Low scattering carrier RCS prediction method based on deep learning

Abstract

The invention relates to the field of electronic design automation software, in particular to a low-scattering carrier RCS prediction method based on deep learning, which comprises the following steps: constructing a deep learning solver RCSAI for solving a low-scattering carrier RCS; the method comprises the following steps: acquiring original data for training RCSAI based on an electromagnetic simulation platform; obtaining a sample set required for training a deep learning solver based on the original data; training and solving a deep learning solver RCSAI of the low-scattering carrier RCS by taking the surface equivalent current as output; based on the RCSAI, the rapid reasoning of the RCS data of the low-scattering carrier is realized. According to the method, the simulation difficulty is reduced, the RCS parameter solving speed is increased, the design optimization process of the electromagnetic device is optimized, and the cost is saved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic design automation (EDA) software, and relates to a low-scattering carrier RCS calculation process optimization and implementation method, in particular to a low-scattering carrier RCS prediction method based on deep learning. BACKGROUND

[0002] In the design and testing of the stealth performance of aircraft components, a low-scattering carrier is needed to eliminate the influence of the component edge and the embedded structure after the component is separated from the aircraft body, so as to accurately simulate the actual installation state of the component. The low-scattering carrier RCS prediction method based on deep learning can accelerate the design and testing process of the stealth performance of aircraft components.

[0003] At present, the existing electromagnetic simulation platform can usually simulate the electric field distribution of the low-scattering carrier and calculate the corresponding RCS value. However, the existing technology has the following deficiencies:

[0004] Efficiency problem: The electromagnetic simulation platform usually uses finite element, boundary element, finite difference and other algorithms, and usually consumes a large amount of time when simulating the low-scattering carrier.

[0005] Insufficient optimization capability: The optimization design usually depends on the RCS parameters of the low-scattering carrier. Therefore, the time-consuming low-scattering carrier simulation will make the low-scattering carrier design optimization process inefficient, thereby increasing the design cycle and cost. SUMMARY

[0006] The present application provides a low-scattering carrier RCS prediction method based on deep learning, which effectively accelerates the low-scattering carrier simulation optimization design process and improves the design efficiency.

[0007] In order to achieve the purpose of the present application, the technical scheme adopted is: a low-scattering carrier RCS prediction method based on deep learning, comprising:

[0008] Step 1, constructing a deep learning solver RCSAI for solving the RCS of the low-scattering carrier;

[0009] Step 2, based on the electromagnetic simulation platform, obtaining the original data of the deep learning solver RCSAI for solving the RCS of the low-scattering carrier;

[0010] Step 3, obtaining the sample set required for training the deep learning solver based on the original data;

[0011] Step 4, training the deep learning solver RCSAI for solving the RCS of the low-scattering carrier with the RCS parameters as the output target;

[0012] Step 5, based on the deep learning solver RCSAI solving the RCS of low scattering carriers, the corresponding RCS data of low scattering carriers is quickly output.

[0013] As an optimization scheme of the present application, in step 1, the steps of constructing the deep learning solver RCSAI solving the RCS of low scattering carriers are as follows:

[0014] Step 1.1, according to the geometric characteristics and training target of the low scattering carrier to be solved, determine the input and output of the deep learning solver RCSAI;

[0015] Step 1.2, based on the dimensions of the input and output, construct the deep learning solver RCSAI solving the RCS of low scattering carriers based on the Transformer architecture.

