Magnetic field programmable inverse design method, structure, and electronic device and medium

Abstract

A magnetic field programmable reverse design method, structure and electronic equipment and medium are disclosed. In the method, the thin film is magnetized by folding to design the magnetic field. Deep learning combines the structure parameters of the folded paper in the magnetic material thin film and the final magnetic field distribution information through the training model, finds a certain function relationship, so that when there is a special magnetic field design requirement, the corresponding folded paper structure geometric parameters can be found through the model to achieve reverse design. A folded paper related geometric parameter model is established in comsol, the structure is magnetized in blocks and regions, the magnetization effect after folding and magnetization is simulated, the simulation results are post-processed in matlab, a large amount of data set can be obtained in a short time by changing the parameters in matlab, and a deep learning model is trained based on the data set model library to perform reverse design.

Description

Technical Field

[0001] This invention relates to the field of magnetic field technology, and in particular to a programmable reverse design method, structure, electronic device, and medium for magnetic fields. Background Technology

[0002] In recent years, magnetic tactile sensors have attracted widespread attention due to their simple structure, high physical robustness, low cost, and wireless penetration. They can be applied in fields such as precision robotic manipulation, flow rate measurement, and medicine. Existing flexible sensors based on the Hall principle mostly embed rigid permanent magnets, and then decouple multidimensional information by arranging the Hall sensor in an array according to the magnetic field around the permanent magnet. This type of sensor is easily damaged when subjected to large impacts. Flexible magnetic materials have attracted widespread attention. Existing technology decouples multidimensional force information by designing the magnetic field of flexible thin films. This designable magnetization provides more possibilities for tactile sensors. At present, magnetic field design is mainly based on centripetal magnetization and sinusoidal magnetization, with a relatively simple distribution. Combining deep learning with design can reveal more possibilities, which requires parametric simulation and reverse design of the magnetic field.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the shortcomings or defects of the existing technology, a programmable reverse design method, structure, electronic device, and medium for magnetic fields are provided. Based on a magnetic material thin film with a paper-cutting structure, the magnetic field is designed by magnetizing it through folding. Deep learning combines the structural parameters of the paper-cutting structure in the magnetic material thin film with the final magnetic field distribution information through model training to find a functional relationship. Thus, when there are special magnetic field design requirements, the corresponding geometric parameters of the paper-cutting structure can be found through the model.

[0005] The objective of this invention is achieved through the following technical solutions.

[0006] A programmable magnetic field reverse design method includes,

[0007] In COMSOL Multiphysics, a finite element analysis software platform for multiphysics modeling, simulation, and APP design, a magnetic thin film model with adjustable geometric parameters based on a paper-cutting structure was built. In the magnetic thin film model, the magnitude and direction of the residual magnetic field strength after the magnetic film itself is magnetized were set in blocks and regions to simulate the magnetic field effect of magnetization after folding.

[0008] A connection was established between the finite element analysis software COMSOL Multiphysics and the commercial mathematical computing software MATLAB using the COMSOL Multiphysics software. In MATLAB, the geometric parameters of the magnetic thin film model were retrieved using the mphopen(), mphinterp(), and mphplot() functions. Simulation results were obtained to acquire magnetic field information at multiple magnetic field points, including coordinate information (x, y, z), information on the magnetic field intensity component Bz, and spatial information on the magnetic field intensity gradient. , , The magnetic field information of multiple magnetic field points forms a set of data. By iteratively changing the geometric parameters through the mathematical calculation software MATLAB, a dataset containing multiple sets of data is obtained. The dataset is divided into training set and test set according to the proportion.

[0009] The network is constructed in two parts: a first part exploring the relationship between geometric parameters and magnetic field information, and a second part exploring the relationship between local and global magnetic field information. The first part takes geometric parameters as input and outputs multiple magnetic field points and their magnetic field information. Flipping and rotation operations are added to the magnetic field information to enhance the network's generalization ability. The first part is a fully connected network with a selected number of fully connected hidden layers. The model optimizer is set to Adam, and the learning rate is 1e. -3 The second part of the network takes as input multiple magnetic field points within a square area and their magnetic field information as input, and outputs the magnetic field points and their magnetic field information as output. Both the input and output magnetic field information undergo flipping and rotation operations to enhance the network's generalization ability. The second part of the network uses a convolutional neural network, with the selected number of convolutional layers, pooling layers, and fully connected layers. The model optimizer is AdamW, and the learning rate is 1e. -3 ;

[0010] Two sub-networks are trained separately. During training, network weights are first initialized using random initialization. In each iteration, a batch of data (32 samples) is selected from the training dataset and input into the network to calculate the predicted output. The difference between the predicted and true outputs is calculated using a loss function, and the gradient of the loss with respect to the network weights is calculated using backpropagation. Optimizers (Adam and AdamW) are used to update the network weights based on the gradient. When the training loss no longer decreases significantly within a certain number of iterations, and the performance metrics on the validation set stabilize or meet expectations, the network training is considered converged. To ensure convergence, the training loss and performance metrics on the test set are periodically observed during training. Convergence is achieved by changing the learning rate, increasing regularization, and adjusting the network structure. Through multiple iterations, the network gradually learns the complex relationship between geometric parameters and the final magnetic field distribution, as well as the correlation between local and global magnetic field information. Given the required local magnetic field information, the network can predict the global magnetic field information and corresponding geometric parameters based on the learned patterns and relationships, thus obtaining a magnetic field that meets the requirements.

[0011] In the method described, the trained network is given local magnetic field points in the test set. The local magnetic field points are used as input to the second part of the network, which generates all the magnetic field points. This result is then used as input to the first part of the network, which obtains the geometric parameters. The geometric parameters are compared with the true values ​​to evaluate the magnetic thin film model.

[0012] In the method described, the geometric parameters and the true values ​​are evaluated by calculating the mean square error of the magnetic thin film model.

[0013] In the method described, the first part of the network is a fully connected network, which includes an input layer, four sequentially connected fully connected layers, and an output layer.

[0014] In the method described, the second part of the network adopts a convolutional neural network, which includes an input layer, two convolutional layers, a pooling layer, two fully connected layers, two deconvolutional layers and an output layer.

[0015] In the method described, the geometric parameters include the crease width t, the crease indentation thickness d, the distance a from the fold point to one side, and the angle α between the horizontal line and the fold line at the fold point.

[0016] In the method described, the dataset is divided into a training set and a test set in an 8:2 ratio.

[0017] A magnetic thin film structure is obtained using the method described above.

[0018] An electronic device includes a memory, a processor, and instructions stored in the memory and executable on the processor, wherein the processor executes the instructions to implement the steps of the method.

[0019] A computer-readable storage medium storing computer instructions, which, when executed by a machine, implement the steps of the method.

[0020] Compared with the prior art, the beneficial effects of this invention are as follows:

[0021] This invention uses parametric simulation to obtain the impact of parameter changes on magnetic field distribution, thereby optimizing the design. For example, it can find the parameter combination that maximizes the magnetic field strength or the most uniform magnetic field distribution. In actual magnetic field device design, conducting physical experiments to test the effects of different designs usually requires a lot of time and cost. However, this invention allows for rapid testing of a large number of designs on a computer through simulation, greatly saving time and cost. It also improves design quality: previous designs relied heavily on the designer's experience and intuition, resulting in high design thresholds and a large workload. Deep learning technology can learn the complex relationship between magnetic field parameters and simulation results, which can help us to make more accurate designs.

[0022] The description provided is merely an overview of the technical solution of this invention. In order to make the technical means of this invention clearer and more understandable, so that those skilled in the art can implement it according to the contents of the specification, and to make the described and other objects, features and advantages of this invention more obvious and understandable, specific embodiments of this invention are described below. Attached Figure Description

[0023] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0024] In the attached diagram:

[0025] Figure 1 This illustration shows a schematic diagram of a magnetic thin film model with adjustable geometric parameters based on a paper-folding and cutting structure, which is part of a programmable magnetic field reverse design method provided in this application.

[0026] Figure 2 A flowchart illustrating the programmable reverse design method for magnetic fields provided in an embodiment of this application is shown.

[0027] The present invention will be further explained below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0028] The following will refer to the appendix. Figures 1 to 2 Specific embodiments of the invention will be described in more detail below. While specific embodiments of the invention are shown in the accompanying drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0029] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0030] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments, and the accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0031] To better understand, such as Figures 1 to 2 As shown, a programmable reverse design method for magnetic fields includes,

[0032] In COMSOL Multiphysics, a finite element analysis software platform for multiphysics modeling, simulation, and APP design, a magnetic thin film model with adjustable geometric parameters based on a paper-cutting structure was built. In the magnetic thin film model, the magnitude and direction of the residual magnetic field strength after the magnetic film itself is magnetized were set in blocks and regions to simulate the magnetic field effect of magnetization after folding.

[0033] This study utilizes COMSOL Multiphysics with MATLAB software to establish a connection between the finite element analysis software COMSOL Multiphysics and the commercial mathematical computing software MATLAB. In MATLAB, functions such as mphopen(), mphinterp(), and mphplot() can be used to retrieve the geometric parameters of the magnetic thin film model, obtain simulation results, and acquire magnetic field information at multiple magnetic field points. This includes coordinate information (x, y, z), information on the magnetic field intensity component Bz, and spatial information on the magnetic field intensity gradient. , , The magnetic field information of multiple magnetic field points forms a set of data. By iteratively changing the geometric parameters through the mathematical calculation software MATLAB, a dataset containing multiple sets of data is obtained. The dataset is divided into training set and test set according to the proportion.

[0034] The network is constructed in two parts: a first part exploring the relationship between geometric parameters and magnetic field information, and a second part exploring the relationship between local and global magnetic field information. The first part takes geometric parameters as input and outputs multiple magnetic field points and their magnetic field information. Flipping and rotation operations are added to the magnetic field information to enhance the network's generalization ability. The first part is a fully connected network with a selected number of fully connected hidden layers. The model optimizer is set to Adam, and the learning rate is 1e. -3 The second part of the network takes as input multiple magnetic field points within a square area and their magnetic field information as input, and outputs the magnetic field points and their magnetic field information as output. Both the input and output magnetic field information undergo flipping and rotation operations to enhance the network's generalization ability. The second part of the network uses a convolutional neural network, with the selected number of convolutional layers, pooling layers, and fully connected layers. The model optimizer is AdamW, and the learning rate is 1e. -3 ;

[0035] Two sub-networks are trained separately. During training, network weights are first initialized using random initialization. In each iteration, a batch of data (32 samples) is selected from the training dataset and input into the network to calculate the predicted output. The difference between the predicted and true outputs is calculated using a loss function, and the gradient of the loss with respect to the network weights is calculated using backpropagation. Optimizers (Adam and AdamW) are used to update the network weights based on the gradient. When the training loss no longer decreases significantly within a certain number of iterations, and the performance metrics on the validation set stabilize or meet expectations, the network training is considered converged. To ensure convergence, the training loss and performance metrics on the test set are periodically observed during training. Convergence is achieved by changing the learning rate, increasing regularization, and adjusting the network structure. Through multiple iterations, the network gradually learns the complex relationship between geometric parameters and the final magnetic field distribution, as well as the correlation between local and global magnetic field information. Given the required local magnetic field information, the network can predict the global magnetic field information and corresponding geometric parameters based on the learned patterns and relationships, thus obtaining a magnetic field that meets the requirements.

[0036] In a preferred embodiment of the method, the trained network is given local magnetic field points in the test set. The local magnetic field points are used as input to the second part of the network to generate all the magnetic field points. This result is then used as input to the first part of the network to obtain geometric parameters. The geometric parameters are compared with the true values ​​to evaluate the magnetic thin film model.

[0037] In a preferred embodiment of the method, the geometric parameters and the true values ​​are evaluated by calculating the mean square error of the magnetic thin film model.

[0038] In a preferred embodiment of the method, the first part of the network is a fully connected network, which includes an input layer, four sequentially connected fully connected layers, and an output layer.

[0039] In a preferred embodiment of the method, the second part of the network employs a convolutional neural network, which includes an input layer, two convolutional layers, a pooling layer, two fully connected layers, two deconvolutional layers, and an output layer.

[0040] In a preferred embodiment of the method, the geometric parameters include the crease width t, the crease indentation thickness d, the distance a from the fold point to one side, and the angle α between the horizontal line and the fold line at the fold point.

[0041] In a preferred embodiment of the method, the dataset is divided into a training set and a test set in an 8:2 ratio.

[0042] A magnetic thin film structure is obtained using the method described above.

[0043] An electronic device includes a memory, a processor, and instructions stored in the memory and executable on the processor, wherein the processor executes the instructions to implement the steps of the method.

[0044] A computer-readable storage medium stores computer instructions, which, when executed by a machine, implement the steps of the method described. Optionally, the computer instructions in this application embodiment may also be referred to as application code, and this application embodiment does not specifically limit this.

[0045] In a specific implementation, as one embodiment, the processor may include one or more CPUs. In a specific implementation, as one embodiment, the electronic device may include multiple processors, each of which may be a single-core processor or a multi-core processor.

[0046] Optionally, the electronic device may also include a communication interface and memory, which may include read-only memory and random access memory, and provide operation instructions and data to the processor. A portion of the memory may also include non-volatile random access memory (NVRAM).

[0047] In this embodiment, corresponding operations are executed by calling operation instructions stored in the memory. The processor controls the processing operations of any device in the electronic device; the processor can also be called a central processing unit (CPU).

[0048] In one embodiment, the memory may include read-only memory and random access memory, providing instructions and data to the processor. A portion of the memory may also include NVRAM. For example, in an application, the memory, communication interface, and memory are coupled together via a bus system, which may include, in addition to a data bus, a power bus, a control bus, and a status signal bus. The methods disclosed in the embodiments of this application can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor may be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. Computer instructions can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0049] On the one hand, a computer-readable storage medium is provided, which stores instructions that, when executed, implement the functions performed by the terminal device in the above embodiments.

[0050] On the one hand, a chip is provided that is used in a terminal device. The chip includes at least one processor and a communication interface. The communication interface and at least one processor are coupled together. The processor is used to run instructions to implement the functions performed by the medium and large magnetic field programmable reverse design method in the above embodiments.

[0051] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer instruction product. The computer instruction product includes one or more computer instructions or instructions. When the computer instructions or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user equipment, or other programmable device. The computer instructions or instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions or instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video disc (DVD); or it can be a semiconductor medium, such as a solid-state drive (SSD).

[0052] In one embodiment, the programmable reverse design method for magnetic fields utilizes a magnetic material film based on a paper-folding structure. This film is magnetized through folding to design the magnetic field. Deep learning, through model training, combines the structural parameters of the paper-folding structure in the magnetic material film with the final magnetic field distribution information to find a functional relationship. Therefore, when there is a magnetic field design requirement, the model can find the corresponding geometric parameters of the paper-folding structure, achieving reverse design. The steps are as follows: First, a geometric parameter model related to paper-folding is established in COMSOL. The structure is magnetized in blocks and regions to simulate the magnetization effect after folding. Then, the simulation results are post-processed in MATLAB. Changing parameters in MATLAB can quickly obtain a large dataset. Based on this dataset and model library, a deep learning model is trained for reverse design. A parameterized COMSOL magnetostatic simulation model is obtained, and MATLAB is used for post-processing to obtain the magnetic field distribution changes of the magnetic field intensity component Bz at different locations, and the magnetic field gradient distribution changes, thus building a dataset. The dataset is randomly divided into training and testing sets to train the reverse design model. Finally, when the desired magnetic field effect is input, the corresponding geometric parameters are output, and the corresponding magnetic field is obtained.

[0053] In one embodiment, the magnetic field programmable reverse design method includes:

[0054] 1. Build a geometrically adjustable magnetic thin film model inspired by paper-folding and cutting structures in COMSOL;

[0055] 2. In the model, the magnitude and direction of the residual magnetic field strength are set in blocks and regions to simulate the magnetic field effect of magnetization after folding;

[0056] 3. Establish the connection between COMSOL and MATLAB using COMSOL with MATLAB software;

[0057] 4. In MATLAB, retrieve the four geometric parameters of the model at this point, then retrieve the simulation results to obtain the coordinates (x, y, z) of approximately 256×256×3 information points of the magnetic field, the information of the magnetic field intensity component Bz, and the spatial information of the magnetic field intensity gradient. , , This forms a set of data;

[0058] 5. Obtain large-scale datasets by changing geometric parameters in small iteration steps using MATLAB;

[0059] 6. Divide the dataset into training and testing sets in an 8:2 ratio;

[0060] 7. Construct a network, which is divided into two parts. One part explores the relationship between geometric parameters and magnetic field information, while the other part is responsible for the derivation of the relationship, that is, exploring the management between local magnetic field information and overall magnetic field information.

[0061] 8. The first part of the network takes geometric parameters as input (the first four numbers in an Excel spreadsheet) and outputs 256×256×3×7 magnetic field points and their information. Flipping and rotation operations are added to the magnetic field information to enhance the network's generalization ability. A fully connected network is used, with an appropriate number of fully connected layers and hidden layers. The model optimizer is set to Adam, and the learning rate is 1e. -3 ;

[0062] 9. The second part of the network takes as input points within a square area, including 16×16×7 magnetic field information points, and outputs 256×256×3×7 full magnetic field information points. Both the input and output magnetic field information undergo flipping and rotation operations to enhance the network's generalization ability. The network structure uses a convolutional network, with appropriate numbers of convolutional layers, pooling layers, and fully connected layers. The model optimizer is AdamW, and the learning rate is 1e. -3 ;

[0063] 10. The model is trained using the training set, and the network weights are optimized through methods such as backpropagation and gradient descent. During the training process, the model gradually learns the complex relationship between geometric parameters and the final magnetic field distribution, as well as the correlation between local magnetic field information and global magnetic field information.

[0064] 11. A well-trained network model can serve as a good prediction model. Given the magnetic field information points at a local location (16×16×7) in the test set, the information is used as the input to the second network, which generates all the magnetic field information points (256×256×3×7). This result is then used as the input to the first network, which predicts the simulated geometric parameters. The simulated parameters are compared with the true values ​​to evaluate the model performance. The network evaluation uses the mean square error method.

[0065] 12. After successful model training, in practical applications, based on the required force decoupling needs, local magnetic field information is provided, and then global magnetic field information and corresponding geometric parameters are obtained through the network. Then, actual modeling and manufacturing are carried out to achieve the required reverse design requirements.

