Pore channel grouting steel strand cross-domain prestress identification method based on deep transfer learning

Abstract

The invention provides a duct grouting steel strand cross-domain prestress recognition method based on deep transfer learning. The method comprises the steps that duct grouting steel strand ultrasonic guided wave signals are collected; the ultrasonic guided wave signal of the duct grouting steel strand is input into a pre-trained deep transfer learning model, a prestress predicted value is obtained, a total loss function in the deep transfer learning model is determined by a regression loss function and a domain identification loss function, the regression loss function is used for determining an error between predicted prestress and real prestress, and the domain identification loss function is used for identifying the actual prestress. The domain identification loss function is used for constraining the feature extractor, so that the feature extractor maximizes the extraction of domain invariant features between the ultrasonic guided-wave signal of the duct grouting steel strand and the ultrasonic guided-wave signal of the bare steel strand acquired in the experiment. By implementing the method, the prestressing force prediction of the duct grouting steel strand can be realized, so that the ultrasonic guided wave signals of the duct grouting steel strand in the in-service prestressed structure under different prestressing forces can be obtained in actual engineering to calibrate the sensitivity coefficient.

Description

Technical Field

[0001] The present invention belongs to the field of structural health monitoring, and particularly relates to a cross-domain prestress identification method for grouted steel strands in ducts based on deep transfer learning. Background Art

[0002] With the increase in the service life of prestressed bridges, the prestress loss of the internal steel strands is inevitable. The prestress loss of the steel strands will increase the cracks and deformations of the bridge. When the prestress loss exceeds the safety limit, there is also a risk of bridge collapse.

[0003] The ultrasonic guided wave technology shows high application potential in the identification of the effective prestress of steel strands. However, there are still some problems to be solved: the current ultrasonic guided wave-based methods all require obtaining signals under different prestress states in advance for the calibration of the sensitivity coefficient. However, for the in-service prestressed structures in actual engineering, it is unrealistic to obtain the ultrasonic guided wave signals of the grouted steel strands in the ducts under different prestress states for the calibration of the sensitivity coefficient. The above problems limit the further application of the ultrasonic guided wave technology in the prestress identification of grouted steel strands in ducts. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a cross-domain prestress identification method and device for grouted steel strands in ducts based on deep transfer learning to meet the need for realizing the prestress prediction of grouted steel strands in ducts.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] According to the first aspect, the present invention provides a cross-domain prestress identification method for grouted steel strands in ducts based on deep transfer learning, including: collecting ultrasonic guided wave signals of grouted steel strands in ducts; inputting the ultrasonic guided wave signals of grouted steel strands in ducts into a pre-trained deep transfer learning model to obtain the prestress prediction value of the grouted steel strands in ducts. Among them, the total loss function in the deep transfer learning model is determined by a regression loss function and a domain discrimination loss function. The regression loss function is used to determine the error between the predicted prestress and the true prestress, and the true prestress is the prestress corresponding to the ultrasonic guided wave signal of the bare steel strand collected in the experiment. The domain discrimination loss function is used to constrain the feature extractor to maximize the extraction of domain-invariant features between the ultrasonic guided wave signals of the grouted steel strands in ducts and the ultrasonic guided wave signals of the bare steel strands collected in the experiment.

[0007] As an optional implementation manner, the total loss function is:

[0008] L = L MSE -βL CE ;

[0009] Wherein, L represents the total loss function, LMSE Denote the regression loss function as \(L\). CE Denote the domain discrimination loss function as \(D\), and \(\beta\) represents the gradient reversal coefficient, which is used to balance the proportions of the regression loss and the domain discrimination loss in the total loss. Let \(n\) denote the number of samples, \(y\) true denote the true prestress, \(\hat{y}\) pre denote the predicted prestress, \(y\) represents the sample source label, taking values of 0 or 1, where 0 represents that the sample is the ultrasonic guided wave signal of the bare steel strand collected by experiment, and 1 represents that the sample is the ultrasonic guided wave signal of the grouted steel strand in the duct. is the probability predicted by the model, ranging between (0, 1). The closer it is to 0, the greater the probability that the sample is the ultrasonic guided wave signal of the bare steel strand collected by experiment.

