Satellite structure focus response real-time indirect measurement method
Through the combination of the finite element model and the LSTM-VAE neural network model, the problem of difficult monitoring of the satellite structure's focus response is solved, real-time monitoring of the satellite structure's focus response is realized, and the reliability and accuracy of satellite service in orbit are improved.
Patent Information
- Application Number
- CN202411085112.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-07
AI Technical Summary
During the launch and orbital operation of satellite structures, the response to concerns is difficult to suppress due to vibration loads, the existing vibration-absorbing and vibration isolation methods are limited, the damping performance of traditional materials is poor, the numerical simulation calculation efficiency is low, and it is difficult to achieve vibration in precision instrument parts. Response to real-time monitoring.
Using a combination of finite element model and LSTM-VAE neural network model, a high-precision proxy model is constructed through sensor installation and classification, dynamic experimental data acquisition and model correction to realize real-time monitoring and measurement of satellite structure focus response.
Real-time monitoring of satellite structure concern response is realized, data support is provided for optimized design and health monitoring of satellite structures, and the reliability and accuracy of satellite service in orbit are improved.
Smart Images

Figure CN119004665A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of on-orbit monitoring of spacecraft, and in particular relates to a real-time indirect measurement method for the response of a satellite structure focus point. Background Art
[0002] Satellites play an extremely important role in modern society. With the continuous development of science and technology, the application scope of satellites is becoming more and more extensive, and human beings are becoming more and more dependent on satellites. Artificial satellites provide global communications, weather forecasts, earth observation, resource exploration, navigation and scientific research services for human beings. They can also support the development of transportation, environmental protection, disaster response, national defense and security. Artificial satellites have become an important tool and support for human exploration and development, promoting the continuous development and progress of mankind.
[0003] As the demand for remote sensing satellite functions and performance continues to increase, the complexity of satellite structures also increases, which may lead to inadequate ground testing of satellites, resulting in an increase in early failures of satellites in orbit. According to statistics, about 60% of serious failures occur in the first year of satellite orbiting, and many failures occur shortly after the remote sensing satellite enters orbit. In addition, satellites are affected by thermal loads such as thermal radiation, thermal convection, and thermal conduction in space. The long-term reciprocating changes in thermal loads will cause the expansion and contraction of the satellite's main structure and key components, accelerate the damage of the key structures of the satellite system (such as load-bearing structures), shorten the service life of remote sensing satellites in orbit, and reduce the quality of remote sensing satellites in orbit. Routine health management of spacecraft is usually achieved through a combination of two methods: (1) providing robust, damage-resistant design and structural integrity margin to prevent potential failures; (2) relying on regular maintenance and long-term inspections to determine necessary repairs. However, due to the complexity of the environment and loads and the uncertainty of material properties and model assumptions, structural safety margins may lead to unnecessary heavy structures, which may not necessarily ensure the safety of satellites or mission success rates. In addition, since the predetermined maintenance plan cannot take into account the actual environment and operation, there may be a risk of premature failure during flight. Taking an overly conservative approach, such as increasing the frequency of scheduled maintenance, can effectively extend the satellite's service time in orbit, but it will also increase huge maintenance costs. In general, with the advancement of technology and the increase in application needs, the requirements for the accuracy and stability of satellite platforms are also constantly increasing to ensure their successful execution in various complex missions and long-term reliable operation.
