A multi-fidelity data fusion method, device, equipment, medium and product

Abstract

The present application discloses a multi-fidelity data fusion method, apparatus, device, medium and product, relating to the field of structural health monitoring. The method includes conducting a structural damage expansion test on a specimen to obtain a set of damage state parameter values and a first set of structural response data; conducting a static simulation of the above test on each simulation model in the first set of simulation models to obtain a second set of structural response data; conducting a static simulation of the above test on each simulation model in the second set of simulation models to obtain a third set of structural response data; training a first neural network according to the third set of structural response data and each preset damage state parameter value to obtain a low-fidelity network; training a second neural network according to the first and second sets of structural response data and the damage state parameter values at each moment to obtain a feature mapping network; and constructing a structural damage assessment model based on the trained networks. The present application is applicable to the training of a structural damage assessment model.

Description

Technical Field

[0001] The present application relates to the field of structural health monitoring, and particularly to a multi-fidelity data fusion method, device, equipment, medium and product. Background Art

[0002] Structural health monitoring technology uses the monitoring data collected during the operation of a pre-established structural damage assessment model to evaluate the damage state of a structure. The training of such structural damage assessment models requires a large amount of reliable high-fidelity damage data as support. However, for large and complex structures, it is unrealistic to carry out a large number of destructive damage tests to obtain high-fidelity damage data for training the model. In contrast, damage simulation based on a structural model is more feasible and can generate a large amount of damage data in a short time at a low cost. However, the fidelity of these simulated damage data is relatively low, resulting in the problems of lack of high-fidelity data and low accuracy of low-fidelity data, which is not conducive to the training of structural damage assessment models.

[0003] Currently, existing multi-fidelity data fusion algorithms are usually applied to model training in forward-solving problems. However, the structural damage assessment problem belongs to an inverse problem, and existing multi-fidelity data fusion algorithms are not applicable to the training of structural damage assessment models. Summary of the Invention

[0004] The purpose of the present application is to provide a multi-fidelity data fusion method, device, equipment, medium and product, which is applicable to the training of structural damage assessment models.

[0005] To achieve the above purpose, the present application provides the following solutions.

[0006] In a first aspect, the present application provides a multi-fidelity data fusion method, including: performing a structural damage expansion test on a specimen, and collecting the damage state parameter values of the specimen at each moment during the structural damage expansion test and the structural response data of a preset position of the specimen under each damage state parameter value, to obtain a damage state parameter value set and a first structural response data set.

[0007] Performing a static simulation of the structural damage expansion test on each simulation model in the first simulation model set, and obtaining the structural response data of a preset position of each simulation model in the first simulation model set, to obtain a second structural response data set; the first simulation model set includes multiple simulation models obtained by respectively processing the simulation model of the specimen according to each damage state parameter value in the damage state parameter value set.

[0008] Perform static simulation on the structural damage expansion test of each simulation model in the second simulation model set, obtain the structural response data at the preset positions of each simulation model in the second simulation model set, and obtain the third structural response data set; the second simulation model set includes multiple simulation models obtained by processing the simulation model of the sample according to each preset damage state parameter value.

[0009] Train the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network.

[0010] Train the second neural network according to the first structural response data set, the second structural response data set, and the damage state parameter values at each moment to obtain a feature mapping network.

[0011] Construct a structural damage assessment model based on the low-fidelity network and the feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence.

[0012] Optionally, training the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network specifically includes: using the structural response data in the third structural response data set as the input, using each preset damage state parameter value as the output, and aiming at minimizing the value of the first loss function, training the first neural network to obtain a low-fidelity network.

[0013] Optionally, training the second neural network according to the first structural response data set, the second structural response data set, and the damage state parameter values at each moment to obtain a feature mapping network specifically includes: constructing an overall network; the overall network includes a second neural network and a low-fidelity network connected in sequence.

