A compressive stress / damage identification method and device based on transfer learning admittance features

Abstract

The present application belongs to the technical field of structural health monitoring, and discloses a method and equipment for recognizing compressive stress / damage based on transfer learning admittance features, which comprises the following steps: (1) constructing a structural health monitoring system of a piezoelectric sensing system based on piezoelectric admittance technology, and then collecting original admittance data of the structure under different compressive stress states within a predetermined frequency range; (2) enhancing the original admittance data based on Python language; (3) dividing part of the enhanced admittance data into multiple sub-frequency bands, calculating the RMSD value of each sub-frequency band, performing matrix multiplication on the RMSD value to obtain a similarity matrix, and then remodeling the similarity matrix into a two-dimensional data form; (4) inputting the similarity matrix in the form of two-dimensional data into an RB-DANN network model, and then the RB-DANN network model realizes data transfer between two domains and simultaneously realizes compressive stress / damage recognition; wherein the RB-DANN network model is constructed based on residual principle and field adaptation. The present application reduces the cost.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field related to structural health monitoring, and more particularly, to a compression stress / damage identification method and device based on transfer learning admittance features. BACKGROUND

[0002] In recent years, with the continuous development of isolation technology, base isolation has been widely used in building and bridge structures; it dissipates seismic energy to reduce the seismic response transmitted to the upper structure to protect the corresponding structure. Rubber bearings, with their excellent elasticity, energy dissipation capacity, durability and economy, have become the most commonly used and efficient key components in base isolation. However, when subjected to large axial pressure from the upper structure, the performance of the rubber isolation bearing may be affected. First, the rubber material itself has a certain deformability, and under axial pressure, the rubber isolation bearing may compress and deform, causing the stiffness of the bearing to increase, thereby affecting its original isolation effect. With the continuous action of pressure, the material properties of the rubber may be damaged by fatigue, and even indentations, cracks and other phenomena may occur. In addition, if the axial pressure is too large, exceeding the design load capacity of the rubber isolation bearing, it may cause the rubber bearing to degrade, such as permanent deformation, stiffness reduction, material fatigue, and potential damage to the rubber and steel interface, and in severe cases, the bearing may fail, affecting the stability of the overall structure and even causing structural damage. In actual engineering, most structural damage is a gradual process and cannot be identified by the naked eye. However, as the damage intensifies, the safety, suitability and durability of the structure will be affected, thereby affecting life and property safety. Rubber isolation bearings are concealed works in building engineering, making it more difficult to monitor their damage. Monitoring the compression stress of the rubber bearing is an important measure to reflect its structural performance, and by monitoring the change in compression stress, it can be determined whether the rubber bearing has reached its ultimate bearing capacity and has been damaged, providing effective early warning information and important reference information for post-maintenance and damage warning of the structure.

[0003] The rubber support pressure stress / damage identification based on the piezoelectric admittance (reciprocal of impedance) technology verifies the feasibility of the method by analyzing the electric conductance spectrum curve characteristics and differences of the rubber support under different pressure stresses. The entire piezoelectric sensing collection system is composed of a main structure, piezoelectric material (PZT), an impedance meter and a signal processing computer. The piezoelectric sheet which acts as both a sensor and a brake is pasted on the main structure, and the corresponding excitation voltage is applied through the impedance meter to make the piezoelectric sheet vibrate and thus excite the main structure to vibrate, and the mechanical impedance of the two is coupled. Therefore, the piezoelectric sheet impedance spectrum curve in a certain frequency range can be collected to reflect the changes of the mechanical impedance of the main structure, and thus reflect the changes of the internal performance of the structure, such as stiffness, mass, damping, etc. When analyzing the change characteristics of the electric conductance spectrum under different working conditions, the electric conductance spectrum under the undamaged state is generally selected as the reference standard for pressure stress identification, and the electric conductance spectrum under other working conditions is compared to judge the changes of the internal performance of the main structure, so as to achieve the purpose of monitoring the structure.

