin factor, and kurtosis.
The signal characteristic parameters of part of the training set are shown in table 1:
TABLE 1 partial training set Signal characteristic parameters
Step 302: and carrying out standardization processing on the extracted characteristic data.
In the step, Z standardization processing is adopted to the 10 energy characteristic data according to a formula (1), the standardized data mean value is 0, and the variance is 1;
wherein,
i＝1,2,…,40；
j＝1,2,…,10；
for the average value of the j-th feature before normalization, var (x j ) Is the standard deviation, x of the j-th feature before normalization ij For the value of the j-th characteristic of the i-th signal before normalization, +.>The value of the j-th feature of the i-th signal after normalization; after the data in the training set is normalized, part of the results are shown in table 2:
table 2 normalized training set Signal characteristic parameters
Step 4: and (3) performing dimension reduction treatment on the training set energy feature matrix by using principal component analysis, and then primarily selecting m principal components.
Specifically, the method comprises the following substeps:
firstly, for the normalized training set acoustic signal energy characteristics, calculating correlation coefficients between every two characteristics, and constructing a correlation coefficient matrix, as shown in table 3:
TABLE 3 correlation matrix for signal characteristic parameters of experimental training set
Solving the correlation coefficient matrix by Jacobian methodIs a plurality of eigenvalues lambda of (1) j A plurality of corresponding feature vectors and a plurality of Z j ＝(Z j1 ,Z j2 ,…,Z j10 ) Where j=1, 2, …,10;
by means of characteristic values lambda j Corresponding feature vector multiple Z j The principal component analysis method is adopted to carry out 10 energy characteristic parameters after the normalization of the training set acoustic signalsProcessing to obtain 10 principal components, and calculating the contribution rate q of principal component j according to formula (2) j ；
Where j=1, 2, …,10; setting the initial cumulative contribution rate to 90%, and determining 2 principal components according to a cumulative contribution rate calculation formula (3):
wherein m is the number of main components; and the contribution rates of the main component 1 and the main component 2 are 86.799% and 9.019%, respectively, and the cumulative contribution rate reaches 95.818%, as shown in table 4:
TABLE 4 principal component contribution Rate and cumulative contribution Rate
Calculating each principal component according to the feature vector corresponding to the principal component feature value by the formula (4):
F r ＝a r1 x 1 * +a r2 x 2 * +…+a rp x p * (4)
wherein r=1, 2, …, m; a, a r1 、a r2 、…、a rp The value in the feature vector corresponding to the principal component r; x is x 1 * 、x 2 * 、…、x p * Is the energy characteristic parameter after the standardized treatment.
Principal component F 1 、F 2 ：
F 1 ＝0.0760×Z 1 +0.0814×Z 2 +0.0312×Z 3 -0.0098×Z 4 -0.0098×Z 5 -0.0224×Z 6 -0.0254×Z 7 +0.1140×Z 8 +0.0546×Z 9 +0.0781×Z 10
F 2 ＝0.2358×Z 1 +0.2527×Z 2 +0.0969×Z 3 -0.0305×Z 4 -0.0305×Z 5 -0.065×Z 6 -0.0790×Z 7 +0.3538×Z 8 +0.1695×Z 9 +0.2422×Z 10
Step 5: training set n using m principal components 1 The individual acoustic signals are k=2K-means clustered.
The method specifically comprises the following steps:
step 501: setting the iteration times of a K-means clustering algorithm and an initial clustering center;
step 502: and (3) carrying out K (K=2) mean clustering according to the training set signals, namely dividing the training set signals into classes corresponding to the cluster centers closest to the training set signals according to the Euclidean distance between the determined 2 principal components and the initial cluster center and the nearest criterion.
Step 503: and taking the average value corresponding to all the objects in each category as a clustering center of the category, and updating the clustering center.
Step 504: judging whether the clustering center is changed or not and whether the iteration times are reached, and if the clustering center is unchanged or the iteration times are reached, outputting a clustering result; otherwise, continuing the iteration.
