Virtual test method based on multi-source data fusion

Through the virtual experimental method of multi-source data fusion, combined with finite element model and machine learning algorithm, the problem of limited acquisition of offshore test data is solved, the maximum data utilization and high-accuracy physics reproduction are achieved, and the offshore test is provided. More trusted data support.

CN118940569AActive Publication Date: 2024-11-12BEIJING INST OF TECH

Patent Information

Application Number
CN202410980707.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12
Estimated Expiration
2044-07-22

AI Technical Summary

Technical Problem

The existing offshore test data is limited, making it difficult to conduct accurate simulation tests, and the test cost is high, the data acquisition amount is small, the physical process is unclear, and the relative position of the blowing point and the target cannot be accurately obtained.

Method used

Using a virtual test method of multi-source data fusion, combining real test data with simulation tests, using finite element models and machine learning algorithms (support vector machines) for data processing and prediction model training, establish a virtual test platform to maximize data use.

Benefits of technology

Improve the value of experimental data, realize the comprehensiveness, reproducibility and repeatability of data, provide high-accuracy physics reproduction and intuitive process reproduction, and provide rich in weapon equipment development and scientific research. Data information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940569A_ABST
    Figure CN118940569A_ABST
Patent Text Reader

Abstract

The invention discloses a virtual test method based on multi-source data fusion, which adopts a mode of combining real test data and a simulation test to realize a virtual test, and comprises the steps of test data, batch simulation, machine model training, reverse prediction, virtual test and the like. And a comprehensive, reproducible, repeated analysis, physical field reproduction and multi-source data fusion virtual test method is constructed. According to the method, on the basis of limited test data, a virtual test is constructed in combination with batch numerical simulation calculation and a machine learning algorithm. The invention aims to solve the problems of difficulty and less acquisition of the marine test data, and can realize that the marine test data is changed from less to more, point data is changed into field data, and the utilization value of the real test data (only a small amount) is maximized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of ship offshore testing and equipment, and in particular relates to a virtual testing method based on multi-source data fusion. Background Art

[0002] During the course of sailing, ships may be hit by collisions, large wind and wave loads, or fuel gas explosions, which seriously threaten the stability and safety of core devices such as the ship's propulsion system, communication equipment, and data devices. Through virtual test methods, it is possible to provide important and valuable data foundation for evaluating the ship's ability to resist complex sea conditions and improve its protection capabilities.

[0003] At present, from the perspective of measurement data and scientific research, the amount of data that can be obtained from sea trials is still relatively limited. It mainly obtains relevant videos, impact environment, strain, temperature, water level and other information of limited measurement points. It is foreseeable that it will be difficult to make major breakthroughs in the measurement instruments and measurement technology in the foreseeable future. Therefore, a single sea experiment has the disadvantages of high test cost, high test difficulty, difficulty in obtaining data, small amount of data obtained, and unclear physical process of sea trials. In addition, since sea trials cannot accurately obtain the relative position of the explosion point and the target, for simulation tests, there is a lack of accurate working conditions, so it is difficult to conduct a highly convincing simulation test.

[0004] In order to solve the above-mentioned shortcomings, the present invention adopts currently feasible technical means, takes into account the above-mentioned situation, and adopts a way of combining real test data with simulation tests to realize virtual tests, and constructs a comprehensive, reproducible, repeatedly analyzed, physical field reproduced, and multi-source data fusion virtual test method. Summary of the invention

[0005] The purpose of this invention is to provide a new method for constructing virtual experiments based on limited test data, combined with batch numerical simulation calculations and machine learning algorithms. This method can realize the "conversion from small to large" of offshore test data and maximize the research value of a small amount of real test data.

[0006] The technical solution adopted by the present invention is a virtual test method based on multi-source data fusion.

[0007] A virtual test method based on multi-source data fusion, the steps of which are:

[0008] S01: Obtain valid data of the measurement points through experiments or existing experiments, including acceleration measurement data, pressure measurement data, strain measurement data, breach data, etc., and process the data to establish simplified eigenvalues;

[0009] S02: According to the target ship information, a finite element model is established, and random working conditions are established. The random working conditions are calculated in batches, and the valid data obtained from the corresponding test in the results are extracted, namely acceleration measurement data, pressure measurement data, strain measurement data, breach data, etc., and the same data processing method as the test data is adopted to obtain simplified eigenvalues, thereby establishing measurement information corresponding to various working conditions, normalizing the original data, and forming a database used for machine learning.

[0010] S03: Using a machine learning algorithm (support vector machine), with the result data as features (i.e., independent variable X) and the operating condition information as the target (i.e., dependent variable Y), the basic database generated in step S02 is trained. The breach data needs to be trained with a classification model using a machine learning classifiable algorithm, and other data needs to be trained with a regression model using a regressive algorithm, so as to establish corresponding machine learning prediction models for all operating condition information conditions (such as spatial coordinates, drug dosage, water depth), and perform model accuracy evaluation and parameter optimization.

[0011] S04: using the prediction model for the corresponding working condition information established in step S03, taking the data obtained in step S01 as a test set, and predicting the corresponding working condition information during the test, i.e., the amount of explosives, the coordinates of the explosive blasting point, etc.

[0012] S05: Compare the working conditions obtained in step S04 with all known working conditions of the test, that is, compare the dosage, explosion point coordinates, etc. The spatial coordinates of the explosion point of the test are analyzed through videos, pictures and other materials to obtain a distribution range. The predicted coordinate value is considered reasonable within the distribution range, and the error range of other predicted values ​​(drug dosage, etc.) is within 20%. For the predicted values ​​with obvious differences, return and execute steps S03, S04, and S05 until the predicted value is within the error range.

[0013] S06: The working condition information predicted in step S04 is used as input conditions and input into the finite element simulation calculation model established in step S02, and the test is simulated and calculated to obtain the simulation results of the test. The corresponding post-processing software is used to process the simulation results, form corresponding video animation materials, physical field data materials, etc., to visualize the data, and the result file is packaged to form the final virtual experiment platform.

[0014] Preferably, the measurement data processing in steps S01 and S02 establishes simplified characteristic values, and the specific method is:

[0015] S01-01: For the acceleration data of a certain measuring point, convert it into shock spectrum data to obtain spectrum velocity, spectrum displacement, and spectrum acceleration. Specifically, the shock spectrum is obtained according to the following steps:

[0016] (1) Calculate the acceleration spectrum A(ω) as follows:

[0017]

[0018] In the above formula, is the acceleration time domain curve of the measuring point measured in the test; τ is the time variable; t is the time point when the acceleration curve ends; ω is the circular frequency.

[0019] (2) Calculate the velocity spectrum V(ω) as follows:

[0020]

[0021] (3) Calculate the displacement spectrum D(ω) as follows:

[0022]

[0023] (4) Plot the spectrum lines in a logarithmic coordinate system according to the following relationship to obtain the impact spectrum.

[0024] logV=logω+logD,logV=-logω+logA(5)

[0025] S01-02: For the strain data of a certain measuring point, average the stabilized strain data.

[0026] S01-03: For the pressure data of a certain measuring point, if it is an underwater test, extract its initial shock wave peak value, initial shock wave pulse width, bubble pulsation pressure peak value, bubble pulsation pressure pulse width, and bubble pulsation period; if it is an above-water test, extract the shock wave peak value, pulse width, and quasi-static pressure value.

[0027] S01-04: For a certain breach area data, get the equivalent breach radius r e , and extract the maximum length l of the real breach max ; The equivalent breach radius r e Calculate as follows:

[0028] πr e 2 =S real (6)

[0029] In the formula, S real The breach area is the actual measurement or simulation calculation.

[0030] Preferably, the finite element simulation calculation in steps S02 and S06 specifically refers to the explosion fluid-solid coupling simulation calculation of a specific working condition, which is usually performed using the ALE or CEL method, and the simulation calculation steps mainly include:

[0031] (1) Hull structure modeling, water and air domain modeling, and explosive domain modeling;

[0032] (2) Assignment of material properties of hull structure, water medium, air medium, and explosive medium;

[0033] (3) Setting of fluid domain boundary conditions;

[0034] (4) Fluid-structure interaction settings;

[0035] (5) Solution time and result output settings;

[0036] (6) Perform solution calculation.

