A method for preparing three-dimensional cobalt-platinum microstructures based on graph neural network

Abstract

The present invention discloses a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network, which belongs to the technical field of preparing cobalt-platinum microstructures. The method comprises: collecting cobalt-platinum deposition parameters in the preparation of three-dimensional cobalt-platinum microstructures, constructing a data set and performing data preprocessing; establishing a graph causal network in combination with a graph neural network, graph attention and Granger causality mechanism; inputting the preprocessed cobalt-platinum deposition parameters into the graph causal network for processing, inferring the atomic ratio of the three-dimensional cobalt-platinum microstructure, identifying the optimal cobalt-platinum deposition parameters, and completing the preparation of the three-dimensional cobalt-platinum microstructure based on the optimal cobalt-platinum deposition parameters. In this method, the atomic ratio of the three-dimensional cobalt-platinum microstructure is inferred by combining a graph neural network, graph attention and causal inference, and the optimal cobalt-platinum deposition parameters are identified; thereby facilitating enhanced control over deposition conditions, improving the efficiency and accuracy of the cobalt-platinum microstructure preparation process, reducing the number of trial and error in traditional experiments, and improving material performance.

Description

Technical Field

[0001] The present invention relates to the technical field of cobalt-platinum microstructure preparation, and in particular to a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network. Background Art

[0002] At present, the research on micro-magnets has made in-depth progress in many material fields, such as Nd-Fe-B, Sm-Co, ferrite, Co-Pt, etc. Existing manufacturing technologies can control the thickness and composition of the film within a certain range, especially in the manufacture of single-layer films on hard substrates or in the formation of multi-layer structures by combining magnetic layers with non-magnetic layers. The research directions of these technologies are mainly focused on the manufacture of large-area magnetic films or magnetic arrays, and rarely pay attention to magnetic structures of specific shapes. With the increasing requirements for material compliance in the field of flexible robots, the research demand for flexible magnets is becoming increasingly urgent. In order to further enhance the functionality and adaptability of flexible magnets, the manufacturing research of micro-magnetic structures with specific shapes has become an important direction.

[0003] Since its introduction by John D. Madden and Ian W. Hunter in 1996, localized electrochemical deposition (LECD) has become a promising technology for micromagnet fabrication. LECD uses a sharp electrode to apply voltage in an electrolyte solution close to the deposition surface to induce localized deposition, thereby achieving precise control over the shape and size of the material. Although there have been many improvements to this technology, such as ultrasound-assisted LECD and electrolyte pillar LECD, to increase deposition speed and pattern complexity, traditional methods are not efficient enough when dealing with complex multivariable parameter optimization problems.

[0004] In recent years, machine learning techniques have shown significant potential in materials science to solve these optimization problems. For example, Katirci and Danaci used NSGA-II and machine learning algorithms to optimize nickel electroplating and successfully achieved high-precision coating quality prediction and improvement; Frydrych et al. used LSTM networks to optimize the crystal plasticity parameters of electroplated copper and achieved accurate prediction and control under cyclic deformation. However, the application of these techniques in the electrochemical deposition of magnetic materials has not been fully explored.

[0005] In the electrochemical deposition process, a large number of experiments are usually required to verify the optimal deposition parameters to ensure the accuracy of the preparation of cobalt-platinum microstructures. Optimizing electrochemical deposition parameters is crucial to the manufacture of cobalt-platinum microstructures. Current optimization methods often rely on experience and can only model the relationship between the current process and data. They lack the ability to predict the parameter results that are not seen in the model with interpretable logic. They lack effective mathematical models and interpretability, and it is difficult to explore better process methods. Therefore, a new method is needed to solve these optimization problems, especially in the modeling of multivariate interactions and complex causal relationships. Causal inference and graph neural networks (GNNs) provide an innovative solution that can provide mathematical interpretability and powerful causal modeling capabilities for complex systems, which is of great significance for exploring the optimal parameters of the experimental process. Unlike traditional machine learning methods, graph causal networks can clearly reveal the causal relationship between input features (such as pH value, ion ratio, etc.) and output results (such as deposition layer thickness, magnetic properties). This causal inference can not only help understand the mutual influence between variables, but also provide optimization solutions based on causal relationships to avoid simple correlation analysis.