[0004] As an important part of the satellite system, the satellite structure can carry and protect precision instruments and ensure the integrity and stability of the entire system. It plays a vital role in satellite design and operation and maintenance. There may be dynamic coupling between the satellite structure and the payload bracket, which will generate a large amount of vibration loads during launch and on-orbit operation. These vibrations will not only affect the service life of the equipment, but also greatly reduce the measurement accuracy of precision instruments. However, in the field of aerospace, the existing vibration reduction and isolation methods are limited by the poor damping performance of traditional materials, low efficiency of numerical simulation calculations, and lack of systematic configuration optimization design methods, which makes it difficult to suppress the vibration response of the installation site (focus point) of precision instruments. Due to space constraints, the dynamic response of the focus point is difficult to measure directly using sensors. In summary, there is an urgent need for a real-time indirect measurement method that can realize the response of the focus point of the satellite structure, paving the way for research on satellite structure optimization design, health monitoring, reliability assessment, etc. Summary of the invention
[0005] In order to overcome the problems existing in the above-mentioned prior art, the present invention provides a real-time indirect measurement method for the focus point response of the satellite structure, so as to obtain the focus point response of the satellite structure in real time, and provide necessary data support for further research of relevant engineering personnel.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is:
[0007] A real-time indirect measurement method for satellite structure point of interest response, the method comprising the following steps:
[0008] Step S1, installation and classification of sensors, a limited number of sensors are arranged on the satellite structure, and the data obtained by the sensors are divided into two categories: measurable point data and focus point data;
[0009] Step S2, satellite structure dynamics data acquisition, obtain the dynamic data of each measuring point of the satellite structure through dynamics experiments, and use the sensors on the vibration table to obtain excitation data. During the experiment, each sensor needs to collect data synchronously;
[0010] Step S3: construct a finite element model of the satellite structure. Use simulation software to build a finite element model of the satellite structure. Modify key parameters in the model through experimental data. Use the excitation data measured in the dynamics experiment as input to obtain virtual data corresponding to the installation position of each sensor in the satellite structure. Specifically:
[0011] S31, construct a three-dimensional model of the satellite structure and perform discretization processing, determine the finite element nodes corresponding to the positions of the actual satellite structure measurement points, and regard them as virtual measurement points;
[0012] S32. Carry out sine frequency sweep and free vibration attenuation experiments to obtain experimental values of the natural frequency and overall attenuation rate of the structure, which are used to modify the key parameters of the model;
[0013] S33, using an optimization algorithm to correct key parameters of the satellite structure finite element model;
[0014] The key parameters of the finite element model are corrected using the measured natural frequency values and the free decay rate of the entire structure. The response surface method is used to correct the key parameters. The natural frequency values of the satellite structure are used to correct the physical properties of the materials in the model, including the elastic modulus, Poisson's ratio and mass, while the free decay data are used to identify the overall damping of the structure.
[0015] S34, using the collected vibration table response as input, and using the modified finite element model to calculate the virtual dynamic data of each corresponding measuring point;
[0016] Step S4: Use the LSTM-VAE neural network model to establish an error distribution model between the measured data and virtual data of the focus point, and then use the virtual data of the focus point under different stimuli as input to generate high-quality dynamic response data of the focus point; specifically:
[0017] S41. Establish LSTM-VAE neural network model;
[0018] The LSTM-VAE model replaces the feedforward neural network in VAE with LSTM. LSTM-VAE combines the local feature learning ability of VAE with the temporal modeling ability of LSTM neural network. The LSTM-VAE model can learn and capture the complex temporal dependencies in time series data, so that it can generate data with similar temporal dependencies.
[0019] S42, taking the measured focus point response as output and the focus point virtual data calculated by the finite element model as input, and obtaining a generator model between the input and the output by adjusting the LSTM-VAE neural network model hyperparameters;
[0020] The virtual response of the satellite structure finite element model calculation is defined as x, and the measured response is defined as y; according to the formula Standardize the data;
[0021] The dataset is divided into a training set and a validation set, and the hyperparameters in the LSTM-VAE neural network model are adjusted to perform supervised training with the goal of minimizing the difference between the output and the measured response; the attention point response generated by the LSTM-VAE model is defined as ;
[0022] S43, quantitatively evaluate the accuracy of the generator model. If the accuracy does not meet the requirements, return to S42 to re-set a new hyperparameter combination for training;
[0023] The validation set is input into the constructed LSTM-VAE generator model, and the vibration response of the improved point of interest is calculated; the performance of the LSTM-VAE generator model is quantitatively evaluated using correlation coefficient, root mean square error, standard root mean square error, and cosine similarity.
