Concrete interface debonding detection method and device, electronic equipment and storage medium

Abstract

The present application relates to a kind of concrete interface debonding detection method, device, electronic equipment and storage medium, comprising: obtaining source structure and target structure, source structure is the structure of providing debonding detection diagnosis knowledge, target structure is the structure to be detected.Structure and target structure waveguide signal are extracted in the time-frequency domain information of wavelet transform.Based on the time-frequency domain information, artificial neural network based on domain self-adapting field is constructed and obtained, and the feature learning and parameter optimization of source model are carried out by artificial neural network, to obtain source structure result prediction network.Based on the statistical alignment of the pair of data space in domain self-adapting field, target model is guided to carry out adaptive training, to obtain target structure feature extraction network.Source structure result prediction network and target structure feature extraction network are combined to complete the debonding diagnosis of the waveguide signal of the structure to be detected.It has strong universality, and reduces the sensitivity of debonding detection to environmental parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of interface debonding of reinforced concrete structures, and particularly relates to a concrete interface debonding detection method and device, an electronic device and a storage medium. BACKGROUND

[0002] The interface debonding problem of reinforced concrete structures has always been a major factor leading to structural damage and even large-scale catastrophic events, and thus has become a focus of attention in the field of structural health monitoring (SHM). In recent years, the detection of reinforced concrete structure interface debonding based on guided wave propagation (GWP) technology has attracted the attention of a large number of researchers. Due to the inherent sensitivity of guided waves to their waveguide boundary conditions, the phenomenon of interface debonding of reinforced concrete can be clearly reflected in the guided wave signal. Therefore, many scholars have devoted their efforts to exploring the relationship between the guided wave signal characteristics and the structure debonding state. The existing research results can be mainly divided into two categories according to the analysis mechanism, which are statistical model method based on signal statistical characteristics and physical model method based on guided wave propagation law.

[0003] The statistical model method mainly seeks the statistical characteristics of the guided wave signal of the structure under different damage conditions, establishes a statistical fitting relationship between the signal characteristics and the structure damage, and realizes the analysis of the guided wave signal. For example, the signal is decomposed into a set of dimensionless parameters by using statistical methods such as empirical mode decomposition and wavelet transform, and then the dimensionless parameters are converted into a damage index (DI) without clear physical meaning by using statistical methods, and a fitting relationship between the damage index and the structure damage state is established, so as to realize the analysis of the guided wave signal. The physical model method mainly focuses on the propagation behavior of guided waves in the structure, and finds out the appropriate damage analysis excitation frequency or mode of guided waves through the theoretical calculation of waveguide dispersion. This selection focuses on the abnormal law of waves in certain modes, and is based on this to analyze the damage of the structure. For example, the difference in group velocity of guided waves on the steel and steel structure is used to calculate the size of the debonding section on the structure by the time of flight (TOF) of different wave packets.

[0004] Both of the aforementioned methods exhibit high stability and effectiveness under specific structures. However, when applied to real-world engineering projects, various problems or limitations arise, rendering them unsuitable for real-world structures. This is because both methods are overly sensitive to material, structural, and environmental parameters. Therefore, even a slight difference between a parameter of the applied structure and the parameter of the developed method can cause the method to fail. For statistical modeling methods, a change in a structural parameter renders the original model fitting inaccurate in describing the relationship between signal and damage characteristics in the new structure, leading to a significant decrease in method performance. Physical model-based methods rely primarily on anomalous features (wave velocity or propagation attenuation rate of a specific mode) in waveguide dispersion curves from theoretical analysis for structural damage analysis. Changes in relevant parameters alter the dispersion patterns, rendering the original analytical rules inapplicable, thus reducing the method's accuracy or even yielding completely erroneous results. Furthermore, these problems are difficult to resolve by re-experimenting on new structures and constructing new method parameters. This is partly because laboratory conditions—including material selection, sample preparation standards, and testing environments—are far superior to real-world conditions, and the integrity and uniformity of the structures differ from real-world infrastructure. On the other hand, the performance differences of different batches of raw materials and the unavoidable differences during mixing and pouring will further lead to differences in the structure in different parts.

[0005] In summary, traditional methods for detecting debonding at concrete interfaces are highly sensitive to environmental parameters and are difficult to perform debonding detection and analysis on reinforced concrete interfaces across structural structures. Summary of the Invention

[0006] Therefore, it is necessary to provide a concrete interface debonding detection method, device, electronic equipment, and storage medium that is less sensitive to environmental parameters and can realize cross-structure debonding detection of reinforced concrete interfaces.

[0007] This invention provides a method for detecting debonding at concrete interfaces, the method comprising:

[0008] Obtain a source structure and a target structure, wherein the source structure is a structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected;

[0009] Time-frequency domain information is extracted from the guided wave signals of the source and target structures using wavelet transform;

[0010] Based on the time-frequency domain information, an artificial neural network based on domain adaptation is constructed and obtained. The artificial neural network is then used to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network.

[0011] Based on the adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain, the target model is trained in a guided adversarial adaptive manner to obtain the target structural feature extraction network.