[0006] Therefore, it is considered to provide a three-dimensional cobalt-platinum microstructure preparation method based on graph neural network, which can model the electrochemical deposition parameters and realize the prediction of the atomic ratio of cobalt and platinum, thereby enhancing the control of deposition conditions, reducing the large number of experiments required in traditional experiments, and improving the efficiency of the cobalt-platinum microstructure preparation process. Summary of the invention

[0007] In view of this, the present invention provides a method for preparing three-dimensional cobalt-platinum microstructures based on graph neural networks, which at least solves some of the above-mentioned technical problems. In this method, electrochemical deposition parameters are optimized by combining graph neural networks, graph attention and causal inference. Through Granger causal inference, a causal relationship network in the preparation process of cobalt-platinum microstructures can be established, so as to have a deeper understanding of the influence of various parameters (such as the ratio of cobalt-platinum ions, solution pH, nozzle scanning speed and deposition voltage) on the atomic ratio of cobalt and platinum, thereby facilitating enhanced control over deposition conditions, and facilitating achieving an optimal ratio of 50:50 between the atoms of cobalt and platinum, which helps to improve the preparation efficiency and accuracy of cobalt-platinum microstructures and improve material properties.

[0008] To achieve the above object, the technical solution adopted by the present invention is:

[0009] In a first aspect, the present invention provides a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network, the method comprising:

[0010] Collect the cobalt-platinum deposition parameters in the preparation of three-dimensional cobalt-platinum microstructures, construct data sets and perform data preprocessing;

[0011] Combine graph neural network, graph attention and Granger causality mechanism to establish graph causal network;

[0012] The pre-processed cobalt-platinum deposition parameters are input into the graph causal network for processing, the atomic ratio of the three-dimensional cobalt-platinum microstructure is inferred, the optimal cobalt-platinum deposition parameters are identified, and the three-dimensional cobalt-platinum microstructure is prepared based on the optimal cobalt-platinum deposition parameters.

[0013] In an optional embodiment, the collected cobalt-platinum deposition parameters include: cobalt-platinum ion ratio, solution pH value, nozzle scanning speed and deposition voltage.

[0014] In an optional embodiment, the preprocessing operation includes: missing value processing, outlier processing and standardization.

[0015] In an optional embodiment, the established graph causal network specifically includes: a graph convolution layer, a graph attention layer, a causal inference module and a prediction layer; when processing the cobalt platinum deposition parameters, the cobalt platinum deposition parameters are converted into a graph structure, each feature is a node of the graph, and there are edges between each node; wherein:

[0016] The graph convolution layer is used to capture the dependencies between nodes and the global structure information of the graph, and aggregate the neighbor information of each node according to the structure of the graph;

[0017] The graph attention layer is used to calculate the importance of neighbors of different nodes to weight the features of aggregated neighbor nodes, and to give high weights to key neighbor nodes through the attention mechanism;

[0018] The causal inference module uses Granger causality to capture the dynamic relationship between features, obtains the causal relationship between input variables and output variables, and obtains causal attention through linear transformation to act on the prediction layer, so that it can infer output variable values ​​that are more in line with causal laws.

[0019] In an optional embodiment, in the graph convolution layer, each node Through its parent node's variables and a noise term To express:

[0020] ;

[0021] in, Is a node The parent node set of All variables that are relevant; is a causal function that indicates how the parent node affects the current node; is the error term, representing unmodeled external factors or noise.

[0022] In an alternative embodiment, the node The status of The layer is represented as , the current status is updated as follows:

[0023] ;

[0024] in, Is a node The set of neighboring nodes represents other variables that affect the node; is the adjacency matrix element, indicating the strength of the relationship between nodes, and They are layer weights and biases, is the activation function; Representation Node All adjacent nodes feature aggregation.

[0025] In an alternative embodiment, the node The status of The layer is represented as , the current status is updated as follows:

[0026] ;

[0027] in, Is a node The set of neighboring nodes represents other variables that affect the node; is the adjacency matrix element, indicating the strength of the relationship between nodes, and They are layer weights and biases, is the activation function; Representation Node All adjacent nodes feature aggregation.