[0024] Step S5: construct a high-precision proxy model of the measured measurable point data and the generated focus point data, input the real-time measurable point data into the proxy models of the two, and realize the real-time measurement of the focus point response.
[0025] Furthermore, the step S1 specifically includes:
[0026] S11, select the measuring point and install the sensor;
[0027] S12. Classify sensors according to their installation positions: measurable point sensors and focus point sensors; measurable point data are used as input for the proxy model, and measured focus point data are output data.
[0028] Furthermore, the step S2 specifically includes:
[0029] S21, fix the satellite structure on the vibration table using a clamp;
[0030] The satellite structure is connected to the vibration table through a fixture, and a satellite structure dynamics experimental system is established, and the vibration table is used to realize the input of dynamic loads;
[0031] S22, installing an acceleration sensor on the vibration table to collect excitation data applied to the satellite structure during random vibration;
[0032] Install a sensor on the vibration table to collect the response data of the vibration table during random vibration; the vibration direction collected by the sensor is consistent with the vibration direction of the vibration table during the experiment;
[0033] S23. Carry out random vibration experiments and use a high-speed acquisition system to synchronously collect the dynamic response of each measuring point and the excitation data of the vibration table;
[0034] S24. Classify the data collected by the sensor into: focus point data and measurable point data.
[0035] Furthermore, the step S5 specifically includes:
[0036] S51. Build LSTM neural network model;
[0037] S52, establishing a proxy model of the measured measurable point response and the virtual focus point response through the LSTM neural network model;
[0038] S53, quantitative evaluation of the accuracy of the proxy model. If the model accuracy does not meet the requirements, return to S52 and reselect a new LSTM neural network hyperparameter combination for training;
[0039] S54. Real-time measurable point response data is used as input to the proxy model to achieve real-time indirect measurement of the response of the focus point.
[0040] Beneficial effects of the present invention:
[0041] 1) Installation and classification of sensors: A limited number of sensors are arranged on the satellite structure, and the data obtained by the sensors are divided into two categories: measurable point data and focus point data. The satellite load-bearing structure has limited space, and the focus point response is often difficult to measure directly in real working conditions. The sensor data installed at this location is defined as focus point data. The location where the sensor can be installed is directly measured, and this type of sensor data is defined as measurable point data.
[0042] 2) Satellite structure dynamics data acquisition: dynamic data of each measuring point of the satellite structure is obtained through dynamic experiments, and excitation data is obtained using sensors. During the experiment, each sensor needs to collect data synchronously.
[0043] 3) Construct a finite element model of the satellite structure using CAE simulation software (such as ANSYS), modify the key parameters in the model through experimental data, and use the excitation data measured in the dynamic experiment as input to obtain virtual data corresponding to the position of the satellite structure sensor.
[0044] 4) The long short-term memory variational autoencoder (LSTM-VAE) neural network model is used to establish the error distribution relationship between the measured data and virtual data of the focus point. Then the virtual data under different stimuli are used as input to generate high-quality dynamic response data of the focus point to solve the problem that the focus point response is difficult to obtain.
[0045] 5) Build high-precision proxy models of measurable point data (actual measurement) and focus point data (generated), input real-time measurable point data into the proxy models of the two, and realize real-time measurement of focus point responses.
[0046] This solution can not only realize the real-time monitoring of the response of the focus point, but also lay the foundation for the health monitoring of the satellite structure. The present invention takes into account that the response of the focus point of the satellite structure is difficult or impossible to measure in engineering practice, and can be applied to the indirect measurement of the response of the focus point of the actual satellite structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 The present invention is a flow chart of a real-time indirect measurement method of the response of a satellite structure point of interest.
[0048] Figure 2 Schematic diagram of the satellite structure in an embodiment of the present invention.
