Full-field stress inversion method and device for precision parts of spacecraft

Abstract

The invention provides a full-field stress inversion method and device for a spacecraft precision part. The full-field stress inversion method for the spacecraft precision part comprises the following steps: acquiring real-time isolated point strain data of a target part under various actual working conditions; the target isolated point training data are imported into a trained inversion model, a target full-field stress model of the target part is generated, the inversion model comprises a first neural network and a second neural network which are connected in series, the output factor of the first neural network is six-dimensional load data of the target part, and the output factor of the second neural network is six-dimensional load data of the target part; the target full-field stress model is a finite element simulation model for the target part. According to the method, the mapping relation between the measurement data of the strain gauges pasted on the surfaces of the target parts and the whole-field stress distribution is established, the series-connection double-neural-network architecture model is established by means of the strong learning ability of the neural network model, the external load borne by the target parts under the current working condition is effectively recognized, and the reliability of the whole-field stress distribution is improved. And more accurate whole-field stress distribution is obtained through inversion calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of structural strain testing, and in particular to a full-field stress inversion method and device for precision parts of spacecraft. Background Art

[0002] During the manufacturing process of spacecraft products, assembly stress is considered a key factor causing micro-deformation and prestress in components, which can impact product performance and service life. Spacecraft products are characterized by high reliability requirements, harsh service environments, and complex external load and stress conditions. The prestress introduced by assembly stress can lead to rapid degradation of component life during service, thus impacting the mission reliability of spacecraft products. Therefore, structural strain measurement of spacecraft products is necessary to comprehensively assess the micro-deformation and prestress caused by assembly stress.

[0003] At present, although the existing structural strain measurement methods can measure the surface strain of the structure to a certain extent, they are insufficient in the dimension of the measurement data, and it is difficult to obtain the stress distribution in the narrow area of ​​the complex structure. For example, the resistance strain gauge mainly measures the surface stress of the structure and has a limited depth. In addition, the result of the strain gauge measurement is the average strain of the measurement grid area. When the measured area is close to the grid size and there is a large stress gradient change, it is easy to cause a large measurement error. Relying on simulation analysis to calculate the stress of the part, although it can evaluate the strength of the part more comprehensively and accurately and give the stress value of the stress concentration area of ​​the part, its accuracy is limited by the sophistication of the model and the rationality of the parameter setting. It is difficult to fully simulate the complex stress conditions of the parts during the actual assembly process and cannot fully cover all working conditions. Summary of the Invention

[0004] In view of this, the present invention provides a full-field stress inversion method for spacecraft precision parts. One or more embodiments of this specification also relate to a full-field stress inversion apparatus for spacecraft precision parts, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art.

[0005] According to a first aspect of the present invention, a full-field stress inversion method for spacecraft precision parts is provided, comprising:

[0006] Acquire real-time isolated point strain data of the target part under various actual working conditions, where the isolated point strain data is collected by strain gauges attached to multiple preset positions of the target part;

[0007] The target isolated point training data is imported into the trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series. The output factor of the first neural network is the input factor of the second neural network, and the output factor of the first neural network is the six-dimensional load data of the target part. The six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force on the target part. The target full-field stress model is a finite element simulation model for the target part.

[0008] In some embodiments, the step of training the inversion model includes:

[0009] Obtaining isolated point training data and full-field strain distribution of the target part under multiple standard load conditions. The isolated point training data is collected by strain gauges attached to at least one preset position on the target part, and the full-field strain distribution is obtained by measuring the strain of the target part under each standard load condition using a non-contact dynamic displacement measurement method.

[0010] Optimize the full-field strain data based on the isolated point training data to generate the optimized strain distribution;

[0011] Optimizing the preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model;

[0012] Simulate the optimized simulation model according to various preset standard loads to generate a training data set;

[0013] A series of dual neural network architecture models are constructed, and the dual neural network architecture model is trained according to the training data set until it meets the training requirements to obtain an inversion model.

[0014] In some embodiments, optimizing full-field strain data based on isolated point training data to generate an optimized strain distribution includes:

[0015] Filter and remove outliers from isolated point training data and full-field strain data;

[0016] Obtain a pair of data with the largest error between the isolated point training data and the corresponding data in the full-field strain data, and calculate the first error rate of the pair of data;

[0017] If the first error rate of the current time is greater than a preset first threshold, the parameters of the non-contact dynamic displacement measurement are adjusted, the full-field strain data is regenerated, and the first error rate of the next time is calculated after filtering and removing outliers;

[0018] If the first error rate of the current time is not greater than the first threshold, it is determined that the full-field strain data of the current time is the optimized strain distribution.

