CN112214927A - Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method - Google Patents
Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method Download PDFInfo
- Publication number
- CN112214927A CN112214927A CN202010998145.0A CN202010998145A CN112214927A CN 112214927 A CN112214927 A CN 112214927A CN 202010998145 A CN202010998145 A CN 202010998145A CN 112214927 A CN112214927 A CN 112214927A
- Authority
- CN
- China
- Prior art keywords
- fragment cloud
- model
- fragment
- cloud
- collision
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of computers, in particular to a machine learning-based rapid simulation method for ultra-high-speed collision fragment cloud, which comprises the following steps of S10, a fragment cloud key feature extraction method; s20, constructing a fragment cloud rapid simulation model; s30, applying a strategy to the collision condition in the model; and S40, a model accuracy evaluation method. The invention extracts the fragment cloud distribution characteristics by a characteristic engineering method; then, establishing a fragment cloud rapid simulation model based on a variational self-encoder network structure, and taking the distribution characteristics and collision parameters of the fragment cloud as input data; meanwhile, a loss function is designed, the hierarchical structure of the neural network is guided to realize complex function approximation between input data and output data, and then a nonlinear relation between a collision parameter and high-dimensional fragment cloud distribution characteristics is established; in order to evaluate the accuracy of the obtained fragment cloud rapid generation model, a model error evaluation method is established, and various error standards between a model output result and model training data are calculated.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a machine learning-based method for quickly simulating an ultra-high-speed collision fragment cloud.
Background
Fragment cloud is a high-speed motion substance formed by fragments of the crushed projectile and target plate fragments after the projectile impacts the target plate at high speed under certain impact parameters in the ultra-high-speed impact process, and the high-speed motion substance is formed by fragments, melt and vapor and is shaped like a cloud cluster on a macroscopic scale; the fragment cloud model is a mathematical physical model for describing fragment cloud generated by the projectile colliding a target plate at a high speed; the collision parameters refer to relevant parameters of the shot and the target before collision, and comprise: the shot incident speed, the shot incident angle, the shot and target collision coordinate, the materials and the shapes of the shot and the target plate and the like; the fragment cloud distribution characteristics include:
mass distribution: refers to the distribution of material mass in the debris cloud in space, expressed as the spatial density of the mass;
velocity profile: the distribution of the speed of any point on the fragment cloud in the space;
fragment size characteristics in fragment clouds: mean size, size distribution, etc. of the fragments;
as the theory is not mature, the problem of fragment cloud caused by ultra-high speed collision mainly adopts two methods of experiment and numerical simulation as quantitative research means; the ultra-high speed collision experiment needs to use a large-scale implementation device, the cost of the collision experiment is high, and meanwhile, the experiment preparation, the data acquisition and the result processing all need long time, so the time needed by a single experiment is usually in the unit of days; therefore, the numerical simulation method is widely used as an effective supplement of an experimental method due to the advantage of low cost; the smooth particle method (SPH), as a meshless method, geometrically disperses shots and target plates using a large number of discrete particles, and thus can simulate large deformation and breakage phenomena of an object, which is commonly used for ultra-high speed collision simulation; however, the smooth particle method also requires a cyclic traversal of discrete particles, performs a large number of numerical calculations, and increases the calculation time exponentially with the increase in the number of discrete particles used; generally, the time required for ultra-high speed crash simulation of a single condition is in hours; in engineering and experimental design, debris cloud distribution characteristics under a plurality of specified collision parameters need to be rapidly explored, and the problems of long time period exist in both experimental and numerical simulation methods.
Disclosure of Invention
The invention aims to provide a machine learning-based rapid simulation method for ultra-high-speed collision fragment clouds, which aims to solve the problems that in the existing engineering and experimental design provided in the background art, the distribution characteristics of the fragment clouds under a plurality of specified collision parameters are required to be rapidly visited, and the experimental and numerical simulation methods have long time periods.
In order to achieve the above purpose, the invention adopts the technical scheme that: a machine learning-based method for quickly simulating the ultra-high-speed collision fragment cloud comprises the following steps,
s10, a fragment cloud key feature extraction method; s20, constructing a fragment cloud rapid simulation model; s30, applying a strategy to the collision condition in the model; and S40, a model accuracy evaluation method.