[0049] Figure 3 It is a schematic diagram of a dynamic experiment of a satellite structure in an embodiment of the present invention.
[0050] Figure 4 It is a schematic diagram of the LSTM-VAE neural network model structure in an embodiment of the present invention.
[0051] Figure 5 These are the measured and virtual focus point responses in the embodiments of the present invention, wherein the virtual focus point response in (a) is directly obtained by calculating the finite element model, and the virtual focus point response in (b) is further improved by the LSTM-VAE model.
[0052] Figure 6 It is a schematic diagram of the LSTM neural network model structure in an embodiment of the present invention.
[0053] Figure 7 is a time course diagram of the focus point response calculated by the agent model in the embodiment of the present invention. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0055] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.
[0056] like Figure 1 As shown, the real-time indirect measurement method of the satellite structure focus point response provided by this embodiment includes the following steps:
[0057] Step S1, installation and classification of sensors: a limited number of sensors are arranged on the satellite structure, and the data obtained by the sensors are divided into two categories: measurable point data and focus point data.
[0058] This step completes the installation and classification of sensors. The specific implementation process is as follows:
[0059] S11, select the measuring point and install the sensor;
[0060] Determine the satellite structure based on engineering practice ( Figure 2) The locations where sensors can be arranged and the locations where precision instruments are to be installed (points of interest), and then the sensors can be installed.
[0061] S12. Classification by sensor installation location
[0062] According to the installation location of the sensor, the data collected by the sensor is divided into two categories: measurable point data and focus point data. The measurable point data is used as the input of the proxy model, and the measured focus point data is the output data.
[0063] Step S2, satellite structure dynamics data acquisition, obtain the dynamic data of each measuring point of the satellite structure through dynamics experiments, and use sensors to obtain excitation data. During the experiment, each sensor needs to collect data synchronously.
[0064] This step completes the data collection work of each measuring point. The specific implementation process is as follows
[0065] S21. Fix the satellite structure on the vibration table using a clamp.
[0066] The satellite structure is connected to the vibration table through a fixture to establish a satellite structural dynamics experimental system, and the vibration table is used to input dynamic loads. In order to avoid the influence of the experimental fixture on the structural vibration, the structural strength and natural frequency of the fixture are much greater than the relevant dynamic properties of the satellite structure.
[0067] S22. Install acceleration sensors on the vibration table to collect excitation data applied to the satellite structure during random vibration.
[0068] Install sensors on the vibration table to collect the response data of the vibration table during random vibration. The vibration direction collected by the sensor is consistent with the vibration direction of the vibration table during the experiment. The collected vibration table response data can be used as the input load of S34.
[0069] S23. Carry out random vibration experiments and use a high-speed acquisition system to synchronously collect the dynamic response of each measuring point and the excitation data of the vibration table;
[0070] Four acceleration sensors are installed on the surface of the satellite structure outer wall, with an operating frequency of 20 to 9000 Hz. A vibration table is used to apply lateral random vibration excitation. The power spectrum density of the random vibration setting is shown in Table 1.
[0071] Table 1 Power spectral density of satellite random vibration experiment settings
[0072]
[0073] Figure 3This is a schematic diagram of the satellite structural dynamics experiment, in which three accelerometers are installed in the lower part of the structure and defined as measurable points, and one accelerometer is installed in the upper area of the structure and defined as the point of interest. In addition, one accelerometer is installed on the vibration table to measure the dynamic response of the vibration table. High-speed acquisition equipment is used to synchronize data acquisition of the above sensors.
[0074] S24, data classification, classifying the data collected by the sensor into: focus point data and measurable point data.
[0075] The data collected during the S23 process are classified into: measurable point data, focus point data and vibration table data.
[0076] Step S3, construct a finite element model of the satellite structure, use CAE simulation software (such as ANSYS) to construct a finite element model of the satellite structure, modify the key parameters in the model through experimental data, use the excitation data measured in the dynamic experiment as input, and obtain virtual data of points corresponding to the positions of satellite structure sensors.