[0014] Using the structural response data in the first structural response data set as the input of the second neural network, using the structural response data in the second structural response data set as the output of the second neural network, using the damage state parameter values at each moment as the output of the low-fidelity network, and aiming at minimizing the value of the second loss function, training the second neural network in the overall neural network to obtain a feature mapping network.

[0015] Optionally, the first loss function is: , where represents the total number of preset damage state parameter values, represents the output obtained by inputting the th structural response data in the third structural response data set into the first neural network, represents the preset damage state parameter value corresponding to the th structural response data in the third structural response data set, Represents the square of the absolute value calculation.

[0016] Optionally, the second loss function is: , where represents the total number of damage state parameter values at each moment collected, represents the output obtained by inputting into the low-fidelity network, the damage state parameter value corresponding to the th structural response data in the first structural response data set, represents the output obtained by inputting the th structural response data in the first structural response data set into the second neural network, the th structural response data in the second structural response data set,

[0017] Optionally, when the damage state parameter value is the crack length, a structural damage expansion test is performed on the specimen, and the damage state parameter values at each moment on the specimen during the structural damage expansion test and the structural response data at the preset position of the specimen under each damage state parameter value are collected to obtain a damage state parameter value set and a first structural response data set. Specifically: a structural crack expansion test is performed on the specimen, and the crack lengths at each moment on the specimen during the structural crack expansion test and the structural response data at the preset position of the specimen under each crack length are collected to obtain a damage state parameter value set and a first structural response data set.

[0018] In a second aspect, the present application provides a multi-fidelity data fusion device, including: a test module for performing a structural damage expansion test on a specimen and collecting the damage state parameter values at each moment on the specimen during the structural damage expansion test and the structural response data at the preset position of the specimen under each damage state parameter value to obtain a damage state parameter value set and a first structural response data set.

[0019] A first simulation module for performing a static simulation of the structural damage expansion test on each simulation model in the first simulation model set to obtain the structural response data at the preset position of each simulation model in the first simulation model set, thereby obtaining a second structural response data set; the first simulation model set includes multiple simulation models obtained by respectively processing the simulation model of the specimen according to each damage state parameter value in the damage state parameter value set.

[0020] The second simulation module is used to perform static simulation on the structural damage expansion test of each simulation model in the second simulation model set, obtain the structural response data at the preset positions of each simulation model in the second simulation model set, and obtain the third structural response data set; the second simulation model set includes multiple simulation models obtained by processing the simulation model of the sample according to each preset damage state parameter value.

[0021] The first training module is used to train the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network.

[0022] The second training module is used to train the second neural network according to the first structural response data set, the second structural response data set, and the damage state parameter values at each moment to obtain a feature mapping network.

[0023] The structural damage assessment model construction module is used to construct a structural damage assessment model according to the low-fidelity network and the feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence.

[0024] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the multi-fidelity data fusion method described in any one of the above.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the multi-fidelity data fusion method described in any one of the above.

[0026] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the multi-fidelity data fusion method described in any one of the above.

[0027] According to the specific embodiments provided by this application, this application has the following technical effects: This application provides a multi-fidelity data fusion method, device, equipment, medium and product. In the forward problem, the data features of the high-fidelity and low-fidelity data sets used for model training have the same spatial distribution; while in the structural damage assessment problem, the data features of the high-fidelity and low-fidelity data sets used for model training often have different spatial distributions. Existing multi-fidelity data fusion algorithms are difficult to effectively handle the differences in the spatial distribution of high-fidelity and low-fidelity data features, resulting in their difficulty in being directly applied to the training of structural damage assessment models. The structural response data obtained from experiments is high-fidelity data, and the data obtained from simulations is low-fidelity data. The feature mapping network of this application is trained using high-fidelity structural response data from experiments and low-fidelity structural response data from simulations with the same damage state parameter values, and can realize the mapping of the feature space of high-fidelity data to the feature space of low-fidelity data, solving the problem that existing multi-fidelity data fusion algorithms are difficult to effectively handle the differences in the spatial distribution of high-fidelity and low-fidelity data features, and is applicable to the training of structural damage assessment models. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0029] Figure 1 It is a flowchart of the multi-fidelity data fusion method provided by an embodiment of this application.