[0066] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0067] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A programmable reverse design method for magnetic fields, characterized in that, It includes the following steps, In COMSOL Multiphysics, a finite element analysis software platform for multiphysics modeling, simulation, and APP design, a magnetic thin film model with adjustable geometric parameters based on a paper-cutting structure was built. In the magnetic thin film model, the magnitude and direction of the residual magnetic field strength after the magnetic film itself is magnetized were set in blocks and regions to simulate the magnetic field effect of magnetization after folding. A connection was established between the finite element analysis software COMSOL Multiphysics and the commercial mathematical computing software MATLAB using the COMSOL Multiphysics software. In MATLAB, the geometric parameters of the magnetic thin film model were retrieved using the mphopen(), mphinterp(), and mphplot() functions. Simulation results were obtained to acquire magnetic field information at multiple magnetic field points, including coordinate information (x, y, z), information on the magnetic field intensity component Bz, and spatial information on the magnetic field intensity gradient. , , The magnetic field information of multiple magnetic field points forms a set of data. By iteratively changing the geometric parameters through the mathematical calculation software MATLAB, a dataset containing multiple sets of data is obtained. The dataset is divided into training set and test set according to the proportion. The network was constructed in two parts: a first part exploring the relationship between geometric parameters and magnetic field information, and a second part exploring the relationship between local and global magnetic field information. The first part takes geometric parameters as input and outputs multiple magnetic field points and their magnetic field information. Flipping and rotation operations were added to the magnetic field information to enhance the network's generalization ability. The first part is a fully connected network with a selected number of fully connected hidden layers. The model optimizer was set to Adam, and the learning rate was 1e. -3 The second part of the network takes as input multiple magnetic field points within a square area and their magnetic field information as input, and outputs the magnetic field points and their magnetic field information as output. Both the input and output magnetic field information undergo flipping and rotation operations to enhance the network's generalization ability. The second part of the network uses a convolutional neural network, with the selected number of convolutional layers, pooling layers, and fully connected layers. The model optimizer is AdamW, and the learning rate is 1e. -3 ; Two sub-networks are trained separately. During network training, network weights are first initialized using random initialization. In each iteration, a batch of data is selected from the training dataset and input into the network. The network's predicted output is calculated, and the difference between the predicted output and the true output is calculated using the loss function. The gradient of the loss with respect to the network weights is calculated using the backpropagation algorithm, and the network weights are updated using the optimizer based on the gradient. When the training loss no longer decreases significantly within the number of iterations, and the performance indicators on the validation set reach stability or meet the expected requirements, the network training is considered to have converged. To ensure network convergence, the training loss and performance indicators on the test set are observed periodically during network training. Network convergence is achieved by changing the learning rate, increasing regularization, and adjusting the network structure. Through multiple iterations of training, the network gradually learns the complex relationship between geometric parameters and the final magnetic field distribution, as well as the correlation between local magnetic field information and global magnetic field information. Given the required local magnetic field information, the network can predict the global magnetic field information and the corresponding geometric parameters based on the learned patterns and relationships, thereby obtaining a magnetic field that meets the requirements.

2. The method according to claim 1, characterized in that, The trained network is given local magnetic field points in the test set. These local magnetic field points are used as input to the second part of the network, which generates all the magnetic field points. This result is then used as input to the first part of the network, which obtains the geometric parameters. The geometric parameters are compared with the true values ​​to evaluate the magnetic thin film model.

3. The method according to claim 2, characterized in that, The geometric parameters and true values ​​were evaluated by calculating the mean square error of the magnetic thin film model.

4. The method according to claim 1, characterized in that, The first part of the network is a fully connected network, which includes an input layer, four sequentially connected fully connected layers, and an output layer.

5. The method according to claim 1, characterized in that, The second part of the network uses a convolutional neural network, which includes an input layer, two convolutional layers, one pooling layer, two fully connected layers, two deconvolutional layers, and one output layer.

6. The method according to claim 1, characterized in that, The geometric parameters include the crease width t, the crease indentation thickness d, the distance a from the fold point to one side, and the angle α between the horizontal line and the fold line at the fold point.

7. The method according to claim 1, characterized in that, The dataset is divided into training and test sets in an 8:2 ratio.

8. A magnetic thin film structure, characterized in that, It is obtained by the method described in any one of claims 1-7.

9. An electronic device, characterized in that, It includes a memory, a processor, and instructions stored in the memory and executable on the processor, wherein the processor, when executing the instructions, implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores computer instructions, which, when executed by a machine, implement the steps of the method as described in any one of claims 1-7.

Images