Further, the S10 method for extracting key features of a fragment cloud includes analyzing a historical result generated by the smooth particle method, converting a result file described by fragments therein into a key feature file of a fragment cloud described by spatial distribution, performing structured grid division on all spatial regions where fragment distribution may exist, and counting the following physical quantities in each hexahedral grid: the number of fragments, the total mass of the fragments, the average speed of the fragments in the X direction, the average speed of the fragments in the Y direction, the average speed of the fragments in the Z direction, the momentum sum of the fragments and the kinetic energy sum of the fragments, and the key characteristics of the fragment cloud described by the spatial distribution are constructed.
Further, the S20 method for constructing the fragment cloud rapid simulation model includes that the method constructs the fragment cloud rapid simulation model based on a variational self-encoder network structure, wherein a key feature of the fragment cloud described according to spatial distribution is used as an input layer, a simulation result and a mean square error of data on the input layer are used as a loss function, and a model optimization direction is controlled so that a model output result gradually approaches data on the input layer, thereby enabling the fragment cloud rapid simulation model to obtain a spatial distribution feature of the fragment cloud.
Further, in the step S30, applying a policy to the collision condition in the model, the method adds the collision condition as additional data to the input layer data, outputs the prediction result through the fragment cloud rapid simulation model, and adds the mean square error between the prediction result and the input collision condition to the loss function to achieve the purpose of controlling the generation of the fragment cloud through the collision condition, and meanwhile, in order to improve the control effect of the collision condition, increases the weight of the collision condition in the loss function, and the weight coefficient is equal to the average value of the number of meshes in each direction when the structured mesh partition is performed on the space region of the fragment cloud.
Further, in the S40, the method for evaluating model accuracy comprehensively evaluates accuracy and reliability of the model by calculating different error criteria between training set data used in the training process of the model and the output result of the model, where the error criteria include: the total mass error proportion of the fragment cloud, the structural similarity of a mass distribution image of the fragment cloud, the structural similarity of an average velocity distribution image of the fragment cloud, and a dimensionless mass contrast curve of the fragment cloud along the X axis.
The invention has the beneficial effects that:
the method comprises the steps of extracting fragment cloud distribution characteristics through a characteristic engineering method; then, establishing a fragment cloud rapid simulation model based on a variational self-encoder network structure, and taking the distribution characteristics and collision parameters of the fragment cloud as input data; meanwhile, a loss function is designed, the hierarchical structure of the neural network is guided to realize complex function approximation between input data and output data, and then a nonlinear relation between a collision parameter and high-dimensional fragment cloud distribution characteristics is established; in order to evaluate the accuracy of the obtained fragment cloud rapid generation model, a model error evaluation method is established, and the accuracy and the reliability of the model are comprehensively evaluated by calculating various error standards between the output result of the model and the model training data.
Drawings
Fig. 1 is a network structure diagram of a conditional variation self-encoder of a machine learning-based rapid simulation method for an ultra-high-speed collision fragment cloud.
Fig. 2 is a flowchart of a fragment cloud key feature extraction algorithm of a machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method.
Fig. 3 is a schematic diagram of a fragment cloud fast generation model network structure of an ultra-high-speed collision fragment cloud fast simulation method based on machine learning.
FIG. 4 is a sample use case diagram in model training and accuracy evaluation of a machine learning-based ultra-high-speed collision fragment cloud rapid simulation method.
Detailed Description
The following detailed description of the present invention is given for the purpose of better understanding technical solutions of the present invention by those skilled in the art, and the present description is only exemplary and explanatory and should not be construed as limiting the scope of the present invention in any way.
As shown in fig. 1 to 4, the specific structure of the present invention is: a machine learning-based method for quickly simulating the ultra-high-speed collision fragment cloud comprises the following steps,
s10, a fragment cloud key feature extraction method; s20, constructing a fragment cloud rapid simulation model; s30, applying a strategy to the collision condition in the model; and S40, a model accuracy evaluation method.