[0030] Figure 2 It is a schematic diagram of the principle of the multi-fidelity data fusion method provided by an embodiment of this application.

[0031] Figure 3 It is a schematic diagram of the first neural network structure provided by this application.

[0032] Figure 4 It is a schematic diagram of the second neural network structure provided by this application.

[0033] Figure 5 It is a structural diagram of the test piece provided by an embodiment of this application.

[0034] Figure 6 It is a diagram showing the error reduction during the training process of the first neural network provided by an embodiment of this application.

[0035] Figure 7 It is a diagram of the prediction results of the first neural network in the test set provided by an embodiment of this application.

[0036] Figure 8It is the error decline graph during the training process of the second neural network in an embodiment of the present application.

[0037] Figure 9 It is the prediction result graph of the damage state parameters in the structural damage assessment scenario by simulating the multi-fidelity data fusion method provided in an embodiment of the present application.

[0038] Figure 10 It is the neural network architecture graph for comparing model effects.

[0039] Figure 11 It is the prediction result graph of each comparison model.

[0040] Figure 12 It is the structural schematic diagram of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0041] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0042] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0043] In an exemplary embodiment, as Figure 1 shown, a multi-fidelity data fusion method is provided, including the following steps 201 to step 206.

[0044] Step 201: Conduct a structural damage expansion test on the sample piece, and collect the damage state parameter values of the sample piece at each moment during the structural damage expansion test and the structural response data at the preset positions of the sample piece under each damage state parameter value , to obtain a damage state parameter value set and a first structural response data set. Specifically, multiple groups of high-fidelity data are obtained by conducting a structural damage expansion test on a sample with a damaged structure. The high-fidelity data includes damage state parameter values and corresponding structural response data.

[0045] Step 202: Conduct a static simulation of the structural damage expansion test on each simulation model in the first simulation model set, and obtain the structural response data at the preset positions of each simulation model in the first simulation model set , a second structural response data set is obtained; the first simulation model set includes multiple simulation models obtained by processing the simulation model of the sample according to each damage state parameter value in the damage state parameter value set. The second structural response data set, the damage state parameter value set, and the first structural response data set form a multi-fidelity data set. Specifically, the loads corresponding to the damage state parameter values at each moment in step 201 are sampled, and a static simulation of the structure with damage states is carried out. During the static simulation, the structural damage parameter values in the simulation model are the same as those measured in the experiment, and the magnitudes of the static loads in the simulation correspond one-to-one with the loads when the structural damage parameter values are collected in the experiment.

[0046] Step 203: Perform a static simulation of the structural damage propagation test on each simulation model in the second simulation model set to obtain the structural response data at the preset positions of each simulation model in the second simulation model set , a third structural response data set is obtained; the second simulation model set includes multiple simulation models obtained by processing the simulation model of the sample according to each preset damage state parameter value respectively. The third structural response data set and each preset damage state parameter value form a low-fidelity data set. Specifically, global sampling is carried out within the reasonable value range of the damage state parameters to obtain preset damage state parameter values, and the corresponding low-fidelity structural response data are obtained through simulation.

[0047] Step 204: Train the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network.

[0048] Step 205: Train the second neural network according to the first structural response data set, the second structural response data set, and the damage state parameter values at each moment to obtain a feature mapping network.

[0049] Step 206: Construct a structural damage assessment model according to the low-fidelity network and the feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence. By taking the structural response data obtained by the airborne health monitoring sensor as the high-fidelity signal features and inputting them into the structural damage assessment model, the structural damage assessment can be realized.

[0050] Implementing the above steps 201 to 206 can be used for the training of the structural damage assessment model.

[0051] In another exemplary embodiment of the present application, a low-fidelity network is obtained by training a first neural network according to a third structural response data set and each preset damage state parameter value, which specifically includes: using the structural response data in the third structural response data set as the input, using each preset damage state parameter value as the output, and training the first neural network with the goal of minimizing the value of a first loss function to obtain the low-fidelity network.