[0012] The source structure result prediction network is combined with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0013] In one embodiment, the debonding diagnostic knowledge of the source structure is greater than the debonding diagnostic knowledge required by the target structure, and the acquisition of the source structure and the target structure includes:

[0014] Wave propagation tests were conducted on a concrete structure in a known debonding state to obtain the corresponding first guided wave signal, and the source structure was obtained based on the first guided wave signal and the known debonding state.

[0015] A wave propagation test is performed on the target structure to obtain a second guided wave signal of the target structure, which is used to determine the debonding state of the target structure.

[0016] In one embodiment, the extraction of time-frequency domain information from the wavelet transform of the guided signals of the source structure and the target structure includes:

[0017] Continuous wavelet transform is performed using wavelet basis functions of Morlet wavelets, and the wavelet type is correlated with the guided wave signal in the corresponding time-frequency domain;

[0018] The scale of the wavelet transform output spectrum is adjusted to the same order of magnitude by writing a Python script to adapt to the feature analysis of the artificial neural network.

[0019] In one embodiment, the source model includes a source structure feature extraction network and a source structure result prediction network, wherein the source structure feature extraction network adopts a convolutional neural network architecture to adapt to the convolutional neural network skeleton layer with two-dimensional convolutional kernels, and the source structure result prediction network adopts a fully connected network architecture.

[0020] The process of constructing and obtaining an artificial neural network based on the time-frequency domain information and using the artificial neural network to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network includes:

[0021] The convolutional neural network architecture is subjected to feature learning and parameter optimization through the artificial neural network to obtain the source structure feature extraction network.

[0022] The source structure feature extraction network is used to extract features from the target structure to obtain the features to be analyzed.

[0023] The fully connected layer constructed through the fully connected network architecture performs feature analysis on the features to be analyzed and outputs the analysis results.

[0024] In one embodiment, the step of constructing and obtaining an artificial neural network based on the time-frequency domain information and using the artificial neural network to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network further includes:

[0025] The source model is trained by supervised model training to learn features and optimize parameters, and the initial parameters of the source model are used as the parameters of the source model network that has completed model training to train the source model.

[0026] The loss function is minimized based on backpropagation and gradient descent, so that the source model learns the spatial distribution and mapping relationship of the data in the source structure.

[0027] In one embodiment, the guided adversarial adaptive training of the target model based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain to obtain a target structural feature extraction network includes:

[0028] Obtain obfuscation labels by distinguishing the domain labels of the source structural feature extraction network and the target structural feature extraction network, and use the obfuscation labels to make the target structural feature extraction network have a data distribution structure consistent with the source structural feature extraction network;

[0029] The maximum mean difference between the source structural feature extraction network and the target structural feature extraction network is obtained, and the maximum mean difference is minimized by gradient descent, so that the target structural feature extraction network and the source structural feature extraction network have statistical similarity.

[0030] In one embodiment, the target model feature extraction network and the source model feature extraction network use the same architecture.

[0031] The present invention also provides a concrete interface debonding detection device, the device comprising:

[0032] The first acquisition module is used to acquire a source structure and a target structure, wherein the source structure is a structure that provides debonding detection and diagnostic knowledge, and the target structure is a structure to be detected;

[0033] The extraction module is used to extract time-frequency domain information of the guided wave signals of the source structure and the target structure by performing wavelet transform.

[0034] The first training module is used to construct and obtain an artificial neural network based on the time-frequency domain information, and to perform feature learning and parameter optimization on the source model through the artificial neural network to obtain the source structure result prediction network.

[0035] The second training module is used to perform guided adversarial adaptive training on the target model based on the adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain, so as to obtain the target structural feature extraction network.

[0036] The debonding diagnosis module is used to combine the source structure result prediction network with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0037] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the concrete interface debonding detection method as described above.

[0038] The present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the concrete interface debonding detection method as described above.

[0039] The aforementioned concrete interface debonding detection method, device, electronic equipment, and storage medium acquire source and target structures that provide debonding detection and diagnostic knowledge. They then extract time-frequency domain information from the guided wave signals of both structures using wavelet transform, constructing and acquiring a corresponding domain-adaptive artificial neural network. This artificial neural network is then used to learn features and optimize parameters of the source model to obtain a source structure result prediction network. Next, based on adversarial and statistical alignment with a pair of data spaces in the adaptive domain, guided adaptive training is performed on the target model corresponding to the target structure to obtain a corresponding target structure feature extraction network. Finally, the source structure result prediction network and the target structure feature extraction network are combined to complete the debonding diagnosis of the guided wave signal corresponding to the target structure. This method no longer relies on theoretical analysis of a specific feature point or artificially statistical fitting relationship to construct the correspondence between guided wave signals and structural debonding. Instead, it is based on data science, using the overall distribution of guided wave signals from a macroscopic perspective and mapping between data spaces to find universal features across different data distributions. This reduces the sensitivity of debonding detection to environmental parameters and enables debonding detection and analysis of reinforced concrete interfaces across different structures. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is one of the schematic diagrams of the concrete interface debonding detection method provided by the present invention;

[0042] Figure 2 This is the second schematic diagram of the concrete interface debonding detection method provided by the present invention.