[0019] In some embodiments, optimizing a preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model includes:

[0020] Obtain a pair of data with the largest error between the optimized strain distribution and the corresponding data in the finite element simulation model, and calculate the second error rate of the pair of data;

[0021] When the second error rate is greater than a preset second threshold, obtaining a set of abnormal data with all error rates greater than the second threshold;

[0022] If the distribution of the abnormal data set is concentrated in at least one area, adjusting and re-optimizing the at least one area until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model;

[0023] If the distribution of the abnormal data set is not concentrated, the entire finite element simulation model is adjusted and re-optimized until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model.

[0024] In some embodiments, simulating the optimized simulation model according to a plurality of preset standard loads to generate a training data set includes:

[0025] Set a preset number of load intervals at equal intervals within the yield limit range of each degree of freedom of the target part;

[0026] Collect full-field strain data of the optimized simulation model under different load combinations, where the load combinations are combinations of different load conditions and different load intervals;

[0027] The full-field strain data is divided into training set and test set in proportion;

[0028] The training set and test set are cleaned and normalized to obtain the training data set.

[0029] In some embodiments, when the target part has a symmetrical structure, a mirror image sample is generated through spatial transformation to expand the training data set.

[0030] According to a second aspect of the present invention, there is provided a full-field stress inversion device for spacecraft precision parts, comprising:

[0031] an acquisition module configured to acquire real-time isolated point strain data of a target part under various actual working conditions, wherein the isolated point strain data is acquired by strain gauges attached to a plurality of preset positions of the target part;

[0032] The generation module is configured to import the target isolated point training data into the trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series, the output factor of the first neural network is the input factor of the second neural network, and the output factor of the first neural network is the six-dimensional load data of the target part, the six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force on the target part, and the target full-field stress model is a finite element simulation model for the target part.

[0033] In some embodiments, the step of training the inversion model includes:

[0034] Obtaining isolated point training data and full-field strain distribution of the target part under multiple standard load conditions. The isolated point training data is collected by strain gauges attached to at least one preset position on the target part, and the full-field strain distribution is obtained by measuring the strain of the target part under each standard load condition using a non-contact dynamic displacement measurement method.

[0035] Optimize the full-field strain data based on the isolated point training data to generate the optimized strain distribution;

[0036] Optimizing the preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model;

[0037] Simulate the optimized simulation model according to various preset standard loads to generate a training data set;

[0038] A series of dual neural network architecture models are constructed, and the dual neural network architecture model is trained according to the training data set until it meets the training requirements to obtain an inversion model.

[0039] In some embodiments, optimizing full-field strain data based on isolated point training data to generate an optimized strain distribution includes:

[0040] Filter and remove outliers from isolated point training data and full-field strain data;

[0041] Obtain a pair of data with the largest error between the isolated point training data and the corresponding data in the full-field strain data, and calculate the first error rate of the pair of data;

[0042] If the first error rate of the current time is greater than a preset first threshold, the parameters of the non-contact dynamic displacement measurement are adjusted, the full-field strain data is regenerated, and the first error rate of the next time is calculated after filtering and removing outliers;

[0043] If the first error rate of the current time is not greater than the first threshold, it is determined that the full-field strain data of the current time is the optimized strain distribution.

[0044] In some embodiments, optimizing a preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model includes:

[0045] Obtain a pair of data with the largest error between the optimized strain distribution and the corresponding data in the finite element simulation model, and calculate the second error rate of the pair of data;

[0046] When the second error rate is greater than a preset second threshold, obtaining a set of abnormal data with all error rates greater than the second threshold;

[0047] If the distribution of the abnormal data set is concentrated in at least one area, adjusting and re-optimizing the at least one area until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model;

[0048] If the distribution of the abnormal data set is not concentrated, the entire finite element simulation model is adjusted and re-optimized until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model.

[0049] In some embodiments, simulating the optimized simulation model according to a plurality of preset standard loads to generate a training data set includes:

[0050] Set a preset number of load intervals at equal intervals within the yield limit range of each degree of freedom of the target part;

[0051] Collect full-field strain data of the optimized simulation model under different load combinations, where the load combinations are combinations of different load conditions and different load intervals;

[0052] The full-field strain data is divided into training set and test set in proportion;

[0053] The training set and test set are cleaned and normalized to obtain the training data set.

[0054] In some embodiments, when the target part has a symmetrical structure, a mirror image sample is generated through spatial transformation to expand the training data set.

[0055] According to a third aspect of the present invention, there is provided a computing device comprising:

[0056] memory and processor;

[0057] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned full-field stress inversion method for precision parts of spacecraft are realized.

[0058] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, which stores computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned full-field stress inversion method for precision parts of spacecraft.

