Neural network-based random fracture metal grid shielding effectiveness prediction method

Abstract

The invention discloses a random fracture metal grid shielding effectiveness prediction method based on a neural network, and the method comprises the steps: carrying out the simulation of different fracture conditions of a metal grid, obtaining the simulation data of the electromagnetic shielding effectiveness of the metal grid under different fracture conditions, constructing a data set, and constructing a metal grid electromagnetic shielding effectiveness prediction model based on the neural network. And training by using the constructed data set, identifying fracture data of the metal grid through graphical identification software, inputting the fracture data into the trained metal grid electromagnetic shielding effectiveness prediction model, and predicting the electromagnetic shielding effectiveness of the metal grid. According to the scheme, the influence of the relative fracture density of the metal grid on the shielding effectiveness is predicted based on the neural network, the shielding effectiveness change of the corresponding fractured metal grid can be rapidly predicted, and the influence of the metal grid process fracture on the electromagnetic shielding is determined.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of transparent electromagnetic shielding, and particularly relates to a method for predicting shielding effectiveness of a randomly fractured metal mesh based on a neural network. BACKGROUND

[0002] Nowadays, metal mesh is widely used as electromagnetic shielding material because it can have relatively low surface resistance and high light transmittance. It has been widely used in the fields of infrared / optical windows, electronic devices, aerospace, etc. Among various common methods for preparing metal mesh, electrohydrodynamic jet printing technology is one of the preferred processes for preparing high-precision micron-level metal mesh. Although electrohydrodynamic jet printing and other processes can print micron-level metal mesh structures on different substrates, especially flexible and curved substrates, when the mesh line width is reduced to below 10 μm, the metal mesh line is prone to various process defects, including full fracture, cracks, and hollowing. In addition, the process may also cause uneven lines, such as the overall thickening or thinning of the entire mesh line or the uneven thickening or thinning of the same mesh line. Such changes will cause deviations between the actual results and the theoretical values of the shielding effectiveness.

[0003] Traditional methods usually rely on electromagnetic simulation software to model and simulate the metal-defective mesh. Although this method can provide reliable shielding effectiveness prediction, it consumes a large amount of time and computing resources. Due to the random nature of defects, it is almost impossible to establish a structure model that is exactly the same as the actual defect in the software every time. Currently, there is no research on how to efficiently obtain the shielding effectiveness of the fractured metal mesh with micron-level line width caused by the process.

[0004] Previously, the influence of different parameter changes of the metal mesh on its shielding effectiveness was studied by the control variable method. It was found that for the complete metal mesh, the surface resistance has a greater impact on the electromagnetic shielding effectiveness at low frequencies, and as the frequency increases, the change has almost no effect on the electromagnetic shielding effectiveness. The change in line width uniformity has almost no effect on the electromagnetic shielding effectiveness. For the fractured metal mesh, when the relative fracture density is constant, the electromagnetic shielding effectiveness is independent of the fracture position, sample size, and period, and the shielding effectiveness is related to the relative fracture density. Therefore, using the relative fracture density to study the shielding effectiveness of the metal mesh can simplify the research parameters and is not limited by the size of the mesh. SUMMARY

[0005] The present application aims to provide a method for predicting the shielding effectiveness of a randomly fractured metal mesh based on a neural network, using the relative fracture density of the metal mesh as a neural network training parameter to predict its shielding effectiveness, which can quickly and accurately predict the impact of process errors on the SE of the metal mesh and increase the size freedom of the prediction.

[0006] The specific technical scheme for achieving the object of the present application is:

[0007] A random fracture metal mesh shielding effectiveness prediction method based on a neural network, comprising the following steps:

[0008] Step 1, classify the cases of metal mesh fracture due to process, and obtain the corresponding data respectively;

[0009] Step 2, simulate different fracture cases of the metal mesh to obtain the electromagnetic shielding effectiveness simulation data of the metal mesh in different fracture cases, and construct a data set;

[0010] Step 3, construct a metal mesh electromagnetic shielding effectiveness prediction model based on a neural network, and train it using the constructed data set;

[0011] Step 4, identify the fracture data of the metal mesh by using a graphical recognition software, input the fracture data into the trained metal mesh electromagnetic shielding effectiveness prediction model, and predict the electromagnetic shielding effectiveness of the metal mesh.

[0012] Compared with the prior art, the present application has the following advantages:

[0013] (1) The scheme of the present application is aimed at the various typical process fractures of the metal mesh, such as mesh fracture and uneven line width, which may occur due to the limitations of the process in actual preparation, and clearly shows the influence of the process fracture of the metal mesh on its electromagnetic shielding;

[0014] (2) The scheme of the present application uses a neural network to predict the electromagnetic shielding effectiveness of the fractured metal mesh, and no longer relies on electromagnetic simulation software to model and simulate the fractured metal mesh, thereby saving a large amount of time and computing resources and solving the problem that it is almost impossible to establish a structure model in the software that is completely consistent with the actual fracture each time;

[0015] The scheme of the present application is based on a neural network to predict the influence of the relative fracture density of the metal mesh on the shielding effectiveness, which can quickly predict the shielding effectiveness change of the corresponding fractured metal mesh, and also can use the shielding effectiveness results of a small size mesh for training and predicting the shielding effectiveness of various size meshes. In this way, not only the speed of data collection can be improved, but also the prediction is no longer limited by the size.

[0016] The present application will be further described below in conjunction with the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The figure is a flowchart of the random fracture metal mesh shielding effectiveness prediction method based on a neural network of the present scheme.

[0018] Figure 2A schematic diagram of a complete metal mesh structure in the embodiment of the present application.

[0019] Figure 3 A schematic diagram of irregular fracture of a metal mesh in the embodiment of the present application.

[0020] Figure 4 A schematic diagram of random fracture case (a) and the corresponding complete mesh shape (b) in the embodiment of the present application.

[0021] Figure 5 A neural network workflow diagram in the embodiment of the present application.

[0022] Figure 6 An update flow diagram of Adam in a complete time step in the embodiment of the present application.

[0023] Figure 7 A Loss versus epoch diagram in the embodiment of the present application.

[0024] Figure 8 SE results of metal mesh simulation and prediction using relative fracture density as a neural network input parameter in the embodiment of the present application.

[0025] Figure 9 SE results of metal mesh prediction and test using relative fracture density as a neural network input parameter in the embodiment of the present application. DETAILED DESCRIPTION

[0026] EMBODIMENT

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. The described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work under the premise that the embodiments fall within the scope of protection of the present application.

[0028] As shown in the present application and claims, unless the context clearly indicates otherwise, the words “one”, “an”, “a”, and / or “the” do not specifically refer to the singular, but also include the plural. Generally, the terms “comprise” and “include” only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0029] ​The relative arrangement of parts and steps, numerical expressions, and values set forth in the examples herein are not intended to limit the scope of the present application unless otherwise specifically stated. It is to be understood that the drawings are not necessarily to scale as the dimensions of the parts shown are for the purpose of illustration and description only and not to limit the scope of the application. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art can not be discussed in detail but are intended to be part of the scope of the present application. In all examples shown and discussed herein, any specific value is to be interpreted as illustrative only and not as a limitation. Thus, other examples of exemplary embodiments can have different values. It is noted that like numbers and letters refer to like elements throughout the several views of the drawings and as such, discussion of these elements throughout the several views does not need to be repeated in connection with each view.

[0030] In conjunction Figure 1 A neural network-based method for predicting the shielding effectiveness of a randomly fractured metal mesh includes the following steps:

[0031] Step 1, classify the cases of metal mesh fracture due to process, and obtain the corresponding data respectively;

[0032] The metal mesh includes square mesh, triangular mesh, pentagonal mesh, and hexagonal mesh.

[0033] In conjunction Figure 2 For the metal mesh structure used in this embodiment, it is a typical metal mesh 2-dielectric 1 structure with m*m number of meshes in a single period. Since the shielding effectiveness of the complete metal mesh is independent of the value of m, in order to save simulation resources, the mesh model with m=5 is used for simulation research. The dielectric constant εr=5.5, the thickness h=1.1mm. The metal mesh surface resistance is r. The unit period p=0.4mm, and w is the mesh line width. This embodiment mainly illustrates the feasibility of using relative fracture density as a neural network input parameter to predict the shielding effectiveness of fractured metal mesh when XX polarization is used.

[0034] The cases of metal mesh fracture due to process include complete fracture, crack, and hollowing.

[0035] As Figure 3 shown, several metal mesh irregular fracture schematic diagrams are shown, which are divided into three cases of complete fracture 3, crack 4, and hollowing 5. The relative fracture density n is defined, which represents the fracture density value in a single m*m mesh period, and the infinite mesh is periodically arranged with a fracture density of n in m*m mesh.

[0036] In this scheme, the relative fracture density is defined as the ratio of the fracture area to the area of the complete metal mesh as the fracture data of the metal mesh. Figure 4A mesh fracture case (a) and its corresponding complete mesh shape (b) are shown, and the relative fracture density n is defined as the ratio of the fracture area 6 to the complete metal mesh area 7.

[0037] Step 2, simulate different fracture cases of the metal mesh, obtain the electromagnetic shielding effectiveness simulation data of the metal mesh in different fracture cases, and construct a data set;

[0038] In this embodiment, MATLAB is used in combination with CST to simulate different fracture cases of the metal mesh, the fracture position and size of the metal mesh are randomly set, the relative fracture density is converted for neural network input parameters, and the corresponding metal mesh shielding effectiveness curve is collected as a training set.

[0039] In this example, the metal mesh model parameters are set as m=5, p=0.4, w=0.02, r is a random function with a Gaussian distribution in the interval (0.05~2)Ω / sq, and the peak value is 0.2Ω / sq; the position, size, and number of fractures are random, and the fracture density is converted according to the area of the fracture for training and prediction.

[0040] Step 3, construct a metal mesh electromagnetic shielding effectiveness prediction model based on a neural network, and train the constructed data set;

[0041] In this embodiment, 500 groups of neural network models are used to train and predict the metal electromagnetic shielding effectiveness, and the prediction model uses a deep neural network model based on the TensorFlow / Keras framework, including a convolutional neural network, a recurrent neural network, and a Transformer network;

[0042] In this example, the network is composed of an input layer, a hidden layer, and an output layer, and the neural network workflow diagram is as shown in Figure 5 The input data is first processed in the input layer, then propagated through multiple hidden layers, each layer aggregates the weighted input from the previous layer, applies an activation function, and passes the converted information to the next layer until the last layer produces the expected result.

[0043] For example, the metal mesh electromagnetic shielding effectiveness prediction model adopts a L2-regularization + batch normalization + ReLU feedforward neural network as a baseline model;

[0044] In the data preprocessing stage, the input features X and the target variables y are respectively standardized by Z-score, and the input and output are all converted to N(0, 1) to accelerate gradient descent and suppress saturation;

[0045] The output layer adopts linear activation to maintain the global value of the regression amount, and all weights are initialized using HeNormal, that is, Wᵢⱼ~N(0,2 / fan_in), which is matched with the activation function ReLU to maintain the signal variance at each layer to be approximately 1, thereby alleviating gradient disappearance;

[0046] In addition, in order to suppress overfitting, L2 regularization is introduced at each layer of the model.

[0047] In addition, batch statistics are used in the training stage, and moving average is used in the inference stage, so that the network is not sensitive to weight initialization and allows a higher learning rate. The loss function used by the metal mesh electromagnetic shielding effectiveness prediction model in training is the MES error function, that is, the mean squared error (MSE):

[0048]

[0049] Where: N is the number of samples, k is the output dimension, is the true value of the i-th sample and the j-th output, is the model prediction value of the i-th sample and the j-th output.

[0050] In addition, the neural model architecture adopts a fully connected network structure, and the model is trained by an Adam optimizer. Adam stabilizes the direction by momentum, adjusts the scale of the adaptive learning rate, and is supplemented by bias correction to ensure initial stability. A complete time step update flowchart is shown in FIG. 6. First, the current gradient gtis calculated, then the first moment estimate (momentum) mtand the second moment estimate (adaptive learning rate term) vtare updated. Since is initialized to 0, at the beginning of training, and will be biased to 0. Therefore, Adam needs to perform bias correction next, and finally update the model parameters θ. In addition, the initial learning rate α of the neural network in this embodiment is 1.5×10⁻³.

[0051] In this embodiment, the loss of the model on the training set changes with the epoch during training as shown in Figure 7 The loss decreases to a stable value as the number of training increases, and finally approaches 0.

[0052] Step 4: Identify the fracture data of the metal mesh by using a graphical recognition software (such as ImageJ graphical recognition software), input the fracture data into the trained metal mesh electromagnetic shielding effectiveness prediction model, and predict the electromagnetic shielding effectiveness of the metal mesh.

[0053] In this embodiment, 10 groups of random fracture grid models are generated using MATLAB combined with CST and simulated, random data is input into the trained neural network model for prediction, and the prediction results are compared with the simulation results. The 10 groups of curves have 97.7% of the data points with an error of 0%~10%, 2.3% of the data points with an error of 10%~15%, and an average error of 3.47%. The prediction and simulation SE curves of one group of metal grid are as shown in Figure 8 .

[0054] In this embodiment, the sample is a 7cm*7cm random fracture grid, the grid unit period p=0.4mm, and the line width w=20μm. The ImageJ image recognition software is used to identify the pattern of the fractured metal grid sample to obtain the fracture area, which is defined as the difference between the complete metal grid area and the area identified by the software.

[0055] In this embodiment, the prediction results of the sample are compared with the test results, and 95.2% of the data points have an error of 0%~10%, 3.3% of the data points have an error of 10%~15%, 1.5% of the data points have an error of 15%~20%, and the average error is 3.96%. The prediction and test SE curves of one group of metal grid are as shown in Figure 9 .

[0056] In summary, the use of relative fracture density as an input parameter of the neural network to predict the electromagnetic shielding effectiveness of the metal grid can accurately and quickly obtain the required grid shielding effectiveness, significantly reducing the dependence on tedious modeling and simulation, and improving the work efficiency. The present application can systematically and quickly and accurately predict the influence of process errors on the SE of the metal grid, and its use is not limited to samples made by the EHD method, which paves the way for the future large-scale production and application of metal grid shielding products.

[0057] The present application also provides a neural network-based random fracture metal grid shielding effectiveness prediction system, comprising the following modules:

[0058] The data acquisition module is used for classifying the fracture of the metal grid caused by the process, and acquiring the corresponding data respectively, simulating different fracture conditions of the metal grid, obtaining the electromagnetic shielding effectiveness simulation data of the metal grid in different fracture conditions, and constructing a data set;

[0059] The model construction module is used for constructing a metal grid electromagnetic shielding effectiveness prediction model based on a neural network, and training the constructed data set;

[0060] The metal mesh electromagnetic shielding effectiveness prediction module is configured to: input the fracture data of the metal mesh into a trained metal mesh electromagnetic shielding effectiveness prediction model through graphical recognition software, and predict the electromagnetic shielding effectiveness of the metal mesh.

[0061] The present application also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0062] Step 1: classifying the fracture of the metal mesh caused by the process, and obtaining corresponding data respectively;

[0063] Step 2: simulating different fracture conditions of the metal mesh to obtain electromagnetic shielding effectiveness simulation data of the metal mesh in different fracture conditions, and constructing a data set;

[0064] Step 3: constructing a metal mesh electromagnetic shielding effectiveness prediction model based on a neural network, and training the model using the constructed data set;

[0065] Step 4: inputting the fracture data of the metal mesh into a trained metal mesh electromagnetic shielding effectiveness prediction model through graphical recognition software, and predicting the electromagnetic shielding effectiveness of the metal mesh.

[0066] The present application also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0067] Step 1: classifying the fracture of the metal mesh caused by the process, and obtaining corresponding data respectively;

[0068] Step 2: simulating different fracture conditions of the metal mesh to obtain electromagnetic shielding effectiveness simulation data of the metal mesh in different fracture conditions, and constructing a data set;

[0069] Step 3: constructing a metal mesh electromagnetic shielding effectiveness prediction model based on a neural network, and training the model using the constructed data set;

[0070] Step 4: inputting the fracture data of the metal mesh into a trained metal mesh electromagnetic shielding effectiveness prediction model through graphical recognition software, and predicting the electromagnetic shielding effectiveness of the metal mesh.

[0071] The above-described embodiments only express the one embodiment of the present application, which is described in a more specific and detailed manner, but should not be construed as limiting the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the patent of the present application should be subject to the appended claims.

Claims

1. A method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks, characterized in that, Includes the following steps: Step 1: Classify the cases of metal mesh breakage due to the process and obtain the corresponding data for each case; Step 2: Simulate different fracture conditions of the metal mesh to obtain simulation data of the electromagnetic shielding effectiveness of the metal mesh under different fracture conditions, and construct a dataset; Step 3: Construct a prediction model for the electromagnetic shielding effectiveness of metal mesh based on neural networks, and train it using the constructed dataset; Step 4: Identify the fracture data of the metal mesh using graphical recognition software, input the fracture data into the pre-trained metal mesh electromagnetic shielding effectiveness prediction model, and predict the electromagnetic shielding effectiveness of the metal mesh.

2. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 1, characterized in that, The metal mesh may break due to the process, including complete breakage, cracks, and hollowing out. The relative fracture density is defined as the ratio of the fracture area to the area of ​​the intact metal mesh as the fracture data of the metal mesh.

3. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 1, characterized in that, The metal mesh includes square mesh, triangular mesh, pentagonal mesh, and hexagonal mesh.

4. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 2, characterized in that, In step 2, when simulating different fracture conditions of the metal mesh, MATLAB is used in conjunction with CST for simulation. The fracture location and size of the metal mesh are randomly set and converted into relative fracture density for neural network input parameters. The corresponding metal mesh shielding effectiveness curves are collected as training sets.

5. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 1, characterized in that, The electromagnetic shielding effectiveness prediction model for the metal mesh in step 3 uses a deep neural network model based on the TensorFlow / Keras framework, including convolutional neural networks, recurrent neural networks, and Transformer networks.

6. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 5, characterized in that, The electromagnetic shielding effectiveness prediction model for the metal mesh adopts a feedforward neural network with L2 regularization, batch normalization, and ReLU as the baseline model. In the data preprocessing stage, the input feature X and the target variable y are standardized by Z-score to accelerate gradient descent and suppress saturation. The output layer uses linear activation to maintain the global value of the regressor. All weights are initialized using HeNormal, i.e., Wᵢⱼ~N(0,2 / fan_in), which is combined with the ReLU activation function to maintain the signal variance approximately 1 in each layer and alleviate gradient vanishing. In addition, L2 regularization is introduced in each layer of the model to suppress overfitting.

7. The method for predicting the shielding effectiveness of randomly fractured metal mesh based on neural networks according to claim 5, characterized in that, The loss function used in training the electromagnetic shielding effectiveness prediction model for the metal mesh is the MES error function.

8. A prediction system for the shielding effectiveness of randomly fractured metal mesh based on neural networks, characterized in that, Includes the following modules: Data acquisition module: This module is used to classify the fractures of metal meshes caused by the manufacturing process, acquire the corresponding data for each fracture, simulate different fracture conditions of the metal mesh, obtain simulation data of the electromagnetic shielding effectiveness of the metal mesh under different fracture conditions, and construct a dataset. Model building module: used to build a prediction model for the electromagnetic shielding effectiveness of metal mesh based on neural networks, and to train it using the built dataset; Metal Mesh Electromagnetic Shielding Effectiveness Prediction Module: This module uses graphical recognition software to identify fracture data in the metal mesh, inputs the fracture data into a pre-trained metal mesh electromagnetic shielding effectiveness prediction model, and then predicts the electromagnetic shielding effectiveness of the metal mesh.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-7.

10. A computer-storable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images