CN-122017048-A - Composite board interface defect classification and identification method and system based on echo feature fusion

CN122017048ACN 122017048 ACN122017048 ACN 122017048ACN-122017048-A

Abstract

The invention belongs to the technical field of material ultrasonic detection, and relates to a composite board interface defect classification and identification method and system based on echo feature fusion, wherein the method comprises the steps of extracting ultrasonic echo multidimensional acoustic features; the method comprises the steps of calculating the aliasing degree of local acoustic features, generating dynamic acoustic cost weight by combining acoustic impedance values, the aliasing degree of the local acoustic features and the sample scarcity degree, reconstructing a random forest model, utilizing information gain indexes and cost weight to construct cost sensitive voting factors, inputting multidimensional acoustic features of the composite board to be tested to perform weighted summation to obtain classification results, evaluating the feature aliasing degree more accurately, comprehensively considering the sample scarcity and acoustic features, ensuring the importance of high-risk defect features in decision making, effectively solving the problem of misjudgment, and improving the accuracy and reliability of classification and identification of the interface defects of the composite board.

Inventors

ZHANG HANGYONG
GUO LONGCHUANG
LIU YAN
WANG YA
WANG ZHE
WANG TIANXIANG
SHI JIN
Che Longquan

Assignees

宝钛金属复合材料有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. The composite board interface defect classification and identification method based on echo feature fusion is characterized by comprising the following steps of: acquiring ultrasonic echo signals of all scanning points of a composite board interface, extracting multidimensional acoustic features of the ultrasonic echo signals to form a training sample set, and adding category labels; Extracting the neighbor relation of each sample in the training sample set in the feature space, and calculating the aliasing degree of the local acoustic features by combining the dispersion variance of the ultrasonic echo signals; obtaining a standard acoustic impedance value of a composite board matrix and an actual inversion acoustic impedance value of a composite board interface, and generating a dynamic acoustic cost weight by combining the local acoustic characteristic aliasing degree and the scarcity degree of the class of the corresponding sample in global data; Tracing node splitting records of each decision tree in the random forest model, extracting information gain indexes obtained by each decision tree when dividing samples, and constructing cost sensitive voting factors by combining the information gain indexes and dynamic acoustic cost weights to finish reconstruction of the random forest model; And inputting each multi-dimensional acoustic feature of the composite board interface to be tested into the reconstructed random forest model, and carrying out weighted summation on each multi-dimensional acoustic feature according to the cost sensitive voting factors to obtain a final classification result of the scanning point corresponding to each multi-dimensional acoustic feature so as to realize defect classification identification of the composite board interface.
2. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1, wherein the steps of extracting the multi-dimensional acoustic features of the ultrasonic echo signals to form a training sample set and adding class labels include: Performing short-time Fourier transform and wavelet packet decomposition on an ultrasonic echo signal, and extracting an initial acoustic feature set comprising an echo peak value, a center frequency offset, a high-frequency attenuation coefficient and a frequency band energy duty ratio; performing standardized processing on the initial acoustic feature set to eliminate dimension differences, so as to obtain multidimensional acoustic features; And adding defect class labels to each multi-dimensional acoustic feature according to the verification result of the destructive test, and mapping the multi-dimensional acoustic features with the added class labels into a feature space to form a training sample set.
3. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1, wherein the local acoustic feature aliasing degree satisfies: ; In the formula, Is the first The degree of aliasing of the local acoustic features of the individual samples, According to the first The number of neighbor samples chosen for the neighbor relation of the individual samples in the feature space, And (3) with Respectively the first Sample and the first The multi-dimensional acoustic features of the individual neighbor samples, And (3) with Respectively normalized first Sample and the first The neighboring samples correspond to the dispersion variance of the ultrasonic echo signal of the scan point, To prevent the super-parameters with denominator 0, As a function of the natural index of refraction, Is a euclidean distance symbol.
4. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1, wherein the dynamic acoustic cost weight satisfies: ; In the formula, Is the first The dynamic acoustic cost weights of the individual samples, Is the first The degree of aliasing of the local acoustic features of the individual samples, Is the standard acoustic impedance value of the composite board matrix, Is the first The actual inversion acoustic impedance values of the individual samples at the composite plate interface, To train the number of samples within the sample set, Is the first The number of categories to which each sample belongs, As a natural logarithmic function.
5. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1, wherein the cost sensitive voting factors satisfy the following conditions: ; In the formula, Is the first Cost sensitive voting factors of the decision tree in the forest prediction stage, Is the first A training sample set of decision trees, To train the first in the sample set The dynamic acoustic cost weights of the individual samples, Is the first The decision tree is being divided into the first The information gain index acquired by the node containing the sample in the splitting process, Is a hyperbolic tangent activation function.
6. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1 or 3, wherein the acquisition mode of the dispersion variance of the ultrasonic echo signal is as follows: performing short-time Fourier transform on ultrasonic echo signals of all scanning points of the composite board interface, and calculating initial standard deviations of frequency components in all time windows; and calculating the ratio of the initial standard deviation to the global dispersion variance standard deviation of the training sample set to obtain the dispersion variance of the dimensionless processed ultrasonic echo signal.
7. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1 or 4, wherein the actual inversion acoustic impedance value of the composite board interface is obtained by the following steps: Measuring the reflection amplitude and the incidence amplitude of the interface at the scanning point corresponding to each sample by using ultrasonic detection equipment, and calculating to obtain the reflectivity; and substituting the reflectivity into a reflection coefficient formula to carry out inversion calculation to obtain an actual inversion acoustic impedance value of the composite board interface.
8. The composite board interface defect classification and identification method based on echo feature fusion according to claim 1, wherein the obtaining the final classification result of each multi-dimensional acoustic feature corresponding to the scanning point comprises: Comparing and calculating a random forest to judge the weighted summation result of each defect category of the multi-dimensional acoustic feature, judging the defect category corresponding to the highest weighted summation result as the true defect category of the scanning point corresponding to the multi-dimensional acoustic feature, and obtaining the final classification result of the scanning point corresponding to the multi-dimensional acoustic feature; The true defect class includes at least a normal bond type, a micro inclusion type, and a high-risk defect type.
9. The composite board interface defect classification and identification method based on echo feature fusion according to claim 8, wherein the method for realizing defect classification and identification of the composite board interface comprises the following steps: based on the final classification results of all scanning points of the interface of the composite board to be tested, executing composite board level judgment according to a safety priority principle: in response to the existence of only the normal combination type, judging the composite board to be tested as a qualified product; in response to detecting at least one high-risk defect type, judging that the composite board to be tested is unqualified; judging the composite board to be tested as a qualified product when only the micro inclusion type exists and the proportion is not higher than a threshold value; and in response to the condition that only the tiny inclusion type exists and the proportion is higher than the threshold value, judging the composite board to be detected as a degradation product.
10. The composite board interface defect classification and identification system based on echo feature fusion is characterized by comprising a processor and a memory, wherein the memory stores computer program instructions, and the computer program instructions realize the composite board interface defect classification and identification method based on echo feature fusion according to any one of claims 1-9 when the computer program instructions are executed by the processor.