Step 6: judging whether the accuracy of the training set acoustic signal clustering result meets the requirement or not, if so, obtaining a detection model, and outputting a clustering center T; otherwise, returning to the step 4 to adjust the number of the main components.
In this step, when it is determined that the accuracy rate does not meet the requirement, the number of principal components may be adjusted by adjusting the initial cumulative contribution rate in step 4.
The final clustering results obtained are shown in table 5:
table 5 training aggregate class results
Table 5 illustrates that in the case where the initial cumulative contribution rate is 90% and 2 principal components are determined, the clustering accuracy is low, and at this time, the number of principal components can be continuously adjusted by adjusting the initial cumulative contribution rate; the number of principal components can be adjusted, and the related parameters of the clustering algorithm, such as the iteration times, can be adjusted.
In this example, the initial cumulative contribution rate was adjusted to 97%, and the contribution rates of 3 principal components and principal component 1, principal component 2, and principal component 3 were 86.799%, 9.019%, and 1.977%, respectively, in this order, and the cumulative contribution rate reached 97.795%, as shown in table 6:
TABLE 6 principal component contribution and cumulative contribution
At this time, the main component G is obtained according to the above formula (4) 1 、G 2 And G 3 ：
G 1 ＝-0.3370×Z 1 -0.3357×Z 2 -0.3323×Z 3 -0.3323×Z 4 +0.3265×Z 5 +0.3238×Z 6 +0.3194×Z 7 +0.3191×Z 8 +0.3123×Z 9 -0.2006×Z 10
G 2 ＝0.0747×Z 1 +0.0558×Z 2 +0.1495×Z 3 +0.495×Z 4 +0.2632×Z 5 -0.0126×Z 6 +0.3011×Z 7 +0.1338×Z 8 +0.2832×Z 9 +0.8297×Z 10
G 3 ＝-0.0472×Z 1 -0.1685×Z 2 +0.2068×Z 3 +0.2068×Z 4 -0.0944×Z 5 +0.0202×Z 6 +0.1551×Z 7 +0.7057×Z 8 -0.5933×Z 9 +0.0022×Z 10
The training set signals were K (k=2) averaged using these 3 principal components to obtain clustering results as shown in table 7:
TABLE 7 training aggregate class results
Table 7 illustrates that in the case of adjusting the cumulative contribution rate to 97% and determining 3 principal components, the clustering accuracy is already high, at this time, the training set signal cluster scatter diagram is shown in fig. 5, and the output cluster center T is shown in table 8:
TABLE 8 clustering center T
And step 6, after the detection model is obtained through the steps, verifying the accuracy of the detection model, wherein the verification steps are as follows:
a total of n has been obtained by the above steps 3 The individual acoustic signals are used as a test set; in the step, the same p energy characteristics are extracted from each acoustic signal of a test set, and an energy characteristic matrix of the test set is constructed; firstly, reducing the dimension of the energy characteristic matrix of the test set through the detection model, and then, taking the clustering center T as an initial clustering center of the test sound collecting signal, and carrying out K-means clustering of K=2 on the test sound collecting signal to verify the accuracy of the training model; judging whether the accuracy of the test aggregation result meets the requirement, and if so, adopting a training model; otherwise, returning to the step 4 to adjust the number of the main components.