[0059] According to a fifth aspect of the present invention, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned full-field stress inversion method for precision parts of spacecraft.

[0060] At least one embodiment of the present invention, on the one hand, by establishing a mapping relationship between the strain gauge measurement data attached to the surface of the target part and the full-field stress distribution, relying on the powerful learning ability of the neural network model, establishes a series dual neural network architecture model, effectively identifies the external load on the target part under the current working condition, and inversely calculates to obtain a more accurate full-field stress distribution. On the other hand, to ensure the validity of the neural network model training data, by designing a loading test under standard working conditions, the strain gauge measurement data and the non-contact DIC full-field strain measurement data are collected, a basic database is established, and the finite element simulation model is corrected with high precision. On the other hand, based on the corrected finite element model, the stress data under different working conditions are calculated as the input factor of the neural network model, and the adhesive strain gauge measurement data, DIC strain measurement data, and finite element simulation model calculation data are used for multiple verifications, which effectively improves the precision and accuracy of the inverse calculation and provides strong support for the improvement of part structure design. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of a full-field stress inversion method for spacecraft precision parts provided by the present invention;

[0062] Figure 2 This is a flow chart of the training steps of the inversion model in the full-field stress inversion method for spacecraft precision parts provided by the present invention;

[0063] Figure 3 This is a structural schematic diagram of an inversion model in a full-field stress inversion method for precision parts of spacecraft provided by the present invention;

[0064] Figure 4 This is a simplified structural diagram of a full-field stress inversion device for precision parts of spacecraft provided by the present invention;

[0065] Figure 5 This is a structural block diagram of a computing device provided by the present invention. DETAILED DESCRIPTION

[0066] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0067] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms of "a" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The modifications of "one" and "a plurality" mentioned in this disclosure are illustrative and not restrictive, and those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0068] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0069] First, the terms involved in one or more embodiments of this specification are explained.

[0070] DIC: Digital Image Correlation, digital image correlation method.

[0071] Adam optimizer: Adaptive Moment Estimation, Adam, adaptive moment estimation algorithm.

[0072] During the manufacturing process of spacecraft products, assembly stress is considered a key factor causing micro-deformation and prestress in components, which can impact product performance and service life. Spacecraft products are characterized by high reliability requirements, harsh service environments, and complex external load and stress conditions. The prestress introduced by assembly stress can lead to rapid degradation of component life during service, thus impacting the mission reliability of spacecraft products. Therefore, structural strain measurement of spacecraft products is necessary to comprehensively assess the micro-deformation and prestress caused by assembly stress.

[0073] At present, although the existing structural strain measurement methods can measure the surface strain of the structure to a certain extent, they are insufficient in the dimension of the measurement data, and it is difficult to obtain the stress distribution in the narrow area of ​​the complex structure. For example, the resistance strain gauge mainly measures the surface stress of the structure and has a limited depth. In addition, the result of the strain gauge measurement is the average strain of the measurement grid area. When the measured area is close to the grid size and there is a large stress gradient change, it is easy to cause a large measurement error. Relying on simulation analysis to calculate the stress of the part, although it can evaluate the strength of the part more comprehensively and accurately and give the stress value of the stress concentration area of ​​the part, its accuracy is limited by the sophistication of the model and the rationality of the parameter setting. It is difficult to fully simulate the complex stress conditions of the parts during the actual assembly process and cannot fully cover all working conditions.

[0074] In view of this, the present invention provides a full-field stress inversion method for spacecraft precision parts. One or more embodiments of this specification also relate to a full-field stress inversion apparatus for spacecraft precision parts, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art.

[0075] See also Figure 1 , Figure 1 A flowchart of a full-field stress inversion method for precision parts of spacecraft provided according to some embodiments of this specification is shown, which specifically includes the following steps.

[0076] Step 101: Acquire real-time isolated point strain data of a target part under various actual working conditions.

[0077] In some embodiments, the execution subject (such as a preset computing device) of the full-field stress inversion method for precision parts of spacecraft can be connected to the target device via a wired connection or a wireless connection, and then obtain the real-time isolated point strain data of the target part under various actual working conditions. The real-time isolated point strain data is collected by strain gauges attached to multiple preset positions of the target part. The above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G / 6G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0078] Target parts can refer to specific precision components within a spacecraft requiring stress analysis, typically critical structural members subject to complex loads. Various actual operating conditions can refer to actual operational processes such as assembly, testing, environmental testing, and inspection. Isolated point strain data can be obtained by using discretely distributed strain gauges to measure local strain, reflecting the deformation response at specific locations on a part. Strain gauges can be attached to resistive sensors on the surface of a part, measuring local strain through changes in resistance. Preset locations can be based on stress concentration areas identified through finite element simulation or empirically determined, such as key monitoring points like hole edges and welds.

