Large bearing structure digital twinning test method and related device

Abstract

The invention discloses a large-scale bearing structure digital twinning test method and a related device, which utilize limited measurement point data to map structure global stress response, improve the prediction capability of a model under a complex loading condition through a multi-task transfer learning mechanism, and provide a visualization method to render a model calculation result. The problem that a traditional method is insufficient and incomplete in structural state prediction accuracy under complex conditions is solved, and the usability of the digital twinning technology is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of digital twins of large-scale structural static tests, and in particular to a digital twin test method and related devices for large-scale load-bearing structures. Background Art

[0002] The liquid rocket engine frame is the main load-bearing component of the engine, and is responsible for the important task of transmitting the engine thrust. Once the rocket engine frame fails, it will directly cause the rocket propulsion system to fail, which will lead to the failure of the flight mission and bring huge economic losses or even serious casualties. Therefore, static tests need to be carried out on large frame structures before the flight mission to ensure the smooth completion of the flight mission.

[0003] However, using rocket engine mounts to conduct static tests is accompanied by high costs and high risks, and actual measured data on the engine mounts is difficult to obtain. The above problems have greatly limited the smooth development of large-scale structural health status prediction technology.

[0004] The traditional static test monitoring method is to evaluate the health status of the structure by placing strain sensors at key parts of the structure, simulating actual working conditions, and analyzing the test data. For large load-bearing structural parts, this method faces two main problems: (1) The structure is large in size. In order to accurately monitor the health status of the structure, a large number of sensors need to be placed, which seriously reduces the efficiency of the delivery process. (2) The structure is complex in size. On some curved surfaces and narrow spaces, it is difficult to place sensors or the measurement accuracy is seriously insufficient, which will seriously affect the monitoring reliability.

[0005] At present, the use of neural network technology to establish a digital twin model of the structure to achieve structural health monitoring has been widely studied, and the full-field stress response is mapped by inputting the load sensor signal. However, this method only considers the establishment of such a mapping, and fails to consider how to effectively optimize the overall prediction effect of the model based on the measured signal in actual application. Summary of the invention

[0006] The purpose of the present invention is to provide a digital twin test method and related devices for a large load-bearing structure to overcome the defects of the prior art. The transfer learning method based on the multi-task learning mechanism of the present invention can simultaneously extract truly effective information from a large amount of simulation information and a small amount of measured signals, thereby achieving high-accuracy and high-resolution structural state response prediction, and mapping the full-field stress response of the structure through a small number of measuring point signals, which has important guiding significance for realizing rapid verification of large load-bearing structures with the same structure and accurate identification of structural abnormal states.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme: The digital twin test method of large load-bearing structures includes the following steps: Step 1: Perform finite element numerical simulation according to the working conditions in the static test of the rocket engine frame. After obtaining the simulation results of the frame structure under different working conditions, extract the full-field stress response of the frame structure simulation data to obtain a simulation data set. At the same time, establish a measured data set collected by the static test of the frame, and construct a simulation and measured point data set based on the simulation data set and the measured data set; Step 2: Use the full-field stress response of the simulation data in step 1 to train the autoencoder. Then, take out the decoder part of the autoencoder as the generator. Then use the simulated measured point data set to train a simple connector from the measured point dimension to the autoencoder latent space. Combine the generator and the simple connector to obtain a preliminary stress state prediction digital twin model. Step 3: Use the measured data set in step 1 to input, adjust and verify the preliminary stress state prediction digital twin model to obtain a revised digital twin model; Step 4: Design a full-field stress visualization method for the model, update the predicted stress in real time in the form of cloud rendering, deploy the modified digital twin model, infer the full-field response of the modified digital twin model based on the measured data set, and visualize the results; Step 5: Replace the measured data set input into the modified digital twin model with the strain data of the measuring points collected in real time during the static test, and then perform the self-iterative optimization of the model through the multi-task transfer learning mechanism, continuously input the strain data of the measuring points into the modified digital twin model, and iterate and update the accuracy of the modified digital twin model in real time.

[0008] Furthermore, the full-field stress response of the rack structure simulation data in step 1 is expressed as , the measured data set collected by the frame static test is expressed as ,in and Represent the simulation data and measured data of each working condition respectively, and n represents the total number of test samples; and , , where N represents the node, and the superscripts s and e represent the simulation and experiment, respectively. represents the number of nodes in the simulation results, Represents the number of structure sensor measurement points, .