[0043] Figure 3 This is the third schematic diagram of the concrete interface debonding detection method provided by the present invention;

[0044] Figure 4 This is the fourth schematic diagram of the concrete interface debonding detection method provided by the present invention;

[0045] Figure 5 This is the fifth schematic diagram of the concrete interface debonding detection method provided by the present invention;

[0046] Figure 6 This is the sixth schematic diagram of the concrete interface debonding detection method provided by the present invention;

[0047] Figure 7 This is a schematic diagram of the detection system structure of the concrete interface debonding detection method in a specific embodiment of the present invention;

[0048] Figure 8 A schematic diagram of a three-dimensional reinforced concrete beam model for the concrete interface debonding detection method provided in a specific embodiment of the present invention;

[0049] Figure 9 A schematic diagram of the artificial neural network for the concrete interface debonding detection method in a specific embodiment of the present invention;

[0050] Figure 10 A schematic diagram of the source model training process for the concrete interface debonding detection method in a specific embodiment of the present invention;

[0051] Figure 11 This is a schematic diagram of the target model training process for the concrete interface debonding detection method in a specific embodiment of the present invention.

[0052] Figure 12 a is a T-SNE diagram of the original data distribution of the source structure in the concrete interface debonding detection method provided in a specific embodiment of the present invention;

[0053] Figure 12 b is a T-SNE diagram of the data distribution after feature mapping of the source model in a concrete interface debonding detection method provided in a specific embodiment of the present invention.

[0054] Figure 12 c is a T-SNE diagram of the original data distribution of the target structure in the concrete interface debonding detection method provided in a specific embodiment of the present invention.

[0055] Figure 12 d is a T-SNE diagram of the data distribution after feature mapping of the source model of the target structure in the concrete interface debonding detection method provided in a specific embodiment of the present invention.

[0056] Figure 12 e is a T-SNE diagram of the data distribution after feature mapping of the target model following training of the target structure in the concrete interface debonding detection method provided in a specific embodiment of the present invention.

[0057] Figure 13 A comparison diagram of the debonding detection performance of the detection system of the concrete interface debonding detection method in a specific embodiment of the present invention and the traditional debonding detection system;

[0058] Figure 14 This is a schematic diagram of the concrete interface debonding detection device provided by the present invention.

[0059] Figure 15 An internal structural diagram of the computer device provided by the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] The following is combined Figures 1-15 The present invention describes a concrete interface debonding detection method, apparatus, electronic device, and storage medium.

[0062] like Figure 1 As shown, in one embodiment, a method for detecting debonding at concrete interfaces includes the following steps:

[0063] Step S110: Obtain the source structure and the target structure. The source structure is the structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected.

[0064] Specifically, the server obtains the source structure and the structure to be tested from the concrete structure, which provide knowledge for debonding detection and diagnosis.

[0065] Among them, the source structure is a concrete structure with a known concrete interface debonding state. The debonding state of the concrete interface is diverse, and it provides debonding detection and diagnosis knowledge for the concrete structure to be tested. That is, the debonding state of the concrete structure to be tested is within the range of the known concrete interface debonding state.

[0066] Step S120: Extract time-frequency domain information of the guided wave signals of the source structure and the target structure by wavelet transform.

[0067] Specifically, the server extracts the time-frequency domain information of the guided wave signals of the source and target structures by performing wavelet transform, in order to obtain the corresponding time-frequency domain information.

[0068] Step S130: Construct and obtain an artificial neural network based on domain adaptation based on time-frequency domain information, and perform feature learning and parameter optimization on the source model through the artificial neural network to obtain the source structure result prediction network.

[0069] Specifically, the server constructs and obtains an artificial neural network for the corresponding adaptive domain based on the time-frequency domain information extracted in step S120, and performs feature learning and parameter optimization on the source model through the artificial neural network to obtain the corresponding source structure result prediction network.

[0070] Step S140: Based on the adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptation domain, the target model is trained in guided adversarial adaptation to obtain the target structural feature extraction network.

[0071] Specifically, the server performs guided adversarial adaptive training on the target model corresponding to the target structure based on the adversarial alignment theory and statistical alignment theory of a pair of data spaces in the domain adaptation domain in step S130.

[0072] Step S150: Combine the source structure result prediction network with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0073] Specifically, the server combines the source structure result prediction network obtained in step S130 with the target structure feature extraction network obtained in step S140, and performs debonding detection on the concrete structure to be detected based on the guided wave signal of the concrete structure to be detected, thus completing the debonding diagnosis of the guided wave signal of the target structure.

[0074] The aforementioned concrete interface debonding detection method acquires the source structure and the structure to be detected, providing diagnostic knowledge for debonding detection. It then extracts time-frequency domain information from the guided wave signals of both structures using wavelet transform, constructing and obtaining a corresponding domain-adaptive artificial neural network. This artificial neural network is then used to learn features and optimize parameters of the source model to obtain a source structure result prediction network. Next, based on adversarial and statistical alignment with a pair of data spaces in the adaptive domain, guided adaptive training is performed on the target model corresponding to the structure to be detected to obtain a corresponding target structure feature extraction network. Finally, the source structure result prediction network and the target structure feature extraction network are combined to complete the debonding diagnosis of the guided wave signal corresponding to the structure to be detected. This method no longer relies on theoretical analysis of a specific feature point or artificially statistical fitting relationship to construct the correspondence between guided wave signals and structural debonding. Instead, it is based on data science, using the overall distribution of guided wave signals from a macroscopic perspective and mapping between data spaces to find universal features across different data distributions. This reduces the sensitivity of debonding detection to environmental parameters and enables cross-structure debonding detection analysis of reinforced concrete interfaces.