[0079] Step 102: Import the target isolated point training data into the trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series, the output factor of the first neural network is the input factor of the second neural network, and the output factor of the first neural network is the six-dimensional load data of the target part, and the six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force on the target part, and the target full-field stress model is a finite element simulation model for the target part.

[0080] The inversion model can refer to a series of neural network architectures designed specifically for the target part, which uses machine learning to map discrete strain data into mechanical responses. It should be noted that each type of part corresponds to a separate inversion model. When targeting a different type of part, separate modeling and training are required to generate an inversion model corresponding to the other part. The first neural network can refer to a deep learning model responsible for inverting six-dimensional loads from strain data, with output including forces and moments. The second neural network can refer to a deep network that receives six-dimensional load inputs and reconstructs full-field stresses, achieving a nonlinear mapping of loads to stress fields. The six-dimensional load data can include a composite load representation of three-dimensional spatial forces (Fx, Fy, Fz) and three-dimensional moments (Mx, My, Mz). The three-dimensional force data can describe the force components borne by the part in the three orthogonal directions of X / Y / Z. The three-dimensional moment data can represent the moment components of the part about the three axes of X / Y / Z. The target full-field stress model can be a finite element model optimized by the neural network, which can accurately predict the global stress distribution of the part. The finite element simulation model can be a physical field calculation model established based on the numerical discretization method, which can simulate the mechanical behavior under complex boundary conditions.

[0081] As a specific example, the data collected by the strain gauge of the target part under assembly conditions is: [-3.07×10-6, -4.41×10-5, 2.58×10-6, 5.32×10-5, -2.61×10-5, 5.97×10-5].

[0082] The measured strain data is input into the neural network model, and the output external load on the target part is L[-25.01N, 25.05N, 99.75N, 0.7N·m, 0.7N·m, 0.99N·m].

[0083] The maximum stress value in the full-field stress inversion model is 737.86 MPa, and the coordinates of the maximum stress point are (-0.48 mm, 7.31 mm, 288.93 mm).

[0084] The beneficial effects of one of the embodiments of the present specification include at least the following: on the one hand, by establishing a mapping relationship between the strain gauge measurement data attached to the surface of the target part and the full-field stress distribution, relying on the powerful learning ability of the neural network model, a series dual neural network architecture model is established to effectively identify the external load on the target part under the current working condition, and inversely calculate to obtain a more accurate full-field stress distribution. On the other hand, in order to ensure the validity of the neural network model training data, by designing a loading test under standard working conditions, strain gauge measurement data and non-contact DIC full-field strain measurement data are collected, a basic database is established, and the finite element simulation model is corrected with high precision. On the other hand, based on the corrected finite element model, the stress data under different working conditions are calculated as the input factor of the neural network model, and the adhesive strain gauge measurement data, DIC strain measurement data, and finite element simulation model calculation data are used for multiple verifications, which effectively improves the precision and accuracy of the inverse calculation and provides strong support for the improvement of part structure design.

[0085] Continue to see Figure 2 , Figure 2 A flowchart of the training steps of an inversion model in a full-field stress inversion method for precision parts of spacecraft provided in accordance with some embodiments of this specification is shown, which specifically includes the following steps.

[0086] Step 201: Obtain isolated point training data and full-field strain distribution of a target part under a variety of standard load conditions, wherein the isolated point training data is collected by a strain gauge attached to at least one preset position of the target part, and the full-field strain distribution is obtained by measuring the strain of the target part under various standard load conditions using a non-contact dynamic displacement measurement method.

[0087] Full-field strain distribution may refer to the full-field strain field data of the part surface obtained by non-contact measurement, reflecting the overall deformation state. Standard load conditions may be pre-set combinations of typical stress states, including standard types such as tension, compression, bending, and torsion. Non-contact dynamic displacement measurement may refer to a real-time deformation monitoring method that does not require physical contact and is achieved using optical, laser, or infrared technologies. Preferably, non-contact dynamic displacement measurement may use non-contact DIC. Isolated point training data is similar to the aforementioned isolated point data, and is data generated during training.

[0088] Step 202 : Optimize the full-field strain data based on the isolated point training data to generate an optimized strain distribution.

[0089] In some optional implementations, the full-field strain data is optimized based on the isolated point training data to generate an optimized strain distribution, including: filtering and removing outliers from the isolated point training data and the full-field strain data; obtaining a pair of data with the largest error between the isolated point training data and the corresponding data in the full-field strain data, and calculating a first error rate of the pair of data; if the first error rate of the time is greater than a preset first threshold, adjusting the parameters of the non-contact dynamic displacement measurement, regenerating the full-field strain data, filtering and removing outliers, and then calculating the first error rate for the next time; if the first error rate of the time is not greater than the first threshold, determining that the full-field strain data of the time is an optimized strain distribution.