[0016] As an optimization scheme of the present application, in step 2, it specifically includes the following steps:

[0017] Step 2.1, based on the geometric modeling tool of the electromagnetic simulation platform, construct the geometric model of the low scattering carrier for the target scene;

[0018] Step 2.2, based on the grid generation tool of the electromagnetic simulation platform, generate the grid of the corresponding geometric model;

[0019] Step 2.3, based on the algorithm solving of the electromagnetic simulation platform, obtain the equivalent current J on the surface of the low scattering carrier, and calculate the far-zone scattering field expression as:

[0020] E s =-jωμN t

[0021] Wherein: represents the complex unit, ω=2πf represents the angular frequency, μ represents the magnetic conductivity, N t represents the transverse component of N relative to direction:

[0022]

[0023] Wherein: e represents the natural constant, ∈ represents the dielectric constant, r represents the far field, and ∫·dS' represents the integration on the entire geometric body surface;

[0024] Based on the far-zone scattering field E s and the given incident field E i , the RCS parameter σ of the corresponding low scattering carrier can be calculated:

[0025]

[0026] Wherein: θ i ,φi , θ, φ represent the incident field incident elevation angle, incident field incident azimuth angle, far zone scattering field scattering elevation angle, far zone scattering field scattering azimuth angle; the RCS parameter maps the scattering characteristics of the low-scattering carrier.

[0027] As an optimization scheme of the present application, in step 3, the step of obtaining the sample set required by the training deep learning solver based on the original data is: according to the original data obtained in step 2, a data set is constructed in the form of Parameter-Mesh-J. For each data pair Parameter-Mesh, there is a surface equivalent current J, and according to the surface equivalent current J, the corresponding RCS can be calculated.

[0028] As an optimization scheme of the present application, in step 4, specifically includes the following steps:

[0029] Step 4.1, according to the Parameter-Mesh-J data obtained in step 3, determine the input and output dimensions of the deep learning solver RCSAI for solving low-scattering carrier in step 1, RCSAI is composed of a first multi-layer perception layer, a plurality of self-attention layers and a second multi-layer perception layer, the first multi-layer perception layer is responsible for Parameter-Mesh data dimension reduction and data fusion, and the second multi-layer perception layer is responsible for outputting surface equivalent current data;

[0030] Step 4.2, the geometric parameters and grid data of the input of RCSAI are EM-Parm and Mesh respectively, and the output of RCSAI is surface equivalent current J Pred , which is defined as:

[0031] J Pred = RCSAI(EM-Parm, Mesh)

[0032] The loss function RCSLoss used to train RCSAI is defined as follows:

[0033]

[0034] Where |·| represents the modulus of a complex vector, N is the number of grid points, represents the output of the i-th grid of RCSAI, J i represents the actual surface equivalent current on the i-th grid (the i-th component of the surface equivalent current J), and RCSLoss depicts the average value of the surface equivalent current error on all grids;

[0035] Step 4.3, based on the Adams algorithm, the learnable parameters in the MLP, Self-Attention, and MLP layers in RCSAI are updated iteratively;

[0036] Step 4.4: When the error on the training set is less than a given threshold, or the maximum number of iterations is reached, the training process of RCSAI ends.

[0037] As an optimized solution of the present invention, in step 5, the step of quickly predicting the corresponding RCS data for a low-scattering carrier based on a deep learning solver for solving the RCS of the low-scattering carrier is as follows: A geometric model of the low-scattering carrier is constructed using the geometric modeling tool of the electromagnetic simulation platform; a mesh corresponding to the geometric model is generated using the mesh generation tool of the electromagnetic simulation platform; geometric and electromagnetic parameters are extracted and used as input parameters to the deep learning solver RCSAI for the low-scattering carrier RCS; the trained deep learning solver RCSAI quickly predicts the corresponding RCS result based on the input parameters; and the obtained RCS result is used to quickly optimize the low-scattering carrier.

[0038] This invention offers several advantages: it solves the problem of time-consuming and inefficient simulation of the RCS of low-scattering carriers using traditional methods. Simulating the RCS of low-scattering carriers typically requires solving a large-scale system of linear equations using direct or iterative methods, a process that is usually quite time-consuming. This invention, based on a trained deep learning solver RCSAI, can quickly infer the equivalent surface current information from the input geometric parameters and corresponding mesh information, thereby efficiently calculating the RCS value of the target carrier and ultimately accelerating the optimization design process. This invention reduces simulation difficulty, speeds up RCS parameter solving, optimizes the electromagnetic device design optimization process, and saves costs. Attached Figure Description