[0037] Preferably, the raw data normalization in step S02 is specifically calculated according to the following formula:

[0038]

[0039] Among them, X nom is the normalized data, X is the original data of the variable to be normalized, X min , X max is the minimum and maximum value of the variable data.

[0040] Preferably, the machine learning algorithm in step S03 specifically refers to that, for breach data, first, classification training is performed using a support vector machine algorithm according to whether there is a breach or not, and then regression training is performed using a support vector machine algorithm for the specific breach data with a breach; other measurement data are all subjected to regression training using a support vector machine algorithm according to the specific data.

[0041] Among them, the support vector machine classification / regression prediction takes a certain physical quantity of interest as the output variable, i.e., the target set, and takes the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the underwater explosion depth of the explosive as the input variables, i.e., the feature set. By training the data set, a regression / regression model of a certain physical quantity of interest and the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the water depth of the explosive is obtained, and its form is as follows:

[0042] x i =f(X,Y,Z,M TNT ,d)(8)

[0043] Among them, x i is a physical quantity of interest; X is the x-coordinate of the explosive point; Y is the y-coordinate of the explosive point; Z is the z-coordinate of the explosive point; M TNT is the TNT equivalent of the explosive; d is the underwater explosion depth of the explosive, which is only used for underwater explosions and is always taken as 0 for air explosions; f represents the support vector machine classification / regression model.

[0044] Preferably, the accuracy assessment and parameter optimization of the model in step S03 specifically refers to using cross validation and grid search with mean square error as the loss function. In combination with genetic algorithm, the model parameters C and G of the support vector machine are optimized to improve the prediction accuracy of the model. The mean square error MSE calculation formula is:

[0045]

[0046] Where n is the number of samples, is the predicted value, y i is the true value.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] 1. Reproduce the comprehensive physical field of the real experiment through virtual means, so as to extract more measurement data and improve the data value of a single experiment. That is, through a single experiment, obtain comprehensive, reproducible and repeatedly analyzed test data.

[0049] 2. The present invention adopts machine learning algorithm prediction and test measurement data proofreading optimization, which has higher accuracy.

[0050] 3. The present invention can provide intuitive and highly reliable process reproduction based on real physical fields for experiments, and also provide rich data for weapon equipment development and scientific research. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A system architecture diagram of an example of the present invention;

[0052] Figure 2 It is an operation flow chart of an example of the present invention;

[0053] Figure 3 It is a principle logic diagram of an example of the present invention;

[0054] Figure 4 It is a three-fold impact spectrum diagram of an example of the present invention;

[0055] Figure 5 It is a graph of underwater explosion pressure of an example of the present invention;

[0056] Figure 6 It is a curve diagram of the explosion pressure in the cabin of an example of the present invention;

[0057] Figure 7 This is a diagram of underwater explosion modeling according to an example of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] like Figure 1 , Figure 2 As shown, the virtual test method based on multi-source data fusion mainly consists of a test part, a machine learning prediction model part, and a virtual test platform part. Figure 3 have:

[0060] S01: The experimental part mainly organizes, analyzes and simplifies the experimental data to form a typical data set that can be used for subsequent machine learning model training.

[0061] Obtain valid data of the measuring points through experiments or existing experiments, including acceleration measurement data, pressure measurement data, strain measurement data, breach data, etc., and process the data to establish simplified eigenvalues;

[0062] S02: According to the target ship information, a finite element model is established, and random working conditions are established. The random working conditions are calculated in batches, and the valid data obtained from the corresponding test in the results are extracted, namely acceleration measurement data, pressure measurement data, strain measurement data, breach data, etc., and the same data processing method as the test data is adopted to obtain simplified eigenvalues, thereby establishing measurement information corresponding to various working conditions, normalizing the original data, and forming a database used for machine learning.

[0063] The measurement data processing in steps S01 and S02 establishes simplified characteristic values, and the specific method is:

[0064] S01-01: For the acceleration data of a certain measuring point, convert it into shock spectrum data to obtain spectrum velocity, spectrum displacement, and spectrum acceleration. Specifically, the shock spectrum is obtained according to the following steps:

[0065] (1) Calculate the acceleration spectrum A(ω) as follows:

[0066]

[0067] In the above formula, is the acceleration time domain curve of the measuring point measured in the test; τ is the time variable; t is the time point when the acceleration curve ends; ω is the circular frequency.