[0010] As an alternative implementation, the deep transfer learning model includes: a dual-channel input module for inputting the ultrasonic guided wave time-domain signal and the ultrasonic guided wave frequency-domain signal; a convolutional layer for extracting features from the ultrasonic guided wave time-domain signal and the ultrasonic guided wave frequency-domain signal; and an adaptive feature splicing module for splicing the features of the ultrasonic guided wave time-domain signal and the ultrasonic guided wave frequency-domain signal.

[0011] As an alternative implementation, the adaptive feature splicing module performs splicing through the following formula:

[0012] \(G = \text{concat}[\alpha\cdot T, (1 - \alpha)\cdot F]=[\alpha\cdot t\) 1 ,\cdots,\alpha\cdot t\) n ,(1 - \alpha)\cdot f\) 1 ,\cdots,(1 - \alpha)\cdot f\) m ;

[0013] where \(G\) is the new spliced feature with a length of \(m + n\), \(\text{concat}\) is the feature splicing operation, \(\alpha\) is the weight parameter, \(T\) is the time-domain feature vector, \(T = [t\) 1 ,t\) 2 ,\cdots,t\) n , \(t\) n is the eigenvalue of the ultrasonic guided wave time-domain signal feature, \(F\) is the frequency-domain feature vector, \(F = [f\) 1 ,f\) 2 ,\cdots,f\) m , \(f\) m is the eigenvalue of the ultrasonic guided wave frequency-domain signal feature.

[0014] As an alternative implementation, the training process of the deep transfer learning model includes: collecting ultrasonic guided wave signals of post-tensioned tendons in ducts; collecting ultrasonic guided wave signals of bare tendons experimented under different tensile forces and determining the corresponding prestresses, where the total lengths of the post-tensioned tendons in ducts and the bare tendons are the same, and the sampling frequencies and sampling lengths of the ultrasonic guided wave signals are the same; inputting the ultrasonic guided wave signals of the post-tensioned tendons in ducts, the ultrasonic guided wave signals of the bare tendons experimented under different tensile forces, the corresponding prestresses, and the sample source labels into the deep transfer learning model to be trained, where the sample source labels are used to mark whether the input ultrasonic guided wave signals are the ultrasonic guided wave signals of the post-tensioned tendons in ducts or the ultrasonic guided wave signals of the bare tendons; calculating the gradients of the total loss function with respect to the network parameters in the deep transfer learning model according to the chain rule, and updating the network parameters according to the gradients until the end condition is reached, so as to obtain the trained deep transfer learning model.

[0015] As an alternative implementation, the gradient reversal coefficient β = 20.

[0016] According to a second aspect, the present invention provides a cross-domain prestress identification device for post-tensioned tendons in ducts based on deep transfer learning, including: a signal acquisition device for collecting ultrasonic guided wave signals of post-tensioned tendons in ducts; a prediction device for inputting the ultrasonic guided wave signals of the post-tensioned tendons in ducts into a pre-trained deep transfer learning model to obtain a prestress prediction value of the post-tensioned tendons in ducts, where the total loss function in the deep transfer learning model is determined by a regression loss function and a domain discrimination loss function, the regression loss function is used to determine the error between the predicted prestress and the true prestress, the true prestress is the prestress corresponding to the ultrasonic guided wave signals of the bare tendons collected in the experiment, and the domain discrimination loss function is used to constrain the feature extractor to maximize the extraction of domain-invariant features between the ultrasonic guided wave signals of the post-tensioned tendons in ducts and the ultrasonic guided wave signals of the bare tendons collected in the experiment.

[0017] According to a third aspect, an embodiment of the present invention provides an electronic device, the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the steps of the cross-domain prestress identification method for post-tensioned tendons in ducts based on deep transfer learning according to the first aspect or any implementation manner of the first aspect.

[0018] According to a fourth aspect, an embodiment of the present invention provides a computer storage medium, on which a computer instruction is stored, and when the instruction is executed by a processor, the steps of the cross-domain prestress identification method for post-tensioned tendons in ducts based on deep transfer learning according to the first aspect or any implementation manner of the first aspect are implemented.