[0052] In another exemplary embodiment of the present application, a feature mapping network is obtained by training a second neural network according to a first structural response data set, a second structural response data set, and the damage state parameter values at each moment, which specifically includes: constructing an overall network; the overall network includes a second neural network and a low-fidelity network connected in sequence.

[0053] Using the structural response data in the first structural response data set as the input of the second neural network, using the structural response data in the second structural response data set as the output of the second neural network, using the damage state parameter values at each moment as the output of the low-fidelity network, and training the second neural network in the overall neural network with the goal of minimizing the value of a second loss function to obtain the feature mapping network.

[0054] In another exemplary embodiment of the present application, the first loss function is: , where represents the total number of preset damage state parameter values, represents the output obtained by inputting the th structural response data in the third structural response data set into the first neural network, represents the preset damage state parameter value corresponding to the th structural response data in the third structural response data set, represents calculating the square of the absolute value.

[0055] In another exemplary embodiment of the present application, the second loss function is: , where represents the total number of damage state parameter values at each moment collected, represents the output obtained by inputting into the low-fidelity network, represents the damage state parameter value corresponding to the th structural response data in the first structural response data set, represents a weight coefficient, represents the output obtained by inputting the th structural response data in the first structural response data set into the second neural network, indicating the th structural response data in the second set of structural response data, representing the square of the calculated absolute value. In this loss function, corresponds to the prediction error , while corresponds to the feature mapping error .

[0056] In another exemplary embodiment of the present application, when the damage state parameter value is the crack length, a structural damage propagation test is performed on the sample, and the damage state parameter values at each moment on the sample and the structural response data at the preset position of the sample under each damage state parameter value during the structural damage propagation test are collected to obtain a set of damage state parameter values and a first set of structural response data. Specifically: a structural crack propagation test is performed on the sample, and the crack lengths at each moment on the sample and the structural response data at the preset position of the sample under each crack length during the structural crack propagation test are collected to obtain a set of damage state parameter values and a first set of structural response data.

[0057] In another exemplary embodiment of the present application, when the damage state parameter value is the crack length, a static simulation of the structural damage propagation test is performed on each simulation model in the first set of simulation models, and the structural response data at the preset position of each simulation model in the first set of simulation models is obtained to obtain a second set of structural response data. Specifically: a first set of simulation models is constructed; the first set of simulation models includes multiple simulation models obtained by inserting cracks with different crack lengths in the set of damage state parameter values into the sample respectively.

[0058] A static simulation of the structural crack propagation test is performed on each simulation model in the first set of simulation models, and the structural response data at the preset position of each simulation model in the first set of simulation models is obtained to obtain a second set of structural response data.

[0059] In another exemplary embodiment of the present application, when the damage state parameter value is the crack length, a static simulation of the structural damage propagation test is performed on each simulation model in the second set of simulation models, and the structural response data at the preset position of each simulation model in the second set of simulation models is obtained to obtain a third set of structural response data. Specifically, it includes: constructing a second set of simulation models; the second set of simulation models includes multiple simulation models obtained by inserting cracks with each preset crack length into the sample respectively.

[0060] A static simulation of the structural crack propagation test is performed on each simulation model in the second set of simulation models, and the structural response data at the preset position of each simulation model in the second set of simulation models is obtained to obtain a third set of structural response data.

[0061] In another exemplary embodiment of the present application, when the damage state parameter value is the crack length, a low-fidelity network is obtained by training a first neural network according to the third structural response data set and each preset damage state parameter value. Specifically: a low-fidelity network is obtained by training a first neural network according to the third structural response data set and each preset crack length.

[0062] In another exemplary embodiment of the present application, when the damage state parameter value is the crack length, a feature mapping network is obtained by training a second neural network according to the first structural response data set, the second structural response data set, and the damage state parameter value at each moment. Specifically: a feature mapping network is obtained by training a second neural network according to the first structural response data set, the second structural response data set, and the crack length at each moment.