Specifically, the signal energy characteristics of the test set are extracted, and the Z normalization treatment is carried out, and the data after partial normalization are shown in the table 9:
table 9 standardized test set signal characteristic parameters
Substituting the data in Table 9 into the training set to obtain 3 energy characteristic main components of the test set; taking the clustering center T as an initial clustering center of the test set, and carrying out K (K=2) mean clustering on the main components of the 3 energy characteristics of the test set, wherein the clustering result is shown in table 10:
table 10 test set signal clustering results
In table 10, the clustering type 1 is a defect-free type, the clustering type 2 is a debonding type, which shows that the error between the clustering result and the actual result is smaller, the defect-free signal detection accuracy reaches 93.8%, the debonding signal detection accuracy reaches 100%, the total signal detection accuracy of the test set reaches 96.43%, and at this time, the signal clustering scatter diagram of the test set is shown in fig. 6;
through the process, a verified detection model is established, the model has no report missing for the signals with interface debonding defects, and has less false report for the non-defective signals, and the overall detection accuracy is higher, so that the detection model can be used for detecting the interface defects of the multi-layer bonding structure in engineering practice, and the steps are as follows:
step S1: collecting an acoustic signal of a multilayer bonding structure to be detected;
step S2: extracting the same p energy characteristics from each acoustic signal, and constructing an energy characteristic matrix;
step S3: and (3) firstly analyzing the energy feature matrix by using a principal component through a detection model, reducing the dimension, and then carrying out K-means clustering with the K being 2 on acoustic signals corresponding to the energy feature matrix after the dimension reduction, so as to detect the defect area and the defect-free area of the multi-layer bonding structure.
According to the interface debonding defect detection method based on principal component analysis and K-means clustering analysis, through repeated iterative training on the multi-layer bonding structure acoustic signal training set, a multi-layer bonding structure interface debonding defect detection model is built, and high recognition rate of multi-layer bonding structure interface debonding defect detection is achieved.
In summary, the above embodiments are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. The method for detecting the interface debonding defect of the multi-layer bonding structure is characterized by comprising the following steps of:
step S1: collecting an acoustic signal of a multilayer bonding structure to be detected;
step S2: extracting the same p energy characteristics from each acoustic signal, and constructing an energy characteristic matrix;
step S3: the energy feature matrix is firstly subjected to dimension reduction by using a principal component analysis through a detection model, and K-means clustering with K being 2 is carried out on acoustic signals corresponding to the energy feature matrix after dimension reduction, so that a defect area and a defect-free area of the multi-layer bonding structure are detected;
the detection model is obtained by the following steps:
step 1: establishing an artificial defect calibration sample of a multi-layer bonding structure, and collecting n acoustic signals of a defect area and a defect-free area of the artificial defect calibration sample;
step 2: firstly discarding a total of sound signals at the initial acquisition stage and the final acquisition stage, and removing sound signals containing coarse errors in the residual signals; then the rest n 1 The personal sound signals are used as training sets;
step 3: extracting the same p energy features from each acoustic signal of the training set, and constructing an energy feature matrix of the training set;
step 4: performing dimension reduction treatment on the training set energy feature matrix by using principal component analysis, and then primarily selecting m principal components;
step 5: training set n using m principal components 1 K=2K means clustering is carried out on the individual sound signals;
step 6: judging whether the accuracy of the training set acoustic signal clustering result meets the requirement or not, if so, obtaining a detection model, and outputting a clustering center T; otherwise, returning to the step 4 to adjust the number of the main components;
the step 4 comprises the following steps:
step 401, performing a Z normalization process on the data of the training set energy feature matrix according to formula (1):
wherein,
i＝1,2,…,n 1 ；
j＝1,2,…,p；
for the average value of the j-th feature before normalization, var (x j ) Is the standard deviation, x of the j-th feature before normalization ij For the value of the j-th characteristic of the i-th signal before normalization, +.>The value of the j-th feature of the i-th signal after normalization;
step 402, calculating a correlation coefficient matrix of the normalized training set energy feature matrix, and solving a plurality of feature values lambda of the correlation coefficient matrix j A plurality of corresponding feature vectors Z j ＝(Z j1 ,Z j2 ,…,Z jp ) Where j=1, 2, …, p;
step 403, calculating the contribution rate q of the jth principal component according to formula (2) j ：
Setting an initial cumulative contribution rate to 90%, and determining the number m of required main components according to a cumulative contribution rate calculation formula (3);
calculating m principal components from formula (4):
F r ＝a r1 x 1 * +a r2 x 2 * +…+a rp x p * (4)
wherein r=1, 2, …, m; a, a r1 、a r2 、…、a rp The value in the feature vector corresponding to the principal component r;
x 1 * 、x 2 * 、…、x p * the p energy characteristic parameters after the normalization treatment.