Preferably, the S10 method for extracting key features of a fragment cloud includes analyzing a historical result generated by a smooth particle method, converting a result file described by fragments therein into a key feature file of a fragment cloud described by spatial distribution, performing structured grid division on all spatial regions where fragment distribution may exist, and counting the following physical quantities in each hexahedral grid: the number of fragments, the total mass of the fragments, the average speed of the fragments in the X direction, the average speed of the fragments in the Y direction, the average speed of the fragments in the Z direction, the momentum sum of the fragments and the kinetic energy sum of the fragments, and the key characteristics of the fragment cloud described by the spatial distribution are constructed.
Specifically, training data and test data of the fragment cloud rapid generation model are derived from numerical simulation results output by using a smooth particle method for ultrahigh-speed collision simulation. The numerical simulation result is stored in a table form, the fragment number is taken as a vertical axis, and physical information such as the mass, the space coordinate, each axial speed, each axial characteristic size and the like of the fragment is taken as a horizontal axis;
taking a round bullet high-speed collision target plate in a thin-wall cylinder shape as an example, the bullet is crushed and penetrates through the cylinder, then a spindle-shaped incident fragment cloud is generated in the incident speed direction, and meanwhile, a backsplash fragment cloud is generated above a collision point; in the numerical simulation result, according to different collision parameters, the size of the fragment cloud and the number difference of fragments contained in the fragment cloud are large, and the quantity difference among the fragments is abnormally up to 3 orders of magnitude; meanwhile, the number of fragments generated by collision is large, and the problem dimensionality is too large by directly using all fragment data as input of a fast generation model, so that the fast solution is not facilitated; therefore, the faster solving strategy is to convert the description mode of the fragment cloud from the fragment as a unit to the description according to the spatial distribution by taking the whole fragment cloud as a research object;
the algorithm logic is as follows: 1. according to an empirical formula, estimating a spatial region in which fragment clouds are likely to be distributed after high-speed collision, wherein the spatial region is used as a fragment cloud space; 2. carrying out structured grid division on the fragment cloud space, and considering that the sample size of a numerical simulation result which is usually used as model training data is less, dividing the fragment cloud space into hexahedral grids with the area of 20 multiplied by 20 or 30 multiplied by 30; 3. traversing all grids, counting the number of fragments, the total mass, the total momentum, the average speed in each axial direction, the maximum speed and the maximum kinetic energy of a space region occupied by the grids, and recording the fragment cloud characteristics of 6 channels of the fragment cloud; 4. respectively carrying out data normalization operation on 6 channel data of fragment cloud characteristics, wherein the total mass channel and the total momentum channel need to be subjected to logarithmic operation before normalization operation so as to optimize data distribution;
through the extraction of the key features of the fragment cloud, the numerical simulation result of the fragment cloud is converted into matrix data of 20 multiplied by 20 or 30 multiplied by 30, each point in the matrix has 6 channels of data, and the data respectively correspond to the physical quantities such as the number, the total mass, the total momentum, the axial average speed, the maximum kinetic energy and the like of fragments in the corresponding region. Through the method, the fragment cloud characteristic description based on the spatial region distribution is realized, and the data are used as the input of the fragment cloud rapid generation model.
Preferably, the S20 method for constructing the fragment cloud rapid simulation model includes that the method constructs the fragment cloud rapid simulation model based on a variational self-encoder network structure, wherein a key feature of the fragment cloud described according to spatial distribution is used as an input layer, a simulation result and a mean square error of data of the input layer are used as a loss function, and a direction of model optimization is controlled so that a model output result gradually approaches data of the input layer, thereby enabling the fragment cloud rapid simulation model to obtain a spatial distribution feature of the fragment cloud.