[0063] In another exemplary embodiment of the present application, the structural response data is stress, strain or displacement.

[0064] The present application provides a specific embodiment, as Figure 2 shown, to illustrate the above multi-fidelity data fusion method in detail, which specifically includes the following steps 1 to 4.

[0065] Step 1: Construct a data set for model training through a structural damage expansion test and damage simulation.

[0066] Step 1.1: As Figure 5 shown, the test piece is a simple sample with an artificially prefabricated notch, and its material is 2A12-T4 aluminum alloy. The size of the sample is 50mm 200mm 3mm, and there is an artificially prefabricated notch with a size of 1mm 6mm 3mm in the middle of the longitudinal direction of the sample. 4 FBG strain sensors are pasted on the sample, numbered 1 to 4. The 1st FBG strain sensor is 50mm away from the upper end of the test piece and is centered horizontally; the 2nd to 4th FBG strain sensors are 85mm away from the upper end of the test piece and are evenly distributed horizontally.

[0067] Step 1.2: Construct a multi-fidelity dataset through experiments and static simulations. First, conduct a structural crack propagation experiment. The load is a tensile cyclic load applied to both ends of the test piece, and the stress ratio of the load cycle is 0.1. When measuring the crack length and the structural strain response, the load is maintained at 55% of the peak value of the cyclic load. During the experiment, the crack length continuously increases, and the crack length is measured at regular intervals. A total of 34 groups of crack lengths and the corresponding high-fidelity measured strain values at the corresponding moments are obtained. Then, establish a simulation model corresponding to the test piece, and insert cracks with the lengths measured above on the expected crack propagation path behind the prefabricated notch, as Figure 5 shown. Obtain 34 simulation models, perform static simulations under a constant load (i.e., 55% of the cyclic load), and obtain low-fidelity simulation strain values corresponding one-to-one to the experimental data, that is, 34 strain values corresponding to the crack lengths collected above. Finally, combine the high-fidelity measured strain values, crack lengths, and low-fidelity simulation strain values to form a multi-fidelity dataset.

[0068] Step 1.3: Construct a low-fidelity dataset through static simulations. According to the experimental results of the above Step 1.2, it can be known that the maximum crack length before the instability fracture of the sample is about 35 mm - 37 mm. Therefore, randomly sample 350 values in the range of [1 mm, 35 mm] to obtain the preset crack lengths. Through batch simulations, extract the low-fidelity simulation strain values at the same positions as the real sensors. According to the preset crack lengths inserted into the model and the low-fidelity simulation strain values corresponding to each preset crack length, construct a low-fidelity dataset.

[0069] Step 2: Construct and train the first neural network to obtain a low-fidelity network . The structure and training method of the first neural network are as Figure 3 shown. The input of the first neural network is the low-fidelity simulation strain values in the low-fidelity dataset, and the output is the predicted crack length. The number of neurons in the input layer and output layer of this network is determined by the number of channels of the structural response data, while other hyperparameters such as the number of internal hidden layers and the number of neurons in each layer are set according to needs by conventional methods. According to the specific situation of this embodiment, the input layer of the network has 4 neurons, corresponding to 4 FBG strain sensors respectively; the output layer has 1 neuron, corresponding to the crack length; set 1 hidden layer, and this layer contains 4 neurons. The network uses the tanh activation function, uses the Adam training algorithm, sets the initial value of the learning rate to 0.01, and adopts a training strategy with a decreasing learning rate. Randomly divide the low-fidelity dataset into three parts according to the ratio of 0.8:0.1:0.1, which are used as the training set, validation set, and test set of the low-fidelity network respectively. The error reduction curve during the training process is as Figure 6 shown, and the effect of the network on the test set is as Figure 7As shown, it can be seen that the network is fully trained and converges, and the prediction effect in the test set is good. The root mean square error of the predicted structural crack length is 0.0807 mm.