[0009] Furthermore, the construction of the simulation and measured point data set according to the simulation data set and the measured data set is specifically as follows: Obtain the measurement points that are consistent with the stress change trend of the measured sensor points from the simulation data set, and construct the simulation and measured point data set .

[0010] Furthermore, the finite element numerical simulation in step 1 is specifically as follows: According to the load size, application position and boundary conditions during the test, design the finite element model of the frame structure in the finite element analysis software, set the grid size according to the twin requirements; submit the calculation job in the finite element analysis software; After the calculation is completed, the analysis results under each loading condition are stored as RPT files, and the finite element model calculation result information is stored as INP files. Then, Python is used to extract the node information, and the simulation results of the rack structure under different working conditions are obtained. The Mises stress information in the simulation results is extracted and saved as a deep learning network training form. When extracting Mises stress information, the measurement point matching is completed at the same time. Specifically, find the simulation model nodes that match the measured sensor measurement points during the actual test in the simulation model, select the nodes on the simulation model from three aspects: spatial position, stress change trend and numerical value, to ensure the correct matching of the measurement points, thereby establishing a simulation data set, a measured data set and a real measurement point data set corresponding to the measured points in the simulation data set.

[0011] Furthermore, the step 2 is specifically as follows: Train an autoencoder using a simulated dataset; Use the decoder of the autoencoder as the generator; The latent space data of the trained autoencoder part is saved, and the simulated measured point data set is input to output the latent space data, and a simple mapper from the low-dimensional input signal to the latent space is obtained. Finally, the simple mapper and the generator are assembled to obtain a preliminary stress state prediction digital twin model that maps the measured point data to the full-field stress response.

[0012] Furthermore, the design model full-field stress visualization method is specifically as follows: Receive binary information stream from real-time acquisition based on TCP / IP protocol through Python, and decode the binary information stream to obtain real-time measured strain data; The strain data is input into the modified digital twin model to obtain the full-field stress prediction results of the rack structure. In the visualization part, the .vtk files of the rack structure are batch displayed in the form of Unstructured Grid, and stress values ​​are assigned to each node. The cloud map display effect is generated by customizing the color bar, and the cloud map is updated after each test data is read, thus realizing digital twins.

[0013] Furthermore, the model self-iterative optimization process is specifically as follows: Each measured sensor signal is divided into three types of signals, namely input signal, adjustment signal and verification signal. First, the full-field prediction of the finite element simulation model and the full-field prediction of the digital twin correction model are compared to construct loss function 1; at the same time, the adjustment signal, that is, the signal used to adjust the training model, is compared to construct loss function 2; finally, the input signal in the model prediction result and the collected input signal are compared to construct loss function 3; weights are assigned to the three loss functions and then summed up to obtain a comprehensive loss function, which is used to optimize and correct the digital twin model. Finally, the optimization result of the corrected digital twin model is verified according to the verification signal.

[0014] Large-scale load-bearing structure digital twin test system, including: Dataset establishment module: used to perform finite element numerical simulation according to the working conditions in the rocket engine frame static test, obtain the simulation results of the frame structure under different working conditions, extract the full-field stress response of the frame structure simulation data, and obtain the simulation data set. At the same time, establish the measured data set collected by the frame static test, and construct the simulation measured point data set based on the simulation data set and the measured data set; Model construction module: used to train the autoencoder with the full-field stress response of simulation data, then take out the decoder part of the autoencoder as the generator, and then use the simulated measured point data set to train the simple connector from the measured point dimension to the autoencoder latent space, combine the generator and the simple connector to obtain the preliminary stress state prediction digital twin model; Model correction module: used to input, adjust and verify the preliminary stress state prediction digital twin model using the measured data set to obtain a corrected digital twin model; Visualization module: used to design the model's full-field stress visualization method, update the predicted stress in real time in the form of cloud rendering, deploy the modified digital twin model, infer the full-field response of the modified digital twin model based on the measured data set, and visualize the results; Self-iterative optimization module: It is used to replace the measured data set input into the modified digital twin model with the strain data of the measuring points collected in real time during the static test, and then perform the self-iterative optimization process of the model through the multi-task transfer learning mechanism, continuously input the strain data of the measuring points into the modified digital twin model, and iterate and update the accuracy of the modified digital twin model in real time.