2. The method of claim 1, further comprising the operation of verifying the accuracy of the detection model, the verifying step being as follows:
a: collecting the total n of the defect area and the defect-free area of the artificial defect calibration sample 2 A personal acoustic signal;
b: b acoustic signals at the initial stage and the final stage of acquisition are abandoned, the acoustic signals containing coarse errors in the residual signals are removed, and then the residual n 3 The individual acoustic signals are used as verification sets;
c: extracting the same p energy characteristics from each acoustic signal of the verification set, and constructing an energy characteristic matrix of the verification set;
d: firstly, reducing the dimension of the verification set energy feature matrix through the detection model;
e: taking the clustering center T output in the step 6 as an initial clustering center of the verification sound collecting signal, and carrying out K-means clustering of K=2 on the verification sound collecting signal so as to verify the accuracy of the training model;
f: judging whether the accuracy of the gathering result meets the requirement or not, and if so, adopting a training model; otherwise, returning to the step 4 to adjust the number of the main components.
3. The method of claim 2, wherein the adjusting the number of principal components is adjusting the number of principal components by adjusting an initial cumulative contribution rate.
4. A method according to any one of claims 1-3, wherein the p energy features comprise root mean square values, standard deviations, absolute means, square root magnitudes, peak-to-peak values, waveform factors, peak factors, pulse factors, margin factors, and kurtosis.
5. The method of claim 1, wherein the step of discarding unstable acoustic signals at the beginning and end of acquisition and removing acoustic signals containing coarse errors from the remaining signals is performed before extracting the same p energy features for each acoustic signal.
6. A method according to any one of claims 1-3, wherein the multi-layer bonded structure to be tested is a solid cylinder comprising an outer layer steel structure, an intermediate layer rubber structure and an inner core medicament structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111543787.2A CN114235957B (en) | 2021-12-16 | 2021-12-16 | Method for detecting interface debonding defect of multi-layer bonding structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111543787.2A CN114235957B (en) | 2021-12-16 | 2021-12-16 | Method for detecting interface debonding defect of multi-layer bonding structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114235957A CN114235957A (en) | 2022-03-25 |
| CN114235957B true CN114235957B (en) | 2024-02-06 |
Family
ID=80757577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111543787.2A Active CN114235957B (en) | 2021-12-16 | 2021-12-16 | Method for detecting interface debonding defect of multi-layer bonding structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114235957B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH095305A (en) * | 1995-06-16 | 1997-01-10 | Daido Steel Co Ltd | Method and device for inspecting defect |
| CN106153725A (en) * | 2016-08-17 | 2016-11-23 | 西安长大公路养护技术有限公司 | A kind of pavement distress detection equipment and detection method thereof |
| RU2627539C1 (en) * | 2016-11-14 | 2017-08-08 | Акционерное общество "Обнинское научно-производственное предприятие "Технология" им. А.Г. Ромашина" | Method for non-destructive testing of adhesive joint of monolithic sheets made of polymeric composite materials |
| CN108872380A (en) * | 2018-06-04 | 2018-11-23 | 航天特种材料及工艺技术研究所 | The bonding defect detecting method of multi-layer bonded component |
-
2021
- 2021-12-16 CN CN202111543787.2A patent/CN114235957B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH095305A (en) * | 1995-06-16 | 1997-01-10 | Daido Steel Co Ltd | Method and device for inspecting defect |
| CN106153725A (en) * | 2016-08-17 | 2016-11-23 | 西安长大公路养护技术有限公司 | A kind of pavement distress detection equipment and detection method thereof |