[0019] An embodiment of the present invention provides a cross - domain prestress identification method for grouted steel strands in ducts based on deep transfer learning. By establishing a deep transfer learning model, the prediction of the prestress of grouted steel strands in ducts is realized, making it possible to obtain ultrasonic guided wave signals of grouted steel strands in in - service prestressed structures under different prestress states in actual engineering for sensitivity coefficient calibration. While learning the mapping relationship between ultrasonic guided wave signals and prestress from source - domain signals and corresponding prestresses, this model extracts domain - invariant features between the source domain and the target domain, and can efficiently realize the prestress prediction of grouted steel strands in ducts.

[0020] Other advantages, objectives, and features of the present invention will be described in the following specification, and to some extent, they are obvious to those skilled in the art, or those skilled in the art can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To make the objectives, technical solutions, and beneficial effects of the present invention clearer, the following drawings are provided for illustration:

[0022] Figure 1 It is a specific example flowchart of a cross - domain prestress identification method for grouted steel strands in ducts based on deep transfer learning in the present invention;

[0023] Figure 2 It is a source - domain and target - domain prestress identification result diagram obtained after removing the domain discriminator from the deep transfer learning model in the present invention;

[0024] Figure 3 It is a source - domain and target - domain prestress identification result diagram obtained by using the deep transfer learning model with the retained domain discriminator in the deep transfer learning model of the present invention;

[0025] Figure 4 It is a target - domain prestress identification error diagram corresponding to removing and retaining the domain discriminator in the deep transfer learning model of the present invention;

[0026] Figure 5 It is an overall process schematic diagram of a cross - domain prestress identification method for grouted steel strands in ducts based on deep transfer learning in the present invention;

[0027] Figure 6 It is a diagram showing the change of the training loss of the deep transfer learning model in the present invention with the number of training rounds;

[0028] Figure 7a It is a diagram showing the change of the RMSE of the target - domain prestress identification result with the gradient reversal coefficient within the search range of {0.01, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100, 500} in the present invention;

[0029] Figure 7b This is a graph showing the variation of the RMSE of the prestress identification results in the target domain with the gradient reversal coefficient within the search range of {10, 20, 30, 40, 50} for the present invention;

[0030] Figure 8 This is a principle block diagram of a specific example of the electronic device in the embodiment of the present invention. Detailed implementation manners

[0031] Next, the technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] In the description of the present invention, it should be noted that, unless otherwise clearly defined and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be a direct connection or an indirect connection through an intermediate medium, and it may also be the communication inside two components. It may be a wireless connection or a wired connection. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0033] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0034] The embodiment of the present invention provides a method for identifying the cross-domain prestress of the duct grouting steel strand based on deep transfer learning, as Figure 1 shown, including:

[0035] S101, collecting the ultrasonic guided wave signals of the duct grouting steel strand;

[0036] S102, inputting the ultrasonic guided wave signals of the duct grouting steel strand into a pre-trained deep transfer learning model to obtain the prestress prediction value of the duct grouting steel strand. Among them, the total loss function in the deep transfer learning model is determined by a regression loss function and a domain discrimination loss function. The regression loss function is used to determine the error between the predicted prestress and the true prestress, and the true prestress is the prestress corresponding to the ultrasonic guided wave signal of the bare steel strand collected by the experiment. The domain discrimination loss function is used to constrain the feature extractor to maximize the extraction of domain-invariant features between the ultrasonic guided wave signals of the duct grouting steel strand and the ultrasonic guided wave signals of the bare steel strand collected by the experiment.