[0028] In the second aspect, the present invention also provides a three-dimensional cobalt-platinum microstructure preparation device, which includes: a working platform, an electrochemical deposition system, a supply device and a control device; when used, the preparation device executes the above-mentioned three-dimensional cobalt-platinum microstructure preparation method based on a graph neural network, infers the atomic ratio of the three-dimensional cobalt-platinum microstructure, identifies the optimal cobalt-platinum deposition parameters, and completes the three-dimensional cobalt-platinum microstructure preparation based on the optimal cobalt-platinum deposition parameters.

[0029] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0030] 1. The present invention provides a method for preparing three-dimensional cobalt-platinum microstructures based on graph neural networks, which can predict the atomic ratio of cobalt and platinum and identify the optimal cobalt-platinum deposition parameters, thereby facilitating enhanced control over deposition conditions, reducing the large number of tests required in traditional experiments, and helping to improve the efficiency of the cobalt-platinum microstructure preparation process.

[0031] 2. The method of the present invention optimizes electrochemical deposition parameters by combining graph neural networks, graph attention and causal inference. Through Granger causal inference, a causal network in the preparation process of cobalt-platinum microstructures can be established, so as to have a deeper understanding of the influence of various parameters (such as cobalt-platinum ion ratio, solution pH, nozzle scanning speed and deposition voltage) on the atomic ratio of cobalt and platinum, thereby facilitating enhanced control over deposition conditions, and facilitating achieving an optimal atomic ratio of 50:50 between cobalt and platinum, which helps to improve the preparation efficiency and accuracy of cobalt-platinum microstructures and improve material properties.

[0032] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0033] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0036] Figure 1 A schematic flow chart of a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network provided in an embodiment of the present invention.

[0037] Figure 2 Schematic diagram of the working principle of the method for preparing three-dimensional cobalt-platinum microstructure based on graph neural network provided in an embodiment of the present invention.

[0038] Figure 3 A schematic diagram of the principle of a graph causal network provided by an embodiment of the present invention.

[0039] Figure 4 A schematic diagram of the structure of a three-dimensional cobalt-platinum microstructure preparation device provided in an embodiment of the present invention.

[0040] Figure 5 A schematic diagram of the effect of preparing a three-dimensional cobalt-platinum microstructure provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0042] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0043] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] See also Figure 1 As shown, an embodiment of the present invention provides a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network, and the method mainly includes:

[0045] 1. Collect the cobalt-platinum deposition parameters in the preparation of three-dimensional cobalt-platinum microstructures, construct a data set and perform data preprocessing;

[0046] 2. Combine graph neural network, graph attention and Granger causality mechanism to establish graph causal network;

[0047] 3. Input the pre-processed cobalt-platinum deposition parameters into the graph causal network for processing, infer the atomic ratio of the three-dimensional cobalt-platinum microstructure, identify the optimal cobalt-platinum deposition parameters, and then complete the three-dimensional cobalt-platinum microstructure preparation based on the optimal cobalt-platinum deposition parameters.

[0048] The present invention proposes a method for preparing three-dimensional cobalt-platinum microstructures based on graph neural networks. Through machine learning optimization, a data-driven method is used to identify the optimal deposition parameters, which provides a reliable method for optimizing the synthesis process of complex materials. This method opens up new possibilities for manufacturing three-dimensional microstructures with complex shapes and structures, and is suitable for fields such as microelectromechanical systems (MEMS), microrobots, and data storage devices. It is conducive to enhancing the control of deposition conditions and achieving excellent material properties.

[0049] The principle and implementation mode of the method of the present invention are described in detail below:

[0050] In order to improve the accuracy and efficiency of the cobalt-platinum microstructure preparation process, the present invention introduces a method combining graph neural network and causal inference to optimize deposition parameters. Through Granger causal inference, the present invention can establish a causal relationship network in the cobalt-platinum microstructure preparation process, so as to have a deeper understanding of the influence of various parameters (such as cobalt-platinum ion ratio, solution pH, nozzle scanning speed and deposition voltage) on the ratio of cobalt and platinum atoms, so as to achieve the optimal ratio of 50:50 for the atomic ratio of cobalt and platinum, which is crucial to improving the magnetic properties of the material. The prepared cobalt-platinum microstructure exhibits high coercivity and remanence comparable to that of bulk cobalt-platinum magnets.

[0051] In an embodiment of the present invention, the optimization process starts with data collection and preprocessing. Experimental data from previous cobalt-platinum deposition experiments were collected, including variables such as the ratio of cobalt-platinum ions, solution pH, nozzle scanning speed, and deposition voltage. The target variable is the atomic ratio of cobalt-platinum. The data is preprocessed to handle missing values ​​and outliers, and standardized to ensure the consistency of the machine learning model. On this basis, the graph causal network helps the present invention model the causal relationship between each input variable and the target output, and clarifies which factors are decisive factors and which are potential influencing factors.

[0052] Some previous studies were mostly based on machine learning regression models, including linear regression (LR), Ridge regression, ElasticNet regression, Lasso regression, random forest (RF), gradient boosting regressor (GBR), extreme gradient boosting (XGB), Extra Trees regressor (ETR) and AdaBoost regressor (ADA). However, due to the high cost of obtaining experimental data, the amount of data is usually small, and there is noise disturbance between the parameters and results recorded in the experiment, it is difficult for machine learning models to learn the essential connection between experimental parameters and experimental results, and they are prone to overfitting and cannot effectively infer the optimal experimental parameters that have not been seen.

[0053] In the method of the present invention, the graph neural network (GNN) is combined with the Granger causal mechanism, which has significant advantages over traditional machine learning methods (such as Extra Trees) in processing complex relationships and structured data. GNN can naturally capture high-order interactions and dynamic dependencies in graph data, which are often difficult to achieve with traditional methods. By introducing the Granger causal mechanism, the model further improves the interpretability, can identify the causal relationship between features, and reduce the impact of false correlations and irrelevant dependencies. This combination effectively realizes representation learning, integrates spatial and dynamic characteristics with causal information, and significantly improves prediction accuracy, generalization ability for unseen scenes, and the ability to interpret potential patterns of data. Compared with the limitations of traditional models that are highly dependent on feature engineering and difficult to model nonlinear graph structure relationships, this framework provides a scalable and interpretable solution for complex learning tasks, and can give hypothetical suggestions through reverse causal generation based on the expected output.

[0054] In the embodiment of the present invention, the machine learning algorithm includes the following contents:

[0055] In order to improve the accuracy and efficiency of the cobalt-platinum microstructure preparation process, the present invention introduces a graph causal network to optimize deposition parameters, including the cobalt-platinum ion ratio, solution pH, nozzle scanning speed, and deposition voltage. These parameters affect the final atomic ratio of cobalt to platinum, and there is a complex causal relationship between these variables. Through the graph causal network, the present invention can accurately model the dependency between these variables in the causal graph, thereby achieving more accurate process control and parameter optimization.

[0056] The goal of the graph causal network is to predict the quality of the Co-Pt microstructure based on the input deposition parameters through causal reasoning. (such as solution pH, deposition voltage, etc.) can be obtained through the variables of its parent node and a noise term To express:

[0057] ;

[0058] in, Is a node The parent node set of All variables related, for example, the relationship between the ratio of cobalt to platinum ions and the pH of the solution. is a causal function that indicates how the parent node affects the current node. For example, the pH of the solution may affect the ratio of cobalt and platinum ions through a nonlinear function. is the error term, representing unmodeled external factors or noise.

[0059] Graph Causal Network combines Graph Neural Network (GNN) and causal inference. In Graph Neural Network, information is propagated through the graph structure, and the update of each node is based on the information of its neighbor nodes and the current hidden state. Through the reasoning of multi-layer Graph Neural Network (GNN), the model can iteratively update the state of each node, thereby learning the causal relationship between nodes.

[0060] In a graph neural network, suppose a node , its status is The layer is represented as , the update rules of the current state are as follows:

[0061] ;

[0062] in, Is a node The set of neighboring nodes represents other variables that affect the node; is the adjacency matrix element, indicating the strength of the relationship between nodes, and They are layer weights and biases, is the activation function; Representation Node All adjacent nodes By iterating between multiple layers of the network, the graph neural network learns the causal relationship between nodes and uses the causal relationship to infer the ideal output.