[0070] Step 3: Construct and train a second neural network to obtain a feature mapping network . This network realizes the mapping of strain from high-fidelity to low-fidelity. The input and output of this network have the same number of neurons, and each neuron corresponds to an FBG strain sensor. When the first neural network is trained, the network parameters are fixed, and the second neural network is connected in series in front of. The structure and training method of the second neural network are as Figure 4 shown. The input of the second neural network is the high-fidelity measured strain value, and the output is the low-fidelity strain value; the low-fidelity strain value output by this network is input to the low-fidelity network to obtain the crack length output by the low-fidelity network. The input and output layers of the second neural network are determined by the number of channels of the structural response data, while other hyperparameters of the network are set in the conventional way. According to the specific situation of this embodiment, both the input layer and the output layer of the network are 4 neurons, corresponding to 4 FBG strain sensors respectively. The network is set with 1 hidden layer, which contains 8 neurons. The network uses the tanh activation function, uses the Adam training algorithm, the initial value of the learning rate is set to 0.01, and a training strategy of learning rate decay is adopted. To facilitate the verification of the model effect, only 90% of the data in the multi-fidelity dataset is used as the training set and the validation set in this embodiment. The result of the error decline curve during the training process is as Figure 8 shown, and it can be seen that the network is fully trained and converges.

[0071] Step 4: Construct a structural damage assessment model based on the low-fidelity network and the feature mapping network, and deploy the structural damage assessment model online, and apply the structural damage assessment model to structural damage assessment. Use the remaining 10% of the data in the multi-fidelity dataset that has not participated in model training and validation as the test set to test the effect of the model. Use the high-fidelity measured strain value in the test set as the input, and the output of the structural damage assessment model is the predicted crack length value, and its prediction effect is as Figure 9 shown, and the root mean square error of the predicted structural crack length is 0.6715 mm.

[0072] To verify the advantage of the structural damage assessment model ( Figure 11 denoted as FM-MFNN in Figure 11 ) proposed in this application in the inverse problem scenario of structural damage assessment, use the most common multi-fidelity Gaussian process regression model ( Figure 11The (denoted as C-MFNN in the text) model is compared with the model constructed in this application. The multi-fidelity Gaussian process regression model is constructed using the most common co-kriging algorithm, and the commonly used radial basis kernel function and zero-mean hypothesis are used. The observation noise is taken as 0.01, and the hyperparameters of the kernel function are optimized by maximum likelihood estimation. The architecture of the classical multi-fidelity neural network is as shown in Figure 10 and consists of two networks, namely the low-fidelity network and the high-fidelity correction network . has the same structure as the low-fidelity network described in step 2 above. Its input is the low-fidelity simulation strain value, it contains 4 neurons, and its output is the predicted crack length value, which contains 1 neuron; 's input is 's output and the high-fidelity measured strain value, it contains 5 neurons, and its output is the predicted crack length value, which contains 1 neuron. To avoid the influence of randomness such as training set division and model parameter initialization, the training and test verification of the above three models are each repeated 100 times, and the average value of their prediction errors is used to compare the model accuracy. In addition, the interquartile range (IQR) index of the model prediction error is used to evaluate the stability of the model. The prediction effects of the structure damage assessment model constructed in this application and the above two existing multi-fidelity algorithm models are as shown in Figure 11 . It can be seen that the average prediction error of the structure damage assessment model proposed in this application is 0.72 mm, and the IQR range is 0.37; while the average prediction error of the multi-fidelity Gaussian process regression model is 0.78 mm, and the IQR range is 0.69; the average prediction error of the classical multi-fidelity neural network is 2.43 mm, and the IQR range is 1.70233084. It can be seen that both the average prediction error and the IQR range of the model proposed in this application are significantly lower than those of the existing algorithm models, indicating that the model proposed in this application has better prediction accuracy and model stability.