Specifically, based on a conditional variation autoencoder network structure (CVAE), a deep neural network structure for automatic image generation is used for establishing a fragment cloud rapid simulation model;
according to the structural design requirement of the CVAE network, an encoder module of the fragment cloud rapid simulation model uses the key features and collision parameters of the fragment cloud extracted from the numerical simulation result as model input; the fragment cloud is characterized by a 3-dimensional matrix of 6 channels, the 3-dimensional matrix is converted into 6-channel data with the size of 80 multiplied by 100 or 150 multiplied by 180 through matrix deformation after a model is input, the 6-channel data passes through a two-dimensional convolution network layer and a batch normalized network layer combination layer of 6 layers and then passes through two layers of full-connection layer networks, and finally a hidden space layer with the degree of freedom of 300 and the probability density distribution conforming to Gaussian distribution is achieved;
according to the design requirements of a CVAE network structure, a decoder module of a fragment cloud rapid simulation model uses 300-dimensional data and collision parameters randomly sampled from a hidden space as input data; inputting data, passing through two full-connection layers, passing through a combination layer of 5 two-dimensional deconvolution layers and a batch standard layer, and finally outputting 6-channel data with the size of 80 × 100 or 150 × 180; then converting the data into three-dimensional matrix data of 20 multiplied by 20 or 30 multiplied by 30 of 6 channels through matrix deformation;
an optimization objective function used by a fragment cloud rapid simulation model consists of two parts: the first part represents the k.l. distance between the probability density distribution and the gaussian distribution of the hidden space; the second part represents the similarity between the key features of the fragment cloud extracted from the numerical simulation result and the features of the fragment cloud generated by the fragment cloud rapid simulation model; in the model training process, the model parameter training direction is guided by optimizing the objective function, so that the fragment cloud rapid simulation model learns the fragment cloud distribution rule under various collision parameters, and the key characteristics of the fragment cloud spatial distribution are obtained.
Preferably, in step S30, the method includes adding the collision condition as additional data to the input layer data, outputting the prediction result through the fragment cloud rapid simulation model, and adding the mean square error between the prediction result and the input collision condition to the loss function to achieve the purpose of controlling generation of the fragment cloud through the collision condition, and meanwhile, to improve the control effect of the collision condition, increasing the weight occupied by the collision condition in the loss function, where the weight coefficient is equal to the average value of the number of meshes in each direction when the space region of the fragment cloud is subjected to structured mesh division.
Specifically, in the process of constructing the fragment cloud rapid simulation model, the input data of the encoder module should include not only the key features of the fragment cloud in the form of a 6-channel 3-dimensional matrix, but also the collision parameters at that time; the collision parameters comprise the projectile incidence speed, the incidence angle and the collision point coordinates, and can be subjected to vectorization description; the collision parameters serve as supplementary key features of the fragment cloud, and when a loss function of the fragment cloud rapid simulation model is calculated, the collision parameters also serve as a part of the key features of the original fragment cloud, and a formula is input for calculation;
the collision parameter is used as a single numerical value, and when the loss function is calculated, the scale factor is smaller than that of the original fragment cloud key feature matrix data, so that the weight of the collision parameter in the loss function needs to be adjusted; after the collision parameters are input into the model, copying the numerical values of each type of collision parameters into a column, and adding the column into a converted fragment cloud key characteristic matrix; at this time, the original 6-channel matrix with the size of 80 × 100 or 150 × 180 is increased to a 6-channel matrix with the size of 80 × (100+3) or 150 × (180+ 3); through the design, the effect of the collision parameters can be quickly applied to the fragment cloud quick simulation model on the premise of fully utilizing the original network structure.
Preferably, S40, the method for evaluating model accuracy comprehensively evaluates accuracy and reliability of the model by calculating different error criteria between training set data used in the training process of the model and the output result of the model, where the error criteria include: the total mass error proportion of the fragment cloud, the structural similarity of a mass distribution image of the fragment cloud, the structural similarity of an average velocity distribution image of the fragment cloud, and a dimensionless mass contrast curve of the fragment cloud along the X axis.