[0063] In the embodiment of the present invention, the optimization goal is to maximize the consistency between the cobalt-platinum microstructure and the true output, so the model parameters (specifically including the network parameters of the graph convolution layer, the graph attention layer, and the fully connected layer of the final regression parameters) are adjusted by minimizing the mean square error and the absolute value error, thereby improving the accuracy of the prediction. The optimization process can be expressed as:

[0064] ;

[0065] in, represents the parameters of the model, Express expectations, represents the parameter-based model forward inference function, represents the input features, represents the true label.

[0066] After training, all possible parameters are searched at the granularity level allowed by the experiment to obtain the optimal cobalt-platinum atomic ratio, thereby realizing the search for the optimal parameters based on logical inference.

[0067] When training a graph neural network, the training data first needs to be converted into a graph structure, with each feature being treated as a node of the graph, and edges between each node. In order not to miss the possible causal relationship between features, a fully connected graph is constructed in the present invention, with an edge between each node. In order to control variables in the experiment, the input feature parameters are fixed discrete categories, so the data labels need to be uniquely encoded.

[0068] In a specific embodiment, Figure 2 and Figure 3 As shown in the figure, the specific structure of the graph causal network includes: graph convolution layer, graph attention layer, causal inference module and output layer. Among them:

[0069] Graph convolution automatically aggregates the neighbor information of each node according to the structure of the graph. For the task of analyzing the microstructure of cobalt platinum, the input features themselves are discrete categories, and the relationship between each node may be very complex. The graph convolution layer can propagate information through the graph structure, capture the dependencies between nodes and the global structural information of the graph, and improve the representation between features.

[0070] The graph attention mechanism weights and aggregates the features of neighboring nodes by calculating the importance of neighbors of different nodes, rather than simply summing them up, so that the model can adaptively focus on the neighboring nodes that are most important for target prediction. In complex graph data, some neighboring nodes may contribute much more to the prediction results than other nodes. Graph attention gives these key neighbors higher weights through the attention mechanism, thereby improving the accuracy of the aggregation process.

[0071] Traditional graph neural networks often only focus on the similarity and adjacency between node features, and do not necessarily explicitly consider the causal relationship between node features. However, the features in many practical problems are not simply interrelated, but have a clear causal mechanism. Therefore, in the causal module, Granger causality is used to capture the dynamic association between features. It is based on the assumption that if the past value of a time series can help predict the current value of another time series, then it can be considered that the first sequence "Granger causes" the change of the second sequence. Although Granger causality is usually used for time series data, the present invention can still draw on its ideas to process non-time series data. In this task, the input features are discrete category labels, and the historical information is calculated with the current information by analogy with the lagged features to evaluate the relationship between these interactive features and the target variable, which is closer to the logic of humans inferring the relationship between variables based on historical information during the experiment.

[0072] The Granger causality hypothesis is as follows: if the variable For variables have a causal relationship (these two variables are the input features and the true labels described above), then Past values ​​of can help predict The current value of , otherwise the causal relationship does not hold.

[0073] The causal model assumes the following formula:

[0074] ;

[0075] in, yes time The value of yes time The value of and is the number of delayed terms, and represents the time term of Granger causality analysis, and correspond and The value of is the error term, represents the constant term, i.e., the underlying offset that cannot be explained by the causal variable, represents the coefficient of the lagged dependent variable, reflecting the impact of past output on current output, The coefficient of the lagged independent variable reflects the impact of past input on current output;

[0076] To test Is Granger caused , the non-causal model formula is:

[0077] ;

[0078] Non-causal models do not consider variables For variables If the residual sum of squares of the causal model is significantly smaller than that of the non-causal model, it indicates that the variable For variables There is Granger causality. Whether there is a causal relationship can be determined by F test or other statistical test methods.

[0079] After the Granger causal inference module, the causal relationship between the input variable and the output variable is obtained. After linear transformation, the causal attention is applied to the prediction layer to help infer the output variable value that is more in line with the causal law.

[0080] After multiple rounds of iterations, the model learns the causal characteristics between disordered data features, which is equivalent to using neural networks to help experimenters analyze data patterns. Eventually, the model converges to obtain the optimal causal logical relationship under the current data set, and infers and analyzes the theoretically optimal experimental parameters based on this relationship.

[0081] Furthermore, in the embodiment of the present invention, five-fold cross validation is also used to verify the model performance. Experimental data, including parameters such as solution pH, nozzle scanning speed, deposition voltage, etc., are used to predict the Co:Pt atomic ratio. The verification results are shown in Table 1:

[0082] Table 1: Model performance in five-fold cross validation

[0083]

[0084] As can be seen from Table 1, according to the mean square error (MSE), mean absolute error (MAE) and determination coefficient ( ) The performance of each model is evaluated, and the present invention performs best. It should be noted that: in the general linear equation determination coefficient calculation formula, , the closer it is to 1, the better the fitting degree is, and there is no negative value domain. The model in this paper belongs to a nonlinear neural network model, in which The calculation formula is , when the residual sum of squares is greater than the total variance sum of squares, negative values ​​may appear. The total variance sum of squares is the sum of the squares of the difference between the predicted result and the true result of each predicted value, while the residual sum of squares is the sum of the squares of the difference between the original observed data and the mean of the observed data. For some unevenly distributed data, the offset of the numerator term is easily magnified, so negative values ​​appear. .

[0085] From the description of the above embodiments, those skilled in the art can know that: the present invention provides a method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network, which is used to model electrochemical deposition parameters and predict the atomic ratio of cobalt and platinum. The method uses a data-driven method to identify the optimal deposition parameters, significantly reducing the large number of experiments required in traditional experiments, and improving the efficiency of the cobalt-platinum microstructure preparation process. The method infers the atomic ratio of a three-dimensional cobalt-platinum microstructure through a graph neural network, graph attention, Granger causal attention and a nonlinear prediction layer, provides a mathematical basis for optimizing the optimal deposition parameters, improves the interpretability of the neural network in simulating the electrochemical deposition process, and promotes the realization of an ideal ratio of 50:50 between the atoms of cobalt and platinum, which is conducive to improving the potential for the preparation of high-performance cobalt-platinum microstructures, which is crucial for the application of MEMS, micro-robots and data storage devices.

[0086] Further, see Figure 4As shown, the present invention also provides a three-dimensional cobalt-platinum microstructure preparation device, which includes: a working platform, an electrochemical deposition system, a supply device and a control device; when the preparation device is used, a three-dimensional cobalt-platinum microstructure preparation method based on a graph neural network described in the above embodiment is executed to infer the atomic ratio of the three-dimensional cobalt-platinum microstructure, identify the optimal cobalt-platinum deposition parameters, and complete the three-dimensional cobalt-platinum microstructure preparation based on the optimal cobalt-platinum deposition parameters.

[0087] In this example, by using electrolytes and micronozzles to control the deposition process, the prepared cobalt-platinum microstructures exhibit high coercivity and remanence comparable to bulk cobalt-platinum magnets. The integration of hardware and software ensures precise control of the deposition process, which is divided into three main components: mechanical displacement, electrochemical deposition, and deposition pressure control. Communication between modules is achieved through the pyserial serial communication method.

[0088] For the cobalt-platinum alloy magnetic structure, a graphite electrode and an ITO conductive glass electrode were used as the anode and cathode, respectively. The electrolyte solution contained cobalt and platinum ions specially formulated for the deposition process.

[0089] Combining the comprehensive theoretical derivation of the thermodynamics and kinetics of the electrochemical deposition process, as well as rigorous simulation and experimental verification, the electromechanical system and its parameter range for 3D printing micromagnetic structures were established. The process of realizing the electromechanical system of the micromagnetic structure 3D printing ultimately forms a predetermined magnetic pattern, with effects such as Figure 5 As shown, Figure 5 Part (a) is the sample preparation route planning diagram. Figure 5 Part (b) in the middle is the first deposition diagram of the micromagnetic structure 3D printing. Figure 5 Part (c) is a preview of all deposition modes.