[0073] This application first constructs a data set for model training through structure damage expansion tests and damage simulations; then uses low-fidelity data to train the low-fidelity network; further, uses the multi-fidelity data set to train the feature mapping network to achieve the fusion of multi-fidelity data; finally, deploys the model online to achieve accurate assessment of the structure damage state. It can be used to fuse multi-fidelity data in this inverse problem of structure damage assessment.

[0074] This application has the advantages of low demand for high-fidelity data and high accuracy of structure damage assessment. It can effectively fuse multi-fidelity data in the scenario of this type of inverse problem of structure damage assessment, and has high engineering application value in the field of structural health monitoring where high-fidelity data is lacking and the accuracy of low-fidelity data is poor.

[0075] This application trains a second neural network using a second loss function, which can ensure the effectiveness of the network training process.

[0076] Based on the same inventive concept, an embodiment of this application also provides a multi-fidelity data fusion device for implementing the multi-fidelity data fusion method involved above. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the multi-fidelity data fusion device provided below can refer to the limitations on the multi-fidelity data fusion method in the above text and will not be repeated here.

[0077] In an exemplary embodiment, a multi-fidelity data fusion device is provided, including: a test module, configured to perform a structural damage propagation test on a sample, and collect the damage state parameter values of the sample at each moment during the structural damage propagation test and the structural response data at a preset position of the sample under each damage state parameter value, to obtain a set of damage state parameter values and a first set of structural response data.

[0078] A first simulation module, configured to perform a static simulation of the structural damage propagation test on each simulation model in the first set of simulation models, and obtain the structural response data at the preset position of each simulation model in the first set of simulation models, to obtain a second set of structural response data; the first set of simulation models includes multiple simulation models obtained by respectively processing the simulation model of the sample according to each damage state parameter value in the set of damage state parameter values.

[0079] A second simulation module, configured to perform a static simulation of the structural damage propagation test on each simulation model in the second set of simulation models, and obtain the structural response data at the preset position of each simulation model in the second set of simulation models, to obtain a third set of structural response data; the second set of simulation models includes multiple simulation models obtained by respectively processing the simulation model of the sample according to each preset damage state parameter value.

[0080] A first training module, configured to train a first neural network according to the third set of structural response data and each preset damage state parameter value to obtain a low-fidelity network.

[0081] A second training module, configured to train a second neural network according to the first set of structural response data, the second set of structural response data, and the damage state parameter values at each moment to obtain a feature mapping network.

[0082] A structural damage assessment model construction module, configured to construct a structural damage assessment model according to the low-fidelity network and the feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence.

[0083] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 12 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store multi-fidelity data fusion data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a multi-fidelity data fusion method.

[0084] Those skilled in the art can understand that Figure 12 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the above method embodiments are implemented.

[0085] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above method embodiments are implemented.

[0086] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above method embodiments are implemented.

[0087] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0088] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0089] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0090] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, 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, it should be considered as the scope described in this specification.

[0091] In this article, specific examples are used to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A multi-fidelity data fusion method, characterized in that: The multi-fidelity data fusion method comprises: Performing a structural damage extension test on the sample, and collecting damage state parameter values ​​of the sample at each moment during the structural damage extension test and structural response data of a preset position of the sample under each damage state parameter value, to obtain a damage state parameter value set and a first structural response data set; Performing static simulation of the structural damage extension test on each simulation model in the first simulation model set, acquiring structural response data of a preset position of each simulation model in the first simulation model set, and obtaining a second structural response data set; the first simulation model set includes a plurality of simulation models obtained by processing the simulation model of the sample according to each damage state parameter value in the damage state parameter value set; Performing static simulation of the structural damage extension test on each simulation model in the second simulation model set, obtaining structural response data of a preset position of each simulation model in the second simulation model set, and obtaining a third structural response data set; the second simulation model set includes a plurality of simulation models obtained by processing the simulation model of the sample according to each preset damage state parameter value; Training the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network; According to the first structural response data set, the second structural response data set and the damage state parameter value at each moment, the second neural network is trained to obtain a feature mapping network; A structural damage assessment model is constructed according to a low-fidelity network and a feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence.