[0004] However, this rubber support pressure stress / damage identification method has certain limitations. It only analyzes the pressure stress and damage of the main structure from a qualitative point of view, which often depends on experience and is affected by subjective factors. Due to the large amount of admittance data, it requires a lot of time and labor cost to collect it. In recent years, with the continuous development of deep learning, relevant means can be provided to quantitatively analyze the admittance data and ensure high accuracy. However, for the same structure, the admittance data collected by different piezoelectric sheets requires multiple network models to be trained multiple times to achieve the purpose of stress / damage identification, which reduces the utilization rate of network resources and increases the economic and time cost. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a pressure stress / damage identification method and device based on transfer learning admittance features, which aims to solve the problem that the existing pressure stress / damage identification method requires multiple network models to be trained multiple times for the same structure and different piezoelectric sheets to collect admittance data, resulting in high cost.

[0006] To achieve the above purpose, according to one aspect of the present application, a pressure stress / damage identification method based on transfer learning admittance features is provided, which comprises the following steps:

[0007] (1) Constructing a structural health monitoring system of a piezoelectric sensing system based on piezoelectric admittance technology, and then collecting original admittance data of the structure in a predetermined frequency range under different pressure stress states;

[0008] (2) Enhancing the original admittance data based on Python language;

[0009] (3) The partially enhanced admittance data is divided into multiple sub-frequency bands, the RMSD value of each sub-frequency band is calculated, the RMSD value is subjected to matrix multiplication to obtain a similarity matrix, so as to reshape into a form of two-dimensional data;

[0010] (4) The similarity matrix in the form of two-dimensional data is input into the RB-DANN network model, and then the RB-DANN network model realizes data migration between two domains and simultaneously realizes stress / damage identification; wherein the RB-DANN network model is constructed based on the residual principle and domain adaptation.

[0011] Further, the original admittance data includes source domain data and target domain data; and the predetermined frequency band is 40Hz-500kHz.

[0012] Further, step (2) includes the following sub-steps:

[0013] 2.1, the mean value of the admittance data under each working condition is calculated, and standard admittance data is constructed, and the corresponding calculation formula is:

[0014]

[0015] In the formula, represents the jth dimension value of the ith working condition data of the original admittance data; n represents the number of admittance data under the ith working condition, is the constructed standard admittance data;

[0016] 2.2, the relative difference value and the RMSD value of the existing admittance data under each working condition and the standard admittance data under the working condition are calculated;

[0017] 2.3, the array conforming to the Gaussian random noise is generated based on the original admittance data, and the relative difference value and the RMSD value are multiplied, and finally added to the original admittance data to obtain the enhanced admittance data.

[0018] Further, the corresponding calculation formula of the relative difference value and the RMSD value is:

[0019]

[0020]

[0021] In the formula, is the relative difference value of the jth dimension of the admittance data under the ith working condition; RMSD i represents the mean square deviation under the jth working condition.

[0022] Further, the corresponding calculation formula of the enhanced admittance data is:

[0023] ε i ~ N(0, σ); ε' i= epsilon i X RMSD i

[0024]

[0025] In the formula, epsilon i represents randomly generated noise; epsilon i ' is noise after multiplication with RMSD i X aug represents generated enhanced admittance data.

[0026] Further, the processed source domain data is labeled with 0 to 11, and the target domain data is not labeled, and the data is divided into a training set and a test set according to a ratio of 8:2.

[0027] Further, the RB-DANN network model comprises a feature extraction block, a label prediction block and a domain adaptation block, the feature extraction block comprises four residual blocks, two pooling layers and a flattening layer, the residual block is composed of two paths, one is composed of two convolution layers, the convolution layer 1 is transmitted to the convolution layer 2 through a batch normalization layer and an activation layer, and the convolution layer 2 is transmitted to the pooling layer / output layer through a batch normalization layer and an activation layer; the other is composed of a convolution layer and an activation layer.

[0028] Further, the label prediction block comprises two fully connected layers, the first fully connected layer performs linear transformation on the one-dimensional data output by the feature extraction block, and then performs Relu function activation; the second fully connected layer only performs linear transformation, and then outputs the compressive stress / damage identification result; the domain adaptation block is composed of a domain classifier and a gradient reversal layer, the domain classifier comprises six fully connected layers, the first fully connected layer to the fifth fully connected layer perform linear transformation on the one-dimensional data output by the feature extraction block, and then perform Relu function activation; the sixth fully connected layer only performs linear transformation, and then outputs the domain adaptation result; wherein the residual block is connected through a jump connection.