[0015] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the large-scale load-bearing structure digital twin test method when executing the computer program.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the large-scale load-bearing structure digital twin test method.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: 1) Based on the deep learning multi-task transfer learning mechanism, this paper proposes to use simulation data sets and measured data sets to jointly optimize the model, achieving high-accuracy and high-dimensional global stress prediction, and solving the problem that the overall prediction accuracy decreases instead of increases during the optimization process of general digital twin models.

[0018] 2) The digital twin model for predicting the full-field stress state constructed by the present invention can invert the global stress response of the structure through limited measuring points, which solves the problem of node dimension disaster of general digital twin models. At the same time, it solves the problems of low verification efficiency and insufficient accuracy in the early stage of project delivery, and reduces the difficulty of establishing a model for learning low-dimensional mapping to high latitude.

[0019] 3) The present invention proposes a twin result visualization method, which extracts mesh information from the structural model INP file and aligns it with the prediction results of the spliced ​​model. Finally, it is visualized and output through Pyvista to obtain a digital twin model that can respond in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings in the specification are used to provide further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0021] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is the principle diagram of measuring point matching in the method of the present invention; Figure 3 It is the prediction effect diagram of the present invention; Figure 4 are the prediction results of the twin model, where (a) is the prediction result of the component structure at the 1st loading level, (b) is the prediction result of the component structure at the 7th loading level, (c) is the prediction result of the component structure at the 17th loading level, and (d) is the prediction result of the component structure at the 24th loading level; Figure 5 Comparison chart of the twin model's input signal prediction results and the measured sensor signal; Figure 6 The diagram shows the predicted results of the measured input signal input into the preliminary component-level twin model at three measuring points; Figure 7The following is a comparison chart of the prediction effects at the measuring points of the two adjustment methods, where (a) is the traditional transfer learning method and (b) is the multi-task learning mechanism adjustment method. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] Embodiment 1 A large-scale load-bearing structure digital twin test method comprises the following steps: Step 1: In the offline stage, finite element numerical simulation is performed according to the working conditions in the static test of the rocket engine frame. After obtaining the simulation results of the frame structure under different working conditions, the full-field stress response of the frame structure simulation data is extracted. , and obtain the simulation data set. At the same time, establish the measured data set collected by the frame static test .in, and They represent the simulation data and experimental data of each working condition respectively. , ,in represents the number of nodes in the simulation results, Represents the number of structural sensor measurement points. Generally speaking , n represents the total number of test samples, N represents the node, and the superscripts s and e represent simulation and test, respectively. Subsequently, it is necessary to complete the measurement point matching, obtain the measurement points that are consistent with the stress change trend of the measured sensor measurement points from the simulation data set, and construct the simulation and measurement point data set , it can be seen that the number of working condition nodes in each simulation measurement point data set is consistent with the number of measurement points.

[0025] Finite element analysis specifically includes: establishing a finite element analysis model according to the geometric dimensions of the research object, defining the assembly and constraint relationships between components completely, determining the load size, application position and boundary conditions during the test, designing a finite element model of the structure in finite element analysis software such as ABAQUS or ANSYS, setting the grid size according to the twin requirements, submitting the calculation job in the finite element analysis software, and saving and accessing the analysis results under each loading condition as an RPT file after the calculation, saving and accessing the finite element model information as an INP file, and then using python to extract the node information, and obtain the simulation results of the rack structure under different working conditions, extracting the Mises stress information in the simulation results and saving it in the form of deep learning network training. When extracting the Mises stress information, it is necessary to complete the measurement point matching, find the measured sensor in the simulation model, select it from three aspects: spatial position, stress change trend and numerical value, ensure the correct matching of the measurement points, and establish a simulation data set, a measured data set and a real measurement point data set corresponding to the measured points in the simulation data set.