| RU2627539C1 (en) * | 2016-11-14 | 2017-08-08 | Акционерное общество "Обнинское научно-производственное предприятие "Технология" им. А.Г. Ромашина" | Method for non-destructive testing of adhesive joint of monolithic sheets made of polymeric composite materials |
| CN108872380A (en) * | 2018-06-04 | 2018-11-23 | 航天特种材料及工艺技术研究所 | The bonding defect detecting method of multi-layer bonded component |
Non-Patent Citations (3)
| Title |
|---|
| UHMWPE/LDPE层合板损伤声发射信号聚类分析研究;王旭 等;武汉纺织大学学报;31(03);第21-25页 * |
| 基于改进LMS算法的复合材料超声检测缺陷识别;王峰林 等;军械工程学院学报;25(05);第66-69页 * |
| 基于改进半监督FCM聚类的复合材料超声检测脱粘缺陷识别;王峰林等;军械工程学院学报;25(06);第49-53页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114235957A (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Godin et al. | Clustering of acoustic emission signals collected during tensile tests on unidirectional glass/polyester composite using supervised and unsupervised classifiers | |
| Yousefi et al. | Damage categorization of glass/epoxy composite material under mode II delamination using acoustic emission data: a clustering approach to elucidate wavelet transformation analysis | |
| Ramasso et al. | Unsupervised consensus clustering of acoustic emission time-series for robust damage sequence estimation in composites | |
| CN107328868B (en) | A kind of Analysis of Acoustic Emission Signal method of quick identification ceramic coating failure type | |
| Heidary et al. | Clustering of acoustic emission signals collected during drilling process of composite materials using unsupervised classifiers | |
| Arumugam et al. | Identification of failure modes in composites from clustered acoustic emission data using pattern recognition and wavelet transformation | |
| CN108318579B (en) | Honeycomb co-curing structure defect ultrasonic A scanning identification method and device | |
| Al-Jumaili et al. | Characterisation of fatigue damage in composites using an Acoustic Emission Parameter Correction Technique | |
| Harizi et al. | Damage mechanisms assessment of Glass Fiber-Reinforced Polymer (GFRP) composites using multivariable analysis methods applied to acoustic emission data | |
| Boussetta et al. | Study of the filament wound glass/polyester composite damage behavior by acoustic emission data unsupervised learning | |
| CN110376289A (en) | A kind of composite fiber braiding layer damnification recognition method based on sound emission means | |
| Karimian et al. | Acoustic emission signal clustering in CFRP laminates using a new feature set based on waveform analysis and information entropy analysis | |
| CN114692302B (en) | Fatigue crack detection method and system based on Gaussian mixture model | |
| CN114235957B (en) | Method for detecting interface debonding defect of multi-layer bonding structure | |
| Chen et al. | Ultrasonic signal classification and porosity testing for CFRP materials via artificial neural network | |
| CN114324582A (en) | HHT (Hilbert-Huang transform) and neural network-based composite plate layering defect air coupling ultrasonic detection method | |
| EP2502063B1 (en) | Cluster analysis system and method to improve sorting performance | |
| CN102621221A (en) | Defect classification method based on phased ultrasonic wave | |
| CN107886071A (en) | A kind of processing method of fibre reinforced composites damage acoustic emission signal | |
| CN115308063A (en) | Concrete strength sampling detection method for existing building | |
| Gift | A critical review of the role of acoustic emission algorithms in the delamination analysis of adhesive joints | |
| Fotouhi et al. | Acoustic emission based method to characterize glass/epoxy composite damages during 3-point bending test | |
| Rahul et al. | Investigating the impact from ultrasonic testing uncertainty in quantifying ply-stack orientation on the probabilistic failure envelope | |
| Kraljevski et al. | Autoencoder-based Ultrasonic NDT of Adhesive Bonds | |
| Yuan et al. | Recognition types of cracked material under uniaxial tension based on improved mel frequency cepstral coefficients (mfcc) |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |