Hanging structure load identification method and system based on transfer learning, medium and equipment

Abstract

The invention discloses a hanging structure load identification method and system based on transfer learning, a medium and equipment, and the method comprises the steps: building a hanging structure finite element model, calculating the vibration response under different impact load excitation, and constructing a simulation data set containing a simulation vibration response signal and a concentrated load signal; collecting a vibration response signal of the monitored hanging structure under the action of the concentrated impact load, collecting a corresponding concentrated load time history, and constructing an experimental data set; constructing a gated convolutional network, and performing pre-training by using the simulation data set to obtain a pre-training model; performing parameter fine tuning on the pre-training model by using the experimental data set to obtain a hanging structure load identification network model; and carrying out reverse identification on the concentrated impact load of the hanging structure by using the hanging structure load identification network model.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring, and particularly relates to a method, system, medium and device for identifying loads of a hanging structure based on transfer learning. Background Art

[0002] In modern engineering equipment and mechanical structures, the hanging structure is a common connection structure, and its safety and reliability are crucial. There is usually a concentrated load at the hanging point between two structural members connected by a hanging structure. Accurately identifying the concentrated load acting on the hanging structure is of great significance for evaluating the health status of the hanging structure connection members, optimizing design parameters, and ensuring operation safety. It is usually very difficult to directly measure the load at the hanging structure by a dynamometer. In recent years, with the rapid development of deep learning technology, deep neural networks have been widely applied to multiple engineering field problems due to their powerful feature extraction ability and end-to-end processing ability. However, in engineering practice, obtaining sufficient training data through experiments usually has high economic costs and labor costs, and deep learning methods such as convolutional neural networks require a large number of training samples to obtain sufficient load identification accuracy. The lack of training data will limit the performance of the neural network. Transfer learning can obtain feature extraction ability through pre-training from source domain data, and then improve the performance of the model in the target domain through transfer learning. A large number of sample data can be obtained at a low cost through a finite element model. However, there is inevitably a certain gap in the feature distribution between simulation data and experimental data. Therefore, how to accurately identify the load under the conditions of a small number of experimental samples and a large number of simulation samples is a challenging problem.

[0003] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present invention, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0004] The present invention provides a method, system, medium and device for identifying loads of a hanging structure based on transfer learning, which can accurately identify the concentrated impact load of the hanging structure, and at the same time compensate for problems such as a small number of experimental samples and difficulty in training a deep neural network.

[0005] To achieve the above object, the present disclosure provides the following technical solutions:

[0006] A method for identifying loads of a hanging structure based on transfer learning includes:

[0007] Step S1, establishing a finite element model of the hanging structure and calculating the vibration response under different impact load excitations, and constructing a simulation data set including simulation vibration response signals and concentrated load signals;

[0008] Step S2, collect the vibration response signals of the monitored hanging structure under the action of concentrated impact loads, and at the same time collect the corresponding concentrated load time history to construct an experimental data set;

[0009] Step S3, construct a gated convolutional network, and use the simulation data set for pre-training to obtain a pre-trained model. The gated convolutional network includes a feature extractor composed of sequentially connected K gated convolutional blocks and a regressor composed of a fully connected layer;

[0010] Step S4, fine-tune the parameters of the pre-trained model using the experimental data set to obtain a load identification network model for the hanging structure;

[0011] Step S5, use the load identification network model of the hanging structure to inversely identify the concentrated impact load of the hanging structure.

[0012] In the described method for identifying the load of a hanging structure based on transfer learning, Step S1 includes

[0013] Step S1.1, establish a three-dimensional geometric model according to the geometric dimensions of the hanging structure;

[0014] Step S1.2, obtain the corresponding material properties according to the materials used in the hanging structure;

[0015] Step S1.3, establish a finite element model of the hanging structure according to the three-dimensional geometric model and the material properties;

[0016] Step S1.4, according to the finite element model of the hanging structure, apply a concentrated impact load and the same constraint conditions as the monitored hanging structure, and calculate the simulated vibration response signals of the response points;

[0017] Step S1.5, use the simulated vibration response signals as network input features, and use the corresponding concentrated load time history as the corresponding sample labels to construct a simulation data set.

[0018] In the described method for identifying the load of a hanging structure based on transfer learning, in Step S2, the vibration response signals collected by strain gauges are used as network input features, and the concentrated impact load time history obtained by a dynamometer is used as the corresponding sample labels to establish an experimental data set.