[0037] Exemplarily, the ultrasonic guided wave signals of the grouted prestressed steel strands in ducts can be collected by an ultrasonic guided wave measuring device. Since the grouted prestressed steel strands in ducts are in-service structures, the prestress corresponding to the ultrasonic guided wave signals of the grouted prestressed steel strands in ducts is unknown. The present invention proposes a deep transfer learning model. By inputting the ultrasonic guided wave signals of the grouted prestressed steel strands in ducts into the pre-trained deep transfer learning model, the corresponding prestress prediction value can be obtained. The construction idea of the deep transfer learning model is as follows:

[0038] Since it is unrealistic to obtain the ultrasonic guided wave signals of the grouted prestressed steel strands in ducts inside actual projects under different prestress states. And in the laboratory, the bare steel strands are tensioned in stages, and according to the ultrasonic guided wave signals corresponding to the tensile forces collected by the ultrasonic guided wave measuring device, while collecting the ultrasonic guided wave signals, the prestress magnitudes corresponding to the ultrasonic guided wave signals of the bare steel strands under different tensile forces can be calculated according to the tensile forces. It can be seen that the prestress magnitudes corresponding to the ultrasonic guided wave signals of the bare steel strands under different tensile forces are known. For the sake of convenience of description, a bare steel strand is tensioned in stages in the laboratory, and after each stage of tensioning is completed, the ultrasonic guided wave signals corresponding to the tensile force are collected by the ultrasonic guided wave measuring device as the source domain signals. Therefore, in the deep transfer learning model of this embodiment, the source domain signals and the prestress corresponding to the source domain signals are used as one of the training data. At the same time, the ultrasonic guided wave signals of the grouted prestressed steel strands in ducts in an unknown prestress state are collected by the same ultrasonic guided wave measuring device and used as the target domain signals. It should be noted that the total lengths of the bare steel strands and the grouted prestressed steel strands in ducts are the same, and the sampling frequencies and sampling lengths of the ultrasonic guided wave signals are also the same. The source domain signals, the corresponding prestress, and the target domain signals are input into the deep transfer learning model, and the model is trained to obtain a trained deep transfer learning model so that the model can predict the corresponding prestress according to the collected ultrasonic guided wave signals of the grouted prestressed steel strands in ducts.

[0039] Specifically, in order to enable this deep transfer learning model to learn the mapping relationship between the ultrasonic guided wave signals and the prestress from the source domain signals and the corresponding prestress, and at the same time be able to extract the domain-invariant features between the source domain and the target domain, two target tasks need to be executed in this deep transfer learning model. The first is the regression task, which is used to reduce the error between the predicted prestress and the true prestress. The true prestress is the prestress corresponding to the ultrasonic guided wave signals of the bare steel strands collected in the experiment. The second task is the domain discrimination task, which constrains the feature extractor through adversarial learning to extract the domain-invariant features between the source domain and the target domain as much as possible. The above two target tasks are reflected in the total loss function of the model. The total loss function is composed of a regression loss function and a domain discrimination loss function. The regression loss function is used to complete the regression task, and the domain discrimination loss function is used to complete the domain discrimination task.

[0040] As an alternative implementation, the total loss function is:

[0041] L = L MSE -βL CE ;

[0042] Among them, L represents the total loss function, L MSE represents the regression loss function, and L CE represents the domain discriminant loss function. β represents the gradient reversal coefficient, which is used to balance the proportions of the regression loss and the domain discriminant loss in the total loss. In the present invention, β is determined by repeated trials, n represents the number of samples, and y true represents the true prestress, and y pre represents the predicted prestress. y represents the sample source label, which is taken as 0 or 1, where 0 represents that the sample is the ultrasonic guided wave signal of the bare steel strand collected in the experiment, and 1 represents that the sample is the ultrasonic guided wave signal of the grouted steel strand in the duct, is the probability predicted by the model, and the range is between (0, 1). The closer it is to 0, the greater the probability that the sample is the ultrasonic guided wave signal of the bare steel strand collected in the experiment.