[0090] An embodiment of the present invention provides a three-dimensional cobalt-platinum microstructure preparation device, whose implementation principle and technical effects are the same as those of the aforementioned method embodiment. For the sake of brief description, parts not mentioned in this embodiment can be referred to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0091] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0092] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0093] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for preparing three-dimensional cobalt-platinum microstructure based on graph neural network, characterized in that: The method includes: Collect the cobalt-platinum deposition parameters in the preparation of three-dimensional cobalt-platinum microstructures, construct data sets and perform data preprocessing; A graph causal network is established by combining graph neural network, graph attention and Granger causality mechanism; the established graph causal network specifically includes: a graph convolution layer, a graph attention layer, a causal inference module and a prediction layer; when processing cobalt platinum deposition parameters, the cobalt platinum deposition parameters are converted into a graph structure, each feature is a node of the graph, and there are edges between each node; wherein: The graph convolution layer is used to capture the dependencies between nodes and the global structure information of the graph, and aggregate the neighbor information of each node according to the structure of the graph; The graph attention layer is used to calculate the importance of neighbors of different nodes to weight the features of aggregated neighbor nodes, and to give high weights to key neighbor nodes through the attention mechanism; The causal inference module uses Granger causality to capture the dynamic association between features, obtains the causal relationship between input variables and output variables, and obtains causal attention through linear transformation to act on the prediction layer, so that it can infer the output variable value that is more in line with the causal law; The pre-processed cobalt-platinum deposition parameters are input into the graph causal network for processing, the atomic ratio of the three-dimensional cobalt-platinum microstructure is inferred, the optimal cobalt-platinum deposition parameters are identified, and the three-dimensional cobalt-platinum microstructure is prepared based on the optimal cobalt-platinum deposition parameters.

2. A method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network according to claim 1, characterized in that: The collected cobalt-platinum deposition parameters include: cobalt-platinum ion ratio, solution pH, nozzle scanning speed and deposition voltage.

3. The method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network according to claim 1, characterized in that: Preprocessing operations include: missing value processing, outlier processing and standardization.

4. The method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network according to claim 1, characterized in that: In the graph convolution layer, each node X i Through its parent node variables and a noise term ∈ i To express: X i =f i (Pa(X i ))+∈ i Among them, Pa(X i ) is node X i The parent node set of X i All relevant variables; f i (·) is a causal function that indicates how the parent node affects the current node; ∈ i is the error term, representing unmodeled external factors or noise.

5. A method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network according to claim 4, characterized in that: Node X i The state of at layer n is represented as The current status is updated as follows: in, is node X i A is a set of neighboring nodes, which represents other variables that affect the node; ik is the adjacency matrix element, indicating the strength of the relationship between nodes, W n and b n are the weight and bias of the nth layer, and σ is the activation function; Represents node X i Aggregation of features of all neighboring nodes k.

6. A method for preparing a three-dimensional cobalt-platinum microstructure based on a graph neural network according to claim 5, characterized in that: The Granger causality hypothesis adopted by the causal inference module is: if variable X has a causal relationship with variable Y, then the past value of X can help predict the current value of Y, otherwise the causal relationship does not hold; specifically, it is judged by a causal model and a non-causal model, where the causal model formula is: Among them, Y t is the value of Y at time t, X t is the value of X at time t, p and q are the number of delay terms, i and j represent the time terms of Granger causality analysis, X t-j and Y t-i Corresponding values ​​of tj and ti; ∈ t is the error term, α is the constant term, i.e., the underlying offset that cannot be explained by the causal variable, and β i represents the coefficient of the lagged dependent variable, reflecting the impact of past output on current output, γ j It represents the coefficient of the lagged independent variable, reflecting the impact of past input on current output. In order to test whether X Granger causes Y, the non-causal model formula is: The non-causal model does not consider the causal effect of variable X on variable Y; if the residual sum of squares of the causal model is significantly smaller than that of the non-causal model, it indicates that variable X has a Granger causal relationship with variable Y.

7. A three-dimensional cobalt-platinum microstructure preparation device, characterized in that: The preparation device includes: a working platform, an electrochemical deposition system, a supply device and a control device; when used, the preparation device executes a three-dimensional cobalt-platinum microstructure preparation method based on a graph neural network as described in any one of claims 1-6, infers the atomic ratio of the three-dimensional cobalt-platinum microstructure, identifies the optimal cobalt-platinum deposition parameters, and completes the three-dimensional cobalt-platinum microstructure preparation based on the optimal cobalt-platinum deposition parameters.

Images