[0077] This step completes the construction of the satellite structure finite element model, parameter correction and calculation of virtual data of the corresponding measurement points. The specific implementation process is as follows
[0078] S31, construct a three-dimensional model of the satellite structure and perform discretization processing, determine the finite element nodes corresponding to the positions of the actual satellite structure measurement points, and regard them as virtual measurement points;
[0079] Construct a satellite finite element model, divide the finite element mesh, and import the obtained satellite finite element model into the finite element simulation software ANSYS Workbench. It should be noted here that the satellite structural dynamics calculation software is not limited to ANSYS Workbench, but can also be other finite element software. Then, determine the corresponding node position in the discretized finite element model according to the position of each measurement point of the satellite structure.
[0080] S32. Carry out sine frequency sweep and free vibration attenuation experiments to obtain experimental values of the natural frequency and overall attenuation rate of the structure, which are used to modify the key parameters of the model;
[0081] The vibration table was used to perform sweep frequency analysis with a frequency range of 20-100 Hz and an acceleration level of 0.8 g. In addition, transient excitation was applied by hammering to obtain free decay response data of the load-bearing structure.
[0082] S33, using an optimization algorithm to correct key parameters of the satellite structure finite element model;
[0083] The key parameters of the finite element model are corrected using the measured natural frequency values and the free decay rate of the entire structure. Considering that it is very time-consuming to directly call the satellite structure finite element model for key parameter correction and identification, in order to solve this problem, the response surface method is used to correct the key parameters. The natural frequency values of the satellite structure are used to correct the physical properties of the materials in the model, including elastic modulus, Poisson's ratio and mass, while the free decay data are used to identify the overall damping of the structure. Table 2 shows the natural frequency values of the satellite structure finite element model measured and corrected by sinusoidal sweep.
[0084] Table 2 The natural frequency values of the satellite structure sine swept frequency measured and the corrected finite element model
[0085]
[0086] S34, using the vibration table response collected in S23 as input, and using the modified finite element model to calculate the virtual dynamic data of each corresponding measuring point.
[0087] The transient dynamics calculation module in ANSYS workbench is used to perform dynamic analysis on the satellite structure, and the shaking table data collected in S23 is used as the input of the model. After the calculation is completed, the dynamic response of the nodes corresponding to the position of each measuring point of the satellite structure is extracted.
[0088] Step S4, using the LSTM-VAE neural network model to establish an error distribution relationship between the measured data and the virtual data of the focus point, so as to generate high-quality dynamic response data of the focus point;
[0089] This step uses the LSTM-VAE neural network to establish a generative model between the measured and virtual data of the measurable points. The specific implementation process is as follows
[0090] S41. Establish an LSTM-VAE neural network model.
[0091] The LSTM-VAE model structure is as follows Figure 4 As shown in the figure, the feedforward neural network in VAE is replaced by LSTM. LSTM-VAE combines the local feature learning ability of VAE and the temporal modeling ability of LSTM neural network. Considering the temporal correlation between data, the LSTM-VAE model is able to learn and capture the complex temporal dependencies in time series data, so that it can generate data with similar temporal dependencies. The main purpose of establishing the LSTN-VAE neural network model is to narrow the gap between virtual data and measured data.
[0092] S42. Taking the measured focus point response as output and the focus point virtual data calculated by the finite element model as input, the generator model between the input and output is obtained by adjusting the LSTM-VAE neural network model hyperparameters.