[0090] Filtering can refer to the use of digital signal processing methods (such as low-pass filtering, median filtering, etc.) to eliminate high-frequency noise components in strain data. Outlier removal can refer to the elimination of measurement data points that deviate significantly from the normal range through statistical methods or physical rationality judgment. The first error rate can refer to the maximum relative deviation between the actual measured value of the isolated point strain gauge and the full-field strain data of the corresponding position. The first threshold can refer to a pre-set data consistency acceptance standard, which is used to determine whether the full-field strain data needs to be recollected. As an example, the first threshold can be 5%. That is, when the first error rate is greater than 5%, it is necessary to adjust the parameters of the non-contact dynamic displacement measurement to regenerate the full-field strain data. The first threshold can also be other values, set as needed. Measurement parameter adjustment can refer to, but is not limited to, adjusting the sampling frequency of the optical measurement system, the exposure time, or the scanning density of the laser displacement meter.

[0091] By mapping and transforming strain gauge measurements and full-field stress distribution, combined with the powerful learning capabilities of neural network models, accurate full-field stress inversion calculations for precision parts are achieved. This inversion method overcomes the limitations of traditional experimental measurement methods, enabling more comprehensive application to a wide range of possible part operating conditions with excellent accuracy. Compared to conventional testing methods in existing technologies, full-field stress inversion calculations using neural network models are suitable for online, real-time monitoring and inversion calculations of various complex part operating conditions, providing strong support for part structural design and optimization.

[0092] Step 203 : Optimize the preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model.

[0093] In some optional implementations, a preset finite element simulation model is optimized according to the optimized strain distribution to obtain an optimized simulation model, including: obtaining a pair of data with the largest error between the optimized strain distribution and the corresponding data in the finite element simulation model, and calculating the second error rate of the pair of data; when the second error rate is greater than a preset second threshold, obtaining a set of abnormal data with all error rates greater than the second threshold; if the distribution of the abnormal data set is concentrated in at least one area, adjusting and re-optimizing the at least one area until the second error rate is no greater than the second threshold, and obtaining the optimized simulation model; if the distribution of the abnormal data set is not concentrated, adjusting and re-optimizing the entire finite element simulation model until the second error rate is no greater than the second threshold, and obtaining the optimized simulation model.

[0094] Optimized strain distribution may refer to high-confidence strain field data after data screening and correction processing, which is used to guide the calibration of the finite element model. The second error rate may refer to the maximum relative deviation percentage between the measured strain and the simulation result in the current iteration step. The second threshold may refer to a pre-set model optimization termination criterion threshold to determine whether it is necessary to continue iterative correction. As an example, the second threshold may be 5%, or 3%, etc., set as needed. The abnormal data set may refer to a data set to be corrected consisting of all strain-simulation data deviation combinations that exceed the second threshold. Regional concentrated distribution may refer to the local aggregation characteristics of abnormal data in space, usually corresponding to stress concentration or boundary condition mismatch areas. Model adjustment may refer to operations to improve simulation accuracy by modifying material parameters, boundary conditions or mesh density.

[0095] Step 204 : simulating the optimized simulation model according to a plurality of preset standard loads to generate a training data set.

[0096] In some optional implementations, the optimized simulation model is simulated according to a plurality of preset standard loads to generate a training data set, including: setting a preset number of load intervals at equal intervals within the yield limit range of each degree of freedom of the target part; collecting full-field strain data of the optimized simulation model under different load combinations, where the load combination is a combination of different load conditions and different load intervals; dividing the full-field strain data into a training set and a test set in proportion; and cleaning and normalizing the training set and the test set to obtain a training data set.

[0097] The yield limit range can refer to the working boundary formed by the critical stress / strain value when the material begins to undergo plastic deformation. The load interval can refer to the interval set equally according to the mechanical properties within the yield limit. As an example, the load interval can divide the yield limit range into 6-8 intervals. For example, assuming that the maximum force applied in the X direction can be 600N, the yield limit is reached at this time. Dividing the yield limit into 6 intervals can result in a total of 6 intervals: 0-100N, 100-200N, ..., 500-600N.

[0098] A load combination can refer to the Cartesian product of a standard operating condition (e.g., tension / bending) and a specific load intensity range. Data cleaning can refer to the removal of abnormal simulation results (e.g., non-convergent data) and data points that do not conform to physical laws. Normalization can refer to linearly mapping strain data to the interval [0, 1] to eliminate dimensional effects.