[0019] In the described method for identifying the load of a hanging structure based on transfer learning, Step S3 includes:

[0020] Step S3.1, construct a gated convolutional network feature extractor by sequentially connecting multiple gated convolutional blocks. Each gated convolutional block contains two parallel dilated convolutional layers, a convolutional layer with a kernel length of 1, two batch normalization layers, and a non-linear activation layer. The operation of the gated convolutional block It is expressed as follows:

[0021] , where x is the input feature of the gated convolutional block, represents a convolutional layer with a kernel length of 1, relu represents the rectified linear unit, represents the dilated convolution operation, W and V are the convolutional kernel weight parameters in the convolutional layer, b and c respectively represent the bias weight parameters of the convolutional layer, represents the dot product operation, represents the sigmoid non-linear function, bn is the batch normalization operation. The dilated convolution in the gated convolutional block expands the receptive field of the convolutional kernel, and at the same time controls the proportion of the features extracted by the convolutional kernel to be passed to the next layer through the sigmoid non-linear function, and maintains the gradient stability of the deep network through the residual connection;

[0022] Step S3.2, construct a regressor, and use the features output by the feature extractor through the fully connected layer to predict the concentrated load, and output the reverse identification result of the concentrated impact load of the hanging structure;

[0023] Step S3.3, construct a gated convolutional neural network. The gated convolutional network includes a feature extractor composed of K gated convolutional blocks connected in sequence and a regressor composed of a fully connected layer,

[0024] Step S3.4, divide the simulation data set into a training set and a validation set, and pre-train the gated convolutional network. The mean squared error MSE is used as the loss function during the pre-training process:

[0025] ,

[0026] where is the number of training samples, represents the sample length, represents the predicted value of the i-th sample, represents the true value of the i-th sample,

[0027] During the pre-training process, first batch the simulation data training samples into the gated convolutional neural network to obtain the corresponding load prediction results. Calculate the MSE loss function value according to the load prediction results and the true load values. Then, according to the derivative chain rule, calculate the gradient values of the weight parameters in the network in the order from the output end to the input end of the network, and update each weight parameter through gradient descent. Through multiple loop iterations, the weight parameters of the gated convolutional neural network are trained and optimized. Calculate the loss function value of the gated convolutional network for the samples in the validation set in each loop iteration, and use the early stopping strategy to determine whether the pre-training process of the network stops, that is, after the validation set loss function does not decrease continuously rounds of training, the training process stops.

[0028] In the described load identification method for a hanging structure based on transfer learning, in step S4, the experimental data set is divided into a training set and a validation set. For the pre-trained model, the convolution kernel weight parameters and the convolution layer bias weight parameters of its feature extractor are frozen, and the parameters are not updated during the transfer process. During the transfer learning process, the experimental data training samples are input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. According to the load prediction results and the true load values, the MSE loss function value is calculated, the gradient value corresponding to the weight parameters in the fully connected layer is calculated, and the weight parameters of the fully connected layer are updated through gradient descent to achieve parameter fine-tuning. Fine-tuning means that when using the response-load experimental data set to perform transfer learning on the learnable parameters of the predictor, fewer training epochs and a smaller learning rate are adopted than in the pre-training process. The transfer learning process also determines whether the training process stops through an early stopping strategy. After the transfer learning is completed, a load identification network model for the hanging structure is obtained.

[0029] In the described load identification method for a hanging structure based on transfer learning, in step S5, the vibration response of the monitored hanging structure under the action of an impact load is collected and input into the load identification network model for the hanging structure to obtain the inverse identification result of the concentrated load.

[0030] In the described load identification method for a hanging structure based on transfer learning, the hanging structure includes a hanging object, a hanging inner lining plate, a hanging point, and a connecting piece.

[0031] A load identification system for a hanging structure based on transfer learning includes

[0032] A finite element modeling unit, which is used to establish a finite element model of the hanging structure and calculate the vibration response under different impact load excitations, and construct a simulation data set containing simulation vibration response signals and concentrated load signals;

[0033] A collection unit, which is used to collect the vibration response signals of the monitored hanging structure under the action of a concentrated impact load, and at the same time collect the corresponding concentrated load time history to construct an experimental data set;

[0034] A construction unit, which is used to construct a gated convolutional network and perform pre-training using the simulation data set to obtain a pre-trained model;

[0035] A fine-tuning unit, which is used to perform parameter fine-tuning on the pre-trained model using the experimental data set to obtain a load identification network model for the hanging structure;

[0036] An identification unit, which is used to perform inverse identification of the concentrated impact load of the hanging structure using the load identification network model for the hanging structure.

[0037] A computer storage medium, the storage medium includes computer instructions, when it runs on a computer, it enables the computer to execute the described method.