[0043] After the deep transfer learning model inputs the ultrasonic guided wave signals of the bare steel strand under different tensile forces, the corresponding prestresses, and the ultrasonic guided wave signals of the grouted steel strand in the duct, on the one hand, it can learn the mapping relationship between the ultrasonic guided wave signal of the bare steel strand and the corresponding prestress, and on the other hand, it can learn the domain-invariant features between the ultrasonic guided wave signal of the bare steel strand and the ultrasonic guided wave signal of the grouted steel strand in the duct. Although there are differences in the structure and acoustic propagation characteristics between the grouted steel strand in the duct and the bare steel strand, there is a certain similarity between them. The deep transfer learning model can partially transfer the knowledge of the mapping relationship between the ultrasonic guided wave signal of the bare steel strand and the prestress learned to the situation of the grouted steel strand in the duct. When the ultrasonic guided wave signal of the grouted steel strand in the duct is input again, the model can match and associate it with the feature pattern learned from the bare steel strand data based on the domain-invariant features already learned. And because the domain-invariant features have cross-domain universality, the model can use these features to infer the prestress corresponding to the ultrasonic guided wave signal of the grouted steel strand in the duct.

[0044] In order to verify the cross-domain prestress identification performance of the deep transfer learning model proposed in the present invention, the prestress identification result of the target domain based on this model is compared with the model that only uses a regressor without using a domain discriminator. This model only removes the domain discriminator based on the deep transfer learning model. As Figure 2 and Figure 3 are the prestress identification results of the source domain and the target domain corresponding to the cases of removing and retaining the domain discriminator respectively. For the case of removing the domain discriminator, as Figure 2As shown, the predicted value in the target domain deviates significantly from the true value, indicating that after removing the domain discriminator, the predicted value in the source domain is very close to the true value while the predicted value in the target domain deviates significantly from the true value. This shows that after removing the domain discriminator, the model does not exhibit cross-domain prestress identification ability, but only shows good prediction performance in the source domain and poor performance in the target domain. For the deep transfer learning model with the domain discriminator retained, such as Figure 3 As shown, the predicted values corresponding to both source domain samples and target domain samples are relatively close to the true values, indicating that the addition of the domain discriminator can effectively reduce the feature differences between the source domain and the target domain, and the deep transfer learning model exhibits good cross-domain prestress identification performance. The target domain prestress identification errors corresponding to removing and retaining the domain discriminator are as shown in Figure 4 As shown, the target domain prestress identification error corresponding to the deep transfer learning model is only 0.0505, while the target domain prestress identification error corresponding to the case of removing the domain discriminator is as high as 0.1319.

[0045] An embodiment of the present invention provides a cross-domain prestress identification method for post-tensioned steel strands in duct grouting based on deep transfer learning. By establishing a deep transfer learning model, the prediction of the prestress of post-tensioned steel strands in duct grouting is realized, making it possible to obtain ultrasonic guided wave signals of post-tensioned steel strands in in-service prestressed structures under different prestress states in actual engineering for sensitivity coefficient calibration. While learning the mapping relationship between ultrasonic guided wave signals and prestress from source domain signals and corresponding prestresses, this model can extract domain-invariant features between the source domain and the target domain, and can efficiently realize the prediction of the prestress of post-tensioned steel strands in duct grouting.

[0046] As an optional implementation manner, the deep transfer learning model includes: a dual-channel input module for inputting ultrasonic guided wave time-domain signals and ultrasonic guided wave frequency-domain signals; a convolutional layer for extracting features from the ultrasonic guided wave time-domain signals and ultrasonic guided wave frequency-domain signals; and an adaptive feature splicing module for splicing the features of the ultrasonic guided wave time-domain signals and ultrasonic guided wave frequency-domain signals.

[0047] Among them, the dual-channel input module is used to input ultrasonic guided wave time-domain signals and ultrasonic guided wave frequency-domain signals. Then, the signal features output from the time-domain and frequency-domain channels can be respectively extracted through convolution, and then the features are flattened to obtain the flattened time-domain feature vector T = [t 1 , t 2 ,... t n and the frequency-domain feature vector F = [f 1 , f 2 ,... f m, where t and f represent the eigenvalues in the feature vectors, n and m represent the number of eigenvalues in the time-domain and frequency-domain feature vectors respectively. Then, a weight parameter α that can be automatically updated during training is constructed. The adaptive feature concatenation of the time-domain and frequency-domain features can be expressed as:

[0048] G = contat[α·T, (1 - α)·F] = [α·t 1 ,..., α·t n , (1 - α)·f 1 ,...,(1 - α)·f m ;

[0049] where G is the new concatenated feature with a length of m + n, contat is the feature concatenation operation, α is the weight parameter, T is the time-domain feature vector, T = [t 1 , t 2 ,...t n , t n is the eigenvalue of the ultrasonic guided wave time-domain signal, F is the frequency-domain feature vector, F = [f 1 , f 2 ,...f m , f m is the eigenvalue of the ultrasonic guided wave frequency-domain signal.