[0099] In some optional implementations, the training set accounts for 60% of the total, and the test set accounts for 40% of the total.

[0100] In some optional implementations, when the target part has a symmetrical structure, the training data set is expanded by generating mirror samples through spatial transformation.

[0101] A symmetric structure can refer to a geometric shape with at least one plane or axis of symmetry, whose mechanical response is invariant under symmetric transformations. Spatial transformations can refer to rigid transformations (such as mirror reflection and rotational symmetry) applied to three-dimensional coordinate data. A mirror sample can refer to a new data sample generated by reflecting the original strain field data through a symmetry plane.

[0102] Step 205: construct a series dual neural network architecture model, train the dual neural network architecture model according to the training data set until it meets the training requirements, and obtain an inversion model.

[0103] In some embodiments, as Figure 3 As shown in the figure, a specific inversion model construction step is shown:

[0104] The first step is to set the input and output features: the strain gauge measurement data of the target part under different working conditions is used as the input factor of the first neural network, and the output factor is the external load and torque value (Fx, Fy, Fz, Mx, My, Mz) of the target part. The output factor of the first neural network is used as the input factor of the second neural network, and the output factor is the full-field stress model of the target part under different working conditions.

[0105] Step 2: Select the hidden layer structure: Set the number of hidden layers to 6 fully connected. The first hidden layer uses the TanH function as the activation function, with 32 neurons. The second hidden layer uses the TanH function as the activation function, with 128 neurons. The third hidden layer uses the LeakyReLU function (with a negative slope parameter of 0.2) as the activation function, with 256 neurons. The fourth hidden layer uses the LeakyReLU function (with a negative slope parameter of 0.1) as the activation function, with 256 neurons. The fifth hidden layer uses the ReLU function as the activation function, with 128 neurons. The sixth hidden layer uses the ReLU function as the activation function, with 32 neurons. The weights and biases of the neural network are updated using the backpropagation algorithm. Input data passes through the neural network from the input layer to the hidden layer and then to the output layer, with a weighted sum activation function calculated at each layer. TanH, LeakyReLU, and ReLU are all commonly used activation functions and will not be explained in detail.

[0106] The third step is learning rate setting: set the initial learning rate to 0.001.

[0107] The fourth step is to construct the loss function: using mean square error as an excellent evaluation indicator of the neural network model, construct a loss function to evaluate the difference between the predicted results and the measured data. If the difference is large, use the Adam optimizer to adjust the network parameters and adaptively adjust the learning rate.

[0108] Step 5: Gradient calculation: Calculate the gradient of the loss function at the output layer, propagate back to each layer, and calculate the gradient of each parameter. Based on the gradient and the learning rate, update the weights and biases of each layer.

[0109] Step 6: Determine the optimal parameter combination: Through hyperparameter search, find the optimal hyperparameters in each converged model and determine the optimal network structure and learning parameters.

[0110] Through the deep integration of deep learning and finite element simulation, combined with the test data and a large amount of training data generated by the simulation model, and through the serial dual neural network architecture model for reverse identification of the load, a serial stress inversion calculation method of "strain measurement"-"external load"-"full-field stress" is proposed. For the strain measurement data, massive data such as the external load and full-field stress of the parts can be given at the same time. It not only solves the shortcomings of traditional methods in terms of neural network training accuracy and robustness under complex working conditions of precision parts, but also supports rapid model migration and engineering application promotion as a standardized process. In addition, through multiple comparisons, verifications, and model corrections among strain gauge measurements, DIC measurements, and finite element simulation model calculations under standard working conditions, the strain gauge measurement data under standard working conditions is first used to compare and correct the DIC full-field measurement data, and then the DIC measurement data is used to correct the finite element simulation model. This effectively improves the accuracy of the finite element model calculation and solves the problem that the strain at key positions of some precision parts with narrow structures cannot be directly measured. Multiple verifications under various working conditions ensure the accuracy of the stress inversion calculation, accurately obtain the full-field stress distribution of the part, quickly identify the weak links and stress concentration areas of the part, and improve the overall reliability level of the part.

[0111] Corresponding to the above method embodiment, this specification also provides an embodiment of a full-field stress inversion device for precision parts of spacecraft, Figure 4 FIG1 shows a schematic diagram of the structure of a full-field stress inversion device for precision parts of spacecraft provided by some embodiments of this specification. Figure 4 As shown, the device includes:

[0112] An acquisition module 401 is configured to acquire real-time isolated point strain data of a target part under various actual working conditions, wherein the isolated point strain data is collected by strain gauges attached to multiple preset positions of the target part;

[0113] The generation module 402 is configured to import the target isolated point training data into the trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series, the output factor of the first neural network is the input factor of the second neural network, and the output factor of the first neural network is the six-dimensional load data of the target part, the six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force on the target part, and the target full-field stress model is a finite element simulation model for the target part.