[0038] An electronic device, the electronic device includes:

[0039] A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein,

[0040] When the processor executes the program, it implements the described method.

[0041] Compared with the prior art, the present invention has the following advantages: Compared with traditional deep learning-based methods, the present invention can fully explore the information of finite element simulation data, reduce the demand for experimental data, and lower the data acquisition cost. Through transfer learning, it can improve the inverse identification accuracy of the concentrated impact load of the hanging structure under the condition of a small number of experimental data samples. Description of the Drawings

[0042] By reading the following detailed description of the preferred specific embodiments, various other advantages and benefits of the present invention will become clear to those of ordinary skill in the art. The accompanying drawings of the specification are only for the purpose of showing the preferred embodiments, and are not considered to be a limitation of the present invention. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. And throughout the drawings, the same reference numerals are used to represent the same components.

[0043] In the drawings:

[0044] Figure 1 is the flowchart of the method of the present invention;

[0045] Figure 2 is the overall structure diagram of the gated convolutional network;

[0046] Figure 3 is a schematic diagram of identifying the concentrated impact load of the simulated hanging structure in an embodiment;

[0047] Figs. 4(a) and 4(b) are schematic diagrams of the load identification results of the gated convolutional network for the same test sample before and after transfer learning, where Fig. 4(a) is the load identification result of the pre-trained model, and Fig. 4(b) is the load identification result of the load identification network of the hanging structure after transfer learning.

[0048] The following further explains the present invention in conjunction with the drawings and embodiments. Detailed Embodiments

[0049] Specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although specific embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be fully conveyed to those skilled in the art.

[0050] It should be noted that in the specification and claims, certain terms are used to refer to specific components. Those skilled in the art should understand that technicians may use different terms to refer to the same component. The specification and claims do not use the difference in terms as a way to distinguish components, but use the difference in the functions of components as the criterion for distinction. As mentioned throughout the specification and claims, "comprising" or "including" is an open-ended term and should be interpreted as "including but not limited to". The subsequent description in the specification is the preferred embodiment for implementing the present invention, but the description is for the purpose of the general principles of the specification and is not intended to limit the scope of the present invention. The protection scope of the present invention shall be subject to what is defined by the appended claims.

[0051] For the convenience of understanding the embodiments of the present invention, the following will further explain with specific embodiments as examples in conjunction with the accompanying drawings, and each accompanying drawing does not constitute a limitation on the embodiments of the present invention.

[0052] As Figures 1 to 4(b) shown, the method for identifying the load of the hanging structure based on transfer learning includes the following steps:

[0053] Step S1, establish a finite element model of the hanging structure and calculate the vibration response under different impact load excitations, and construct a simulation data set including simulation vibration response signals and concentrated load signals;

[0054] Step S2, collect the vibration response signals of the monitored hanging structure under the action of concentrated impact loads, and at the same time collect the corresponding concentrated load time history, and construct an experimental data set;

[0055] Step S3, construct a gated convolutional network, and use the simulation data set for pre-training to obtain a pre-trained model. The network includes a feature extractor composed of K gated convolutional blocks connected in sequence and a regressor composed of a fully connected layer;

[0056] Step S4, use the experimental data set to fine-tune the parameters of the pre-trained model to obtain a load identification network model for the hanging structure;

[0057] Step S5, use the load identification network model of the hanging structure to inversely identify the concentrated impact load of the hanging structure.

[0058] In the preferred embodiment of the method for identifying the load of a hanging structure based on transfer learning, step S1 includes:

[0059] Step S1.1: Establish a three-dimensional geometric model according to the geometric dimensions of the hanging structure.

[0060] Step S1.2: Obtain the corresponding material properties according to the materials used in the hanging structure.

[0061] Step S1.3: Establish a finite element model of the hanging structure according to the three-dimensional geometric model and the material properties.

[0062] Step S1.4: Apply a concentrated impact load and the same constraint conditions as the monitored hanging structure according to the finite element model of the hanging structure, and calculate the simulated vibration response signal of the response point.

[0063] Step S1.5: Construct a simulation data set by using the simulated vibration response signal as the network input feature and the corresponding concentrated impact load time history as the corresponding sample label.

[0064] In the preferred embodiment of the method for identifying the load of a hanging structure based on transfer learning, in step S2, the vibration response signal collected by the strain gauge is used as the network input feature, and the concentrated impact load time history obtained by the dynamometer is used as the corresponding sample label to establish an experimental data set.