Specifically, in order to evaluate the accuracy of the fragment cloud rapid simulation model, a fragment cloud numerical simulation result needs to be divided into training set data and test set data; the test set data is only used for evaluating the accuracy of the model and accounts for 10 to 30 percent of the total sample proportion;
when the model accuracy is evaluated, comparing an output result of the fragment cloud rapid simulation model with the fragment cloud under the same collision parameters in the test set data; the error indicators used in the evaluation strategy include: the image structure similarity of the fragment number distribution diagram of the central section of the fragment cloud, the image structure similarity of the mass distribution diagram of the central section of the fragment cloud, the average absolute value error of mass distribution along the axial direction, the average absolute value error of average momentum distribution, the average absolute value error of maximum speed distribution and the average absolute value error of maximum kinetic energy distribution; and quantitatively evaluating the accuracy of the physical quantity of each channel generated by the fragment cloud rapid simulation model through the error indexes.
The working principle and the using process of the method are as follows: aiming at a fragment cloud numerical simulation result of the existing smooth particle method, firstly, a fragment cloud key feature extraction method is needed to be used for converting the numerical simulation result; as shown in the flowchart of the fragment cloud key feature extraction algorithm in fig. 2: 1) calculating the possible distribution area of the fragments after collision according to an empirical formula; 2) enlarging the fragment area obtained by calculation of an empirical formula by 20 percent to ensure that all fragments fall into the calculation area; 3) scaling the fragment cloud size by dividing the fragment space coordinates (x, y, z) by the incident velocity v; 4) carrying out structured grid division on the zoomed fragment cloud space, and dividing 20 or 30 units in each axial direction; 5) traversing hexahedral meshes, counting the number, total mass, total momentum, each axial average velocity, maximum velocity and maximum kinetic energy of the chips contained in a space region corresponding to each mesh, filling the number, total mass, total momentum, each axial average velocity, maximum velocity and maximum kinetic energy into matrix data of 20 × 20 × 20 or 30 × 30 × 30, wherein each point in the matrix corresponds to 6-channel data; 6) and normalizing the physical quantity of each channel, wherein the data of the three channels, namely the total mass, the total momentum and the maximum kinetic energy, are subjected to logarithmic processing before normalization processing, so that the data distribution rule is optimized.
In specific implementation, reading a calculation result file of a smooth particle method through a read _ csv function of a pandas library of python; and then, a numpy library is used for declaring a 20 × 20 × 20 × 6 or 30 × 30 × 30 × 6 matrix array, traversing the structured grids generated by the fragment cloud space division, counting the 6-channel physical quantity of the contained fragments in each grid, and filling data into the numpy matrix array.
The data processing process is executed on all the numerical simulation results, all the fragment cloud key feature results are combined into a numpy matrix, and matrix data are stored in a binary file mode through a python's folder library, so that data loading is conveniently carried out during fragment cloud rapid simulation model training; and the collision parameters corresponding to the numerical simulation result are also organized in a numpy matrix mode and are stored as a binary file through a folder.
As shown in fig. 3, the network structure of the fragment cloud rapid generation model is constructed by using a two-dimensional convolution layer function, a two-dimensional deconvolution layer function, a batch normalization function, a full link layer function, and an optimized target function interface provided by a keras library according to the network structure diagram of the fragment cloud rapid simulation model; the open source machine learning framework tensorflow is used as a background computing kernel of the keras to provide support for training and predicting a fragment cloud rapid simulation model.
As shown in fig. 3, in the network structure of the fragment cloud rapid generation model, in the training process of the fragment cloud rapid simulation model, 80% of all samples are randomly extracted as training set data, and input into the model for training; and randomly extracting 15% of training set data in the training process as verification set data; and after the model training is finished, all residual samples which do not participate in the model training are used as test set data for model accuracy evaluation.
After the data preparation and model building work is completed, high-performance GPU resources are needed to execute model training work; when the fragment cloud rapid simulation model is trained, software resources need to meet installation and operation requirements of a Keras library and a GPU implementation library of TensorFlow, and GPU hardware resources need to support installation and operation of NVIDIACUDA software.
As shown in the use condition of the sample in the model training and accuracy evaluation in fig. 4, after the model training is completed, the accuracy of the fragment cloud rapid simulation model needs to be evaluated; all residual samples which do not participate in model training are used as test set data, the same collision parameters are input into the model for fragment cloud prediction, and therefore a model prediction result is generated; the error indexes designed according to the evaluation strategy comprise: comparing the 6-channel physical quantity in the prediction result of the rapid simulation model with the numerical simulation result generated by the smooth particle method; the error indexes quantitatively evaluate the accuracy of the prediction result from two aspects of image similarity and physical quantity average absolute value error; through comparison of the fragment cloud rapid simulation model generation result and untrained data, the universality of the model for other collision parameters can be predicted and evaluated.
When performing a prediction task using a fragment cloud rapid simulation model, as shown in fig. 3, only the decoder module using the model is needed; generating 300-dimensional random data by random sampling of Gaussian distribution, and inputting the 300-dimensional random data and the specified collision parameters into a data inlet of a decoder module at the same time; in a common personal desktop computer, 6-channel physical quantities which can describe the key features of the fragment cloud under the corresponding collision condition can be quickly obtained within second-level time by using a small amount of CPU resources and memory resources.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
The principles and embodiments of the present invention are explained herein using specific examples, which are presented only to assist in understanding the method and its core concepts of the present invention. The foregoing is only a preferred embodiment of the present invention, and it should be noted that there are objectively infinite specific structures due to the limited character expressions, and it will be apparent to those skilled in the art that a plurality of modifications, decorations or changes may be made without departing from the principle of the present invention, and the technical features described above may be combined in a suitable manner; such modifications, variations, combinations, or adaptations of the invention using its spirit and scope, as defined by the claims, may be directed to other uses and embodiments.
Claims (5)
1. A machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method is characterized by comprising the following steps of: comprises the following steps of (a) carrying out,
s10, a fragment cloud key feature extraction method; s20, constructing a fragment cloud rapid simulation model; s30, applying a strategy to the collision condition in the model; and S40, a model accuracy evaluation method.
2. The machine learning-based ultra-high speed collision fragment cloud rapid simulation method according to claim 1, characterized in that: the S10 method for extracting the fragment cloud key feature includes analyzing a historical result generated by the smooth particle method, converting a result file described by the fragment into a fragment cloud key feature file described by spatial distribution, performing structured grid division on all spatial regions where the fragment distribution may exist, and counting the following physical quantities in each hexahedral grid: the number of fragments, the total mass of the fragments, the average speed of the fragments in the X direction, the average speed of the fragments in the Y direction, the average speed of the fragments in the Z direction, the momentum sum of the fragments and the kinetic energy sum of the fragments, and the key characteristics of the fragment cloud described by the spatial distribution are constructed.
3. The machine learning-based ultra-high speed collision fragment cloud rapid simulation method according to claim 1, characterized in that: s20, constructing a fragment cloud rapid simulation model based on a variational self-encoder network structure, wherein the method takes the key features of the fragment cloud described according to spatial distribution as an input layer, takes the mean square error of a simulation result and data of the input layer as a loss function, and controls the optimization direction of the model to enable the output result of the model to gradually approach the data of the input layer, so that the fragment cloud rapid simulation model obtains the spatial distribution features of the fragment cloud.
4. The machine learning-based ultra-high speed collision fragment cloud rapid simulation method according to claim 1, characterized in that: and S30, applying a strategy to the collision condition in the model, adding the collision condition as additional data into the data of the input layer, outputting a prediction result through the fragment cloud rapid simulation model, adding the prediction result and the mean square error of the input collision condition into a loss function so as to achieve the purpose of controlling the generation of the fragment cloud through the collision condition, and meanwhile, in order to improve the control effect of the collision condition, increasing the weight of the collision condition in the loss function, wherein the weight coefficient is equal to the average value of the number of grids in each direction when the space region of the fragment cloud is subjected to structured grid division.