[0114] In some embodiments, the step of training the inversion model includes:

[0115] Obtaining isolated point training data and full-field strain distribution of the target part under multiple standard load conditions. The isolated point training data is collected by strain gauges attached to at least one preset position on the target part, and the full-field strain distribution is obtained by measuring the strain of the target part under each standard load condition using a non-contact dynamic displacement measurement method.

[0116] Optimize the full-field strain data based on the isolated point training data to generate the optimized strain distribution;

[0117] Optimizing the preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model;

[0118] Simulate the optimized simulation model according to various preset standard loads to generate a training data set;

[0119] A series of dual neural network architecture models are constructed, and the dual neural network architecture model is trained according to the training data set until it meets the training requirements to obtain an inversion model.

[0120] In some embodiments, optimizing full-field strain data based on isolated point training data to generate an optimized strain distribution includes:

[0121] Filter and remove outliers from isolated point training data and full-field strain data;

[0122] Obtain a pair of data with the largest error between the isolated point training data and the corresponding data in the full-field strain data, and calculate the first error rate of the pair of data;

[0123] If the first error rate of the current time is greater than a preset first threshold, the parameters of the non-contact dynamic displacement measurement are adjusted, the full-field strain data is regenerated, and the first error rate of the next time is calculated after filtering and removing outliers;

[0124] If the first error rate of the current time is not greater than the first threshold, it is determined that the full-field strain data of the current time is the optimized strain distribution.

[0125] In some embodiments, optimizing a preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model includes:

[0126] Obtain a pair of data with the largest error between the optimized strain distribution and the corresponding data in the finite element simulation model, and calculate the second error rate of the pair of data;

[0127] When the second error rate is greater than a preset second threshold, obtaining a set of abnormal data with all error rates greater than the second threshold;

[0128] If the distribution of the abnormal data set is concentrated in at least one area, adjusting and re-optimizing the at least one area until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model;

[0129] If the distribution of the abnormal data set is not concentrated, the entire finite element simulation model is adjusted and re-optimized until the second error rate is no greater than the second threshold, thereby obtaining an optimized simulation model.

[0130] In some embodiments, simulating the optimized simulation model according to a plurality of preset standard loads to generate a training data set includes:

[0131] Set a preset number of load intervals at equal intervals within the yield limit range of each degree of freedom of the target part;

[0132] Collect full-field strain data of the optimized simulation model under different load combinations, where the load combinations are combinations of different load conditions and different load intervals;

[0133] The full-field strain data is divided into training set and test set in proportion;

[0134] The training set and test set are cleaned and normalized to obtain the training data set.

[0135] In some embodiments, when the target part has a symmetrical structure, a mirror image sample is generated through spatial transformation to expand the training data set.

[0136] The above is a schematic diagram of a full-field stress inversion device for spacecraft precision parts according to this embodiment. It should be noted that the technical solution of this full-field stress inversion device for spacecraft precision parts and the technical solution of the full-field stress inversion method for spacecraft precision parts described above share the same concept. For details not described in detail in the technical solution of the full-field stress inversion device for spacecraft precision parts, please refer to the description of the technical solution of the full-field stress inversion method for spacecraft precision parts described above.

[0137] Figure 5 1 shows a block diagram of a computing device 500 according to some embodiments of the present disclosure. Components of the computing device 500 include, but are not limited to, a memory 501 and a processor 502. The processor 502 is connected to the memory 501 via a bus 503, and a database 505 is used to store data.

[0138] The computing device 500 also includes an access device 504 that enables the computing device 500 to communicate via one or more networks 506. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 504 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0139] In one embodiment of the present specification, the above components of the computing device 500 and Figure 5 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 5 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0140] Computing device 500 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, personal digital assistant, laptop computer, notebook computer, netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer (PC). Computing device 500 may also be a mobile or stationary server.

[0141] The processor 502 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described full-field stress inversion method for spacecraft precision parts. The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of the computing device and the above-described technical solution of the full-field stress inversion method for spacecraft precision parts are based on the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-described full-field stress inversion method for spacecraft precision parts.

[0142] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned full-field stress inversion method for precision parts of spacecraft.

[0143] The above is a schematic diagram of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium is based on the same concept as the technical solution of the above-mentioned full-field stress inversion method for spacecraft precision parts. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-mentioned full-field stress inversion method for spacecraft precision parts.

