Inverse Design Method of Electromagnetic Protection Structure Based on Tri-Periodic Minimal Surface Unit Cell Array

Abstract

The present invention discloses an inverse design method for an electromagnetic protection structure based on a triply periodic minimal surface unit cell array. The method includes: establishing a training data set including a triply periodic minimal surface unit cell structure and its structural performance; preprocessing the training data set; establishing a conditional adversarial generation network, including a generator model G and a discriminator model D; training the conditional adversarial generation network using the training data set as training samples to obtain a trained conditional adversarial generation network; inputting the performance parameter matrix of the electromagnetic protection structure of the triply periodic minimal surface unit cell array to be inversely designed into the trained conditional adversarial generation network to obtain a plurality of triply periodic minimal surface unit cell structures that meet the requirements. This inverse design method can directly obtain triply periodic minimal surface unit cell structures that meet the requirements through the required structural performance parameter matrix, reducing the relevant professional knowledge required by designers and greatly improving the design efficiency.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic protection structure design, and particularly to an inverse design method for an electromagnetic protection structure based on a triple-period minimal surface unit cell array. Background Art

[0002] With the development of wireless power transfer technology, industries such as electric vehicles and electric vehicle charging piles have paid increasing attention to wireless charging technology. Compared with wired charging, wireless charging does not have problems such as easy damage to the plug connection part and easy aging of the line, and has advantages such as being less susceptible to external environmental interference and having strong applicability and flexibility. It is gradually being applied to various fields of production and life. During wireless charging, a part of the magnetic leakage diffuses to the surrounding non-working areas and cannot be effectively magnetically coupled. Especially for the wireless charging system of electric vehicles with a relatively high charging power level, excessive magnetic leakage will not only cause electromagnetic interference to the equipment, but also directly threaten the human safety around the vehicle.

[0003] The existing technology weakens the magnetic leakage by equipping the wireless charging device with electromagnetic shielding technical measures. After adding an aluminum plate shield on the basis of ferrite, although the diffusion of magnetic flux in the direction of the back of the coil is further weakened, a part of the magnetic flux generates an eddy current effect when passing through the aluminum plate, thereby converting the magnetic field energy into heat loss and dissipating it. Due to the counteraction between the reverse magnetic flux generated by the eddy current and the main magnetic flux, the transmission efficiency is reduced. At the same time, the traditional aluminum plate structure has poor heat dissipation ability, and the shielding effectiveness of the magnetic shielding material will decrease with the increase in temperature. And because the wireless charging system is arranged on the chassis of the electric vehicle, the structural arrangement is more complicated. Therefore, problems such as the protection of the wireless charging system and the integration of the vehicle body also need to be solved urgently.

[0004] Minimal surfaces have important application values in geometry and physics. They are surfaces with zero curvature and are widely used in fields such as materials science, architectural design, and biomedical engineering. In particular, the triple-period minimal surface is widely used in the development of new functional materials and structures due to its unique geometric properties and excellent mechanical properties.

[0005] Currently, the design of the triple-period minimal surface mainly relies on numerical calculation and geometric modeling methods. Although these methods can generate high-quality minimal surfaces, their design processes are usually complex, time-consuming, and require high professional knowledge of the designer, which limits the practical application of the triple-period minimal surface in electromagnetic shielding structures.

[0006] With the rapid development of artificial intelligence-related technologies in recent years, one of the disciplines of artificial intelligence, namely deep learning technology, has achieved unprecedented breakthroughs. Deep learning algorithms infer the corresponding relationship between input and output through neural networks, especially having a very good fitting effect on mapping relationships without accurate functional or mathematical corresponding relationships. The conditional generative adversarial network can generate target data with specific characteristics by introducing conditional constraints, which provides a new idea for the generation of complex surfaces. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an inverse design method for an electromagnetic protection structure based on a triply periodic minimal surface unit cell array aiming at the defects involved in the background technology.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] An inverse design method for an electromagnetic protection structure based on a triply periodic minimal surface unit cell array includes the following steps:

[0010] Step 1), establish a training data set including the triply periodic minimal surface unit cell structure and its structural performance;

[0011] The implicit function of the triply periodic minimal surface unit cell structure is:

[0012] ;

[0013] In the formula, ;

[0014] ;

[0015] ;

[0016] In the formula, , , are the level set values of the minimal surfaces , , respectively; , , are the preset weights of the minimal surfaces , , respectively, , ;

[0017] The structural performance of the triply periodic minimal surface unit cell structure includes electromagnetic shielding performance, mechanical performance, and heat dissipation performance;

[0018] The specific steps are as follows:

[0019] Step 1.1), randomly generate N groups , , ; and generate N groups , , through the Latin hypercube sampling strategy, evenly covering the level set space;

[0020] Step 1.2), randomly pair the N groups , , and the N groups , , to form k three-period minimal surface unit structures;

[0021] Step 1.3), use a multi-physics simulation software to simulate the k three-period minimal surface unit structures, and obtain the electromagnetic shielding effectiveness curves, stress-strain curves, and temperature-time curves of the k three-period minimal surface unit structures;

[0022] Step 1.4), take the shape parameter vector of each three-period minimal surface unit structure and the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of this three-period minimal surface unit structure as a set of data, form k sets of data, and use these k sets of data as the training data set;

[0023] Step 2), preprocess each set of data in the training data set. The preprocessing steps for the i-th set of data are as follows:

[0024] Step 2.1), name the shape parameter vector in the i-th set of data as the shape vector ;

[0025] Step 2.2), sequentially select 20 points on the electromagnetic shielding effectiveness curve of the i-th set of data according to a preset first step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0026] Step 2.3), sequentially select 20 points on the stress-strain curve of the i-th set of data according to a preset second step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0027] Step 2.4), sequentially select 20 points on the temperature-time curve of the i-th set of data according to a preset third step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0028] Step 2.5), the column vector , , Concatenate to form the performance matrix of the i-th group of data ;

[0029] Step 3), establish a conditional adversarial generation network, which includes a generator model G and a discriminator model D;

[0030] The generator model G includes 6 fully connected layers connected in sequence. The first fully connected layer serves as the input layer, and the sixth fully connected layer serves as the output layer. The ReLU activation function is used for the first to fifth fully connected layers, and the Softmax function is used for the output layer; the input parameter of the input layer is the shape vector of the data in the training dataset , performance matrix , and the output of the output layer is the reversely generated shape parameter vector ;

[0031] The discriminator model D includes 6 fully connected layers connected in sequence. The first fully connected layer serves as the input layer, and the sixth fully connected layer serves as the output layer. The ReLU activation function is used for the first to fifth fully connected layers, and the Sigmoid function is used for the output layer; the input parameters of the input layer are the shape vector of the data in the training dataset , performance matrix and its corresponding reversely generated shape parameter vector , and the output is the determination result of true or false;

[0032] The loss function of the conditional adversarial generation network is:

[0033] ;

[0034] Among them, represents the expectation, D is the discriminator model D, G is the generator model, P data is the data distribution of the training dataset. x is sampled from all the shape vectors in the training dataset, y is sampled from all the performance matrices in the training dataset, and z is sampled from the noise distribution P z of the added Gaussian noise;

[0035] Step 4), use the training dataset as training samples to train the conditional adversarial generation network to obtain the trained conditional adversarial generation network;

[0036] Step 5), obtain its performance parameter matrix according to the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of the electromagnetic protection structure of the three-period minimal surface unit cell array that needs inverse design ;

[0037] Step 5.1), select 20 points in sequence on the electromagnetic shielding effectiveness curve of the electromagnetic protection structure of the three-period minimal surface unit cell array to be inversely designed according to a preset first step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0038] Step 5.2), select 20 points in sequence on the stress-strain curve of the electromagnetic protection structure of the three-period minimal surface unit cell array to be inversely designed according to a preset second step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0039] Step 5.3), select 20 points in sequence on the temperature-time curve of the electromagnetic protection structure of the three-period minimal surface unit cell array to be inversely designed according to a preset third step length threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0040] Step 5.4), splice the single-column vectors , , into a performance matrix ;

[0041] Step 6), input the performance parameter matrix of the electromagnetic protection structure of the three-period minimal surface unit cell array to be inversely designed into the trained conditional adversarial generation network to obtain the required three-period minimal surface unit structure.