Description

Composite board interface defect classification and identification method and system based on echo feature fusion Technical Field The invention belongs to the technical field of ultrasonic detection of materials, and relates to a composite board interface defect classification and identification method and system based on echo feature fusion, which are used for detecting impurities and quality defects existing in an interface in the processing process of a metal composite board. Background In the production of specific metal composite plates such as titanium steel, zirconium steel and the like, complex wavy interfaces are generated at material joints by an explosion compounding process, and ultrasonic echo signals are generally adopted in industry for nondestructive inspection in order to detect whether defects such as non-lamination, oxide inclusion or microcracks exist at the composite plate interfaces. However, since the difference of physical characteristics of the interface defects is significant, the acoustic wave characteristics of a single dimension cannot fully characterize the multi-physical field coupling characteristics of the defects, so that the conventional method is difficult to accurately distinguish high-risk defects from common interface fluctuations. The technical scheme of the Chinese patent document with the publication number of CN114994177B is that an ultrasonic defect detection method is applied to a composite board material to be detected, and the method comprises the steps of collecting Lamb wave signals on the composite board material, selecting two mutually orthogonal linear directions on the composite board material to be detected as the step scanning directions, detecting the Lamb wave signals along the step scanning directions in a preset step length K in an ultrasonic detection mode, and collecting the Lamb wave signals on the composite board material. The defect index DI is obtained through the Lamb wave signal, each imaging point of the scanning area defines a virtual synthetic aperture, and the defect position is obtained through the virtual synthetic aperture and the defect index DI. However, the patent document only completes shallow processing of signals, deep association of physical characteristics and defect types in a feature space is not established, weak features of high-risk defects are submerged by normal interface fluctuation, accurate identification of the high-risk defects cannot be achieved, and the miss judgment rate is high. The Chinese patent document with the publication number of CN112633368B discloses a flat vibration motor defect detection system and a detection method based on an improved multi-granularity cascade forest, wherein the multi-granularity cascade forest is combined with cascade CatBoost, firstly, the connection among acquired electric signal data points is processed by using a multi-granularity cascade forest structure to obtain feature vectors, and then the feature vectors are classified by using cascade CatBoost to obtain the defect types of the flat vibration motor. However, the patent document is not optimized for the problem of composite board ultrasonic echo characteristic aliasing, and normal interface fluctuation and high-risk defect characteristics cannot be effectively distinguished when composite board defect detection is processed. The Chinese patent document with the publication number of CN116642946B detects through the vertical incidence of ultrasonic waves to a coating sample, acquires an aliased echo PR of a coupling medium/coating and a coating/substrate interface, performs fast Fourier transform on the PR to obtain an unfolding phase phi R, obtains a first-order partial derivative of frequency f by the phi R to construct an ultrasonic echo phase first-order derivative spectrum UEPDS, extracts extremum frequency in UEPDS, inverts the rigidity of the interface according to the theoretical relationship between the extremum frequency and interface rigidity Kn, and indirectly realizes quantitative measurement of the bonding strength of the coating/substrate interface. However, the patent document is only used for quantitative measurement of interface rigidity, is not applied to classification of composite plate interface defects, does not consider influence of sample scarcity on classification results, and cannot effectively identify high-risk defects in unbalanced industrial data. In the prior art, although partial schemes attempt to fuse by extracting multidimensional features such as time domain, frequency domain and energy, deep association of physical characteristics and defect types in a feature space cannot be established, so that weak features of high-risk defects are submerged by a large number of normal interface fluctuations, meanwhile, as the high-risk defect samples have extremely small proportion, in extremely unbalanced industrial data, the difference of discrimination contribution degrees of different featur