[0050] The embodiment of the present invention provides a method for cross-domain prestress identification of grouted steel strands in ducts based on deep transfer learning, which introduces a dual-channel input and an adaptive feature concatenation strategy to achieve effective concatenation of the time-domain and frequency-domain features of ultrasonic guided wave signals. Through the dual-channel input module, the model can process time-domain and frequency-domain signals simultaneously, making full use of the information of these two modalities, so as to more comprehensively characterize the characteristics of the signals. Through feature concatenation, the model can integrate these two kinds of information, avoid the limitations of single-modal information, and improve the accuracy and robustness of signal analysis.

[0051] As an alternative embodiment, the training process of the deep transfer learning model, as Figure 5 shown, includes:

[0052] S001, collect ultrasonic guided wave signals of grouted steel strands in ducts;

[0053] S002, collect ultrasonic guided wave signals of bare steel strands tested under different tensions and determine the corresponding prestresses. Among them, the total lengths of the grouted steel strands in ducts and the bare steel strands are the same, and the sampling frequencies and sampling lengths of the ultrasonic guided wave signals are the same;

[0054] S003, input the ultrasonic guided wave signals of the post-tensioned strands in grouted ducts, the ultrasonic guided wave signals of the bare strands tested under different tensions, the corresponding prestresses, and the sample source labels into the deep transfer learning model to be trained, where the sample source labels are used to mark whether the input ultrasonic guided wave signals are the ultrasonic guided wave signals of the post-tensioned strands in grouted ducts or the ultrasonic guided wave signals of the bare strands;

[0055] S004, calculate the gradients of the total loss function with respect to the network parameters in the deep transfer learning model according to the chain rule, and update the network parameters according to the gradients until the end condition is reached, obtaining the trained deep transfer learning model.

[0056] Exemplarily, input the ultrasonic guided wave signals of the post-tensioned strands in grouted ducts, the ultrasonic guided wave signals of the bare strands tested under different tensions, the corresponding prestresses, and the sample source labels into the deep transfer learning model to be trained, where the sample source labels are taken as 0 or 1, where 0 represents that the input ultrasonic guided wave signal (sample) is the ultrasonic guided wave signal of the bare strands collected in the experiment, and 1 represents that the input ultrasonic guided wave signal (sample) is the ultrasonic guided wave signal of the post-tensioned strands in grouted ducts. Then, use the chain rule to calculate the gradients of the total loss function with respect to the network parameters, and update the network parameters through backpropagation until one of the following conditions is met, and the training is completed, obtaining the trained deep transfer learning model: (1) the loss function converges to a target value and no longer decreases significantly, and the target value can be 0.001; (2) the preset number of training iterations is reached.

[0057] The change of the training loss of the deep transfer learning model is as Figure 6 shown. In the first 25 rounds of training, the domain discrimination loss gradually increases while the regression loss gradually decreases, which indicates that the features of the source domain and the target domain are becoming increasingly difficult to distinguish, and their distributions are becoming closer and closer. At the same time, the model can perform the prediction task well according to the transferred source domain features. Correspondingly, the total loss also gradually decreases in the first 25 rounds of training. After 25 rounds of training, the downward trend of the total loss tends to level off. Finally, the total loss stabilizes at a very small value, indicating that the deep transfer learning model proposed by the present invention has good training stability.