2. The multi-fidelity data fusion method according to claim 1, characterized in that: The first neural network is trained according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network, specifically including: Taking the structural response data in the third structural response data set as input, taking the preset damage state parameter values ​​as output, and taking the minimum value of the first loss function as the goal, the first neural network is trained to obtain a low-fidelity network.

3. The multi-fidelity data fusion method according to claim 1, characterized in that: According to the first structural response data set, the second structural response data set and the damage state parameter value at each moment, the second neural network is trained to obtain a feature mapping network, specifically including: Constructing an overall network; the overall network includes a second neural network and a low-fidelity network connected in sequence; The structural response data in the first structural response data set is used as the input of the second neural network, the structural response data in the second structural response data set is used as the output of the second neural network, the damage state parameter value at each moment is used as the output of the low-fidelity network, and the goal is to minimize the value of the second loss function. The second neural network in the overall neural network is trained to obtain a feature mapping network.

4. The multi-fidelity data fusion method according to claim 2, characterized in that: The first loss function for: ,in, represents the total number of preset damage state parameter values, Indicates that the third structure response data set The output obtained by inputting the structural response data into the first neural network is Indicates the third structure response data set The preset damage state parameter value corresponding to the structural response data is Calculates the square of the absolute value.

5. The multi-fidelity data fusion method according to claim 3, characterized in that: The second loss function for: ,in, represents the total number of damage state parameter values ​​collected at each moment, Indicates that The output of the low-fidelity network is input, Represents the first structure response data set The damage state parameter value corresponding to the structural response data is: represents the weight coefficient, Indicates that the first structure response data set The output obtained by inputting the structural response data into the second neural network is Indicates the first Structural response data, Calculates the square of the absolute value.

6. The multi-fidelity data fusion method according to claim 1, characterized in that: When the damage state parameter value is the crack length, a structural damage extension test is performed on the sample, and the damage state parameter values ​​of the sample at each moment during the structural damage extension test and the structural response data of the preset position of the sample under each damage state parameter value are collected to obtain a damage state parameter value set and a first structural response data set, specifically: A structural crack growth test is performed on the sample, and the crack lengths at various moments on the sample during the structural crack growth test and the structural response data of a preset position of the sample at various crack lengths are collected to obtain a damage state parameter value set and a first structural response data set.

7. A multi-fidelity data fusion device, characterized in that: The multi-fidelity data fusion device comprises: A test module is used to perform a structural damage extension test on the sample, and collect damage state parameter values ​​of the sample at each moment during the structural damage extension test and structural response data of a preset position of the sample under each damage state parameter value, to obtain a damage state parameter value set and a first structural response data set; The first simulation module is used to perform static simulation of a structural damage extension test on each simulation model in the first simulation model set, obtain structural response data of a preset position of each simulation model in the first simulation model set, and obtain a second structural response data set; the first simulation model set includes a plurality of simulation models obtained by processing the simulation model of the sample according to each damage state parameter value in the damage state parameter value set; The second simulation module is used to perform static simulation of the structural damage extension test on each simulation model in the second simulation model set, obtain the structural response data of the preset position of each simulation model in the second simulation model set, and obtain a third structural response data set; the second simulation model set includes a plurality of simulation models obtained by processing the simulation model of the sample according to each preset damage state parameter value; A first training module is used to train the first neural network according to the third structural response data set and each preset damage state parameter value to obtain a low-fidelity network; A second training module is used to train the second neural network to obtain a feature mapping network according to the first structural response data set, the second structural response data set and the damage state parameter value at each moment; The structural damage assessment model construction module is used to construct a structural damage assessment model based on a low-fidelity network and a feature mapping network; the structural damage assessment model includes a feature mapping network and a low-fidelity network connected in sequence.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-fidelity data fusion method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-fidelity data fusion method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-fidelity data fusion method according to any one of claims 1 to 6 is implemented.

Images