[0093] The virtual response of the satellite structure finite element model calculation point is defined as x, and the measured response is defined as y. According to the formula The data is standardized to eliminate the distribution differences between the measured point data. The data set is divided into a training set and a validation set, and the hyperparameters in the LSTM-VAE neural network model are adjusted to perform supervised training with the goal of minimizing the difference between the output and the measured response. The attention point response generated by the LSTM-VAE model is defined as
[0094] S43, quantitatively evaluate the accuracy of the generator model. If the accuracy does not meet the requirements, return to S42 to re-set a new hyperparameter combination for training;
[0095] The validation dataset divided in the S42 process is input into the constructed LSTM-VAE generator model to calculate the vibration response of the improved point of interest. The performance of the LSTM-VAE generator model is quantitatively evaluated using correlation coefficient, root mean square error, standard root mean square error, and cosine similarity. Figure 5 are the responses of the measured and virtual focus points, where (a) is the virtual focus point response obtained directly by calculating the finite element model, and (b) is the virtual focus point response further improved by the LSTM-VAE model.
[0096] Table 3 Quantitative evaluation results of LSTM-VAE model performance
[0097]
[0098] Table 3 gives the quantitative indicators of the focus response. The calculation of the above indicators is as follows:
[0099] Correlation coefficient:
[0100] Root Mean Square Error:
[0101] Normalized root mean square error:
[0102] Cosine Similarity:
[0103] y i represents the true value of the response of the focus point at time i, represents the virtual value of the improved attention point response at time i. y represents the measured attention point response, Represents the improved attention point response virtual data, whose sample length is n.
[0104] Step S5: construct high-precision proxy models of measurable point data (actual measurement) and focus point data (generated), input real-time measurable point data into the proxy models of the two, and realize real-time measurement of focus point response.
[0105] This step uses the LSTM neural network to establish a proxy model between the measured data of the measurable points and the virtual data of the focus points. The specific implementation process is as follows
[0106] Construct LSTM neural network model structure (such as Figure 6 As shown in Figure 1, the measured data of the measurable points are used as input, and the generated attention point responses are used as output to establish a proxy model for the two. The hyperparameters in the LSTM neural network model are adjusted to perform supervised training with the goal of minimizing the difference between the output and the attention point response, and a nonlinear proxy model between the input and output is constructed. The measured measurable point responses are used as input to the proxy model, and the dynamic responses of the attention points are calculated in a very short time. The time course of the attention point response calculated based on the LSTM proxy model is shown in Figure 1. Figure 7 The quantitative evaluation results of the LSTM proxy model accuracy are listed in Table 4.
[0107] Table 4 Quantitative evaluation results of LSTM proxy model accuracy
[0108]
[0109] In summary, the present invention aims to achieve real-time indirect measurement of satellite structure focus point responses. It utilizes finite element modeling, model correction, generative model and proxy model technology to ultimately overcome the challenge that satellite structure focus point responses are difficult or impossible to measure directly, and can achieve real-time indirect measurement of satellite structure focus point responses.
[0110] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A real-time indirect measurement method for satellite structure point of interest response, characterized in that: The method comprises the following steps: Step S1, installation and classification of sensors, a limited number of sensors are arranged on the satellite structure, and the data obtained by the sensors are divided into two categories: measurable point data and focus point data; Step S2, satellite structure dynamics data acquisition, obtain the dynamic data of each measuring point of the satellite structure through dynamics experiments, and use the sensors on the vibration table to obtain excitation data. During the experiment, each sensor needs to collect data synchronously; Step S3: construct a finite element model of the satellite structure. Use simulation software to build a finite element model of the satellite structure. Modify key parameters in the model through experimental data. Use the excitation data measured in the dynamics experiment as input to obtain virtual data corresponding to the installation position of each sensor in the satellite structure. Specifically: S31, construct a three-dimensional model of the satellite structure and perform discretization processing, determine the finite element nodes corresponding to the positions of the actual satellite structure measurement points, and regard them as virtual measurement points; S32. Carry out sine frequency sweep and free vibration attenuation experiments to obtain experimental values of the natural frequency and overall attenuation rate of the structure, which are used to modify the key