[0058] As an alternative implementation, the gradient reversal coefficient β = 20. In the present invention, β is determined by repeated attempts. As shown in Figure 7, the target domain prediction error changes with the gradient reversal coefficient β. First, the search range of β is set as {0.01, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100, 500}. Within the above search range, the change of the RMSE of the target domain prestress identification result with the gradient reversal coefficient is as Figure 7a shown, and the errors corresponding to the cases where β is 10 and 50 are the smallest. Therefore, in the subsequent search, the search range is set as {10, 20, 30, 40, 50}. AsFigure 7b As shown, the β corresponding to the minimum error is 20. Therefore, in this embodiment, the optimal gradient reversal coefficient β in the deep transfer learning model is set to 20.

[0059] The present invention provides a cross-domain prestress identification device for grouted steel strands in ducts based on deep transfer learning, comprising:

[0060] A signal acquisition device for acquiring ultrasonic guided wave signals of grouted steel strands in ducts; for the specific process, refer to the corresponding part of the above method embodiment and will not be elaborated here.

[0061] A prediction device for inputting the ultrasonic guided wave signals of grouted steel strands in ducts into a pre-trained deep transfer learning model to obtain a prestress prediction value of the grouted steel strands in ducts. Among them, the total loss function in the deep transfer learning model is determined by a regression loss function and a domain discrimination loss function. The regression loss function is used to determine the error between the predicted prestress and the true prestress, and the true prestress is the prestress corresponding to the ultrasonic guided wave signal of the bare steel strand collected in the experiment. The domain discrimination loss function is used to constrain the feature extractor to maximize the extraction of domain-invariant features between the ultrasonic guided wave signals of grouted steel strands in ducts and the ultrasonic guided wave signals of the bare steel strands collected in the experiment. For the specific process, refer to the corresponding part of the above method embodiment and will not be elaborated here.

[0062] The embodiment of the present application also provides an electronic device, such as Figure 8 shown, a processor 501 and a memory 502, where the processor 501 and the memory 502 can be connected through a bus or other means.

[0063] The processor 501 can be a central processing unit (CPU). The processor 501 can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or a combination of the above types of chips.

[0064] The memory 502, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for identifying cross-domain prestress of grouted steel strands in ducts based on deep transfer learning in the embodiments of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory.

[0065] The memory 502 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory 502 may optionally include a memory remotely provided with respect to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0066] The one or more modules are stored in the memory 502 and, when executed by the processor 501, execute the method for identifying cross-domain prestress of grouted steel strands in ducts based on deep transfer learning in the embodiments Figure 1 as shown.

[0067] For specific details of the above electronic device, reference can be made to Figure 1 the corresponding relevant descriptions and effects in the shown embodiments, which will not be elaborated here.

[0068] This embodiment also provides a computer storage medium, and the computer storage medium stores computer-executable instructions that can execute the method for identifying cross-domain prestress of grouted steel strands in ducts based on deep transfer learning in any of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disc, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (abbreviation: HDD), or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memories.

[0069] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A cross-domain prestress identification method for duct grouting steel strand based on deep transfer learning, characterized in that: include: Collect ultrasonic guided wave signals of steel strands for grouting in holes; The ultrasonic guided wave signal of the duct-grouted steel strand is input into the pre-trained deep transfer learning model to obtain the predicted value of the prestress of the duct-grouted steel strand. The total loss function in the deep transfer learning model is determined by the regression loss function and the domain identification loss function. The regression loss function is used to determine the error between the predicted prestress and the actual prestress. The actual prestress is the prestress corresponding to the ultrasonic guided wave signal of the bare steel strand collected in the experiment. The domain identification loss function is used to constrain the feature extractor so that the feature extractor maximizes the extraction of domain-invariant features between the ultrasonic guided wave signal of the duct-grouted steel strand and the ultrasonic guided wave signal of the bare steel strand collected in the experiment.

2. According to the method of cross-domain prestress identification of duct grouting steel strand based on deep transfer learning in claim 1, it is characterized in that: The total loss function is: L=L MSE -βL CE ; Among them, L represents the total loss function, L MSE represents the regression loss function, L CE represents the domain identification loss function, β represents the gradient reversal coefficient, which is used to balance the proportion of regression loss and domain identification loss in the total loss. n represents the number of samples, yt rue represents the real prestress, y pre represents the predicted prestress, y represents the sample source label, which is 0 or 1, where 0 represents that the sample is the ultrasonic guided wave signal of bare steel strand collected in the experiment, and 1 represents that the sample is the ultrasonic guided wave signal of grouted steel strand. is the probability predicted by the model, ranging from (0, 1). The closer it is to 0, the greater the probability that the sample is an ultrasonic waveguide signal of bare steel strands collected in the experiment.