[0029] The application further provides a compressive stress / damage identification system based on the transfer learning admittance feature, the system comprising a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the compressive stress / damage identification method based on the transfer learning admittance feature.

[0030] The application further provides a computer readable storage medium, the computer readable storage medium stores machine executable instructions, and the machine executable instructions enable the processor to implement the compressive stress / damage identification method based on the transfer learning admittance feature when the machine executable instructions are called and executed by the processor.

[0031] Compared with the prior art, the method and device for recognizing compressive stress / damage based on transfer learning admittance characteristics provided by the present application mainly have the following beneficial effects:

[0032] 1. The present application uses the combination of residual idea and field adaptation idea to process the admittance data collected by piezoelectric sheets in different parts of the same structure, and then realizes the recognition of compressive stress / damage in multiple parts through one training, thereby improving the utilization rate of network resources and reducing economic and time costs.

[0033] 2. The labeled source domain data and unlabeled target domain data are used as inputs, and the compressive stress and the domain adaptation result between them are used as outputs; the RB-DANN network continuously trains and tests among the three blocks to predict the compressive stress / damage of the source domain and target domain data and judge the domain adaptation result; the structural health monitoring method provided by the present application has high economy and practicality, and can accurately recognize the structural performance under different parts of the same structure, making the structural health monitoring intelligent and efficient.

[0034] 3. The original admittance data is enhanced based on Python language, which can be enhanced to multiple groups based on the collection of a group of original admittance data under different working conditions, which is of great significance to save the labor cost generated during the collection of admittance data.

[0035] 4. The method provided by the present application realizes the prediction of compressive stress / damage in multiple parts of the structure through one training on the basis of accurately recognizing the compressive stress borne by the structure, which can greatly improve the network resource utilization rate and reduce the time and economic costs in long-term structural health monitoring.

[0036] 5. The method provided by the present application has good effect for recognizing the compressive stress / damage of rubber supports, can recognize the compressive stress borne by the structure under multiple working conditions, and the residual block is connected through jumping to avoid the problems of gradient explosion and gradient disappearance of the network. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a flowchart of a method for recognizing compressive stress / damage based on transfer learning admittance characteristics provided by the present application;

[0038] Figure 2 is a schematic diagram of the principle of a residual block network structure;

[0039] Figure 3 is a schematic diagram of the overall network architecture of a residual-field adaptation neural network;

[0040] Figure 4 is a schematic diagram of the training and learning of a residual-field adaptation neural network;

[0041] Figure 5 is a schematic diagram of the admittance collection system and result monitoring;

[0042] Figure 6 is a schematic diagram of the comparison of the source domain data lossless condition and other different condition admittance data, wherein (a), (b), (c), (d), (e), (f), (g), (h), (i), (j), (k) correspond to condition 2, condition 3, condition 4, condition 5, condition 6, condition 7, condition 8, condition 9, condition 10, condition 11 and condition 12 respectively;

[0043] Figure 7 is a schematic diagram of the comparison of the target domain data lossless condition and other different condition admittance data, wherein (a), (b), (c), (d), (e), (f), (g), (h), (i), (j), (k) correspond to condition 2, condition 3, condition 4, condition 5, condition 6, condition 7, condition 8, condition 9, condition 10, condition 11 and condition 12 respectively;

[0044] Figure 8 (a), (b) in are schematic diagrams of the damage indicators of the source domain data and the target domain data under the full condition respectively;

[0045] Figure 9 is a schematic diagram of the label predictor training and test loss and accuracy, wherein (a), (b), (c), (d) correspond to the training set label loss, the test set label loss, the training set label accuracy and the test set label accuracy respectively;

[0046] Figure 10 is a schematic diagram of the domain adaptation block training and test loss and accuracy, wherein (a), (b), (c), (d) correspond to the training set domain loss, the test set domain loss, the training set accuracy and the test set accuracy respectively. DETAILED DESCRIPTION

[0047] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0048] Please refer to Figure 1 The present application provides a compressive stress / damage identification method based on transfer learning admittance features, which can accurately identify the compressive stress and damage of two domains at one time of network training, greatly improving the utilization rate of network resources.

[0049] The identification method mainly comprises the following steps:

[0050] Step one, based on piezoelectric admittance technology to construct the structure health monitoring system of piezoelectric sensing system, and then collect the original admittance data of the structure in different pressure stress states in the predetermined frequency range.