[0144] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned full-field stress inversion method for precision parts of a spacecraft.

[0145] The above is a schematic diagram of a computer program according to this embodiment. It should be noted that the technical solution of this computer program is based on the same concept as the technical solution of the aforementioned full-field stress inversion method for spacecraft precision parts. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the aforementioned full-field stress inversion method for spacecraft precision parts.

[0146] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0147] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0148] It should be noted that for the aforementioned method embodiments, for ease of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0149] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0150] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of the present invention. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A full-field stress inversion method for spacecraft precision parts, characterized in that: include: Acquiring real-time isolated point strain data of a target part under various actual working conditions, wherein the isolated point strain data is collected by strain gauges attached to multiple preset positions of the target part; The target isolated point training data is imported into the trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series, the output factor of the first neural network is the input factor of the second neural network, the output factor of the first neural network is the six-dimensional load data of the target part, the six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force applied to the target part, and the target full-field stress model is a finite element simulation model for the target part.

2. The method according to claim 1, characterized in that The training steps of the inversion model include: Obtaining isolated point training data and full-field strain distribution of a target part under multiple standard load conditions, wherein the isolated point training data is collected by a strain gauge attached to at least one preset position of the target part, and the full-field strain distribution is obtained by measuring the strain of the target part under each standard load condition using a non-contact dynamic displacement measurement method; Optimizing the full-field strain data according to the isolated point training data to generate an optimized strain distribution; Optimizing a preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model; simulating the optimized simulation model according to a plurality of preset standard loads to generate a training data set; A series dual neural network architecture model is constructed, and the dual neural network architecture model is trained according to the training data set until it meets the training requirements, thereby obtaining the inversion model.

3. The method according to claim 2, characterized in that Optimizing the full-field strain data according to the isolated point training data to generate an optimized strain distribution includes: Filtering and removing outliers on the isolated point training data and the full-field strain data; Obtaining a pair of data with the largest error between the isolated point training data and the corresponding data in the full-field strain data, and calculating a first error rate of the pair of data; If the first error rate of the current time is greater than a preset first threshold, adjusting the parameters of the non-contact dynamic displacement measurement, regenerating the full-field strain data, filtering and removing outliers, and then calculating the first error rate of the next time; If the first error rate of the current time is not greater than the first threshold, it is determined that the full-field strain data of the current time is the optimized strain distribution.

4. The method according to claim 2, characterized in that Optimizing a preset finite element simulation model according to the optimized strain distribution to obtain an optimized simulation model includes: Obtaining a pair of data with the largest error between the optimized strain distribution and the corresponding data in the finite element simulation model, and calculating a second error rate of the pair of data; When the second error rate is greater than a preset second threshold, obtaining a set of abnormal data with all error rates greater than the second threshold; If the distribution of the abnormal data set is concentrated in at least one area, adjusting and re-optimizing the at least one area until the second error rate is no greater than the second threshold, thereby obtaining the optimized simulation model; If the distribution of the abnormal data set is not concentrated, the entire finite element simulation model is adjusted and re-optimized until the second error rate is no greater than the second threshold, thereby obtaining the optimized simulation model.

5. The method according to claim 2, characterized in that The optimized simulation model is simulated according to a plurality of preset standard loads to generate a training data set, including: Set a preset number of load intervals at equal intervals within the yield limit range of each degree of freedom of the target part; Collecting full-field strain data of the optimized simulation model under different load combinations, where the load combination is a combination of different load conditions and different load intervals; Dividing the full-field strain data into a training set and a test set in proportion; The training set and the test set are cleaned and normalized to obtain the training data set.

6. The method according to claim 5, characterized in that When the target part has a symmetrical structure, the training data set is expanded by generating mirror samples through spatial transformation.

7. A full-field stress inversion device for precision parts of spacecraft, characterized in that: include: an acquisition module configured to acquire real-time isolated point strain data of a target part under various actual working conditions, wherein the isolated point strain data is collected by strain gauges attached to a plurality of preset positions of the target part; A generation module is configured to import the target isolated point training data into a trained inversion model to generate a target full-field stress model of the target part, wherein the inversion model corresponds to the target part, and the inversion model includes a first neural network and a second neural network connected in series, the output factor of the first neural network is the input factor of the second neural network, the output factor of the first neural network is the six-dimensional load data of the target part, the six-dimensional load data includes three-dimensional force data and three-dimensional moment data of the external load force on the target part, and the target full-field stress model is a finite element simulation model for the target part.

8. A computing device, characterized in that include: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the full-field stress inversion method for precision parts of spacecraft described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by a processor, the steps of the full-field stress inversion method for spacecraft precision parts described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program or instructions, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images