[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0040] Figure 1 This is a schematic diagram of the process of the present invention;

[0041] Figure 2 This is a schematic diagram of the workflow of the deep learning solver RCSAI;

[0042] Figure 3 A schematic diagram of the double Ogive mesh for a low-scattering carrier;

[0043] Figure 4 This is a graph showing the output RCS parameters. Detailed Implementation

[0044] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0045] like Figure 1 As shown, this invention discloses a deep learning-based method for predicting the RCS of low-scattering carriers, which includes the following steps:

[0046] Step 1, constructing a deep learning solver for solving RCS of a low-scattering carrier;

[0047] The embodiment of the application adopts a deep learning library Pytorch to construct a deep learning solver for solving RCS of a low-scattering carrier, which can realize fast obtaining of corresponding surface equivalent current J and RCS value from input of electromagnetic size, electromagnetic parameter, grid and other information of the low-scattering carrier.

[0048] Step 2, obtaining original data of the deep learning solver for solving RCS of a low-scattering carrier based on an electromagnetic simulation platform.

[0049] In the embodiment of the application, the Rainbow Studio EM software based on a full-wave electromagnetic simulation platform can obtain a data set for training RCSAI, as shown in Figure 1 and Figure 2 The workflow for obtaining the data set is as follows: based on the geometry modeling platform of the full-wave electromagnetic simulation platform Rainbow Studio EM, the target carrier double Ogive is geometrically modeled, and the geometric correctness is checked; based on the geometry modeling, the corresponding grid of the geometry is generated by means of the grid generation tool of Rainbow Studio EM, and the grid quality is checked, referring to Figure 3 ; based on the generated grid, the surface equivalent current J of the target carrier is obtained by means of the algorithm solving of Rainbow Studio EM. According to the surface equivalent current J, the corresponding RCS value σ can be calculated, and the calculation formula is as follows:

[0050] The calculation formula of the far-zone scattering field expression is as follows:

[0051] E s =-jωμN t

[0052] Wherein: represents a complex unit, ω = 2πf represents an angular frequency, μ represents a magnetic permeability, N t represents the transverse component of N relative to direction:

[0053]

[0054] Wherein: e represents a natural constant, ∈ represents a dielectric constant, r represents a far field, and ∫·dS' represents integration on the entire geometric surface;

[0055] Based on the far-zone scattering field E s and the given incident field E i , the RCS parameter σ of the corresponding low-scattering carrier can be calculated:

[0056]

[0057] wherein: θ i ,φ i ,θ,φ represent the incident field incident elevation angle, incident field incident azimuth angle, far zone scattering field scattering elevation angle, far zone scattering field scattering azimuth angle respectively; the RCS parameter maps the scattering characteristics of the low-scattering carrier.

[0058] Step 3, based on the original data, a sample set required for training RCSAI is obtained;

[0059] In the embodiment of the application, based on step 2, a geometric parameter-J data pair can be obtained, and for different geometric parameters, we have the corresponding surface equivalent current J. The geometric parameters here include the radius, length, angle, etc. of the low-scattering carrier. For the geometric parameter in the geometric parameter-J data pair, the geometric parameter can be converted into grid data Mesh according to the electromagnetic simulation platform grid generation tool, so as to convert the geometric parameter-J data pair into Parameter-Mesh-J data pair (as shown in Figure 2 80% of these data pairs are randomly selected as the training set, 10% as the test set, and 10% as the validation set.

[0060] Step 4, training the deep learning solver RCSAI for solving the RCS of the low-scattering carrier with RCS parameter as the output target;

[0061] Step 4.1, according to the Parameter-Mesh-J data pair obtained in step 3, the input and output dimensions of the deep learning solver RCSAI for solving the low-scattering carrier in step 1 can be determined, wherein the input dimension is related to the geometric input, frequency, incident angle, etc., and the output dimension is related to the number of grid points. RCSAI is composed of multiple multilayer perception (MLP) layers, multiple self-attention layers (Self-Attention) and multiple multilayer perception (MLP) layers. The first MLP layer is responsible for Parameter-Mesh data dimension reduction and data fusion, and the last MLP layer is responsible for outputting surface equivalent current data.