[0051] The original admittance data includes source domain data and target domain data. The predetermined frequency range is 40Hz-500kHz. In an embodiment, based on piezoelectric admittance technology to construct the structure health monitoring system of piezoelectric sensing system, under the action of impedance meter 1V excitation voltage, a group of original admittance signals (801 dimensions under each working condition, source domain data and target domain data each 1 group) of the structure in different pressure stress states in the frequency range (40Hz-500kHz) are collected.

[0052] Step two, based on Python language to enhance the original admittance data.

[0053] Step two specifically comprises the following sub-steps:

[0054] 2.1, calculate the mean value of the admittance data under each working condition and construct the standard admittance data, and the corresponding calculation formula is:

[0055]

[0056] In the formula, represents the jth dimension value of the ith working condition data of the original admittance data; n represents the number of admittance data under the ith working condition, is the constructed standard admittance data; i=1, 2…, 12; j=1, 2,…, 801; n=801 in the application.

[0057] 2.2, calculate the relative difference value and RMSD value of the existing admittance data under each working condition and the standard admittance data under the working condition, and the corresponding calculation formula is:

[0058]

[0059] In the formula, is the relative difference value of the jth dimension of the admittance data under the ith working condition; RMSD i represents the mean square deviation under the jth working condition.

[0060] 2.3, based on the original admittance data, generate an array conforming to the Gaussian random noise, multiply the relative difference value and the RMSD value, and finally add the original admittance data to obtain the enhanced admittance data, and the corresponding calculation formula is:

[0061] ε i ~N(0, σ); ε' i = ε iX RMSD i

[0062]

[0063] In the formula, ε i represents a randomly generated noise, and the present embodiment is subject to a Gaussian noise with a mean value of 0 and a standard deviation of 0.00003; ε i is noise multiplied by the RMSD i ; X aug represents the generated enhanced admittance data.

[0064] Step three, divide the part of the enhanced admittance data into multiple sub-frequency bands, calculate the RMSD value of each sub-frequency band, and perform matrix multiplication on the RMSD value to obtain a similarity matrix, so as to reshape it into a two-dimensional data form.

[0065] Specifically, the 50 groups of enhanced admittance data are preprocessed, the unstable frequency bands at the beginning and the end are removed, 784 data are selected therefrom, and the selected data are divided into 56 sub-frequency bands, each of which has 14 data. The RMSD values of the 50 groups of data in the 56 sub-frequency bands are calculated. The similarity matrix is obtained by performing matrix multiplication on the RMSD values, so as to reshape the matrix into a 56x56 form, and the two-dimensional array is used as the input data of the RB-DANN network. The processed source domain data is labeled with 0 to 11, and the target domain data is not labeled. The data is divided into a training set and a test set in a ratio of 8:2.

[0066] Step four, input the similarity matrix in the form of two-dimensional data into the RB-DANN network model, and then the RB-DANN network model realizes data migration between two domains and simultaneously realizes stress / damage identification; wherein the RB-DANN network model is constructed based on the residual principle and domain adaptation.

[0067] The RB-DANN network model includes a feature extraction block, a label prediction block, and a domain adaptation block. The feature extraction block includes four residual blocks, two pooling layers, and a flattening layer. The residual block is composed of two paths. One is composed of two convolution layers. When the convolution layer 1 is transmitted to the convolution layer 2, it passes through a batch normalization layer and an activation layer. When the convolution layer 2 is transmitted to the pooling layer / output layer, there is also a batch normalization layer and an activation layer. The other is composed of a convolution layer and an activation layer.

[0068] The label prediction block includes two fully connected layers. The first fully connected layer performs linear transformation on the one-dimensional data output by the feature extraction block and then performs Relu function activation. The second fully connected layer only performs linear transformation and then outputs the stress / damage identification result.

[0069] The domain adaptation block is composed of a domain classifier and a gradient reversal layer, the domain classifier includes six fully connected layers, the first to fifth fully connected layers perform linear transformation on one-dimensional data output by the feature extraction block and then perform Relu function activation; the sixth fully connected layer only performs linear transformation and then outputs the domain adaptation result.