5. The machine learning-based ultra-high speed collision fragment cloud rapid simulation method according to claim 1, characterized in that: the S40, method for evaluating model accuracy, in which the accuracy and reliability of the model are comprehensively evaluated by calculating different error criteria between training set data used in the model training process and the model output result, where the error criteria include: the total mass error proportion of the fragment cloud, the structural similarity of a mass distribution image of the fragment cloud, the structural similarity of an average velocity distribution image of the fragment cloud, and a dimensionless mass contrast curve of the fragment cloud along the X axis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010998145.0A CN112214927A (en) | 2020-09-21 | 2020-09-21 | Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010998145.0A CN112214927A (en) | 2020-09-21 | 2020-09-21 | Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112214927A true CN112214927A (en) | 2021-01-12 |
Family
ID=74049742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010998145.0A Pending CN112214927A (en) | 2020-09-21 | 2020-09-21 | Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112214927A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116861241A (en) * | 2023-07-11 | 2023-10-10 | 北京天工科仪空间技术有限公司 | Space debris collision probability prediction method, system, terminal and storage medium based on artificial intelligence |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108775998A (en) * | 2018-03-27 | 2018-11-09 | 浙江大学 | A kind of system and method measuring hypervelocity impact panus three-dimensional structure |
-
2020
- 2020-09-21 CN CN202010998145.0A patent/CN112214927A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108775998A (en) * | 2018-03-27 | 2018-11-09 | 浙江大学 | A kind of system and method measuring hypervelocity impact panus three-dimensional structure |
Non-Patent Citations (1)
| Title |
|---|
| 文永 等: "基于深度学习的超高速碰撞碎片云模拟", 《固体力学学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116861241A (en) * | 2023-07-11 | 2023-10-10 | 北京天工科仪空间技术有限公司 | Space debris collision probability prediction method, system, terminal and storage medium based on artificial intelligence |
| CN116861241B (en) * | 2023-07-11 | 2024-02-20 | 北京天工科仪空间技术有限公司 | Space debris collision probability prediction method and system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105550374A (en) | Random forest parallelization machine studying method for big data in Spark cloud service environment | |
| Tüysüz et al. | Hybrid quantum classical graph neural networks for particle track reconstruction | |
| CN109388565B (en) | Software system performance optimization method based on generating type countermeasure network | |
| CN112560966B (en) | Polarized SAR image classification method, medium and equipment based on scattering map convolution network | |
| CN107704883A (en) | A kind of sorting technique and system of the grade of magnesite ore | |
| CN104573279A (en) | Method for quickly generating nuclear radiation shield calculation grids based on deep stripping | |
| CN107579821A (en) | Password dictionary generation method and computer-readable recording medium | |
| CN110987436A (en) | Bearing fault diagnosis method based on excitation mechanism | |
| CN112214927A (en) | Machine learning-based ultrahigh-speed collision fragment cloud rapid simulation method | |
| CN113221200A (en) | Three-dimensional efficient random arrangement method suitable for uncertainty analysis of reactor core particle distribution | |
| Gilan et al. | Sustainable building design: A challenge at the intersection of machine learning and design optimization | |
| Zhang et al. | Using genetic algorithm to optimize artificial neural network: A case study on earthquake prediction | |
| CN103489007A (en) | Method for selecting kernel function of support vector machine based on sample prior information and application | |
| CN111859785B (en) | Fluid feature extraction method, system, computer-readable storage medium and device | |
| Zhao et al. | CNC thermal compensation based on mind evolutionary algorithm optimized BP neural network | |
| CN112347531A (en) | Three-dimensional crack propagation path prediction method and system for brittle marble | |
| CN112819246A (en) | Energy demand prediction method for optimizing neural network based on cuckoo algorithm | |
| CN109885877B (en) | Constraint domain optimization Latin hypercube design method based on clustering algorithm | |
| Mirsoleimani et al. | A two-tier design space exploration algorithm to construct a gpu performance predictor | |
| Feldkamp et al. | Explainable AI For Data Farming Output Analysis: A Use Case for Knowledge Generation Through Black-Box Classifiers | |
| Zhenjun | APPLICATION OF NEURAL NETWORK TECHNOLOGY IN MACHINING ERROR RECOVERY. | |
| CN112307673A (en) | Grid surface quality detection method based on deep learning | |
| Wang et al. | An improved system for efficient shape optimization of vehicle aerodynamics with “noisy” computations | |
| CN112241811A (en) | Method for predicting hierarchical mixed performance of customized product in 'Internet +' environment | |
| WO2019209571A1 (en) | Proactive data modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |