Multi-parameter collaborative design method for wave-absorbing superstructure based on RGB three-channel coding

Abstract

The invention discloses a wave-absorbing superstructure multi-parameter collaborative design method based on RGB three-channel coding, and belongs to the field of electromagnetism. Through collaborative design of material-size-patterned metasurface and fusion of multiple parameters through RGB coding, a design space is expanded from a single dimension to a multi-dimensional space, more potential high-performance wave-absorbing superstructure design schemes can be excavated, and abundant possibilities are provided for innovation of wave-absorbing materials. After the forward prediction model is trained and optimized by a data set sample, the prediction accuracy of the wave-absorbing reflectivity of the wave-absorbing superstructure reaches 95% or above, and a reliable basis is provided for reverse design. Under the assistance of transfer learning, the reverse design model enables the generator model to converge quickly and generates wave-absorbing superstructure design highly matched with the target reflectivity, and efficient design of the multi-parameter wave-absorbing superstructure is achieved. The RGB image coding mode and the reverse design method constructed by the invention are applied to optimization of different wave-absorbing superstructure configuration scenes, and multi-parameter collaborative rapid optimization of the wave-absorbing superstructure is realized.

Description

Technical Field

[0001] The invention relates to a multi-parameter collaborative design method of a wave-absorbing superstructure based on RGB three-channel coding, and belongs to the field of electromagnetics. Background Art

[0002] Stealth performance, as the core capability of the new generation of military equipment, is particularly critical in countering the networking upgrade of modern early warning detection and interception systems. In particular, high-speed aircraft and power systems are limited by aerodynamic shape and functional requirements, and traditional shape stealth technology is difficult to break through. This makes absorbing material technology an important breakthrough in improving radar stealth performance.

[0003] Current absorbing material design faces significant challenges. Traditional methods rely on empirical formulas and repeated experiments, making it difficult to efficiently optimize complex structures and material combinations, hindering the development of high-performance absorbing materials. Furthermore, while absorbing metastructures offer advantages such as broadband absorption and lightweight design, they lack the ability to coordinate the design of their structural dimensions with the metasurface pattern configuration. This leads to unclear multi-physical parameter coupling mechanisms, severely restricting design freedom. With the rise of artificial intelligence (AI), deep learning has shown great potential in materials design. However, the reverse design accuracy of existing deep learning-based absorbing metastructure design methods needs to be improved. Summary of the Invention

[0004] In order to solve the problems mentioned above or in the prior art, the purpose of the present invention is to provide a multi-parameter collaborative design method for absorbing superstructures based on RGB three-channel coding. This method integrates the supersurface pattern shape, material properties and structural size parameters of the absorbing superstructure into RGB color images by constructing a coding system. On this basis, the absorbing superstructure is automatically simulated by combining MATLAB software and numerical simulation software CST Studio Suite 2023 to construct a rich data set. A forward prediction model is constructed and fully trained with data set samples to achieve the prediction of the superstructure's absorbing reflectivity. A reverse design model is constructed and fully trained with data set samples to achieve the rapid design of absorbing superstructures for given absorbing reflectivity.

[0005] In order to achieve the above object, the technical solution adopted in the present invention is:

[0006] The multi-parameter collaborative design method of an absorbing superstructure based on RGB three-channel coding disclosed in the present invention includes the following steps:

[0007] Step 1: Construct a coding system; integrate and encode the metasurface pattern shape, material properties, and structural dimension parameters of the absorbing metastructure into an RGB color image; the absorbing metastructure is composed of a plurality of absorbing unit structures arranged in a two-dimensional periodic manner; the structural dimensions include the unit structure periodic dimension and the structure thickness; the material property is the square resistance of the metasurface resistive film;

[0008] The RGB image consists of three color channels: R, G, and B, each with a size of 32×32. The R channel encodes and maps the square resistance value of the metasurface resistive film to pixel values; the G channel encodes and maps the unit structure period size and structure thickness to pixel values; and the B channel converts the pattern topology into an adjustable digital outline through binary encoding. The RGB image formed by the fusion of the three channels realizes the pixel-level feature correlation between the metasurface pattern shape, material properties, and structural size parameters of the absorbing metastructure.

[0009] MATLAB and CST Studio Suite 2023, a numerical simulation software, were used to automatically simulate the absorbing superstructures and calculate the absorbing reflectivity. The RGB image corresponding to each absorbing superstructure and the calculated absorbing reflectivity form a data pair, and these data pairs form the raw data set. The data in the raw data set were normalized and preprocessed to ensure that the data were of the same magnitude.

[0010] Step 2: Establish and fully train a forward prediction model. This model is based on a convolutional neural network, employing a multi-layer architecture consisting of convolutional layers, pooling layers, fully connected layers, and activation function layers. The input to this forward prediction model is the RGB image from the dataset obtained in Step 1, and the output is the absorbance reflectivity. The convolutional layers capture the image's feature information; the pooling layers compress and reduce the dimensionality of this feature information, improving computational efficiency; the fully connected layers map the feature information to the absorbance reflectivity; and the activation function layer uses the ReLU function to enhance the model's nonlinear expression capabilities.

[0011] The forward prediction model is fully trained using the dataset obtained in step 1, achieving a mean absolute error of less than 0.01 in the absorption reflectivity predicted from the input RGB image.

[0012] Step 3: Establish and fully train the reverse design model. This model is constructed based on a generative adversarial network (GAN). The reverse design model consists of a generator, a discriminator, and the forward prediction model from step 2. The generator takes a random noise vector and the absorber reflectivity from the dataset obtained in step 1 as input, and gradually generates an image through deconvolution layers. The discriminator distinguishes between the generator's image and the real image. The forward prediction model guides the generator's image to accurately reflect the target absorber reflectivity. The generator and discriminator are fully trained using the dataset obtained in step 1 to improve the generator's generation quality. After training, the generator generates an RGB image based on the input target reflectivity.

[0013] Step 4: Input the target reflectivity into the generator trained in step 3 to obtain an RGB image. Reversely decode the image according to the combination method of step 1 to obtain the metasurface pattern, the square resistance value of the metasurface resistive film, and the unit structure period size and structure thickness.

[0014] The method further includes step five, obtaining the metasurface pattern, the square resistance value of the metasurface resistive film, the unit structure period size, and the structure thickness according to step four, establishing a corresponding absorbing metastructure, achieving the target broadband absorbing performance, and applying the absorbing metastructure to the fields of electromagnetic radiation, electromagnetic pollution, and military equipment stealth.

[0015] Beneficial effects:

[0016] 1. The disclosed multi-parameter collaborative design method for absorbing metastructures based on RGB three-channel coding overcomes the bottleneck of isolated optimization of each factor in traditional design through the collaborative design of materials, dimensions, and patterned metasurfaces. By integrating multiple parameters through RGB coding, the design space expands from a single dimension to a multi-dimensional space, uncovering more potential high-performance absorbing metastructure design solutions and providing rich possibilities for innovation in absorbing materials.

[0017] 2. This invention discloses a multi-parameter collaborative design method for absorbing superstructures based on RGB three-channel coding. After training and optimization of the forward prediction model using dataset samples, the model achieves over 95% accuracy in predicting the absorbing reflectivity of the absorbing superstructure, providing a reliable basis for reverse design. Assisted by transfer learning, the reverse design model enables rapid convergence of the generator model and generates an absorbing superstructure design that closely matches the target reflectivity, achieving efficient design of multi-parameter absorbing superstructures.

[0018] 3. The multi-parameter collaborative design method of absorbing superstructures based on RGB three-channel coding disclosed in the present invention constructs an RGB image coding method and an inverse design method, which can be applied to the optimization design of different absorbing superstructure configuration scenarios, and realizes the rapid collaborative design of multi-parameters of absorbing superstructures.

[0019] 4. The multi-parameter collaborative design method of the absorbing superstructure based on RGB three-channel coding disclosed in the present invention, on the basis of achieving the above three beneficial effects, the designed absorbing superstructure can be applied to the fields of electromagnetic radiation, electromagnetic pollution, and military equipment stealth. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 1 is a structural model diagram of the wave-absorbing superstructure unit according to Example 1 of the present invention;

[0021] Figure 2 Schematic diagram of an RGB three-channel encoded image according to embodiment 1 of the present invention;

[0022] Figure 3 Schematic diagram of the network structure of the forward prediction model of Example 1 of the present invention;

[0023] Figure 4 is a forward prediction model training loss curve diagram of Example 1 of the present invention;

[0024] Figure 5 1 is a schematic diagram of the reverse design model structure of Example 1 of the present invention;

[0025] Figure 6 1 is a graph showing the training loss of the reverse design model according to Example 1 of the present invention;

[0026] Figure 7 is a flow chart of reverse design result evaluation of Example 1 of the present invention;

[0027] Figure 8 is the input target reflectivity curve of Example 1 of the present invention;

[0028] Figure 9 is an RGB image obtained by inputting the target reflectivity into the generator of the reverse design model in Example 1 of the present invention;

[0029] Figure 10 This is a comparison diagram of the wave absorption reflectivity curves of the structure corresponding to the generated RGB image and the target wave absorption reflectivity curve in Example 1 of the present invention. DETAILED DESCRIPTION

[0030] In order to better illustrate the purpose and advantages of the present invention, the invention is further described below with reference to the accompanying drawings and examples.

[0031] Example 1:

[0032] like Figure 1As shown in the figure, the multi-parameter collaborative design method for an absorbing superstructure based on RGB three-channel coding disclosed in this embodiment uses a absorbing superstructure unit structure, from bottom to top, consisting of: a metal base plate, a first PMI foam substrate layer, a first PET dielectric layer (square resistor film), a second PMI foam substrate layer, a second PET dielectric layer, and a patterned resistor film. The unit structure period is p. The metal base plate is 0.1 mm thick and serves as a total reflection layer. The first PMI foam substrate layer has a relative dielectric constant ε = 1.1 and a thickness of h. The first PET dielectric layer has a thickness of 0.125 mm and a dielectric constant ε = 3.4. The square resistor film has a square resistance of R1. The second PMI foam substrate layer has a relative dielectric constant ε = 1.1 and a thickness of h. The second PET dielectric layer has a thickness of 0.125 mm. The square resistor film has a square resistance of R2. The pattern shapes are rich and diverse coded images obtained through pattern transformation operations using crosses, square rings, and squares as basic patterns.

[0033] Step 1: Design a coding system; Figure 2 As shown, the metasurface pattern shape, material properties, and structural size parameters of the absorbing superstructure in the absorbing superstructure are integrated and encoded into an RGB color image; the RGB image consists of three color channels: R, G, and B, and the size of each channel is 32×32.

[0034] The R channel encodes and maps the square resistance value of the super-surface resistor film to pixel values; the square resistance value R1 of the lower square resistor film ranges from 80Ω / sq to 180Ω / sq, and the square resistance value R2 of the upper patterned resistor film ranges from 60Ω / sq to 160Ω / sq. The mapping is performed using the following formula:

[0035]

[0036] Where R'1 and R'2 are the pixel values after mapping.

[0037] The G channel maps the unit structure period size and structure thickness code to pixel values; the unit structure period size p ranges from 8mm to 30mm, and the thickness h of the PMI foam substrate ranges from 3mm to 7.5mm. The mapping is performed using the following formula:

[0038]

[0039] Where p' and h' are the pixel values after mapping.

[0040] The B channel encodes the metasurface pattern of the topmost patterned resistive film. Figure 2As shown in the diagram of channel B, the white pixel value is 255, representing the metasurface pattern and the resistive film material; the black pixel value is 0. Based on three basic patterns: cross, square, and square ring, a rich variety of coded patterns are generated through image transformation methods such as noise enhancement and reduction, and rotation. MATLAB software and CST Studio Suite 2023 are used to automatically simulate the absorbing metastructures, and the absorbing reflectivity is calculated. The RGB image corresponding to each absorbing metastructure and the calculated absorbing reflectivity form a data pair. These data pairs constitute the original dataset, which is divided into training and test sets at an 8:2 ratio.

[0041] Step 2: Establish a forward prediction model and train it fully; Figure 3 As shown in the figure, a forward prediction model based on a convolutional neural network is constructed, which adopts a multi-layer convolutional layer, pooling layer, fully connected layer and activation function layer combination architecture; the specific implementation of the network model structure is as follows: the input layer receives a 32×32×3 dimensional RGB encoded image, which corresponds to the multi-parameter fusion feature of the absorbing superstructure. There are 6 convolutional layers, and the convolution kernel size is 3×3. The maximum pooling layer MaxPooling is inserted after the second, fourth and sixth convolutional layers, with a pooling window of 2×2 and a step size of 2 to compress the feature map size. The feature map is expanded into a 2048-dimensional vector by the Flatten layer, and then passes through the first fully connected layer with 1024 neurons and ReLU activation; the second fully connected layer with 512 neurons and ReLU activation; the output layer has 101 neurons, corresponding to the reflectivity prediction value of 0.16GHz interval in the 2-18GHz frequency band. The model training uses 8000 training sets and 2000 test sets, with the Adam optimizer (initial learning rate 1×10 -4 , decaying by 10% every 50 rounds) minimizes the mean absolute error MAE loss function, sets the training batch size to 32 and the training rounds to 600. After training, the model loss curve is as follows Figure 4 As shown, the model shows stable convergence: the MAE of the training set gradually decreases from an initial 0.07 to 0.001. After training, the model is tested, and the test set performance shows an average MAE of 0.0166, and the overall prediction accuracy meets the engineering requirements.

[0042] Step 3: Establish a reverse design model and train it fully; Figure 5As shown in the figure, a reverse design model is constructed based on the generative adversarial network (GAN). The reverse design model consists of a generator, a discriminator, and the forward prediction model of step 2. The generator takes a 100-dimensional random noise vector and a target reflectivity curve as input, and then gradually upsamples through 4 layers of deconvolution, passes through a batch normalization layer and an activation layer, and finally outputs a 32×32×3 RGB image with Tanh activation. The discriminator inputs a 32×32×3 RGB image, which is processed through 4 layers of convolution, batch normalization, and then a LeakyReLU activation function layer, and finally outputs the discrimination probability. The forward prediction model freezes the weights and is embedded in the training process to predict the reflectivity of the generated image. The L1 loss of the predicted reflectivity of the generated image and the target reflectivity is calculated, and the generator is optimized together with the adversarial loss. The model training reuses the 8,000 training sets of the forward model. The target reflectivity curve is used as the input condition of the generator. The generator and the discriminator both use the Adam optimizer, with an initial learning rate of 2×10 -4 With 1×10 -4 , the training rounds are 600, the batch size is 32, and the learning rate decays to 50% of the original value every 100 rounds. During the training process, Figure 6 As shown in Figure 2, the loss curves of the generator and the discriminator show typical adversarial oscillations and then gradually converge. The performance of the generator reverse design is then evaluated. The evaluation process is as follows: Figure 7 As shown, the target curve is input into the generator, which generates an image. The generated image is decoded and then subjected to finite element simulation. The simulation results are compared with the target curve, and the error is calculated. The test set performance shows that the generated images, verified by simulation using CST Studio Suite 2023 software, have an average MAE of 0.06 for the reflectivity curve, demonstrating the ability to accurately generate the absorbing superstructure corresponding to the target curve.

[0043] Step 4: Figure 8 The target reflectivity shown is input into the generator of the reverse design model trained in step 3 to obtain Figure 9 The RGB image shown is reverse decoded according to the combination method of step 1, where the square resistance value of the metasurface resistive film, the unit structure period size, and the structure thickness are decoded by taking the average value of the pixels. The specific values of the metasurface pattern, the square resistance value of the metasurface resistive film, the unit structure period size, and the structure thickness are as follows: p = 26.8mm, h = 11mm, R1 = 100Ω / sq, R2 = 100Ω / sq. A wave-absorbing superstructure model is established and simulated. The obtained wave absorption reflectivity curve is compared with the target wave absorption reflectivity curve. Figure 10 ,The inverse design model generates an absorbing superstructure design that is highly consistent with the target reflectivity, enabling the efficient design of multi-parameter absorbing superstructures.

[0044] Step 5: According to step 4, the metasurface pattern, the square resistance value of the metasurface resistive film, the unit structure period size and the structure thickness are obtained, and the corresponding absorbing metastructure is established to achieve the target broadband absorbing performance, so that the absorbing metastructure can be applied to the fields of electromagnetic radiation, electromagnetic pollution, and military equipment stealth.

[0045] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-parameter collaborative design method for absorbing superstructures based on RGB three-channel coding, characterized by: The following steps are included: Step 1: Construct a coding system; integrate and encode the metasurface pattern shape, material properties, and structural dimension parameters of the absorbing metastructure into an RGB color image; the absorbing metastructure is composed of a plurality of absorbing unit structures arranged in a two-dimensional periodic manner; the structural dimensions include the unit structure periodic dimension and the structure thickness; the material property is the square resistance of the metasurface resistive film; Step 2: Establish a forward prediction model and fully train it; construct a forward prediction model based on a convolutional neural network, using a multi-layer convolutional layer, pooling layer, fully connected layer, and activation function layer combined architecture; the input of the forward prediction model is the RGB image in the data set obtained in step 1, and the output is the absorption reflectivity; the convolution layer is used to capture the feature information in the image; the pooling layer is used to compress the feature information and reduce the dimension and improve the computational efficiency; the fully connected layer maps the feature information to the absorption reflectivity; the activation function layer uses the ReLU function to enhance the nonlinear expression ability of the model; Step 3: Build a reverse design model and fully train it. This model is constructed based on a generative adversarial network (GAN). The reverse design model consists of a generator, a discriminator, and the forward prediction model from step 2. The generator takes a random noise vector and the absorbance reflectivity from the dataset obtained in step 1 as input and gradually generates an image through a deconvolution layer. The discriminator is responsible for distinguishing the images generated by the generator from the real images. The forward prediction model guides the generator to generate images that accurately reflect the target's absorbance reflectivity. The generator and discriminator are fully trained using the dataset obtained in step 1 to improve the generator's generation quality. After training, the generator generates an RGB image based on the input target reflectivity. Step 4: Input the target reflectivity into the generator trained in step 3 to obtain an RGB image. Reversely decode the image according to the combination method of step 1 to obtain the metasurface pattern, the square resistance value of the metasurface resistive film, and the unit structure period size and structure thickness.

2. The multi-parameter collaborative design method for an absorbing superstructure based on RGB three-channel coding according to claim 1, characterized in that: In step one, The RGB image is composed of three color channels: R, G, and B. The R channel encodes and maps the square resistance value of the metasurface resistive film to pixel values; the G channel encodes and maps the unit structure period size and structure thickness to pixel values; and the B channel converts the pattern topology into an adjustable digital outline through binary encoding. The RGB image formed by the fusion of the three channels R, G, and B realizes the pixel-level feature association between the metasurface pattern shape, material properties, and structural size parameters of the absorbing metastructure in the absorbing metastructure. The absorbing superstructures were automatically simulated using MATLAB software and numerical simulation software, and the absorbing reflectivity was calculated. The RGB image corresponding to each absorbing superstructure and the absorbing reflectivity calculated by simulation formed a pair of data, and several pairs of data formed the original data set. Normalize the data in the original dataset to ensure that the data are at the same level.

3. The multi-parameter collaborative design method for an absorbing superstructure based on RGB three-channel coding according to claim 2, characterized in that: The three color channels R, G, and B have a size of 32×32 for each channel.

4. The multi-parameter collaborative design method for an absorbing superstructure based on RGB three-channel coding according to claim 2, characterized in that: CST Studio Suite 2023 was selected as the numerical simulation software.

5. The multi-parameter collaborative design method for an absorbing superstructure based on RGB three-channel coding according to claim 1, 2, 3 or 4, characterized in that: The method further includes step five, obtaining the metasurface pattern, the square resistance value of the metasurface resistive film, the unit structure period size and the structure thickness according to step four, establishing a corresponding absorbing metastructure, and achieving the target broadband absorbing performance.

Images