[0026] Step 2: Construct a digital twin model for stress state prediction based on an autoencoder neural network. Specifically, train an autoencoder with the full-field stress response of the simulation data in step 1, then take out the decoder part of the autoencoder as a generator, and then use the simulation measured point data set to train a simple connector (Simple Mapper) from the measured point dimension to the latent space of the autoencoder. Combine the generator and the simple connector to complete the construction of the preliminary stress state prediction digital twin model. First, we use the input simulation data full-field stress response to train an autoencoder. The autoencoder can automatically compress high-dimensional data into low-dimensional space, and then map the low-dimensional space to the input. By inputting the simulation data full-field stress response training, the decoder part of the autoencoder is the generator we need. The data type is composed of a one-dimensional data sequence. At this time, the input dimension of the generator is different from the measurement point dimension. It is also necessary to establish a simple mapping space from the measurement point dimension to the input dimension. We save the hidden space data of the training autoencoder part, input the measurement point data, output the hidden space data, and obtain such a simple connector. Finally, assemble the generator and the simple connector to obtain a preliminary stress state prediction digital twin model that maps from the measurement point data to the full-field stress response. The training process of the preliminary stress state prediction digital twin model can take each part in the rack structure as the basic unit, or directly input the entire model.

[0027] The stress correction-based full-field stress state digital twin prediction model integrates multiple neural network models to establish a stress prediction tool with high robustness and generalization ability. First, in order to establish a neural network model that maps from low dimensions to high dimensions, the generation model is given priority, and the autoencoder architecture is adopted to automatically construct a mapping model from the measurement point dimension to the prediction dimension. It should be pointed out here that since the number of predicted stress field nodes is often above 10,000, and the number of measured points is small, the potential representation dimension of the autoencoder cannot directly match the measurement point dimension. Therefore, a convolution layer with better feature extraction is used in the network layer, and the convolution kernel parameters are used to process different parts of the input data, reducing the number of parameters of the model and reducing the complexity of the model. At the same time, a simple mapper is also trained in the mapping of the model measurement point dimension to the latent space representation, which fully extracts and represents the complex features in the data. In the encoder part, each layer undergoes a downsampling operation to achieve sufficient dimensionality reduction of the data. In the decoder part, the deconvolution layer is used to implement the generation function of the model. Its specific model parameters are as follows:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040] in, Represents the convolutional layer , LeakyReLU and Softplus are activation functions, represents the deconvolution operation, y is the final output, represents a linear layer, is the maximum pooling layer operation, Represents the output of each layer. The simple mapper model is as follows:

[0041]

[0042]

[0043]

[0044]

[0045]

[0046] in, represents the input dimension, Indicates the output dimension.

[0047] Step 3: Modify the preliminary stress state prediction digital twin model based on the measured data set to improve the accuracy of the full-field stress state prediction; specifically, use the measured data set in step 1 to input, adjust and verify the preliminary stress state prediction digital twin model to obtain a modified digital twin model. During the modification process, the measured data set is also divided into three types of signals, namely input signal, adjustment signal and verification signal. The input signal is input into the model to obtain the full-field response, and then the adjustment signal is used to improve the model prediction accuracy. The verification signal is used to verify the adjustment effect. If the optimization effect is insufficient, readjust it until the model is adjusted to the optimal state.

[0048] Step 4: Design a full-field stress visualization method for the model, update the predicted stress in the form of cloud rendering, deploy the digital twin model, infer and modify the full-field response of the digital twin model based on the measured data set, and visualize the results.

[0049] The binary information stream from real-time acquisition is received through Python based on the TCP / IP protocol, and the binary information stream is decoded to obtain the real-time measured strain data. The strain data is input into the modified digital twin model to obtain the full-field stress prediction results of the rack structure. In the visualization part, the inp file obtained from the finite element model in step 1 is used to extract the component mesh node coordinates and unit information using python, and multiple units are reconstructed into triangular meshes and saved as .vtk files. The .vtk files of the rack structure are displayed in batches in the form of Unstructured Grid, and each node is assigned the stress prediction result obtained in step 3. The cloud map display effect is generated by customizing the color bar, and the cloud map is updated after each test data is read, thereby realizing the digital twin.

[0050] The results are spliced ​​into global stress prediction results, and the model is rendered after the results are aligned with the model nodes to obtain the global stress display results.

[0051] Step 5: During the on-site test, the measured data set input into the modified digital twin model is replaced with the strain data of the measuring points collected in real time during the static test. Then, the model is self-iteratively optimized through a multi-task transfer learning mechanism. The strain data of the measuring points are continuously input into the modified digital twin model, and the model accuracy is updated iteratively in real time.

