Carbon fiber reinforced composite material damage detection method based on Mamba model

Abstract

The invention discloses a carbon fiber reinforced composite material damage detection method based on a Mama model, and relates to the technical field of ultrasonic guided wave detection. The method comprises the following steps: S1, collecting a guided wave signal set on a damaged carbon fiber reinforced composite material; s2, extracting a time characteristic sequence of the guided wave signal set through a guided wave signal processing branch; s3, extracting spectral features of the guided wave signal set through continuous wavelet transform branches; and S4, fusing the time characteristics and the spectrum characteristics of the guided wave signal set, comprehensively obtaining multi-dimensional damage characteristics in the guided wave signals, and carrying out defect detection and positioning on the whole carbon fiber reinforced composite material. And realizing baseline-free guided wave damage detection and positioning of the carbon fiber reinforced composite material under a baseline-free guided wave signal through fusion calculation of time features of Mama, continuous wavelet transform and time-domain spectrum features extracted by a convolutional neural network.

Description

Technical Field

[0001] The invention relates to the technical field of ultrasonic guided wave detection, in particular to a carbon fiber reinforced composite material damage detection method based on a Mamba model. Background Art

[0002] Compared with metal materials, carbon fiber reinforced composite materials have significant advantages such as light weight, high specific strength, high specific stiffness, and corrosion resistance. They are widely used in aerospace and various industrial fields. Due to time-varying loads, low-speed impacts, and harsh operating environments, carbon fiber reinforced composite materials are prone to invisible damage such as delamination and debonding, which seriously threatens the safety of equipment and personnel. Therefore, the study of non-destructive detection and evaluation technology of carbon fiber reinforced polymer composites is of great significance to the economic and reliable operation of aerospace and engineering equipment.

[0003] Structural health monitoring technology based on guided waves is widely considered to be an ideal choice for detecting damage in plate-like composite structures due to its advantages such as long propagation distance, low attenuation, low cost, and high sensitivity to various damages. Guided wave detection is usually implemented through a sensor network composed of piezoelectric transducers, in which the transmitter excites guided waves in the inspected structure and the receiver collects the corresponding signals for subsequent analysis and processing. The characteristics of guided waves will change due to the presence of damage, and may manifest as linear characteristics, such as amplitude attenuation and velocity changes. Different characteristic information corresponds to different damage indices, and special detection algorithms are designed to achieve accurate damage location and assessment.

[0004] Traditional damage identification methods lack effective signal feature extraction technology. Usually, it is necessary to compare the test signal with the baseline signal (the signal from the healthy structure) or subtract the background noise to extract the damage scattering signal and obtain information such as damage flight time and gravitational wave energy attenuation. This damage index extraction method usually only uses a single change in the guided wave characteristics and cannot accurately construct a high-order nonlinear mapping relationship between the guided wave signal and the damage location. Summary of the invention

[0005] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a carbon fiber reinforced composite material damage detection method based on the Mamba model to solve the above-mentioned technical problems.

[0006] To achieve the above object, the present invention provides the following technical solution: a carbon fiber reinforced composite material damage detection method based on the Mamba model, comprising:

[0007] S1: Collecting guided wave signal sets on damaged carbon fiber reinforced composite materials;

[0008] S2: extracting the time feature sequence of the waveguide signal set through the waveguide signal processing branch;

[0009] S3: Extracting the spectral features of the waveguide signal set through the continuous wavelet transform branch;

[0010] S4: The temporal and spectral characteristics of the waveguide signal set are integrated to comprehensively obtain the multi-dimensional damage characteristics in the waveguide signal and perform defect detection and positioning on the entire carbon fiber reinforced composite material.

[0011] The present invention is further configured to use four piezoelectric transducers arranged in a dual-transmit and dual-receive form in step S1 to collect a limited set of waveguide signals.