[0062] Step 4.2, let the geometric parameters and grid data of the input of RCSAI be EM-Parm and Mesh respectively, and the output of RCSAI be:

[0063] J Pred =RCSAI(EM-Parm,Mesh).

[0064] The loss function RCSLoss used to train RCSAI is defined as follows:

[0065]

[0066] where |·| denotes the modulus of a complex vector, N is the number of grid points, represents the output of RCSAI corresponding to the i-th grid, J i represents the actual surface equivalent current on the i-th grid. RCSLoss characterizes the average value of the error of the surface equivalent current on all grids.

[0067] Step 4.3, based on the Adams algorithm, back propagation is carried out, and the learnable parameters in the MLP, Self-Attention, and MLP layers in the RCSAI iteration are updated.

[0068] Step 4.4, when the error on the training set is less than a given threshold, or the maximum number of iterations is reached, the training process of the RCSAI is ended.

[0069] Step 5, based on the deep learning solver RCSAI for solving the RCS of a low-scattering carrier, the RCS data of the low-scattering carrier is quickly predicted;

[0070] In the embodiment of the application, the geometric modeling platform based on the full-wave electromagnetic simulation platform Rainbow Studio EM is used to construct the geometric model of the low-scattering carrier double Ogive, and the geometric correctness is checked. The grid generation tool based on Rainbow Studio EM is used to generate the grid of double Ogive, and the grid quality is checked. The grid data and electromagnetic parameters are extracted, preprocessed and enhanced, and then input into the trained deep learning solver. For the input electromagnetic parameters and grid data, based on the trained deep learning solver, GPU can be used for fast inference. The specific process is as follows: via the first MLP layer, the grid data and electromagnetic parameters are reduced in dimension and fused; the fused data is input into multiple self-attention layers to capture the dependency between elements at different positions in the sequence; finally, via the MLP layer, the data is converted into the surface equivalent current of the geometric body. Since the process of quickly inferring the surface equivalent current of the geometric body does not involve matrix inversion and matrix iteration, compared with the traditional electromagnetic simulation solver, RCSAI is very efficient and fast. Based on the surface equivalent current of the geometric body obtained by RCSAI, the corresponding RCS result can be quickly derived using the calculation formula. Taking the incident angle as the horizontal coordinate and the RCS value as the vertical coordinate, the RCS curve as shown in Figure 4 is obtained. According to the RCS curve, the peak value, mean value, fluctuation, etc. of the curve are recorded, and based on the results, the key scattering sources and problem areas are located, so as to optimize the low-scattering carrier.

[0071] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application, and it should be understood that the above-described specific embodiments are merely examples of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A deep learning-based method for predicting the RCS of low-scattering carriers, characterized in that: include: Step 1: Construct a deep learning solver, RCSAI, to solve the RCS of a low-scattering carrier; Step 2: Based on the electromagnetic simulation platform, obtain the raw data of RCSAI, a deep learning solver for solving the RCS of low-scattering carriers; Step 3: Obtain the sample set needed to train the deep learning solver RCSAI based on the raw data; Step 4: Train the deep learning solver RCSAI to solve the RCS of the low-scattering carrier with the surface equivalent current as the output target; Step 5: Based on RCSAI, quickly predict the corresponding RCS data for low-scattering carriers.

2. The method for predicting the RCS of a low-scattering carrier based on deep learning according to claim 1, characterized in that: In step 1, the steps for constructing the deep learning solver RCSAI for solving the low-scattering carrier RCS are as follows: Step 1.1: Determine the input and output of the solver RCSAI based on the geometric characteristics of the low-scattering carrier to be solved and the training objective; Step 1.2: Based on the input and output of the solver RCSAI, construct a deep learning solver RCSAI for solving the RCS parameters of a low-scattering carrier using the Transformer architecture.