[0052] The self-iterative optimization process of the model is to adaptively assign weights to the simulation information and the measured point information through the constructed loss function, so as to retain the effective value of the two different information sources to the greatest extent, and make the prediction results of the model more accurate. Each measured sensor signal is divided into three types of signals, namely input signal, adjustment signal and verification signal. First, the full-field prediction of the finite element simulation model and the full-field prediction of the digital twin correction model are compared to construct loss function 1; at the same time, the adjustment signal, that is, the signal used to adjust the training model, is compared to construct loss function 2; finally, the input signal in the model prediction result and the collected input signal are compared to construct loss function 3; weights are assigned to the three loss functions and then summed to obtain a comprehensive loss function, which is used to optimize and correct the digital twin model, and finally the optimization result of the corrected digital twin model is verified according to the verification signal.

[0053] The present invention uses simulation and measured data, takes simulation data as the basis, and combines the advantages of wide dimension of simulation data and high precision of measured data through data fusion algorithm to establish a digital twin model, so as to improve the prediction accuracy, ensure the smooth execution of tasks in the healthy state of the structure, and improve the efficiency of project delivery. At the same time, the digital twin model can continuously use simulation data to self-optimize and update the model prediction accuracy. At the same time, in order to avoid the negative migration phenomenon, a multi-task mechanism transfer learning mode is adopted to ensure that the overall prediction of the model can be effectively optimized during the model adjustment process. At the same time, a model full-field state visualization scheme is proposed in the test phase to improve the real-time display performance of the model.

[0054] Embodiment 2 In order to better illustrate the technical effect of the present invention, an example is used to test and verify the present invention at different levels. This embodiment conducts a digital twin experiment based on a static test of a certain type of liquid rocket engine frame. The loading conditions of the static test are shown in the following table:

[0055] First, in order to verify the accuracy of the proposed digital twin technology in inferring the full-field response of the structure based on limited measured data, a part-level digital twin model is first established. Based on the above loading conditions, finite element simulation is performed in Abaqus to obtain the ODB result file and output the RPT file for each loading level. Python is used to extract the data of the specified component and save it in txt file format. The number of lines in each file is the number of nodes of the component, and the data type of each line is:

[0056] At the same time, the component has three measurement sensors, namely S1, SS1 and S2. Next, it is necessary to complete the measurement point matching and determine that the node IDs corresponding to these three sensors in the simulation model are 4816, 2699 and 9387. Use python to extract the data corresponding to the ID in the txt file and save it as a TXT file. This experiment mainly uses the neural network architecture of the autoencoder. In the model correction, the negative transfer phenomenon of traditional transfer learning is demonstrated. After the model training is completed, the real-time response of the structure can be calculated by inputting the measured data set into the model and visualized. In the visualization part, by assigning index values ​​to the VTK files of each component and expressing and rendering them with the Pyvista library, a digital twin model that can respond to the structural status in real time is obtained. The specific model parameters are as follows:

[0057] The proposed preliminary twin model predicts the following Figure 4 shown. Figure 4It shows the effect of the proposed model in fitting finite element data, and the fitting error is less than 0.02%.

[0058] Continue to build a digital twin model that maps global signals with a small number of sensors. The present invention uses the S1 sensor as the input signal, the S2 sensor as the fine-tuning signal, and the S3 sensor as the verification signal. First, compare the difference between the model prediction and the measured signal, such as Figure 5 and Figure 6 , it can be seen that the prediction accuracy of the model at the input measurement point is very high. However, the prediction accuracy at the other two measurement points is not satisfactory.

[0059] The effect of simple migration using SS1 sensor is as follows Figure 7 (a) is shown, while Figure 7 (b) This method is used. The specific errors are shown in the table below. It can be seen that when using the SS1 sensor for transfer learning, the error of S2 does not decrease but increases, indicating that when using new sensor information, the optimization of the model does not have a positive impact on the overall situation. However, using this method can effectively obtain useful knowledge from the new sensor, while reducing the error of the SS1 sensor, the prediction accuracy of S2 is improved.