[0065] In the preferred embodiment of the method for identifying the load of a hanging structure based on transfer learning, step S3 includes:

[0066] Step S3.1: Construct a gated convolutional network feature extractor by sequentially connecting multiple gated convolutional blocks. Each gated convolutional block contains two parallel dilated convolutional layers, a convolutional layer with a kernel length of 1, two batch normalization layers, and a non-linear activation layer. The operation of the gated convolutional block is expressed as follows:

[0067] , where x is the input feature of the gated convolutional block, represents the convolutional layer with a kernel length of 1, relu represents the rectified linear unit, represents the dilated convolution operation, W and V are the convolutional kernel weight parameters in the convolutional layer, b and c respectively represent the bias weight parameters of the convolutional layer, represents the dot product operation, represents the sigmoid non-linear function, bn is the batch normalization operation. The dilated convolution in the gated convolutional block expands the receptive field of the convolutional kernel, and at the same time controls the proportion of the features extracted by the convolutional kernel to be passed to the next layer through the sigmoid non-linear function, and maintains the gradient stability of the deep network through the residual connection.

[0068] Step S3.2: Construct a regressor to predict the concentrated load by using the features output by the feature extractor through a fully connected layer, and output the reverse identification result of the concentrated impact load of the hanging structure.

[0069] Step S3.3: Construct a gated convolutional neural network, which consists of a feature extractor composed of K gated convolutional blocks connected in sequence and a regressor composed of a fully connected layer.

[0070] Step S3.4: Divide the simulation data set into a training set and a validation set, and pre-train the gated convolutional network. The mean squared error (MSE) is used as the loss function during the training process:

[0071] ,

[0072] where is the number of training samples, represents the sample length, represents the predicted value of the i-th sample, represents the true value of the i-th sample. During the pre-training process, the simulation data training samples are first input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. The MSE loss function value is calculated based on the load prediction results and the true load values. Then, according to the derivative chain rule, the gradient values of the weight parameters in the network are calculated in sequence from the output end to the input end of the network, and the weight parameters are updated by gradient descent. Through multiple loop iterations, the weight parameters of the gated convolutional neural network are trained and optimized. The loss function value of the gated convolutional network for the samples in the validation set is calculated for each loop iteration, and an early stopping strategy is used to determine whether the pre-training process of the network stops, that is, the training process stops after the loss function of the validation set does not decrease for consecutive rounds of training.

[0073] In the preferred implementation of the method for identifying the load of a suspension structure based on transfer learning, in step S4, the experimental data set is divided into a training set and a validation set. For the pre-trained model, the convolutional kernel weight parameters and the convolutional layer bias weight parameters of its feature extractor are frozen, and the parameters are no longer updated during the transfer process. During the transfer learning process, the experimental data training samples are input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. According to the load prediction results and the true load values, the MSE loss function value is calculated, the gradient values corresponding to the weight parameters in the fully connected layer are calculated, and the weight parameters of the fully connected layer are updated through gradient descent to achieve parameter fine-tuning. Fine-tuning means that when using the response-load experimental data set to perform transfer learning on the learnable parameters of the predictor, fewer training epochs and a smaller learning rate are adopted than in the pre-training process. The transfer learning process also determines whether the training process stops through an early stopping strategy. After the transfer learning is completed, a suspension structure load identification network model is obtained.

[0074] In the preferred implementation of the method for identifying the load of a suspension structure based on transfer learning, in step S5, the vibration response of the monitored suspension structure under the action of an impact load is collected and input into the suspension structure load identification network model to obtain the reverse identification result of the concentrated load.

[0075] In the preferred implementation of the method for identifying the load of a suspension structure based on transfer learning, the suspension structure includes a suspended object, a suspension lining plate, a hanging point, and a connecting member.

[0076] A suspension structure load identification system based on transfer learning includes

[0077] A finite element modeling unit, which is used to establish a finite element model of the suspension structure and calculate the vibration response under different impact load excitations, and construct a simulation data set containing simulation vibration response signals and concentrated load signals;

[0078] An acquisition unit, which is used to acquire the vibration response signal of the monitored suspension structure under the action of a concentrated impact load, and simultaneously acquire the corresponding concentrated load time history to construct an experimental data set;

[0079] A construction unit, which is used to construct a gated convolutional network and perform pre-training using the simulation data set to obtain a pre-trained model;

[0080] A fine-tuning unit, which is used to perform parameter fine-tuning on the pre-trained model using the experimental data set to obtain a suspension structure load identification network model;

[0081] An identification unit, which is used to perform reverse identification of the concentrated impact load of the suspension structure using the suspension structure load identification network model.

[0082] A computer storage medium, the storage medium comprising computer instructions which, when run on a computer, cause the computer to execute the method described above.