3. The method for predicting the RCS of a low-scattering carrier based on deep learning according to claim 2, characterized in that: Step 2 specifically includes the following steps: Step 2.1: Based on the geometric modeling tool of the electromagnetic simulation platform, construct the geometric model of the low-scattering carrier for the target scene; Step 2.2: Generate the mesh of the corresponding geometric model using the mesh generation tool of the electromagnetic simulation platform; Step 2.3: Solve the algorithm based on the electromagnetic simulation platform to obtain the equivalent current J on the surface of the low-scattering carrier, and then calculate the far-field scattering field E. s Its expression is: E s =-jωμN t in: The complex unit is represented by ω = 2πf, the angular frequency is represented by μ, and the permeability is represented by N. t Indicate that N is relative to the following Lateral component of direction: Where: e represents the natural constant, ∈ represents the dielectric constant, r represents the far field, and ∫·dS ′ This indicates integration over the entire outer surface of the geometry; Based on the far-field scattering field E s and the given incident field E i The RCS parameter σ of the corresponding low-scattering carrier can be calculated: Where: θ i ,φ i θ and φ represent the incident elevation angle, incident azimuth angle, far-field scattering elevation angle, and far-field scattering azimuth angle, respectively; the RCS parameter reflects the scattering characteristics of the low-scattering carrier.

4. The method for predicting the RCS of a low-scattering carrier based on deep learning according to claim 3, characterized in that: In step 3, the geometric parameter-J data pair is obtained according to step 2. For the geometric parameters in the geometric parameter-J data pair, the geometric parameters are converted into mesh data according to the mesh generation tool of the electromagnetic simulation platform, thereby converting the geometric parameter-J data pair into parameter-mesh-J data. 80% of the parameter-mesh-J data is randomly selected as the training set, 10% as the test set, and 10% as the validation set.

5. The method for predicting the RCS of a low-scattering carrier based on deep learning according to claim 4, characterized in that: Step 4 specifically includes the following steps: Step 4.1: Based on the Parameter-Mesh-J data obtained in Step 3, determine the input and output dimensions of the deep learning solver RCSAI for solving the low scattering carrier in Step 1. RCSAI is composed of a first multilayer perceptron layer, multiple self-attention layers, and a second multilayer perceptron layer. The first multilayer perceptron layer is responsible for dimensionality reduction and data fusion of Parameter-Mesh data, and the second multilayer perceptron layer is responsible for outputting the surface equivalent current data. Step 4.2: The geometric parameters and mesh data input to RCSAI are EM-Parm and Mesh, respectively. The output of RCSAI is the surface equivalent current J. Pred Its definition is: J Pred =RCSAI(EM-Parm,Mesh) The loss function RCSLoss used to train RCSAI is defined as follows: Where |·| represents the magnitude of the complex vector, and N is the number of grid points. J represents the output of the i-th grid corresponding to RCSAI. i Represents the actual surface equivalent current on the i-th grid, and RCSLoss characterizes the average value of the surface equivalent current error across all grids; Step 4.3: Perform backpropagation based on the Adams algorithm to update and iterate the learnable parameters in the MLP, Self-Attention, and MLP layers of RCSAI; Step 4.4: When the error on the training set is less than a given threshold, or the maximum number of iterations is reached, the training process of RCSAI ends.

6. The method for predicting the RCS of a low-scattering carrier based on deep learning according to claim 5, characterized in that: In step 5, a geometric model of the low-scattering carrier is constructed based on the geometric modeling tool of the electromagnetic simulation platform; a mesh corresponding to the geometric model is generated based on the mesh generation tool of the electromagnetic simulation platform; geometric parameters and electromagnetic parameters are extracted to obtain Parameter-Mesh-J data pairs, which are then used as input parameters to the deep learning solver RCSAI; RCSAI quickly infers the equivalent surface current J of the low-scattering carrier based on the input Parameter-Mesh data pairs; the RCS result of the low-scattering carrier is quickly predicted using the far-field scattering expression; and the low-scattering carrier is rapidly optimized using the obtained RCS result.

Images