[0012] The present invention is further configured that, in step S2, the time feature sequence of the waveguide signal set is extracted by the waveguide signal processing branch, and the state space model Mamba is used to process the long sequence data:

[0013] Starting from time t = 0, derive the output from the state space model at time points t = 0, 1, 2;

[0014]

[0015] h t represents the state quantity at time t, x t represents the input control quantity at time t, y t represents the system output at time t, A represents the state transfer matrix, and B represents the input control quantity x t For the state quantity h t The influence of C represents the state quantity h t Output y to the system t The influence of B and C is specifically expressed as the input control quantity x t With the state quantity h t The weight matrix maps the input control quantity to the system output. Represents the average value of the weight matrix, using the average value of a set of weight matrices at different time points to smooth the process and reduce the impact of noise;

[0016] Extract the state equation at t = k:

[0017]

[0018] Introduce convolution x for convolution operation:

[0019]

[0020] in, is the input control quantity x t With the system output y t The transfer function matrix between them is composed of the transfer matrix and the weight matrix, and finally outputs the operation result y;

[0021] The waveguide signal set is taken as input data f (1) ∈R B×D , D represents the sequence length of the waveguide signal set, and the data is averaged and pooled to obtain the dimension reduction feature f (2) ∈R B×D′ , for the sequence f (2) Segment f (3) ∈R B×D″ , input state space model Mamba for feature extraction: f (4) =M(f (3) ), where M(·) represents the state-space model Mamba. After the segmented sequence data is input into the state-space model Mamba for feature extraction, the serialized feature f is obtained. (4) ∈R B×D″ ;

[0022] Stack and linearize the serialized features:

[0023] f1 (4) =Reshape(f (4) );f GW =Linear(f1 (4) ),in, is the serialized feature after stacking. Reshape(·) represents the stacking operation on the feature, so that the feature dimension is correctly connected with the thread processing. GW It is the serialized feature after linear processing. Linear(·) means linear processing of serialized features through a linear layer to output a feature vector of fixed size.

[0024] The present invention is further configured that, in step S3, a continuous wavelet transform branch is used to perform time-frequency transformation on the waveguide signal set, and the signal is average-pooled to obtain a dimension reduction feature f (2) ∈R B×D′ , perform continuous wavelet transform on the features and output the features f CWT (1) ∈R B×DCWT :

[0025]

[0026] f CWT (1) =CWT(a,b)[f (2) ];

[0027] Where CWT(·) represents the continuous wavelet transform of the signal, t represents the signal L s (t) is the time variable, ψ(t) is the wavelet mother function, ψ *(t) represents the complex conjugate of ψ(t), a is the scale parameter, which controls the width of the wavelet, and b is the translation parameter, which determines the position of the wavelet in time;

[0028] Get the output f of the wavelet transform CWT (1) , perform average pooling dimensionality reduction and neural network convolution to obtain the spectral feature output of the final wavelet transform branch

[0029] The cross entropy loss function that is more suitable for classification tasks is used in the continuous wavelet transform branch:

[0030]

[0031] in, is the cross entropy loss function, Y represents the true probability that the signal feature belongs to a certain category, and f θ (·) represents the prediction score of an input sample of a certain category, and S(·) represents the Softmax activation function.

[0032] The present invention is further configured that, in step S4, the features of the two branches are fused through a cascade operation, and damage classification is performed using a multi-layer linear layer:

[0033] f Fusion =Concat(f GW ,f CWT );

[0034] f 1 Fusion =Linear1(f Fusion );

[0035] f 2 Fusion =Linear2(f 1 Fusion );

[0036] f Result =Linear3(f 2 Fusion );

[0037] Among them, f Fusion Concat(·) represents the concatenation operation, which is the feature fusion of two branches. is the serialized feature after linear processing, is the spectral feature output of the wavelet transform branch, D1 represents the feature dimension size of the waveguide branch, which is set to 1000, and D2 represents the feature dimension size of the spectrum graph of the continuous wavelet transform branch, which is set to 384. i (·) represents the linear layer number, is the serialized feature after processing by the i-th linear layer, f Result ∈R B×N Represents the output damage classification result, and N represents the number of damage categories.

[0038] The present invention provides a carbon fiber reinforced composite material damage detection method based on the Mamba model. The method collects a waveguide signal set on a damaged carbon fiber reinforced composite material; extracts a time feature sequence of the waveguide signal set through a waveguide signal processing branch; extracts a spectral feature of the waveguide signal set through a continuous wavelet transform branch; integrates the time feature and spectral feature of the waveguide signal set, comprehensively obtains multi-dimensional damage features in the waveguide signal, and performs defect detection and positioning on the entire carbon fiber reinforced composite material. The beneficial effects produced include:

[0039] The proposed detection method integrates the state space model Mamba, continuous wavelet transform, and convolutional neural network to achieve efficient parallel computing. The Mamba method is used to extract time features from long-distance guided wave sequences with lower computational complexity. Continuous wavelet transform enhances the signal feature extraction capability and robustness, while the spectrum of continuous wavelet transform is not a time feature map. Convolutional neural network is used to extract time-frequency features from the guided wave signal after continuous wavelet transform, effectively combining the time features obtained by Mamba with the convolutional neural network to comprehensively obtain multi-scale damage features, extract deeper relationship information in the guided wave signal, adaptively analyze the guided wave signal information, accurately extract feature information related to damage, and achieve high-precision detection of composite material damage. This method realizes damage detection under the conditions of limited guided wave signal set and no baseline signal, directly uses the guided wave signal set of the complete time series as input, expands the engineering application of guided waves in baseline-free environments, and improves the scope of application of ultrasonic guided wave non-destructive testing.

[0040] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0042] Figure 1 A flowchart of a carbon fiber reinforced composite material damage detection method based on the Mamba model is shown as an exemplary embodiment of the present invention;

[0043] Figure 2 A schematic diagram of time feature extraction of a waveguide signal processing branch is shown as an exemplary embodiment of the present invention;

[0044] Figure 3 A schematic diagram of a model of a continuous wavelet transform branch according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0046] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0047] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0048] like Figure 1 As shown, the carbon fiber reinforced composite material damage detection method based on the Mamba model includes:

[0049] S1: Collecting guided wave signal sets on damaged carbon fiber reinforced composite materials;

[0050] S2: extracting the time feature sequence of the waveguide signal set through the waveguide signal processing branch;

[0051] S3: Extracting the spectral features of the waveguide signal set through the continuous wavelet transform branch;

[0052] S4: The temporal and spectral characteristics of the waveguide signal set are integrated to comprehensively obtain the multi-dimensional damage characteristics in the waveguide signal and perform defect detection and positioning on the entire carbon fiber reinforced composite material.

[0053] Specifically, in step S1, four piezoelectric transducers are first arranged on a composite material plate in a dual-transmitting and dual-receiving form, an arbitrary waveform generator generates an excitation signal, which is applied to the excitation piezoelectric transducer through an ultrasonic signal preamplifier, and an oscilloscope collects the ultrasonic signal of the receiving piezoelectric transducer. The ultrasonic signal constitutes a waveguide signal set.

[0054] like Figure 2 As shown, specifically, in step S2, the time feature sequence of the waveguide signal set is extracted, and the output of the state space model at different time points is derived from time t=0 to obtain the state equation at the time point. By introducing the convolution parameter, the convolution is introduced into the calculation and the convolution operation is parallelized. The input waveguide signal set is averaged and pooled to reduce the data dimension and obtain the dimension reduction feature. The dimension reduction data is segmented and input into the Mamba model for feature extraction to obtain the serialized feature. Finally, the serialized feature is stacked and linearized to obtain the waveguide signal set feature.

[0055] The specific stage of waveguide signal set feature extraction is as follows:

[0056] Starting from time t=0, the time points t=0, 1, 2 output from the state-space model are derived.

[0057]

[0058] h t represents the state quantity at time t, x t represents the input control quantity at time t, y t represents the system output at time t, A represents the state transfer matrix, and B represents the input control quantity x t For the state quantity h t The influence of C represents the state quantity h t Output y to the system t The influence of B and C is specifically expressed as the input control quantity x t With the state quantity h t The weight matrix maps the input control quantity to the system output. Represents the average value of the weight matrix, using the average value of a set of weight matrices at different time points to smooth the process and reduce the impact of noise.

[0059] Extract the state equation at t = k:

[0060]

[0061] Extract the state equation at t = k:

[0062]

[0063] Introduce convolution x for convolution operation:

[0064]

[0065] in, is the input control quantity x t With the system output y t The transfer function matrix between them is composed of the transfer matrix and the weight matrix, and finally outputs the operation result y;

[0066] The waveguide signal set is taken as input data f (1) ∈R B×D , D represents the sequence length of the waveguide signal set,

[0067] Perform average pooling on the data to obtain the dimension reduction feature f (2) ∈R B×D′ , for the sequence f (2) Segment f (3) ∈R B ×D″ , input state space model Mamba for feature extraction: f (4) =M(f (3) ), where M(·) represents the state-space model Mamba. After the segmented sequence data is input into the state-space model Mamba for feature extraction, the serialized feature f is obtained. (4) ∈R B×D″ ;

[0068] Stack and linearize the serialized features:

[0069] f1 (4) =Reshape(f (4) );f GW =Linear(f1 (4) ),in, is the serialized feature after stacking. Reshape(·) represents the stacking operation on the feature, so that the feature dimension is correctly connected with the thread processing. GW It is the serialized feature after linear processing. Linear(·) means linear processing of serialized features through a linear layer to output a feature vector of fixed size.

[0070] like Figure 3 As shown, specifically, in step S3, the spectral features of the waveguide signal set are extracted through a continuous wavelet transform branch, the waveguide signal set is average pooled to obtain the dimensionality reduction features of the waveguide signal set, the waveguide signal set after dimensionality reduction is subjected to continuous wavelet transform to obtain the spectral features of the waveguide signal set, the spectral features are dimensionally reduced to obtain the reduced dimensional spectral features, and then convolution and signal segmentation are performed to extract the signal features of the continuous wavelet transform spectral features in the time domain.

[0071] The specific stage of extracting the time domain spectrum features of the waveguide signal set is as follows:

[0072] Perform average pooling on the signal to obtain the dimension reduction feature f (2) ∈R B×D′ , perform continuous wavelet transform on the features and output the features

[0073]

[0074]

[0075] f CWT (1) =CWT(a,b)[f (2) ];

[0076] Where CWT(·) represents the continuous wavelet transform of the signal, t represents the signal L s (t) is the time variable, ψ(t) is the wavelet mother function, ψ * (t) represents the complex conjugate of ψ(t), a is the scale parameter, which controls the width of the wavelet, and b is the translation parameter, which determines the position of the wavelet in time;

[0077] Get the output f of the wavelet transform CWT (1) , perform average pooling dimensionality reduction and neural network convolution, output spectral features of dimension (16, 24), perform signal segmentation, set the dimension to 384, and obtain the spectral feature output of the final wavelet transform branch

[0078] The cross entropy loss function that is more suitable for classification tasks is used in the continuous wavelet transform branch:

[0079]

[0080] in, is the cross entropy loss function, Y represents the true probability that the signal feature belongs to a certain category, and f θ (·) represents the prediction score of an input sample of a certain category, and S(·) represents the Softmax activation function.

[0081] In step S4, the time characteristics and spectral characteristics of the waveguide signal set are integrated to comprehensively obtain the multi-dimensional damage characteristics in the waveguide signal, and to perform defect detection and positioning on the entire carbon fiber reinforced composite material.

[0082] Specifically, the features of the waveguide signal processing branch and the continuous wavelet transform branch are fused through a cascade operation, and a three-layer linear layer operation is performed on the fused waveguide signal set features to obtain the damage classification of carbon fiber reinforced composite materials and complete damage detection and positioning.

[0083] The time characteristics and spectral characteristics of the waveguide signal set are integrated to detect and locate defects in the entire carbon fiber reinforced composite material. The specific stages are as follows:

[0084] f Fusion =Concat(f GW ,f CWT );

[0085] f 1 Fusion =Linear1(f Fusion );

[0086] f 2 Fusion =Linear2(f 1 Fusion );

[0087] f Result =Linear3(f 2 Fusion );

[0088] Among them, f Fusion Concat(·) represents the concatenation operation, which is the feature fusion of two branches. is the serialized feature after linear processing, is the spectral feature output of the wavelet transform branch, D1 represents the feature dimension size of the waveguide branch, which is set to 1000, and D2 represents the feature dimension size of the spectrum graph of the continuous wavelet transform branch, which is set to 384. i (·) represents the linear layer number, is the serialized feature after processing by the i-th linear layer, f Result ∈R B×N Represents the output damage classification result, and N represents the number of damage categories.

[0089] During detection, traditional detection technology uses the healthy waveguide signal set of carbon fiber reinforced composite materials as the baseline signal, trains the neural network model with a large number of healthy waveguide signal sets and damaged waveguide signal sets, and performs damage detection and positioning of carbon fiber reinforced composite materials. As a result, it is impossible to detect and position damage to carbon fiber reinforced composite materials without baseline waveguide signals or in a limited waveguide data set.

[0090] The temporal features of the waveguide signal set were extracted using the Mamba model in the prior art, and seven different regions were randomly selected. A damage point was placed at the center of each region, and there were five damage locations in each region. Four sets of waveguide signal set data were generated for each damage scenario, for a total of 140 waveguide signal data sets.

[0091] Compared with the existing technology, the carbon fiber reinforced composite material guided wave method based on the Mamba model is used for damage detection. 80% of the data set is used for training, 10% of the data set is used for verification, and 10% of the data set is used for testing. The verification accuracy is 100%, the test accuracy is 100%, and the detection time is 84.92 seconds.

[0092] In this embodiment, the extracted waveguide signal sets are all waveguide signals collected under the condition that there are defects on the carbon fiber reinforced composite material, and there is no waveguide signal in a healthy state, that is, the baseline-free waveguide damage detection and positioning of the carbon fiber reinforced composite material based on the Mamba model of this method is performed without the need for a baseline signal. Therefore, a specific formula is used to process and extract the time feature sequence of the waveguide signal set. By introducing convolution parameters, convolution is introduced into the calculation and the convolution operation is parallelized. The input waveguide signal set is averaged and pooled to reduce the data dimension. The Mamba model is input for feature extraction, and the serialized features are stacked and linearized to obtain the characteristics of the waveguide signal set. At the same time, the characteristics of the spectral characteristics of the waveguide signal set in the time domain are extracted by continuous wavelet transform and convolutional neural network, and finally the time characteristics and spectral characteristics of the waveguide signal set are fused to comprehensively obtain the multidimensional damage characteristics in the waveguide signal, and the overall carbon fiber reinforced composite material is defect-detected and positioned. Note that other formulas cannot obtain the deep characteristics and comprehensive information of the damage information in the waveguide signal without a baseline.

[0093] The state space model Mamba is more efficient and has lower computational complexity in processing long sequence waveguide signal sets, which is very beneficial for actual damage detection projects. This method extracts time features from long-distance waveguide sequences through the Mamba method. Continuous wavelet transform enhances the extraction capability and robustness of signal features. Convolutional neural network is used to extract time-frequency features from waveguide signals after continuous wavelet transform. The time features obtained by Mamba are effectively combined with convolutional neural network to comprehensively obtain multi-scale damage features, extract deeper relationship information in waveguide signals, adaptively analyze waveguide signal information, accurately extract feature information related to damage, and realize high-precision detection of composite material damage. This method realizes damage detection under the conditions of limited waveguide signal sets and no baseline signal, directly uses the waveguide signal set of the complete time series as input, expands the engineering application of waveguide in baseline-free environment, and improves the scope of application of ultrasonic guided wave nondestructive testing.

[0094] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0095] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0096] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0097] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0098] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0099] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0100] In the several embodiments provided in the present application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0101] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0102] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0103] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0104] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A carbon fiber reinforced composite material damage detection method based on the Mamba model, characterized in that: include: S1: Collecting guided wave signal sets on damaged carbon fiber reinforced composite materials; S2: extracting the time feature sequence of the waveguide signal set through the waveguide signal processing branch; S3: Extracting the spectral features of the waveguide signal set through the continuous wavelet transform branch; S4: The temporal and spectral characteristics of the waveguide signal set are integrated to comprehensively obtain the multi-dimensional damage characteristics in the waveguide signal and perform defect detection and positioning on the entire carbon fiber reinforced composite material.

2. The carbon fiber reinforced composite material damage detection method based on the Mamba model according to claim 1, characterized in that: In step S1, four piezoelectric transducers are arranged in a dual-transmit and dual-receive form to collect a limited set of guided wave signals.

3. The carbon fiber reinforced composite material damage detection method based on the Mamba model according to claim 1, characterized in that: In step S2, the time feature sequence of the waveguide signal set is extracted through the waveguide signal processing branch, and the state space model Mamba is used to process the long sequence data: Starting from time t = 0, derive the output from the state space model at time points t = 0, 1, 2; h t represents the state quantity at time t, x t represents the input control quantity at time t, y t represents the system output at time t, A represents the state transfer matrix, and B represents the input control quantity x t For the state quantity h t The influence of C represents the state quantity h t Output y to the system t The influence of B and C is specifically expressed as the input control quantity x t With the state quantity h t The weight matrix maps the input control quantity to the system output. Represents the average value of the weight matrix, using the average value of a set of weight matrices at different time points to smooth the process and reduce the impact of noise; Extract the state equation at t = k: Introduce convolution x to perform convolution operation: in, is the input control quantity x t With the system output y t The transfer function matrix between them is composed of the transfer matrix and the weight matrix, and finally outputs the operation result y; The waveguide signal set is taken as input data f (1) ∈R B×D , D represents the sequence length of the waveguide signal set, and the data is averaged and pooled to obtain the dimension reduction feature f (2) ∈R B×D′ , for the sequence f (2) Segment f (3) ∈R B×D″ , input state space model Mamba for feature extraction: f (4) =M(f (3) ), where M(·) represents the state-space model Mamba. After the segmented sequence data is input into the state-space model Mamba for feature extraction, the serialized feature f is obtained. (4) ∈R B×D″ ; Stack and linearize the serialized features: f1 (4) =Reshape(f (4) );f GW =Linear(f1 (4) ),in, is the serialized feature after stacking. Reshape(·) represents the stacking operation on the feature, so that the feature dimension is correctly connected with the thread processing. GW It is the serialized feature after linear processing. Linear(·) means linear processing of serialized features through a linear layer to output a feature vector of fixed size.

4. The carbon fiber reinforced composite material damage detection method based on the Mamba model according to claim 1, characterized in that: In step S3, the continuous wavelet transform branch is used to perform time-frequency transformation on the waveguide signal set, and the signal is averaged and pooled to obtain the dimension reduction feature f (2) ∈R B×D′ , perform continuous wavelet transform on the features and output the features f CWT (1) =CWT(a,b)[f (2) ]; Where CWT(·) represents the continuous wavelet transform of the signal, t represents the signal L s (t) is the time variable, ψ(t) is the wavelet mother function, ψ * (t) represents the complex conjugate of ψ(t), a is the scale parameter, which controls the width of the wavelet, and b is the translation parameter, which determines the position of the wavelet in time; Get the output f of the wavelet transform CWT (1) , perform average pooling dimensionality reduction and neural network convolution to obtain the spectral feature output of the final wavelet transform branch The cross entropy loss function that is more suitable for classification tasks is used in the continuous wavelet transform branch: in, is the cross entropy loss function, Y represents the true probability that the signal feature belongs to a certain category, and f θ (·) represents the prediction score of an input sample of a certain category, and S(·) represents the Softmax activation function.

5. The carbon fiber reinforced composite material damage detection method based on the Mamba model according to claim 1, characterized in that: In step S4, the features of the two branches are fused through a cascade operation, and a multi-layer linear layer is used for damage classification: f Fusion =Concat(f GW ,f CWT ); f 1 Fusion =Linear1(f Fusion ); f 2 Fusion =Linear2(f 1 Fusion ); f Result =Linear3(f 2 Fusion ); Among them, f Fusion Concat(·) represents the concatenation operation, which is the feature fusion of two branches. is the serialized feature after linear processing, is the spectral feature output of the wavelet transform branch, D1 represents the feature dimension size of the waveguide branch, which is set to 1000, and D2 represents the feature dimension size of the spectrum graph of the continuous wavelet transform branch, which is set to 384. i (·) represents the linear layer number, is the serialized feature after processing by the i-th linear layer, f Result ∈R B×N Represents the output damage classification result, and N represents the number of damage categories.

Images