[0068] (2) Calculate the velocity spectrum V(ω) as follows:

[0069]

[0070] (3) Calculate the displacement spectrum D(ω) as follows:

[0071]

[0072] (4) Plot the spectrum lines in a logarithmic coordinate system according to the following relationship to obtain the impact spectrum.

[0073] logV=logω+logD,logV=-logω+logA(5)

[0074] Finally, we get Figure 4 The three-fold impact spectrum is shown.

[0075] S01-02: For the strain data of a certain measuring point, average the stabilized strain data.

[0076] S01-03: For the pressure data of a certain measuring point, if it is an underwater test, the underwater explosion pressure time domain curve is obtained, such as Figure 5 As shown in the figure, the initial shock wave peak value, initial shock wave pulse width, bubble pulsation pressure peak value, bubble pulsation pressure pulse width, and bubble pulsation period are extracted; if it is a water test, the time domain curve of the explosion pressure in the cabin is obtained, such as Figure 6 As shown, the shock wave peak value, pulse width and quasi-static pressure value are extracted.

[0077] S01-04: For a certain breach area data, get the equivalent breach radius r e , and extract the maximum length l of the real breach max ; The equivalent breach radius r e Calculate as follows:

[0078] πr e 2 =S real (6)

[0079] In the formula, S real The breach area is the actual measurement or simulation calculation.

[0080] The "establishing a finite element model based on the target ship information" in step s02 specifically refers to the explosion fluid-solid coupling simulation modeling of specific working conditions. Usually, a simulation model based on the ALE or CEL method is established. The steps mainly include:

[0081] (1) Hull structure modeling, water and air domain modeling, and explosive domain modeling. Figure 7 As shown, the hull structure modeling is mainly based on the actual ship scale, geometry and structural construction, and material properties to establish a real-scale fine finite element model. The water and air domains are generally set as hexahedral domains, which are required to be as large as possible based on the covering structure and explosives. The explosives are set as spheres with a certain radius based on the equivalent TNT equivalent.

[0082] (2) Assignment of material properties of hull structure, water medium, air medium, and explosive medium

[0083] Among them, the hull structure is generally made of steel. Taking Q235 steel as an example, the CS model is used to characterize the dynamic response characteristics of the hull steel under explosion response. The Cowper-Symonds model is a classic model. In LS-DYNA, it is a follow-up model that considers the influence of strain rate. Its dynamic constitutive equation is:

[0084]

[0085] σ0 and σ eff are the static and dynamic yield stresses, respectively, and σ0 is taken as 235MPa; is the strain rate; C and P are Cowper-Symmonds strain rate parameters, which are 40.4 and 5, respectively, to characterize the sensitive characteristics of the material; β is the hardening parameter, β = 0 means only kinematic hardening, and β = 1 means isotropic hardening; is the effective plastic strain;

[0086] E p is the plastic hardening modulus, given by:

[0087]

[0088] Where, E is the elastic modulus, which is 2.1 GPa; E tan is the tangent modulus, which is taken as 1180 MPa.

[0089] The response of water medium under explosion is described by Mie-Gruneisen state equation:

[0090]

[0091] Where, ρ0 is the initial density of water medium; μ is the compressibility coefficient, μ>0 indicates a compressed state, and μ<0 indicates a stretched state; c is the sound velocity of water; S1, S2 and S3 are dimensionless parameters; γ0 is the initial Grüneisen coefficient; a is the first-order volume correction coefficient; and E is the internal energy per unit volume. The parameters of water are shown in Table 1.

[0092] Table 1 Water parameters

[0093]

[0094] Air is described by a linear polynomial state equation:

[0095] p=C0+C1μ+C2μ 2 +C3μ 3 +(C4+C5μ+C6μ 2 )E(10)

[0096] Where p is pressure; Ci (i=0, ..., 3) is a constant with the dimension of pressure; C i (i = 4, ..., 6) is a dimensionless constant; μ = ρ / ρ0-1; E is the internal energy term. The air-related parameters are shown in Table 2.