[0075] like Figure 2 As shown, in one embodiment, the concrete interface debonding detection method provided by the present invention acquires the source structure and the target structure, including the following steps:

[0076] Step S112: Conduct a wave propagation test on a concrete structure in a known debonding state to obtain the corresponding first guided wave signal, and obtain the source structure based on the first guided wave signal and the known debonding state.

[0077] Specifically, the server conducts wave propagation tests on a known Tony installation state concrete structure to obtain the corresponding guided wave signal. To compare the guided wave signal of the concrete structure in the known state with the guided wave signal of the concrete structure to be tested, this guided wave signal is called the first guided wave signal. Based on the first guided wave signal and the corresponding known debonding state, the server obtains the corresponding source structure to provide debonding detection and diagnosis knowledge for the concrete structure to be tested.

[0078] Step S114: Perform a wave propagation test on the target structure to obtain the second guided wave signal of the target structure. The second guided wave signal is used to determine the debonding state of the target structure.

[0079] Specifically, the server performs wave propagation experiments on the target structure to obtain the second guided wave signal of the target structure. This second guided wave signal is the guided wave signal corresponding to the target structure and is used to determine the debonding state of the target structure.

[0080] like Figure 3As shown, in one embodiment, the concrete interface debonding detection method provided by the present invention extracts time-frequency domain information of the guided wave signals of the source structure and the target structure by wavelet transform, including the following steps:

[0081] Step S122: Perform continuous wavelet transform using the wavelet basis functions of Morlet wavelets, and ensure that the wavelet type is correlated with the guided wave signal in the corresponding time-frequency domain.

[0082] Specifically, the server performs continuous wavelet transform using the wavelet basis functions of the Morlet wavelet, and its wavelet type is correlated with the corresponding guided wave signal in the corresponding time-frequency domain.

[0083] Step S124: Adjust the scale of the wavelet transform output spectrum to the same order of magnitude by writing a Python script to adapt to the feature analysis of the artificial neural network.

[0084] Specifically, the server adjusts the scale of the wavelet transform output spectrum to the same order of magnitude by writing Python scripts to make it suitable for feature analysis by artificial neural networks.

[0085] like Figure 4 As shown, in one embodiment, the concrete interface debonding detection method provided by the present invention constructs and obtains an artificial neural network based on domain adaptive domain information based on time-frequency domain information, and performs feature learning and parameter optimization on the source model through the artificial neural network to obtain a source structure result prediction network, including the following steps:

[0086] Step S131: Perform feature learning and parameter optimization on the convolutional neural network architecture using an artificial neural network to obtain the source structure feature extraction network.

[0087] The source model includes a source structure feature extraction network and a source structure result prediction network. The source structure feature extraction network adopts a convolutional neural network architecture to adapt to the convolutional neural network skeleton layer with two-dimensional convolutional kernels, and the source structure result prediction network adopts a fully connected network structure.

[0088] Specifically, the server uses artificial neural networks to learn features and optimize parameters of the convolutional neural network architecture in order to obtain the source structure feature extraction network.

[0089] Step S132: Extract features from the target structure using the source structure feature extraction network to obtain the features to be analyzed.

[0090] Specifically, the server extracts features from the target structure using a source structure feature extraction network to obtain the features to be analyzed in the concrete structure to be detected.

[0091] Step S133: The fully connected layer constructed through the fully connected network architecture performs feature analysis on the features to be analyzed and outputs the analysis results.

[0092] Specifically, the server performs feature analysis on the features to be analyzed obtained in step S132 and outputs the analysis results through a fully connected layer built on a fully connected network architecture.

[0093] like Figure 5 As shown, in one embodiment, the concrete interface debonding detection method provided by the present invention constructs and obtains an artificial neural network based on domain adaptive domain information based on time-frequency domain information, and performs feature learning and parameter optimization on the source model through the artificial neural network to obtain a source structure result prediction network, and further includes the following steps:

[0094] Step S134: The source model is trained by supervised model training to learn features and optimize parameters, and the initial parameters of the source model are used as the network parameters of the source model that has been trained.

[0095] Specifically, the server performs feature learning and parameter optimization on the source model through supervised model training, and uses the initial parameters of the source model as the parameters of the source model network that has already completed model training to train the source model.

[0096] Step S135: Minimize the loss function based on backpropagation and gradient descent so that the source model learns the spatial distribution and mapping relationship of the source structure's data.

[0097] Specifically, the server minimizes the damage function in supervised model training through backpropagation and gradient descent, so that the source model can fully learn the spatial distribution of the source structure's data and the mapping relationships between the data.

[0098] like Figure 6 As shown, in one embodiment, the concrete interface debonding detection method provided by the present invention, based on adversarial alignment and statistical alignment of a pair of data spaces in a domain adaptive domain, performs guided adversarial adaptive training on the target model to obtain a target structural feature extraction network, including the following steps:

[0099] Step S142: Obtain scrambling labels by distinguishing the domain labels of the source structural feature extraction network and the target structural feature extraction network, and use scrambling labels to make the target structural feature extraction network have a data distribution structure consistent with that of the source structural feature extraction network.

[0100] Specifically, the server obtains a confusion label by distinguishing the domain labels of the source structural feature extraction network and the target structural feature extraction network, and uses this confusion label to make the target structural feature extraction network have a data distribution structure consistent with that of the source structural feature extraction network.

[0101] Step S144: Obtain the maximum mean difference between the source structural feature extraction network and the target structural feature extraction network, and minimize the maximum mean difference using the gradient descent method to make the target structural feature extraction network statistically similar to the source structural feature extraction network.

[0102] Specifically, the server obtains the maximum mean difference between the source structural feature extraction network and the target structural feature extraction network, and minimizes this maximum mean difference using the gradient descent method, so that the target structural feature extraction network and the source structural feature extraction network have statistical similarity.

[0103] In a specific embodiment, the present invention provides a method for detecting debonding at concrete interfaces, see [link to relevant documentation]. Figure 7 As shown, a domain-adaptive cross-structure reinforced concrete interface debonding detection system is applied to a reinforced concrete structure health monitoring environment. The system includes the reinforced concrete structure to be tested, a single-point piezoelectric ceramic PZT sheet, an integrated distributed PZT system information source, a data acquisition system, and a domain-adaptive guided wave data and analysis system.

[0104] In this embodiment, considering that all debonding detection methods based on guided wave propagation require a set of original structures with different debonding states as the basis for constructing the signal analysis method, it is necessary to collect a set of such structures to provide diagnostic knowledge for debonding detection. Furthermore, it is also necessary to collect the structures to be detected. The structures providing debonding detection diagnostic knowledge are called source structures, and the structures to be detected are called target structures. The collection of source structures involves conducting wave propagation experiments on structures with known debonding states to obtain guided wave signals, while simultaneously recording the guided wave signals and the debonding states of the structures. The target structure, as the structure to be detected, has an unknown exact debonding state; therefore, wave propagation experiments are conducted, and only its guided wave signals are recorded. Moreover, the debonding diagnostic knowledge provided by the source structures must be greater than the debonding diagnostic knowledge required by the target structure; that is, the debonding state of the target structure must appear in the set of source structures so that the diagnostic knowledge of the source structures can cover the diagnostic detection of the target structure.

[0105] Combination Figure 8 As shown, based on numerical simulation algorithms, a set of three-dimensional reinforced concrete beams with different interface debonding states were established in the commercial finite element software Abaqus, and the guided wave signal was solved using a "dynamic implicit" algorithm. The simulated beam is 500mm long, with a square cross-section of 100mm in length and width, and a 18mm diameter steel bar is placed at the center of the cross-section. The material parameters of the beam are set as shown in Table 1.

[0106] Table 1. Material parameters of the reinforced concrete beam model

[0107]

[0108] This structure contains a total of 10 different debonding states, as shown in Table 2:

[0109] Table 2. Debonding status and damage labels of reinforced concrete beams

[0110]

[0111]

[0112] The structures in Tables 1 and 2 above are used as source structures to provide knowledge for debonding diagnosis.

[0113] In this embodiment, a set of real reinforced concrete beams were cast in the laboratory. Their materials, dimensional parameters, and initial values ​​were consistent with the aforementioned virtual beams. However, parameter fluctuations are unavoidable; therefore, the structure of the real beams cannot be exactly the same as the virtual beams. After ensuring the debonding state of the 10 real beams was consistent with the corresponding numbered virtual beams, guided wave propagation tests were conducted on this set of real beams, and their guided wave signals were recorded. The real beams will serve as the target structure for debonding detection.

[0114] Subsequently, time-frequency domain information extraction based on continuous wavelet transform is performed on the guided wave signals of the source and target structures. The calculation formula for the continuous wavelet transform is as follows:

[0115]

[0116] In the formula, f(t) represents the guided wave signal, f(t)∈L 2 (R), W f (μ, τ) represents the transformation of the guided wave signal at the spatial scale μ and the time scale τ. ω represents the conjugate of the wavelet basis functions, and F(ω) represents the Fourier transform of the guided wave signal. The Fourier transform represents the wavelet basis function.

[0117] The Morlet wavelet is used as the mother wavelet in the continuous wavelet transform, as this wavelet type is correlated with the guided wave signal in both the time and frequency domains. Furthermore, the continuous wavelet transform spectrum of the guided wave signal typically does not have spatial and temporal scales of the same order of magnitude. Therefore, a Python script is needed to adjust the scale of the output spectrum to be of the same order of magnitude to accommodate the subsequent feature analysis by the artificial neural network. In the continuous wavelet transform process for the guided wave signals of the source and target structures, the Morlet wavelet is chosen as the mother wavelet, with a spatial scale of 20 and a time scale of 1000. The wavelet energy spectrum of the guided wave signal is then a (20, 1000) matrix. Therefore, the built-in Python function `reshape` is used to redefine the size of the matrix, modifying it to a (200, 100) matrix.

[0118] Combination Figure 9 As shown, in the process of constructing a domain-adaptive artificial neural network, the source model is first subjected to feature learning and parameter optimization. The optimization objective of the source model is as follows:

[0119]

[0120] In the formula, L CE (·) represents the loss function for a classification task, typically using the cross-entropy function, X. s Y s E represents the source structure guided wave signal and its corresponding structural debonding state, respectively. s P represents the feature extraction network corresponding to the source structure. s The result prediction network represents the source structure.

[0121] It should be noted that the source model consists of two parts: a source structure feature extraction network and a result prediction network. The feature extraction network adopts a convolutional neural network architecture, which is adaptable to the backbone layers of convolutional neural networks using two-dimensional convolutional kernels. The result prediction network adopts a fully connected network architecture, utilizing fully connected layers for feature analysis and result output. Furthermore, the source model is trained using supervised learning methods, combined with… Figure 10 As shown, by minimizing the loss function based on backpropagation and gradient descent, the source model can fully learn the spatial distribution and mapping relationship of the source structure data.

[0122] In this embodiment, a neural network structure as shown in Table 3 was established. A source model was formed by using a feature extraction network and a damage analysis network to learn the data distribution characteristics of the source structure.

[0123] Table 3. Structural details of the constructed artificial neural network for interface debonding analysis

[0124]

[0125]

[0126]

[0127] In Table 3, C represents a convolutional layer, BN represents a batch normalization layer, P represents a pooling layer, DP represents a dropout layer, FC represents a fully connected layer in the damage classification network, FD represents a fully connected layer in the domain discriminator, and the subscripts represent the indices of each layer.

[0128] The T-SNE method is used to tile the high-order data distribution onto a two-dimensional plane for visualization. Figure 12 a and Figure 12b shows the data distribution of guided wave data before and after feature mapping by the source model for different debonding lengths of the source structure, respectively. Here, Category 0, Category 1, Category 2, Category 3, Category 4, and Category 5 represent categories 1, 2, 3, 4, and 5, respectively. It can be seen that the data distribution after feature mapping by the source model is extremely regular, indicating that the source model has fully learned the data distribution of the source structure.

[0129] Then, based on the adversarial alignment and statistical alignment theory of a pair of data spaces in the domain adaptation domain, guided adversarial adaptive training is performed on the target model after the source model is trained. The optimization objective of the target model is:

[0130]

[0131] In the formula, L ADA (·) represents the loss function for adversarial alignment, L DC (·) represents the statistical alignment loss function, where the maximum mean difference is used as the loss function, X. t E represents the guided wave signal of the target structure. t D represents the feature extraction network corresponding to the target structure, and D represents the domain discriminator, used to implement E. s and E t The adversarial training, where λ represents the weight of statistical alignment.

[0132] It should be noted that the feature extraction network of the target model and the feature extraction network of the source model use the same architecture, and their initial network parameters are preset to the parameters of the already trained source model network during training. Figure 11 As shown, the target model is trained using an unsupervised learning method. The training of the target model has two objectives: First, to construct a domain discriminator to distinguish the domain labels of the source and target feature extraction networks. By confusing the domain labels, the target feature extraction network is forced to extract features with a data distribution structure highly consistent with the source feature extraction network—this is adversarial alignment. Second, to construct the maximum mean difference between the outputs of the source and target feature extraction networks. This term mainly describes the maximum difference between the two sets of distributions at each spatial order. Minimizing this difference using gradient descent forces the features extracted by the target network to have statistical similarity—this is statistical alignment. See also... Figure 12 As shown in Figure c, the distribution of the original data of the target structure is obtained using the T-SNE method. Figure 12Figures d and 12e illustrate the data distribution after feature mapping of the target structure data using the source model and the target model, respectively. Here, Category 0, Category 1, Category 2, Category 3, Category 4, and Category 5 represent categories 1, 2, 3, 4, and 5, respectively. It can be seen that because the source model cannot perform cross-structure feature learning, it cannot distinguish the target structure data. Therefore, after mapping with the source model, the data distribution of the target structure remains chaotic. In contrast, the target model, after training with adversarial and statistical alignment, possesses cross-structure decoupling detection capabilities and can distinguish the target structure data to a certain extent.

[0133] Finally, the obtained source structure prediction network and target structure feature extraction network are combined to perform debonding diagnosis on the guided wave signals of the structure to be detected. The constructed neural network is used to test three types of debonding detection tasks at reinforced concrete interfaces: debonding identification, debonding localization, and debonding size quantization. Simultaneously, based on traditional debonding detection methods, a supervised learning-based convolutional neural network is constructed to perform these three tasks. The accuracy rates achieved by the two methods on the three tasks are as follows: Figure 13 As shown in the figure, it can be seen that the concrete interface debonding detection method proposed in this embodiment outperforms the traditional debonding detection method in all three types of tasks.

[0134] The aforementioned concrete interface debonding detection method no longer relies on a specific feature point in theoretical analysis or a manually calculated fitting relationship to construct the correlation between guided wave signals and structural debonding. Instead, it is based on data science to find universal characteristics between different data distributions by nonlinear mapping between data spaces from the overall macroscopic signal distribution. This makes the method less sensitive to the material, size, and environmental parameters of the analyzed structure, thus improving the feasibility of the debonding detection method in practical engineering and giving it broad engineering application prospects.

[0135] The concrete interface debonding detection device provided by the present invention is described below. The concrete interface debonding detection device described below can be referred to in correspondence with the concrete interface debonding detection method described above.

[0136] like Figure 14 As shown, in one embodiment, a concrete interface debonding detection device includes a first acquisition module 1410, an extraction module 1420, a first training module 1430, a second training module 1440, and a debonding diagnosis module 1450.

[0137] The first acquisition module 1410 is used to acquire the source structure and the target structure. The source structure is the structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected.

[0138] The extraction module 1420 is used to extract time-frequency domain information of the wavelet transform of the guided wave signals of the source structure and the target structure.

[0139] The first training module 1430 is used to construct and obtain an artificial neural network based on the domain adaptive domain information based on time and frequency domain information, and to perform feature learning and parameter optimization on the source model through the artificial neural network to obtain the source structure result prediction network.

[0140] The second training module 1440 is used to perform guided adversarial adaptive training on the target model based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain, so as to obtain the target structural feature extraction network.

[0141] The debonding diagnosis module 1450 is used to combine the source structure result prediction network with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0142] In this embodiment, the concrete interface debonding detection device provided by the present invention, the first acquisition module is specifically used for:

[0143] Wave propagation tests were conducted on a concrete structure in a known debonding state to obtain the corresponding first guided wave signal, and the source structure was obtained based on the first guided wave signal and the known debonding state.

[0144] A wave propagation experiment is performed on the target structure to obtain the second guided wave signal of the target structure. The second guided wave signal is used to determine the debonding state of the target structure.

[0145] In this embodiment, the concrete interface debonding detection device provided by the present invention has an extraction module specifically used for:

[0146] Continuous wavelet transform is performed using wavelet basis functions of Morlet wavelets, and the wavelet type is correlated with the guided wave signal in the corresponding time-frequency domain.

[0147] By writing Python scripts to adjust the scale of the wavelet transform output spectrum to the same order of magnitude, it is possible to adapt it to the feature analysis of artificial neural networks.

[0148] In this embodiment, the concrete interface debonding detection device provided by the present invention, the first training module is specifically used for:

[0149] Artificial neural networks are used to learn features and optimize parameters of convolutional neural network architectures to obtain source structure feature extraction networks.

[0150] The source structure feature extraction network is used to extract features from the target structure to obtain the features to be analyzed.

[0151] A fully connected layer, constructed using a fully connected network architecture, performs feature analysis on the features to be analyzed and outputs the analysis results.

[0152] In this embodiment, the first training module of the concrete interface debonding detection device provided by the present invention is further used for:

[0153] The source model is trained by performing feature learning and parameter optimization through supervised model training, and the initial parameters of the source model are used as the parameters of the source model network that has already been trained to train the source model.

[0154] The loss function is minimized using backpropagation and gradient descent, enabling the source model to learn the spatial distribution and mapping relationship of the source structure's data.

[0155] In this embodiment, the second training module of the concrete interface debonding detection device provided by the present invention is specifically used for:

[0156] By distinguishing the domain labels of the source structural feature extraction network and the target structural feature extraction network, confusing labels are obtained, and the confusing labels are used to make the target structural feature extraction network have a data distribution structure consistent with that of the source structural feature extraction network.

[0157] The maximum mean difference between the source structural feature extraction network and the target structural feature extraction network is obtained, and the maximum mean difference is minimized by gradient descent so that the target structural feature extraction network and the source structural feature extraction network have statistical similarity.

[0158] Figure 15 This example illustrates a schematic diagram of the physical structure of an electronic device, which can be a smart terminal. Its internal structure diagram can be as follows: Figure 15 As shown. The electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a concrete interface debonding detection method, which includes:

[0159] Obtain the source structure and the target structure. The source structure is the structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected.

[0160] Time-frequency domain information extraction of guided wave signals from source and target structures using wavelet transform;

[0161] Based on time-frequency domain information, an artificial neural network based on domain adaptation is constructed and obtained. The source model is then used to perform feature learning and parameter optimization through the artificial neural network to obtain the source structure result prediction network.

[0162] Based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptation field, guided adversarial adaptive training is performed on the target model to obtain the target structural feature extraction network.

[0163] The source structure result prediction network is combined with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0164] Those skilled in the art will understand that Figure 15 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0165] On the other hand, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements a method for detecting debonding at concrete interfaces, the method comprising:

[0166] Obtain the source structure and the target structure. The source structure is the structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected.

[0167] Time-frequency domain information extraction of guided wave signals from source and target structures using wavelet transform;

[0168] Based on time-frequency domain information, an artificial neural network based on domain adaptation is constructed and obtained. The source model is then used to perform feature learning and parameter optimization through the artificial neural network to obtain the source structure result prediction network.

[0169] Based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptation field, guided adversarial adaptive training is performed on the target model to obtain the target structural feature extraction network.

[0170] The source structure result prediction network is combined with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0171] In another aspect, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, it implements a concrete interface debonding detection method, the method comprising:

[0172] Obtain the source structure and the target structure. The source structure is the structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected.

[0173] Time-frequency domain information extraction of guided wave signals from source and target structures using wavelet transform;

[0174] Based on time-frequency domain information, an artificial neural network based on domain adaptation is constructed and obtained. The source model is then used to perform feature learning and parameter optimization through the artificial neural network to obtain the source structure result prediction network.

[0175] Based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptation field, guided adversarial adaptive training is performed on the target model to obtain the target structural feature extraction network.

[0176] The source structure result prediction network is combined with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

[0177] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0178] By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0179] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0180] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting debonding at concrete interfaces, characterized in that, The method includes: Obtain a source structure and a target structure, wherein the source structure is a structure that provides knowledge for debonding detection and diagnosis, and the target structure is the structure to be detected; Time-frequency domain information is extracted from the guided wave signals of the source and target structures using wavelet transform; Based on the time-frequency domain information, an artificial neural network based on domain adaptation is constructed and obtained. The artificial neural network is then used to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network. Based on the adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain, the target model is trained in a guided adversarial adaptive manner to obtain the target structural feature extraction network. The source structure result prediction network and the target structure feature extraction network are combined to complete the debonding diagnosis of the guided wave signal of the structure to be detected. The debonding diagnostic knowledge of the source structure is greater than the debonding diagnostic knowledge required by the target structure. The acquisition of the source structure and the target structure includes: Wave propagation tests were conducted on a concrete structure in a known debonding state to obtain the corresponding first guided wave signal, and the source structure was obtained based on the first guided wave signal and the known debonding state. A wave propagation test is performed on the target structure to obtain a second guided wave signal of the target structure, which is used to determine the debonding state of the target structure.

2. The concrete interface debonding detection method according to claim 1, characterized in that, The extraction of time-frequency domain information from the wavelet transform of the guided wave signals of the source and target structures includes: Continuous wavelet transform is performed using wavelet basis functions of Morlet wavelets, and the wavelet type is correlated with the guided wave signal in the corresponding time-frequency domain; The scale of the wavelet transform output spectrum is adjusted to the same order of magnitude by writing a Python script to adapt to the feature analysis of the artificial neural network.

3. The concrete interface debonding detection method according to claim 1, characterized in that, The source model includes a source structure feature extraction network and a source structure result prediction network. The source structure feature extraction network adopts a convolutional neural network architecture to adapt to the convolutional neural network skeleton layer with two-dimensional convolutional kernels. The source structure result prediction network adopts a fully connected network architecture. The process of constructing and obtaining an artificial neural network based on the time-frequency domain information and using the artificial neural network to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network includes: The convolutional neural network architecture is subjected to feature learning and parameter optimization through the artificial neural network to obtain the source structure feature extraction network. The source structure feature extraction network is used to extract features from the target structure to obtain the features to be analyzed. The fully connected layer constructed through the fully connected network architecture performs feature analysis on the features to be analyzed and outputs the analysis results.

4. The concrete interface debonding detection method according to claim 3, characterized in that, The step of constructing and obtaining an artificial neural network based on the time-frequency domain information and adapting the domain to the domain, and then using the artificial neural network to perform feature learning and parameter optimization on the source model to obtain a source structure result prediction network, further includes: The source model is trained by supervised model training to learn features and optimize parameters, and the initial parameters of the source model are used as the parameters of the source model network that has completed model training to train the source model. The loss function is minimized based on backpropagation and gradient descent, so that the source model learns the spatial distribution and mapping relationship of the data in the source structure.

5. The concrete interface debonding detection method according to claim 4, characterized in that, The step of conducting guided adversarial adaptive training on the target model based on adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain to obtain the target structural feature extraction network includes: Obtain obfuscation labels by distinguishing the domain labels of the source structural feature extraction network and the target structural feature extraction network, and use the obfuscation labels to make the target structural feature extraction network have a data distribution structure consistent with the source structural feature extraction network; The maximum mean difference between the source structural feature extraction network and the target structural feature extraction network is obtained, and the maximum mean difference is minimized by gradient descent, so that the target structural feature extraction network and the source structural feature extraction network have statistical similarity.

6. The concrete interface debonding detection method according to any one of claims 3 to 5, characterized in that, The target model feature extraction network and the source model feature extraction network use the same architecture.

7. A concrete interface debonding detection device, based on the concrete interface debonding detection method according to claim 1, characterized in that, The device includes: The first acquisition module is used to acquire a source structure and a target structure, wherein the source structure is a structure that provides debonding detection and diagnostic knowledge, and the target structure is a structure to be detected; The extraction module is used to extract time-frequency domain information of the guided wave signals of the source structure and the target structure by performing wavelet transform. The first training module is used to construct and obtain an artificial neural network based on the time-frequency domain information, and to perform feature learning and parameter optimization on the source model through the artificial neural network to obtain the source structure result prediction network. The second training module is used to perform guided adversarial adaptive training on the target model based on the adversarial alignment and statistical alignment of a pair of data spaces in the domain adaptive domain, so as to obtain the target structural feature extraction network. The debonding diagnosis module is used to combine the source structure result prediction network with the target structure feature extraction network to complete the debonding diagnosis of the guided wave signal of the structure to be detected.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images