[0070] The residual block avoids the problems of gradient explosion and gradient disappearance through the jump connection, and the domain adaptation is realized among the feature extraction block, the label prediction block and the domain adaptation block, so as to achieve the purpose of transfer learning. The labeled source domain data and the unlabeled target domain data are input into the feature extraction block to extract domain-invariant features and stress / damage features; the labeled source domain data is predicted in the label prediction block to realize stress / damage identification; the source domain data and the target domain data realize transfer learning in the domain adaptation block based on the gradient reversal layer. The network learning performance and the generalization ability are judged through the label loss, the domain loss, the label accuracy and the domain accuracy, and the RB-DANN network model meeting the requirements is obtained by adjusting the hyperparameters.

[0071] In the RB-DANN network model training and learning process, the network architecture and the data set division method are used, the divided training set and test set are input into the RB-DANN network model, the labeled source domain data and the unlabeled target domain data are first input into the feature extractor for training and testing to extract the domain-invariant features between the two domains; the labeled source domain data is classified by the label predictor to output the label loss and the accuracy; the domain adaptation block uses the domain classifier to distinguish the features of the two domains in the forward propagation process, and uses the gradient reversal layer to confuse the features between the two domains in the backward propagation process, realizes the domain adaptation and outputs the domain loss and the accuracy.

[0072] According to the RB-DANN network training and learning output result, the hyperparameters such as the learning rate, the training round, the batch size and the loss function in the network are modified, and the data is input into the RB-DANN network for training and testing again, when the label accuracy and the domain adaptation accuracy of the RB-DANN network both reach more than 95%, the model ends the training and testing, and the appropriate RB-DANN network model is obtained.

[0073] In one embodiment, first, the original admittance data (divided into conductance data and susceptance data) of different pressure stress conditions of different parts (source domain data and target domain data) of the rubber support is collected by using a piezoelectric sensing collection system, the conductance data is processed and analyzed in the specific embodiment, and the collected original conductance data is data enhanced, 1 group of admittance data under different conditions is enhanced to 50 groups by using the method of generating Gaussian random noise. Then the enhanced conductance data is preprocessed, that is, the conductance data under each condition is divided into sub-frequency bands, and the corresponding RMSD value is calculated, and the admittance data is converted into a two-dimensional matrix by matrix multiplication operation as the input data of the residual-field adaptive neural network; a residual-field adaptive neural network model is constructed, so as to realize feature learning of multiple parts of the structure and automatically identify the pressure stress of the structure and then perform damage identification. The feasibility and accuracy of the rubber support are verified by the limit compression test, and the results show that the embodiment can accurately identify the pressure stress of the rubber support, and has good practical application.

[0074] Table 1 represents the overall architecture parameter setting of the constructed residual-field adaptive neural network (RB-DANN); Table 2 represents the specific parameter setting of the residual block in the feature extractor, and the details are shown in Table 1 and Table 2.

[0075] Table 1 residual-field adaptive neural network architecture parameter setting

[0076]

[0077]

[0078] Table 2 residual block network parameter setting

[0079]

[0080] Figure 2 is a schematic diagram of the principle of the residual block network, which introduces the feature extraction process of the data set based on the residual block skip connection in the present application. Figure 3 is a schematic diagram of the principle of the residual-field adaptive neural network architecture, which introduces the positions of the three blocks in the network architecture of the present application. Figure 4 is the overall flowchart of the training and learning of the residual-field adaptive network to extract features and output results, and the model network architecture is adjusted according to the flowchart to obtain a suitable network model. Figure 5The system is a data collection system for the admittance of rubber bearings under compressive stress. Two piezoelectric sheets are bonded to the rubber bearing, which is placed on a microcomputer-controlled electro-hydraulic servo universal testing machine. The axial pressure is gradually applied to the rubber bearing. At the same time, the admittance signal on the piezoelectric sheet is measured using an Agilent 4294A commercial impedance analyzer. During the test, the rubber bearing shows obvious external morphological changes under the action of axial pressure. With the increase of axial pressure, the height of the bearing gradually decreases, showing obvious compression deformation. At the same time, the lateral size of the bearing expands, especially under the action of larger pressure, the peripheral part of the bearing becomes wider. With the further increase of pressure, the surface of the bearing appears a slight stress concentration area, the local deformation is more significant, and individual areas may appear a slight depression or protrusion, indicating that the bearing material has been subjected to a large stress concentration. Due to the sensitivity of piezoelectric admittance technology to local changes in structure, its conductance spectrum curve can reflect the changes in structural performance.