[0097] Table 2 Air related parameters

[0098] ρkg / m3 <![CDATA[C0]]> <![CDATA[C1]]> <![CDATA[C2]]> <![CDATA[C3]]> <![CDATA[C4]]> <![CDATA[C5]]> <![CDATA[C6]]> 1.25 0 0 0 0 0.4 0.4 0

[0099] The explosive adopts the JWL state equation,

[0100] The specific form is:

[0101]

[0102] In the formula, V is the relative specific volume of the detonation product; E is the internal energy of the explosive per unit volume, which has the dimension of pressure; A, B, R1, R2, and ω are the five unknown parameters of the JWL state equation. For TNT explosives, the relevant data are shown in Table 3.

[0103] Table 3 Explosive materials and state equation parameters

[0104]

[0105] (3) Setting of fluid domain boundary conditions;

[0106] For explosions above water, the boundary surfaces of the air domain are generally set as non-reflecting boundaries; for underwater explosions, the top and bottom surfaces are generally set as non-reflecting boundaries, and the surrounding areas are set according to whether the water area is open water or confined water. For open water areas, hydrostatic pressure gradient and environmental boundary conditions are set, and for confined water areas, hydrostatic pressure gradient and reflective boundaries are set.

[0107] (4) Fluid-structure interaction settings;

[0108] The outer surface of the hull, the water area, the air area, and the explosives are set as a fluid-solid coupling group to consider the forces between the hull structure and the fluid.

[0109] (5) Solution time and result output settings;

[0110] Set the calculation end time and the frequency of result output, including animation files, explosion pressure, strain, stress, acceleration, etc.

[0111] The random working conditions in step S02, such as Figure 3As shown in Figure 4, the spatial coordinates of the explosives, the amount of explosives and other key factors are randomly sampled based on the strike probability, a series of random working conditions are established, and a series of simulation calculations are performed based on the set simulation model, that is, Monte Carlo simulation is realized, so as to generate a batch simulation result database, which is shown in Table 4, including the explosion point coordinates, equivalent charge, attack speed, shock spectrum generated by acceleration (spectral displacement, spectral velocity, spectral acceleration), pressure peak, maximum strain, breach size, maximum deflection, etc.

[0112] Table 4 Test set and training set data items

[0113]

[0114] The original data in step S02 is normalized. Specifically, the key data required to be extracted for batch simulation, including explosion point coordinates, equivalent charge, attack speed, shock spectrum generated by acceleration (spectral displacement, spectral velocity, spectral acceleration), pressure peak, maximum strain, breach size, maximum deflection, etc., are normalized according to the following formula to form a standard data set for subsequent training.

[0115]

[0116] Among them, X nom is the normalized data, X is the original data of the variable to be normalized, X min , X max is the minimum and maximum value of the variable data.

[0117] S03: Using a machine learning algorithm (support vector machine), with the result data as features (i.e., independent variable X) and the operating condition information as the target (i.e., dependent variable Y), the data set generated in step S02 is trained.

[0118] First, the support vector machine algorithm is used for classification training according to whether there is a breach or not, and then the support vector machine algorithm is used for regression training of the specific breach data; other measurement data are all used for regression training according to the specific data using the support vector machine algorithm.

[0119] Among them, the support vector machine classification / regression prediction takes a certain physical quantity of interest as the output variable, i.e., the target set, and takes the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the underwater explosion depth of the explosive as the input variables, i.e., the feature set. By training the data set, a regression / regression model of a certain physical quantity of interest and the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the water depth of the explosive is obtained, and its form is as follows:

[0120] x i =f(X,Y,Z,M TNT ,d) (13)

[0121] Among them, x i is a physical quantity of interest; X is the x-coordinate of the explosive point; Y is the y-coordinate of the explosive point; Z is the z-coordinate of the explosive point; M TNT is the TNT equivalent of the explosive; d is the underwater explosion depth of the explosive, which is only used for underwater explosions and is always taken as 0 for air explosions; f represents the support vector machine classification / regression model.