[0083] An electronic device, the electronic device comprising:

[0084] A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein

[0085] when the processor executes the program, the method described above is implemented.

[0086] In one embodiment, referring to Figure 1 , the present invention provides a method for identifying the load of a hanging structure based on transfer learning, the method comprising the following steps:

[0087] S1. Establish a finite element model of the hanging structure with high precision and calculate the vibration responses under different impact load excitations, and construct a simulation data set containing simulated vibration response signals and concentrated load signals;

[0088] S2. Use strain gauges to collect the vibration response signals of the monitored hanging structure under the action of concentrated impact loads, and at the same time collect the corresponding concentrated load time history through a force sensor to construct an experimental data set;

[0089] S3. Construct a gated convolutional network, pre-train it using the simulation data set to obtain a pre-trained model, the gated convolutional network includes a feature extractor composed of K sequentially connected gated convolutional blocks and a regressor composed of a fully connected layer;

[0090] S4. Fine-tune the parameters of the pre-trained model using the experimental data set to obtain a hanging structure load identification network model;

[0091] S5. Use the hanging structure load identification network model with fine-tuned parameters to inversely identify the concentrated impact load of the hanging structure.

[0092] Optionally, in step S1, establishing a finite element model of the hanging structure with high precision and calculating the vibration responses under different impact load excitations, and constructing a simulation data set containing simulated vibration response signals and concentrated load signals, specifically includes:

[0093] S1.1. According to the geometric dimensions of the components of the hanging structure, establish a three-dimensional geometric model with accurate shape and size; the components of the hanging structure include hanging objects, hanging inner liners, hanging points, and connecting parts, etc.

[0094] S1.2. According to the materials used for the components of the hanging structure, obtain the corresponding material properties through a material table;

[0095] S1.3 Based on the three-dimensional geometric model in S1.1 and the material properties in S1.2, establish a finite element model of the suspension structure;

[0096] S1.4 Based on the finite element model of the suspension structure established in S1.3, apply a concentrated impact load and the same constraint conditions as the monitored suspension structure, and calculate the simulated vibration response signal of the response point.

[0097] S1.5 Use the simulated vibration response signal calculated by the finite element model of the suspension structure as the network input feature, and use the corresponding concentrated impact load time history as the corresponding sample label to construct a simulation data set.

[0098] Optionally, in step S2, use the vibration response signal of the suspension structure collected by the strain gauge as the network input feature, and use the concentrated impact load time history obtained by the dynamometer as the corresponding sample label to establish an experimental data set.

[0099] Optionally, in step S3, as Figure 2 shown, construct a gated convolutional network, specifically including:

[0100] S3.1 Construct a gated convolutional network feature extractor by sequentially connecting multiple gated convolutional blocks. Each gated convolutional block contains two parallel dilated convolutional layers, a convolutional layer with a kernel length of 1, two batch normalization layers, and a non-linear activation layer. The operation of the gated convolutional block can be expressed as follows:

[0101]

[0102] where x is the input feature of the gated convolutional block, represents the convolutional layer with a kernel length of 1, relu represents the rectified linear unit, represents the dilated convolution operation, W and V are the convolutional kernel weight parameters in the convolutional layer, b and c respectively represent the bias weight parameters of the convolutional layer, represents the dot product operation, represents the sigmoid non-linear function, and bn is the batch normalization operation. The dilated convolution in the gated convolutional block expands the receptive field of the convolutional kernel, simultaneously controls the proportion of the features extracted by the convolutional kernel passed to the next layer through the sigmoid non-linear function, and maintains the gradient stability of the deep network through the residual connection.

[0103] S3.2 Construct a regressor, and use the features output by the feature extractor through the fully connected layer to perform concentrated load prediction, and output the reverse identification result of the concentrated impact load of the suspension structure.

[0104] S3.3. Construct a gated convolutional neural network, which consists of a feature extractor composed of K sequentially connected gated convolutional blocks and a regressor composed of a fully connected layer.

[0105] S3.4. Divide the simulation data set into a training set and a validation set, and pre-train the gated convolutional network. The division ratio of the training set and the validation set is determined according to the actual training effect. The mean squared error (MSE) is used as the loss function during the training process:

[0106]

[0107] where is the number of training samples, represents the sample length, represents the predicted value of the i-th sample, represents the true value of the i-th sample.

[0108] During the pre-training process, the simulation data training samples are first input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. The MSE loss function value is calculated based on the load prediction results and the true load values. Then, according to the chain rule of differentiation, the gradient values of the weight parameters in the network are calculated sequentially from the output end to the input end of the network, and the weight parameters are updated by gradient descent. Through multiple loop iterations, the weight parameters of the gated convolutional neural network are trained and optimized. The loss function value of the gated convolutional network for the samples in the validation set is calculated for each loop iteration, and an early stopping strategy is used to determine whether the pre-training process of the network stops, that is, when the loss function of the validation set does not decrease for consecutive training rounds, the training process stops.

[0109] Optionally, in step S4, the experimental data set is divided into a training set and a validation set. For the pre-trained model, the convolutional kernel weight parameters and the convolutional layer bias weight parameters of its feature extractor are frozen, and the parameters are not updated during the transfer process. During the transfer learning process, the experimental data training samples are input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. The MSE loss function value is calculated based on the load prediction results and the true load values. The gradient values corresponding to the weight parameters in the fully connected layer are calculated, and the weight parameters of the fully connected layer are updated by gradient descent to achieve parameter fine-tuning. Fine-tuning means that when using the response-load experimental data set for transfer learning of the learnable parameters of the predictor, fewer training rounds and a smaller learning rate are used than in the pre-training process. The transfer learning process also determines whether the training process stops through an early stopping strategy. After the transfer learning is completed, a hanging structure load identification network model is obtained. Optionally, in step S5, the vibration response of the monitored hanging structure under the impact load is collected and input into the hanging structure load identification network to obtain the inverse identification result of the concentrated load.

[0110] In one embodiment, with reference to Figure 3 , a simulation hanging structure is used as the research object, and the process of identifying the concentrated impact load is as follows:

[0111] 1. The simulation hanging structure mainly includes two cylindrical parts, a simulation hanging point, a hanging point lining plate, and a test bench bottom plate. The two cylinders are connected by bolts. The inner diameter of the cylinder is 500 mm, and the wall thickness is 3 mm; among them, the height of the upper cylinder is 500 mm, the height of the lower cylinder is 250 mm, and the height of the entire circular structure is 750 mm. The simulation hanging point and the hanging point lining plate are connected and installed on the upper cylinder by bolts. Two reinforcing rings are symmetrically arranged inside the upper cylinder, with a height of 20 mm and a thickness of 5 mm.

[0112] 2. Establish a high-precision finite element model of the hanging structure and calculate the vibration response under different impact load excitations, and construct a simulation data set including simulation vibration response signals and concentrated load signals, specifically including:

[0113] According to the geometric dimensions of each component of the hanging structure, establish a three-dimensional geometric model with accurate shape and size; among them, the components of the hanging structure include hanging objects, hanging inner lining plates, hanging points, and connecting parts.

[0114] According to the materials used for each component of the hanging structure, obtain the corresponding material properties through the material table;

[0115] According to the three-dimensional geometric model and the material properties, establish a finite element model of the hanging structure;

[0116] According to the established finite element model of the hanging structure, apply a concentrated impact load and the same constraint conditions as the monitored hanging structure, and calculate the simulation vibration response signal of the response point.

[0117] Use the simulation vibration response signal calculated by the finite element model of the hanging structure as the network input feature, and use the corresponding concentrated impact load time history as the corresponding sample label to construct a simulation data set, which contains 101 simulation samples in total.

[0118] 3. Apply a concentrated impact load at the simulation hanging point through an impact force hammer, use the vibration response signal of the hanging structure collected by the strain gauge as the network input feature, and use the concentrated impact load time history obtained through the dynamometer in the force hammer as the corresponding sample label to establish an experimental data set, which contains 60 experimental samples in total.

[0119] 4. Construct a gated convolutional network, specifically including:

[0120] Construct a gated convolutional network feature extractor by sequentially connecting multiple gated convolutional blocks. Each gated convolutional block contains two parallel dilated convolutional layers, a convolutional layer with a kernel length of 1, two batch normalization layers, and a non-linear activation layer. The operations of the gated convolutional block can be expressed as follows:

[0121]

[0122] where x is the input feature of the gated convolutional block, represents the convolutional layer with a kernel length of 1, relu represents the rectified linear unit, represents the dilated convolution operation, W and V are the convolutional kernel weight parameters in the convolutional layer, b and c represent the bias weight parameters of the convolutional layer respectively, represents the dot product operation, represents the sigmoid non-linear function, and bn is the batch normalization operation. The dilated convolution in the gated convolutional block expands the receptive field of the convolutional kernel, and at the same time controls the proportion of the features extracted by the convolutional kernel to be passed to the next layer through the sigmoid non-linear function, and maintains the gradient stability of the deep network through the residual connection.

[0123] Construct a regressor, and use the features output by the feature extractor through the fully connected layer to predict the concentrated load, and output the reverse identification result of the concentrated impact load of the hanging structure.

[0124] Construct a gated convolutional neural network, which consists of a feature extractor composed of sequentially connecting K gated convolutional blocks and a regressor composed of a fully connected layer.

[0125] Divide the simulation data set into a training set and a validation set, and pre-train the gated convolutional network. The division ratio of the training set and the validation set is determined according to the actual training effect. The mean squared error MSE is used as the loss function during the training process:

[0126]

[0127] where is the number of training samples, represents the sample length, represents the predicted value of the i-th sample, represents the true value of the i-th sample.

[0128] During the pre-training process, the simulation data training samples are first input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. The MSE loss function value is calculated based on the load prediction results and the true load values. Then, according to the chain rule of differentiation, the gradient values of the weight parameters in the network are calculated sequentially from the output end to the input end of the network, and the weight parameters are updated by gradient descent. Through multiple loop iterations, the weight parameters of the gated convolutional neural network are trained and optimized. The loss function value of the gated convolutional network for the samples in the validation set is calculated in each loop iteration, and an early stopping strategy is used to determine whether the pre-training process of the network stops, that is, the training process stops after the loss function of the validation set does not decrease for consecutive training rounds.

[0129] 5. The experimental data set is divided into a training set and a validation set. The convolutional kernel weight parameters and the convolutional layer bias weight parameters of the feature extractor of the pre-trained model are frozen, and the parameters are not updated during the transfer process. During the transfer learning process, the experimental data training samples are input into the gated convolutional neural network in batches to obtain the corresponding load prediction results. The MSE loss function value is calculated based on the load prediction results and the true load values. The gradient values corresponding to the weight parameters in the fully connected layer are calculated, and the weight parameters of the fully connected layer are updated by gradient descent to achieve parameter fine-tuning. Fine-tuning means that when using the response-load experimental data set to perform transfer learning on the learnable parameters of the predictor, fewer training rounds and a smaller learning rate are used than in the pre-training process. The early stopping strategy is also used to determine whether the training process stops during the transfer learning process. After the transfer learning is completed, the hanging structure load identification network model is obtained.

[0130] 6. The hanging structure load identification network model after the transfer learning is completed is used to identify the concentrated impact load of the hanging structure. The vibration response of the monitored hanging structure under the impact load is collected and input into the hanging structure load identification network to obtain the reverse identification result of the concentrated load.

[0131] To quantitatively evaluate the performance of the hanging structure load identification method based on transfer learning in identifying concentrated impact loads, the following relative error and peak relative error are defined:

[0132]

[0133]

[0134] The fine-tuned load identification network for the hanging structure was tested using 30 experimental test samples. The average RE of the pre-trained model on the 30 test samples was 16.96%, and the average PRE was 13.60%. The average RE of the gated convolutional network after transfer learning on the test set was 6.78%, and the average PRE was 4.42%. Compared with the pre-trained model, the average RE decreased by 60.02%, and the average PRE decreased by 67.50%. Figures 4(a) and 4(b) show the load identification results of the gated convolutional network for the same test sample before and after transfer learning. Among them, Figure 4(a) is the load identification result of the pre-trained model, and Figure 4(b) is the load identification result of the load identification network for the hanging structure after transfer learning. The load identification accuracy after transfer learning has been significantly improved. The present invention successively uses simulation data and a small amount of experimental data to train the deep neural network in stages, and improves the inverse load identification accuracy of the hanging structure through the transfer learning method. It can effectively solve problems such as the high cost of obtaining a large number of experimental samples and the insufficient number of training samples obtained through experiments.

[0135] Although the embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the above specific embodiments and application fields. The above specific embodiments are merely illustrative and guiding, rather than restrictive. Those of ordinary skill in the art can also make many forms under the inspiration of this specification and without departing from the scope protected by the claims of the present invention, and these all belong to the scope of protection of the present invention.

Claims

1. A load identification method for a hanging structure based on transfer learning, characterized in that It includes the following steps: Step S1, establish a finite element model of the hanging structure and calculate the vibration response under different impact load excitations, and construct a simulation data set including simulated vibration response signals and concentrated load signals; Step S2, collect the vibration response signals of the monitored hanging structure under the action of concentrated impact loads, and at the same time collect the corresponding concentrated load time history to construct an experimental data set; Step S3, construct a gated convolutional network, and use the simulation data set for pre-training to obtain a pre-trained model. The gated convolutional network includes a feature extractor composed of K gated convolutional blocks connected in sequence and a regressor composed of a fully connected layer; Step S4, use the experimental data set to fine-tune the parameters of the pre-trained model to obtain a load identification network model for the hanging structure; Step S5, use the load identification network model of the hanging structure to inversely identify the concentrated impact load of the hanging structure.

2. The load identification method of a hanging structure based on transfer learning according to claim 1, characterized in that, Preferably, step S1 includes Step S1.1, establish a three-dimensional geometric model according to the geometric dimensions of the hanging structure; Step S1.2, obtain the corresponding material properties according to the materials used in the hanging structure; Step S1.3, establish a finite element model of the hanging structure according to the three-dimensional geometric model and the material properties; Step S1.4, according to the finite element model of the hanging structure, apply a concentrated impact load and the same constraint conditions as the monitored hanging structure, and calculate the simulated vibration response signal of the response point; Step S1.5, use the simulated vibration response signal as the network input feature, and use the corresponding concentrated impact load time history as the corresponding sample label to construct a simulation data set.

3. A load identification method for a hanging structure based on transfer learning according to claim 1, characterized in that, In step S2, the vibration response signal collected by the strain gauge is used as the network input feature, and the concentrated impact load time history obtained by the dynamometer is used as the corresponding sample label to establish an experimental data set.

4. A load identification method for a suspension structure based on transfer learning according to claim 1, characterized in that Step S3 includes: Step S3.1, construct a gated convolutional network feature extractor by sequentially connecting multiple gated convolutional blocks. Each gated convolutional block contains two parallel dilated convolutional layers, a convolutional layer with a kernel length of 1, two batch normalization layers, and a non-linear activation layer. The operations of the gated convolutional block are expressed as follows: , where x is the input feature of the gated convolution block, represents a convolutional layer with a kernel length of 1, relu represents the rectified linear unit, represents the dilated convolution operation, W and V are the convolutional kernel weight parameters in the convolutional layer, b and c respectively represent the bias weight parameters of the convolutional layer, represents the dot product operation, represents the sigmoid non-linear function, bn is the batch normalization operation. The dilated convolution in the gated convolution block expands the receptive field of the convolutional kernel. At the same time, the sigmoid non-linear function controls the proportion of the features extracted by the convolutional kernel passed to the next layer, and the gradient stability of the deep network is maintained through the residual connection; Step S3.2, construct a regressor, and use the features output by the feature extractor through a fully connected layer to predict the concentrated load, and output the inverse identification result of the concentrated impact load of the hanging structure; Step S3.3, construct a gated convolutional neural network, the gated convolutional network includes a feature extractor composed of K gated convolutional blocks connected in sequence and a regressor composed of a fully connected layer, Step S3.4, divide the simulation data set into a training set and a validation set, and pre-train the gated convolutional network. The mean square error MSE is used as the loss function during the pre-training process: ， where is the number of training samples, represents the sample length, represents the predicted value of the i-th sample, represents the true value of the i-th sample.

5. A load identification method for a hanging structure based on transfer learning according to claim 1, characterized in that, In step S4, the experimental data set is divided into a training set and a validation set.

6. The load identification method of a suspension structure based on transfer learning according to claim 1, characterized in that, In step S5, collect the vibration response of the monitored hanging structure under the impact load, input it into the load identification network model of the hanging structure, and obtain the inverse identification result of the concentrated load.

7. A load identification method for a suspension structure based on transfer learning according to claim 1, characterized in that The hanging structure includes a hanging object, a hanging inner lining board, hanging points and connecting pieces.

8. A load identification system for a hanging structure based on transfer learning, characterized in that, It includes A finite element modeling unit, which is used to establish a finite element model of the hanging structure and calculate the vibration response under different impact load excitations, and construct a simulation data set including simulated vibration response signals and concentrated load signals; A collection unit, which is used to collect the vibration response signals of the monitored hanging structure under the action of concentrated impact loads, and at the same time collect the corresponding concentrated load time history to construct an experimental data set; A building unit, which is used to build a gated convolutional network and obtain a pre-trained model by pre-training with a simulation data set; A fine-tuning unit, which is used to fine-tune the parameters of the pre-trained model with an experimental data set to obtain a hanging structure load identification network model; An identification unit, which is used to inversely identify the concentrated impact load of the hanging structure by using the hanging structure load identification network model.

9. A computer storage medium, characterized in that, The storage medium includes computer instructions, which, when running on a computer, cause the computer to execute the method according to any one of claims 1-7.

10. An electronic device, characterized in that, The electronic device includes: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein When the processor executes the program, the method according to any one of claims 1-7 is implemented.

Images