[0081] According to the specification of “JT / T4-2019-Highway Bridge Plate Rubber Bearing”, the ultimate compressive bearing capacity of rubber is 70 MPa, so 12 working conditions are set for the ultimate compression test of rubber bearing, which are 0 MPa, 15.9 MPa, 31.8 MPa, 47.7 MPa, 63.6 MPa, 79.5 MPa, 95.4 MP, 111.3 MPa, 127.2 MPa, 143.1 MPa, 159 MP, and 174.9 MPa. These working conditions are used to analyze the changes in structural performance of rubber bearings under different compressive stresses and before and after damage. The piezoelectric sheets are bonded to the two sides of the rubber bearing and placed on the microcomputer-controlled electro-hydraulic servo universal testing machine. The original admittance data under different working conditions are obtained by gradually loading according to the above design working conditions, and the admittance data in the frequency band of 40 Hz-500 kHz is selected. The admittance signal points in this frequency band are 801.

[0082] Figure 6 and Figure 7The contrast schematic diagram of the conductance spectrum of the admittance data of the source domain data and the target domain data under the lossless working condition (the compressive stress is 0 MP) and other different working conditions respectively. For the source domain data: (1) the conductance spectrum changes little under working condition 2-working condition 3; the change of the structural performance of the rubber support cannot be seen from the conductance spectrum; (2) the conductance spectrum curve is obviously shifted to the right and up under working condition 4-working condition 12 compared with the conductance spectrum under the lossless state, which indicates that the structural performance of the rubber support changes in the mode of increasing stiffness and decreasing damping under the action of the axial compressive stress, which indicates that the energy dissipation capacity of the rubber support is affected under the action of the compressive stress. For the target domain data: (1) the characteristic change of the conductance spectrum is still small under working condition 2-working condition 3, and the structural performance change of the rubber support cannot be seen from the conductance spectrum; (2) the conductance spectrum curve is also obviously shifted to the right and up under working condition 4-working condition 12, and the reason is consistent with the analysis result of the source domain conductance spectrum curve. It can be seen that the conductance spectrum curve can qualitatively reflect the structural performance change.

[0083] Figure 8 The curve relationship constructed by the conductance data measured under different working conditions of the two piezoelectric sheets is analyzed by using the RMSD and MAPD damage indexes, it can be found that the curve in the figure is in a linear relationship with the corresponding working condition before 79.5 MPa, and the relationship between the damage index and the corresponding working condition tends to be stable or changes abruptly after 79.5 MPa, which can see that the structural performance of the rubber support changes before and after 79.5 MPa, but it is still on the qualitative level of the analysis of the structural performance.

[0084] Figure 9 And Figure 10 The 50 groups of enhanced data are obtained by enhancing the original admittance data of the two piezoelectric sheets of the rubber support under 1 group of different working conditions based on the Gaussian random noise, and the sub-frequency bands under each working condition are preprocessed to obtain 600 groups of 56x56 two-dimensional data as the input data of the RB-DANN network. The input data set is trained and learned by the feature extractor, the label predictor and the domain adapter, and the corresponding damage and accuracy are obtained, the label prediction accuracy of training and testing is 100%; and the domain adaptation accuracy of training and testing can reach 97.29% and 100% respectively, which shows that the RB-DANN network model can better realize label prediction and domain adaptation to achieve the purpose of transfer learning, and the compressive stress / damage identification of different parts of the same structure is realized by one training.

[0085] The application further provides a compressive stress / damage identification system based on transfer learning admittance characteristics, the system comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the compressive stress / damage identification method based on transfer learning admittance characteristics as described above.

[0086] The application further provides a computer readable storage medium, which stores machine executable instructions, and the machine executable instructions, when called and executed by a processor, cause the processor to implement the stress / damage identification method based on the transfer learning mobility feature.