[0060]

[0061] At the same time, according to the following error table, the present invention also uses the whole rack for testing and obtains better results. The whole rack has 72 sensors, which are divided into three groups of signals. It can be seen that this method still effectively improves the global prediction accuracy of the twin model.

[0062]

[0063] Embodiment 3 The present invention provides a large-scale load-bearing structure digital twin test system, comprising: Dataset establishment module: used to perform finite element numerical simulation according to the working conditions in the rocket engine frame static test, obtain the simulation results of the frame structure under different working conditions, extract the full-field stress response of the frame structure simulation data, and obtain the simulation data set. At the same time, establish the measured data set collected by the frame static test, and construct the simulation measured point data set based on the simulation data set and the measured data set; Model construction module: used to train the autoencoder with the full-field stress response of simulation data, then take out the decoder part of the autoencoder as the generator, and then use the simulated measured point data set to train the simple connector from the measured point dimension to the autoencoder latent space, combine the generator and the simple connector to obtain the preliminary stress state prediction digital twin model; Model correction module: used to input, adjust and verify the preliminary stress state prediction digital twin model using the measured data set to obtain a corrected digital twin model; Visualization module: used to design the model's full-field stress visualization method, update the predicted stress in real time in the form of cloud rendering, deploy the modified digital twin model, infer the full-field response of the modified digital twin model based on the measured data set, and visualize the results; Self-iterative optimization module: It is used to replace the measured data set input into the modified digital twin model with the strain data of the measuring points collected in real time during the static test, and then perform the self-iterative optimization process of the model through the multi-task transfer learning mechanism, continuously input the strain data of the measuring points into the modified digital twin model, and iterate and update the accuracy of the modified digital twin model in real time.

[0064] Embodiment 4 The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the large-scale load-bearing structure digital twin test method are implemented.

[0065] Embodiment 5 The present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the large-scale load-bearing structure digital twin test method are implemented.

[0066] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0068] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0069] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims of the invention.

Claims

1. A digital twin test method for large load-bearing structures, characterized in that: The steps include: Step 1: Perform finite element numerical simulation according to the working conditions in the static test of the rocket engine frame. After obtaining the simulation results of the frame structure under different working conditions, extract the full-field stress response of the frame structure simulation data to obtain a simulation data set. At the same time, establish a measured data set collected by the static test of the frame, and construct a simulation and measured point data set based on the simulation data set and the measured data set; Step 2: Use the full-field stress response of the simulation data in step 1 to train the autoencoder. Then, take out the decoder part of the autoencoder as the generator. Then use the simulated measured point data set to train a simple connector from the measured point dimension to the autoencoder latent space. Combine the generator and the simple connector to obtain a preliminary stress state prediction digital twin model. Step 3: Use the measured data set in step 1 to input, adjust and verify the preliminary stress state prediction digital twin model to obtain a revised digital twin model; Step 4: Design a full-field stress visualization method for the model, update the predicted stress in real time in the form of cloud rendering, deploy the modified digital twin model, infer the full-field response of the modified digital twin model based on the measured data set, and visualize the results; Step 5: Replace the measured data set input into the modified digital twin model with the strain data of the measuring points collected in real time during the static test, and then perform the self-iterative optimization of the model through the multi-task transfer learning mechanism, continuously input the strain data of the measuring points into the modified digital twin model, and iterate and update the accuracy of the modified digital twin model in real time.

2. The large-scale load-bearing structure digital twin test method according to claim 1 is characterized in that: The full-field stress response of the rack structure simulation data in step 1 is expressed as , the measured data set collected by the frame static test is expressed as ,in and Represent the simulation data and measured data of each working condition respectively, and n represents the total number of test samples; and , , where N represents the node, and the superscripts s and e represent the simulation and experiment, respectively. represents the number of nodes in the simulation results, Represents the number of structure sensor measurement points, .

3. The large-scale load-bearing structure digital twin test method according to claim 2 is characterized in that: The construction of the simulation and measured point data set according to the simulation data set and the measured data set is specifically as follows: Obtain the measurement points that are consistent with the stress change trend of the measured sensor points from the simulation data set, and construct the simulation and measured point data set .