[0042] The innovation of the present invention is that:

[0043] 1. Based on the conditional adversarial generation network architecture and multi-physics field coupling simulation, the present invention takes the electromagnetic shielding effectiveness curve, stress-strain curve and temperature-time curve as conditional inputs, so that the generated three-period minimal surface unit structure has electromagnetic shielding performance, mechanical properties and heat dissipation performance that meet the requirements. Through multi-physics field coupling, electromagnetic shielding, mechanical strength and heat dissipation effect can be comprehensively optimized in the design stage. In this way, it is ensured that the generated structure has comprehensive performance advantages in practical applications, and the overall reliability and safety of the design are improved; adjust the weights of electromagnetic shielding performance, mechanical properties and heat dissipation performance in training according to user needs, and find the best balance point in the design to meet the special requirements of different application scenarios.

[0044] 2. The present invention uses a conditional generative adversarial network to generate triply periodic minimal surfaces, reducing the complexity and time cost of manual design; moreover, users do not need to have profound mathematical and modeling knowledge, and only need to provide design requirements to generate triply periodic minimal surfaces that meet the requirements. By adjusting the conditional input, diverse triply periodic minimal surface unit structures can be generated to meet the needs of different application scenarios.

[0045] 3. The present invention adds random Gaussian noise to the generator model G and the discriminator model D, enabling the network to generate multiple sets of outputs of different triply periodic minimal surface unit structures under the input of the same electromagnetic shielding performance, mechanical performance, and heat dissipation performance, providing designers with more choices when different structural requirements are needed. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is the overall flowchart of the method described in the present invention;

[0047] Figure 2 is the training flowchart of the conditional generative adversarial network described in the present invention;

[0048] Figure 3 is the structural diagram of the generator model G in the conditional generative adversarial network described in the present invention;

[0049] Figure 4 is the structural diagram of the discriminator model D in the conditional generative adversarial network described in the present invention;

[0050] Figure 5 is a schematic diagram of the application of the triply periodic minimal surface structure inversely designed by the present invention to wireless charging of electric vehicles.

[0051] In the figure, 1 - battery pack, 2 - shielding layer, 3 - ferrite layer, 4 - receiving coil, 5 - transmitting coil. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The following will provide an overall description of the inverse design method of the electromagnetic protection structure based on a triply periodic minimal surface unit array proposed in the present invention in combination with the drawings of the present invention. Taking the shielding layer of an electric vehicle wireless charging device as an example. During the implementation of the embodiment, the overall requirements for the embodiment are as follows:

[0053] 1. The proposed inverse design method of the electromagnetic protection structure based on a triply periodic minimal surface unit array can generate corresponding triply periodic minimal surface unit structures according to the electromagnetic shielding performance, mechanical performance, and heat dissipation performance required by designers;

[0054] 2. For the same requirements of electromagnetic shielding performance, mechanical performance, and heat dissipation performance, multiple different triply periodic minimal surface unit structures can be obtained for designers to choose from, and unit structures with too high processing difficulty are excluded.

[0055] In response to the above requirements, the present invention proposes an inverse design method for an electromagnetic protection structure based on a triple-period minimal surface unit cell array. The specific process of the method is as follows Figure 1 shown.

[0056] Step 1), establish a training data set including the triple-period minimal surface unit cell structure and its structural performance;

[0057] The implicit function of the triple-period minimal surface unit cell structure is:

[0058] ;

[0059] In the formula, ;

[0060] ;

[0061] ;

[0062] In the formula, , , are the level set values of the minimal surfaces , , respectively; , , are the preset weights of the minimal surfaces , , respectively, , ;

[0063] The structural performance of the triple-period minimal surface unit cell structure includes electromagnetic shielding performance, mechanical performance, and heat dissipation performance;

[0064] The specific steps are as follows:

[0065] Step 1.1), randomly generate N groups of , , ; and generate N groups of , , through the Latin hypercube sampling strategy, uniformly covering the level set space;

[0066] Step 1.2), randomly pair the N groups of , , and the N groups of , , to form k triple-period minimal surface unit cell structures;

[0067] Step 1.3), use a multi-physics simulation software to simulate k triply periodic minimal surface unit structures, and obtain the electromagnetic shielding effectiveness curves, stress-strain curves, and temperature-time curves of the k triply periodic minimal surface unit structures;

[0068] Step 1.4), take the shape parameter vector of each triply periodic minimal surface unit structure and the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of this triply periodic minimal surface unit structure as a set of data, form k sets of data, and use these k sets of data as the training data set;

[0069] Step 2), preprocess each set of data in the training data set, and the preprocessing steps for the i-th set of data are as follows:

[0070] Step 2.1), name the shape parameter vector in the i-th set of data as the shape vector ;

[0071] Step 2.2), sequentially select 20 points on the electromagnetic shielding effectiveness curve of the i-th set of data according to a preset first step size threshold, and represent the ordinates of these 20 points as a single-column vector

[0072] Step 2.3), sequentially select 20 points on the stress-strain curve of the i-th set of data according to a preset second step size threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0073] Step 2.4), sequentially select 20 points on the temperature-time curve of the i-th set of data according to a preset third step size threshold, and represent the ordinates of these 20 points as a single-column vector ;

[0074] Step 2.5), splice the column vectors , , to form the performance matrix of the i-th set of data;

[0075] Step 3), establish a conditional adversarial generation network, the conditional adversarial generation network includes a generator model G and a discriminator model D, and the overall training process of the network is as Figure 2 shown.

[0076] The structural diagram of the generator model G is as Figure 3As shown in the figure, in the figure, FC is the fully connected layer, LR is the ReLU function, SM is the Softmax function. The generator model G contains 6 sequentially connected fully connected layers. The first fully connected layer serves as the input layer, and the sixth fully connected layer serves as the output layer. The first to fifth fully connected layers all use the ReLU activation function, and the output layer uses the Softmax function. The input parameter of the input layer is the shape vector of the data in the training dataset , performance matrix , and the output of the output layer is the reversely generated shape parameter vector . Specifically, adding Gaussian noise to the input can solve the problem that ordinary neural networks cannot generate a three-period minimal surface unit structure one-to-many

[0077] The structural diagram of the discriminator model D is as shown in Figure 4 In the figure, FC is the fully connected layer, LR is the ReLU function, S is the Sigmoid function. The discriminator model D contains 6 sequentially connected fully connected layers. The first fully connected layer serves as the input layer, and the sixth fully connected layer serves as the output layer. The first to fifth fully connected layers all use the ReLU activation function, and the output layer uses the Sigmoid function. The input parameters of the input layer are the shape vector of the data in the training dataset , performance matrix and its corresponding reversely generated shape parameter vector , and the output is the determination result of the authenticity of ;

[0078] The loss function of the conditional adversarial generation network is as follows:

[0079] ;

[0080] Among them, denotes expectation, D is the discriminator model D, G is the generator model, P data is the data distribution of the training dataset. x is sampled from all the shape vectors in the training dataset, y is sampled from all the performance matrices in the training dataset, and z is sampled from the noise distribution P z of the added Gaussian noise;

[0081] Step 4), using the training dataset as the training sample to train the conditional adversarial generation network to obtain the trained conditional adversarial generation network;

[0082] Step 5), obtaining its performance parameter matrix according to the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of the electromagnetic protection structure of the three-period minimal surface unit cell array to be inversely designed ;

[0083] Step 5.1), select 20 points in turn on the electromagnetic shielding effectiveness curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset first step length threshold, and express the ordinates of these 20 points as a single column vector ;

[0084] Step 5.2), select 20 points on the stress-strain curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset second step threshold, and express the ordinates of these 20 points as a single column vector ;

[0085] Step 5.3), select 20 points on the temperature-time curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset third step threshold, and express the ordinates of these 20 points as a single column vector ;

[0086] Step 5.4), convert the single column vector , , Stitching into a performance matrix ;

[0087] Step 6), the performance parameter matrix of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed Input into the trained conditional adversarial generative network to obtain the required three-periodic minimal surface unit structure.

[0088] The three-periodic minimal surface unit cells that meet the requirements are arrayed and obtained as follows Figure 5 The shielding layer of the electric vehicle wireless charging device is shown.

[0089] In summary, the inverse design method of the electromagnetic protection structure based on the three-periodic minimal surface unit cell array described in the embodiment of the present invention eliminates the complicated modeling process and improves the design efficiency. Moreover, by changing the input performance parameters, a structure that meets the user's needs can be generated, so that it can be promoted and applied in a wider range of engineering fields; and the three-periodic minimal surface unit cell inversely designed based on the conditional adversarial generation network has good electromagnetic shielding performance, mechanical performance and heat dissipation performance. The electromagnetic shielding layer formed by the unit cell array is between the battery pack and the ferrite, and the cooling medium can flow between the unit cells. The shielding layer can not only shield the electromagnetic leakage between the battery pack and the coil, but also play a role in cooling both. At the same time, the excellent mechanical properties presented by the shielding layer can also provide protection for the battery pack and the wireless charging device.