parameters of the model; S33, using an optimization algorithm to correct key parameters of the satellite structure finite element model; The key parameters of the finite element model are modified using the measured natural frequency value and the free decay rate of the entire structure; The response surface method is used to modify key parameters. The natural frequency value of the satellite structure is used to modify the material properties in the model, including elastic modulus, Poisson's ratio and mass, while the free decay data is used to identify the overall damping of the structure. S34, using the collected vibration table response as input, and using the modified finite element model to calculate the virtual dynamic data of each corresponding measuring point; Step S4: Use the LSTM-VAE neural network model to establish an error distribution model between the measured data and virtual data of the focus point, and then use the virtual data of the focus point under different stimuli as input to generate high-quality dynamic response data of the focus point; specifically: S41. Establish LSTM-VAE neural network model; The LSTM-VAE model replaces the feedforward neural network in VAE with LSTM. LSTM-VAE combines the local feature learning ability of VAE with the temporal modeling ability of LSTM neural network. The LSTM-VAE model can learn and capture the complex temporal dependencies in time series data, so that it can generate data with similar temporal dependencies. S42, taking the measured focus point response as output and the focus point virtual data calculated by the finite element model as input, and obtaining a generator model between the input and the output by adjusting the LSTM-VAE neural network model hyperparameters; The virtual response of the satellite structure finite element model calculation is defined as x, and the measured response is defined as y; according to the formula and Standardize the data; The dataset is divided into a training set and a validation set, and the hyperparameters in the LSTM-VAE neural network model are adjusted to perform supervised training with the goal of minimizing the difference between the output and the measured response; the attention point response generated by the LSTM-VAE model is defined as S43, quantitatively evaluate the accuracy of the generator model. If the accuracy does not meet the requirements, return to S42 to re-set a new hyperparameter combination for training; The validation set is input into the constructed LSTM-VAE generator model, and the vibration response of the improved point of interest is calculated; the performance of the LSTM-VAE generator model is quantitatively evaluated using correlation coefficient, root mean square error, standard root mean square error, and cosine similarity. Step S5: construct a high-precision proxy model of the measured measurable point data and the generated focus point data, input the real-time measurable point data into the proxy models of the two, and realize the real-time measurement of the focus point response.
2. A real-time indirect measurement method for satellite structure point of interest response according to claim 1, characterized in that: The step S1 specifically includes: S11, select the measuring point and install the sensor; S12. Classify sensors according to their installation positions: measurable point sensors and focus point sensors; measurable point data are used as input for the proxy model, and measured focus point data are output data.
3. A real-time indirect measurement method for satellite structure point of interest response according to claim 1, characterized in that: The step S2 specifically includes: S21, fix the satellite structure on the vibration table using a clamp; The satellite structure is connected to the vibration table through a fixture, and a satellite structure dynamics experimental system is established, and the vibration table is used to realize the input of dynamic loads; S22, installing an acceleration sensor on the vibration table to collect excitation data applied to the satellite structure during random vibration; Install a sensor on the vibration table to collect the response data of the vibration table during random vibration; the vibration direction collected by the sensor is consistent with the vibration direction of the vibration table during the experiment; S23. Carry out random vibration experiments and use a high-speed acquisition system to synchronously collect the dynamic response of each measuring point and the excitation data of the vibration table; S24. Classify the data collected by the sensor into: focus point data and measurable point data.
4. A real-time indirect measurement method for satellite structure point of interest response according to claim 1, characterized in that: The step S5 specifically includes: S51. Build LSTM neural network model; S52, establishing a proxy model of the measured measurable point response and the virtual focus point response through the LSTM neural network model; S53, quantitative evaluation of the accuracy of the proxy model. If the model accuracy does not meet the requirements, return to S52 and reselect a new LSTM neural network hyperparameter combination for training; S54. Real-time measurable point response data is used as input to the proxy model to achieve real-time indirect measurement of the response of the focus point.
Citation Information
Patent Citations
Satellite physical field level digital twin model construction method based on deep learning
CN115392077A
System for precision measurement of structure and method therefor
EP3524954A1