4. The large-scale load-bearing structure digital twin test method according to claim 1 is characterized in that: The finite element numerical simulation described in step 1 is specifically as follows: According to the load size, application position and boundary conditions during the test, the finite element model of the frame structure is designed in the finite element analysis software, and the grid size is set according to the twin requirements; Submit calculation jobs in finite element analysis software; After the calculation is completed, the analysis results under each loading condition are stored as RPT files, and the finite element model calculation result information is stored as INP files. Then, Python is used to extract the node information, and the simulation results of the rack structure under different working conditions are obtained. The Mises stress information in the simulation results is extracted and saved as a deep learning network training form. When extracting Mises stress information, the measurement point matching is completed at the same time. Specifically, find the simulation model nodes that match the measured sensor measurement points during the actual test in the simulation model, select the nodes on the simulation model from three aspects: spatial position, stress change trend and numerical value, to ensure the correct matching of the measurement points, thereby establishing a simulation data set, a measured data set and a real measurement point data set corresponding to the measured points in the simulation data set.

5. The large-scale load-bearing structure digital twin test method according to claim 1 is characterized in that: The step 2 is specifically as follows: Train an autoencoder using a simulated dataset; Use the decoder of the autoencoder as the generator; The latent space data of the trained autoencoder part is saved, and the simulated measured point data set is input to output the latent space data, and a simple mapper from the low-dimensional input signal to the latent space is obtained. Finally, the simple mapper and the generator are assembled to obtain a preliminary stress state prediction digital twin model that maps the measured point data to the full-field stress response.

6. The large-scale load-bearing structure digital twin test method according to claim 1 is characterized in that: The full-field stress visualization method of the design model is specifically as follows: Receive binary information stream from real-time acquisition based on TCP / IP protocol through Python, and decode the binary information stream to obtain real-time measured strain data; The strain data is input into the modified digital twin model to obtain the full-field stress prediction results of the rack structure. In the visualization part, the .vtk files of the rack structure are batch displayed in the form of Unstructured Grid, and stress values ​​are assigned to each node. The cloud map display effect is generated by customizing the color bar, and the cloud map is updated after each test data is read, thus realizing digital twins.

7. The large-scale load-bearing structure digital twin test method according to claim 1 is characterized in that: The model self-iterative optimization process is specifically as follows: Each measured sensor signal is divided into three types of signals, namely input signal, adjustment signal and verification signal. First, the full-field prediction of the finite element simulation model and the full-field prediction of the digital twin correction model are compared to construct loss function 1; at the same time, the adjustment signal, that is, the signal used to adjust the training model, is compared to construct loss function 2; finally, the input signal in the model prediction result and the collected input signal are compared to construct loss function 3; weights are assigned to the three loss functions and then summed up to obtain a comprehensive loss function, which is used to optimize and correct the digital twin model. Finally, the optimization result of the corrected digital twin model is verified according to the verification signal.

8. Large-scale load-bearing structure digital twin test system, characterized by: include: Dataset establishment module: used to perform finite element numerical simulation according to the working conditions in the rocket engine frame static test, obtain the simulation results of the frame structure under different working conditions, extract the full-field stress response of the frame structure simulation data, and obtain the simulation data set. At the same time, establish the measured data set collected by the frame static test, and construct the simulation measured point data set based on the simulation data set and the measured data set; Model construction module: used to train the autoencoder with the full-field stress response of simulation data, then take out the decoder part of the autoencoder as the generator, and then use the simulated measured point data set to train the simple connector from the measured point dimension to the autoencoder latent space, combine the generator and the simple connector to obtain the preliminary stress state prediction digital twin model; Model correction module: used to input, adjust and verify the preliminary stress state prediction digital twin model using the measured data set to obtain a corrected digital twin model; Visualization module: used to design the model's full-field stress visualization method, update the predicted stress in real time in the form of cloud rendering, deploy the modified digital twin model, infer the full-field response of the modified digital twin model based on the measured data set, and visualize the results; Self-iterative optimization module: It is used to replace the measured data set input into the modified digital twin model with the strain data of the measuring points collected in real time during the static test, and then perform the self-iterative optimization process of the model through the multi-task transfer learning mechanism, continuously input the strain data of the measuring points into the modified digital twin model, and iterate and update the accuracy of the modified digital twin model in real time.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the large-scale load-bearing structure digital twin testing method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the large-scale load-bearing structure digital twin testing method as described in any one of claims 1 to 7 are implemented.

Images