3. According to the method of cross-domain prestress identification of duct grouting steel strand based on deep transfer learning in claim 1, it is characterized in that: Deep transfer learning models include: A dual-channel input module for inputting ultrasonic guided wave time domain signals and ultrasonic guided wave frequency domain signals; Convolutional layer, used to extract features from ultrasonic guided wave time domain signals and ultrasonic guided wave frequency domain signals; The adaptive feature stitching module is used to perform feature stitching on the ultrasonic guided wave time domain signal and the ultrasonic guided wave frequency domain signal.

4. According to the method of cross-domain prestress identification of duct grouting steel strand based on deep transfer learning in claim 3, it is characterized in that: The adaptive feature splicing module performs splicing through the following formula: G=contat[a·T,(1-a)·F]=[a·t1,...,a·t n ,(1-a)·f1,...,(1-a)·f m ]; Among them, G is the new feature after splicing, with a length of m+n, contat is the feature splicing operation, α is the weight parameter, T is the time domain feature vector, T=[t1,t2,...t n ], t n is the characteristic value of ultrasonic guided wave signal in time domain, F is the characteristic vector in frequency domain, F=[f1,f2,...f m ], f m is the characteristic value of ultrasonic guided wave frequency domain signal.

5. A method for identifying cross-domain prestressing of duct grouting steel strands based on deep transfer learning according to any one of claims 1 to 4, characterized in that: The training process of the deep transfer learning model includes: Collect ultrasonic guided wave signals of duct-grouted steel strands, collect ultrasonic guided wave signals of bare steel strands tested with different tensile forces and determine the corresponding prestress, wherein the total length of the duct-grouted steel strands is the same as that of the bare steel strands, and the sampling frequency and sampling length of the ultrasonic guided wave signals are consistent; The ultrasonic guided wave signal of the duct-grouted steel strand, the ultrasonic guided wave signal of the bare steel strand tested with different tensile forces, the corresponding prestress, and the sample source label are input into the deep transfer learning model to be trained, wherein the sample source label is used to mark the input ultrasonic guided wave signal as the ultrasonic guided wave signal of the duct-grouted steel strand or the ultrasonic guided wave signal of the bare steel strand; The gradient of the total loss function to the network parameters in the deep transfer learning model is calculated according to the chain rule, and the network parameters are updated according to the gradient until the end condition is reached to obtain a trained deep transfer learning model.

6. According to a method for identifying cross-domain prestressing of duct grouting steel strands based on deep transfer learning in claim 2, it is characterized in that: The gradient reversal coefficient β=20.

7. A cross-domain prestressed steel strand identification device for grouting in a hole based on deep transfer learning, characterized in that: include: A signal acquisition device for collecting ultrasonic guided wave signals of the duct grouting steel strand; A prediction device is used to input the ultrasonic guided wave signal of the duct grouting steel strand into a pre-trained deep transfer learning model to obtain the prestress prediction value of the duct grouting steel strand, wherein the total loss function in the deep transfer learning model is determined by the regression loss function and the domain identification loss function, the regression loss function is used to determine the error between the predicted prestress and the actual prestress, the actual prestress is the prestress corresponding to the ultrasonic guided wave signal of the bare steel strand collected in the experiment, and the domain identification loss function is used to constrain the feature extractor so that the feature extractor maximizes the extraction of domain invariant features between the ultrasonic guided wave signal of the duct grouting steel strand and the ultrasonic guided wave signal of the bare steel strand collected in the experiment.

8. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the method for identifying cross-domain prestressing of duct-grouted steel strands based on deep transfer learning as described in any one of claims 1 to 6.

9. A computer storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by the processor, the steps of the cross-domain prestressing identification method of duct grouting steel strands based on deep transfer learning as described in any one of claims 1-6 are implemented.

Images