Claims

1. An inverse design method for electromagnetic protection structure based on a three-periodic minimal surface unit cell array, characterized in that: The following steps are involved: Step 1), establish a training data set including three-periodic minimal surface unit structures and their structural properties; The implicit function of the three-periodic minimal surface unit structure for: ； In the formula, ; ； ； In the formula, , , Minimal Surfaces , , The level set value of , , Minimal Surfaces , , Preset weights, , ; The structural properties of the three-periodic minimal surface unit structure include electromagnetic shielding performance, mechanical properties, and heat dissipation performance; The specific steps are as follows: Step 1.1), randomly generate N groups , , ; And generate N groups through Latin hypercube sampling strategy , , , uniformly covers the level set space; Step 1.2), for N groups , , and N groups , , Random pairing forms k three-periodic minimal surface unit structures; Step 1.3), using multi-physics field simulation software to simulate the k three-period minimal surface unit structures, and obtaining the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of the k three-period minimal surface unit structures; Step 1.4), the shape parameter vector of each three-periodic minimal surface unit structure The electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of the three-period minimal surface unit structure are taken as a set of data to form k sets of data, and the k sets of data are taken as training data sets; Step 2) Preprocess each set of data in the training data set, where the preprocessing steps for the i-th set of data are as follows: Step 2.1), the shape parameter vector in the i-th group of data Named Shape Vector ; Step 2.2), select 20 points on the electromagnetic shielding effectiveness curve of the i-th group of data according to the preset first step length threshold, and express the ordinates of these 20 points as a single column vector ; Step 2.3), select 20 points on the stress-strain curve of the i-th group of data according to the preset second step threshold, and express the ordinates of these 20 points as a single column vector ; Step 2.4), select 20 points on the temperature-time curve of the i-th group of data according to the preset third step threshold, and express the ordinates of these 20 points as a single column vector ; Step 2.5), the column vector , , The performance matrix of the i-th group of data is spliced ​​into ; Step 3), establish a conditional adversarial generation network, wherein the conditional adversarial generation network includes a generator model G and a discriminator model D; The generator model G contains 6 fully connected layers connected in sequence. The first fully connected layer is used as the input layer, and the sixth fully connected layer is used as the output layer. The first to fifth fully connected layers use the ReLU activation function, and the output layer uses the Softmax function. The input parameter of the input layer is the shape vector of the data in the training dataset. , Performance Matrix The output of the output layer is the shape parameter vector generated inversely ; The discriminator model D contains 6 fully connected layers connected in sequence. The first fully connected layer is used as the input layer, and the sixth fully connected layer is used as the output layer. The first to fifth fully connected layers use the ReLU activation function, and the output layer uses the Sigmoid function. The input parameter of the input layer is the shape vector of the data in the training data set. , Performance Matrix and its corresponding inversely generated shape parameter vector , the output is a pair True or false determination result; Loss function for conditional adversarial generation networks for: ； in, represents the expectation, D is the discriminator model D, G is the generator model, P data is the data distribution of the training dataset, x is sampled from all shape vectors in the training dataset, y is sampled from all performance matrices in the training dataset, and z is sampled from the noise distribution P of the added Gaussian noise z Medium sampling; Step 4), using the training data set as a training sample to train the conditional adversarial generation network to obtain a trained conditional adversarial generation network; Step 5), the electromagnetic shielding effectiveness curve, stress-strain curve, and temperature-time curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that is inversely designed as needed are used to obtain its performance parameter matrix ; Step 5.1), select 20 points in turn on the electromagnetic shielding effectiveness curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset first step length threshold, and express the ordinates of these 20 points as a single column vector ; Step 5.2), select 20 points on the stress-strain curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset second step threshold, and express the ordinates of these 20 points as a single column vector ; Step 5.3), select 20 points in turn on the temperature-time curve of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed according to the preset third step threshold, and express the ordinates of these 20 points as a single column vector ; Step 5.4), convert the single column vector , , Stitching into a performance matrix ; Step 6), the performance parameter matrix of the electromagnetic protection structure of the three-periodic minimal surface unit cell array that needs to be inversely designed Input into the trained conditional adversarial generative network to obtain the required three-periodic minimal surface unit structure.

Images