[0087] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for compressive stress / damage identification based on transfer learning admittance features, characterized in that, The method comprises the following steps: (1) Constructing a structural health monitoring system of a piezoelectric sensing system based on a piezoelectric admittance technology, and then collecting original admittance data of a structure in a predetermined frequency range under different pressure stress states; (2) Enhancing the original admittance data based on a Python language; (3) Dividing part of the enhanced admittance data into multiple sub-frequency bands, calculating the RMSD value of each sub-frequency band, performing matrix multiplication on the RMSD value to obtain a similarity matrix, and then remodeling the similarity matrix into a two-dimensional data form; (4) Inputting the similarity matrix in the form of two-dimensional data into an RB-DANN network model, and then the RB-DANN network model realizes data migration between two domains and simultaneously realizes pressure stress / damage identification; wherein the RB-DANN network model is constructed based on a residual principle and a domain adaptation; the two domains are a labeled source domain and an unlabeled target domain; The RB-DANN network model comprises a feature extraction block, a label prediction block and a domain adaptation block; the feature extraction block comprises four residual blocks, two pooling layers and a flattening layer, the label prediction block comprises two fully connected layers, and the domain adaptation block comprises a domain classifier and a gradient reversal layer.

2. The compressive stress / damage identification method based on transfer learning admittance features of claim 1, wherein: The original admittance data comprises source domain data and target domain data; and the predetermined frequency range is 40 Hz-500 kHz.

3. The compressive stress / damage identification method based on transfer learning admittance features of claim 1, wherein: Step (2) comprises the following sub-steps: 2.1, calculating the mean value of the admittance data under each working condition and constructing standard admittance data, and the corresponding calculation formula is: In the formula, represents the original admittance data of the i th dimension value of the j th working condition data; represents the original admittance data of the i th dimension value of the j th working condition data; represents the original admittance data of the i th dimension value of the j th working condition data; represents the original admittance data of the i th dimension value of the j th working condition data; represents the original admittance data of the i th dimension value of the j th working condition data; is the standard admittance data constructed. 2.2, calculating the relative difference value and the RMSD value of the existing admittance data under each working condition and the standard admittance data under the working condition; 2.3, generating an array conforming to the Gaussian random noise based on the original admittance data, multiplying the relative difference value and the RMSD value, and finally adding the original admittance data to obtain the enhanced admittance data.

4. The compressive stress / damage identification method based on transfer learning admittance features of claim 3, wherein: The corresponding calculation formulae of the relative difference value and the RMSD value are: In the formula, is the relative difference of the admittance data of the first dimension under the first working condition; represents the mean square deviation under the first working condition.

5. The compressive stress / damage identification method based on transfer learning admittance features of claim 4, wherein: The corresponding calculation formula of the enhanced admittance data is: In the formula, represents randomly generated noise; is multiplied by noise after multiplication; represents generated enhanced admittance data.

6. The compressive stress / damage identification method based on transfer learning admittance features of claim 2, wherein: The processed source domain data is labeled with 0 to 11, the target domain data is not labeled, and the data is divided into a training set and a test set according to an 8:2 ratio.

7. The compressive stress / damage identification method based on transfer learning admittance features of claim 2, wherein: The residual block is composed of two paths, one of which is composed of two convolutional layers, and the convolutional layer 1 is transmitted to the convolutional layer 2 through a batch normalization layer and an activation layer, and the convolutional layer 2 is transmitted to the pooling layer / output layer through a batch normalization layer and an activation layer; the other one is composed of a convolutional layer and an activation layer.

8. The compressive stress / damage identification method based on transfer learning admittance features of claim 7, wherein: The first fully connected layer of the label prediction block performs linear transformation on the one-dimensional data output by the feature extraction block, and then performs Relu function activation; the second fully connected layer only performs linear transformation, and then outputs the pressure stress / damage identification result; the domain classifier comprises six fully connected layers, the first fully connected layer to the fifth fully connected layer perform linear transformation on the one-dimensional data output by the feature extraction block, and then perform Relu function activation; the sixth fully connected layer only performs linear transformation, and then outputs the domain adaptation result; wherein the residual block is connected through a jump connection.

9. A compressive stress / damage identification system based on transfer learning admittance features, characterized by: The system comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the method for identifying compressive stress / damage based on transfer learning admittance features according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer readable storage medium stores machine executable instructions, and when the machine executable instructions are called and executed by a processor, the machine executable instructions cause the processor to implement the method for identifying compressive stress / damage based on transfer learning admittance features according to any one of claims 1-8.

Images