[0122] The accuracy evaluation and parameter optimization of the model during the model generation process specifically refers to the use of mean square error as the loss function, such as Figure 3 As shown in the figure, cross validation, grid search, and genetic algorithm are used to optimize the model parameters C and G of the support vector machine, thereby improving the prediction accuracy of the model. The mean square error MSE calculation formula is:

[0123]

[0124] Where n is the number of samples, is the predicted value, y i is the true value.

[0125] S04: using the prediction model of the corresponding working condition information established in step S03, taking the data obtained in step S01 as a test set, predicting the corresponding working condition information during the offshore test, i.e., the amount of explosives, the coordinates of the explosive blasting point, etc.

[0126] S05: Compare the working conditions obtained in step S04 with all known working conditions of the test, that is, compare the amount of explosives, explosion point coordinates, etc. The spatial coordinates of the test explosion point are analyzed through videos, pictures and other materials to obtain a distribution range. The coordinate prediction value is considered reasonable within the distribution range, and the other prediction values ​​are considered reasonable within the error range of 20%. For the prediction values ​​with obvious differences, return and execute steps S03, S04, and S05 until the prediction value is within the error range.

[0127] S06: The working condition information predicted in step S04 is used as input conditions and input into the finite element simulation calculation model established in step S02 to simulate the test and obtain the simulation results of the test. The corresponding post-processing software is used to process the simulation results to form corresponding video animation data, physical field data, etc. Figure 2 , Figure 3 As shown, the explosion response process, the whole ship response data, and the physical field data are obtained, the limited measurement point data are expanded into three-dimensional full-field data, the data is visualized, and the result file is packaged to form the final virtual experimental platform.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A virtual test method based on multi-source data fusion, It is characterized in that The following steps are included: S01: Obtain valid data of the measurement points through experiments or existing experiments, including acceleration measurement data, pressure measurement data, strain measurement data, and breach data, and perform data processing to establish simplified eigenvalues; S02: According to the target ship information, a finite element model is established, and random working conditions are established, random working conditions are calculated in batches, and valid data obtained from the corresponding test in the results are extracted, namely acceleration measurement data, pressure measurement data, strain measurement data, breach data, and the same data processing method as the test data is adopted to obtain simplified eigenvalues, thereby establishing measurement information corresponding to various working conditions, normalizing the original data, and forming data used for machine learning; S03: using a machine learning algorithm, taking the result data as a feature, i.e., the independent variable X, and the operating condition information as a target, i.e., the dependent variable Y, to train the basic database generated in step S02, wherein the breach data is trained with a classification model using a machine learning classifiable algorithm, and other data is trained with a regression model using a regression algorithm, thereby establishing corresponding machine learning prediction models for all operating condition information conditions, and performing model accuracy evaluation and parameter optimization; S04: using the prediction model of the corresponding working condition information established in step S03, taking the data obtained in step S01 as a test set, and predicting the corresponding working condition information during the test, i.e., the amount of explosives and the coordinates of the explosive point; S05: Compare the working conditions obtained in step S04 with all known working conditions of the test, that is, compare the charge amount and explosion point coordinates; the spatial coordinates of the explosion point of the test are analyzed through video and picture data to obtain a distribution range, and the coordinate prediction value is considered reasonable if it is within the distribution range, and the other prediction values ​​are considered reasonable if the error range is within 20%; for the prediction values ​​with obvious differences, return and execute steps S03, S04, and S05 until the prediction value is within the error range; S06: Use the working condition information predicted in step S04 as input conditions and input it into the finite element simulation calculation model established in step S02, simulate the experiment and obtain the simulation results of the experiment; use corresponding post-processing software to process the simulation results, form corresponding video animation materials and physical field data materials, visualize the data, and package the result file to form the final virtual experimental platform.

2. A virtual test method based on multi-source data fusion according to claim 1, It is characterized in that The measurement data processing in steps S01 and S02 establishes simplified characteristic values, and the specific method is: S01-01: For the acceleration data of a certain measuring point, convert it into shock spectrum data to obtain spectrum velocity, spectrum displacement, and spectrum acceleration; specifically, the shock spectrum is obtained according to the following steps: (1) Calculate the acceleration spectrum A(ω) as follows: In the above formula, is the acceleration time domain curve of the measuring point measured in the test; τ is the time variable; t is the time point when the acceleration curve ends; ω is the circular frequency; (2) Calculate the velocity spectrum V(ω) as follows: (3) Calculate the displacement spectrum D(ω) as follows: (4) Plot the spectrum lines in a logarithmic coordinate system according to the following relationship to obtain the shock spectrum; logV=logω+logD,logV=-logω+logA(5) S01-02: For the strain data of a certain measuring point, average the stabilized strain data; S01-03: For the pressure data of a certain measuring point, if it is an underwater test, extract its initial shock wave peak value, initial shock wave pulse width, bubble pulsation pressure peak value, bubble pulsation pressure pulse width, and bubble pulsation period; if it is an above-water test, extract the shock wave peak value, pulse width, and quasi-static pressure value; S01-04: For a certain breach area data, get the equivalent breach radius r e , and extract the maximum length l of the real breach max ; The equivalent breach radius r e Calculate as follows: πr e 2 =S real (6) In the formula, S real The breach area is the actual measurement or simulation calculation.

3. A virtual test method based on multi-source data fusion according to claim 1, It is characterized in that The finite element simulation calculation in steps S02 and S06 specifically refers to the explosion fluid-solid coupling simulation calculation of a specific working condition, and the simulation calculation is performed using the ALE or CEL method. The simulation calculation steps mainly include: (1) Hull structure modeling, water and air domain modeling, and explosive domain modeling; (2) Assignment of material properties of hull structure, water medium, air medium, and explosive medium; (3) Setting of fluid domain boundary conditions; (4) Fluid-structure interaction settings; (5) Solution time and result output settings; (6) Perform solution calculation.

4. A virtual test method based on multi-source data fusion according to claim 1, It is characterized in that The original data normalization in step S02 is specifically calculated according to the following formula: Among them, X nom is the normalized data, X is the original data of the variable to be normalized, X min , X max is the minimum and maximum value of the variable data.

5. A virtual test method based on multi-source data fusion according to claim 1, It is characterized in that The machine learning algorithm in step S03 specifically refers to that, for breach data, firstly, classification training is performed using a support vector machine algorithm according to whether there is a breach, and then regression training is performed using a support vector machine algorithm for specific breach data with breaches; other measurement data are all subjected to regression training using a support vector machine algorithm according to specific data; Among them, the support vector machine classification / regression prediction takes a certain physical quantity of interest as the output variable, i.e., the target set, and takes the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the underwater explosion depth of the explosive as the input variables, i.e., the feature set. By training the data set, a regression / regression model of a certain physical quantity of interest and the three-dimensional spatial coordinates of the explosive XYZ, the TNT equivalent of the explosive, and the water depth of the explosive is obtained, and its form is as follows: x i =f(X,Y,Z,M TNT ,d)(8) Among them, x i is a physical quantity of interest; X is the x-coordinate of the explosive point; Y is the y-coordinate of the explosive point; Z is the z-coordinate of the explosive point; M TNT is the TNT equivalent of the explosive; d is the underwater explosion depth of the explosive, which is only used for underwater explosions and is always taken as 0 for air explosions; f represents the support vector machine classification / regression model.

6. A virtual test method based on multi-source data fusion according to claim 1, It is characterized in that The accuracy evaluation and parameter optimization of the model in step S03 specifically refers to using the mean square error as the loss function, adopting cross validation and grid search; and combining the genetic algorithm to optimize the model parameters C and G of the support vector machine, thereby improving the prediction accuracy of the model; wherein the mean square error MSE calculation formula is: Where n is the number of samples, is the predicted value, y i is the true value.

Citation Information

Patent Citations

  • Target detection model training method and device

    CN117011630A

  • Fragment protection structure optimization method based on material point method and support vector machine

    CN117763761A

  • Method and system for calibrating mechanical properties of material to be tested, and storage medium

    WO2023133732A1

  • Laser plate jointing weld seam quality classification method based on multivariate process logistic regression

    WO2024124967A1

Cited By

  • TCN-LSTM-based flight control system multivariate data fusion virtual test verification method

    CN120044914A

  • Test risk digital twinborn early warning method based on multi-domain cooperative monitoring

    CN120822386A

  • A test risk digital twin early warning method for multi-domain collaborative monitoring

    CN120822386B