Screening method for rare earth giant magnetostrictive materials based on machine learning

Abstract

The present invention discloses a method for screening rare earth giant magnetostrictive materials based on machine learning. The method comprises the following steps: obtaining the saturation magnetostriction coefficient of a magnetic material, establishing a dataset of multi-component material compositions and corresponding saturation magnetostriction coefficients, and determining the material's characteristic attributes and target attributes; dividing the dataset into a training set and a test set, constructing a machine learning model based on the training set, and validating the model using the test set to ultimately obtain an optimal model for predicting the material's saturation magnetostriction coefficient; pre-designing the composition ratios of each element in the multi-component magnetostrictive material, establishing a virtual magnetostrictive material library, and predicting the saturation magnetostriction coefficients of each composition in the virtual material library based on the optimized machine learning model. The method then screens the composition regions and boundaries of high-performance rare earth magnetostrictive materials. The present invention provides a necessary solution for the composition design of rare earth giant magnetostrictive materials.

Description

Technical Field

[0001] The present invention relates to the field of material property prediction research, and in particular to a rare earth giant magnetostrictive material screening method based on machine learning. Background Art

[0002] Rare earth giant magnetostrictive materials, due to their excellent magnetostrictive properties, high energy density, and high output power, have become the preferred materials for devices such as sensors, actuators, transducers, and sonar. In these rare earth magnetostrictive material devices, if the material's saturation magnetostriction coefficient is low, it often leads to low response sensitivity and energy conversion efficiency, limiting the application range of magnetostrictive devices. In addition, even if a high-performance magnetostrictive material composition is designed, composition segregation often occurs during the preparation process, causing the material's magnetostriction coefficient to deviate from the ideal value. To this end, we not only need to screen out material composition points with high saturation magnetostriction coefficients, but also need to determine the composition regions formed by these high-performance composition points and their boundaries. However, the traditional "trial and error" material screening model not only has a long R&D cycle and high costs, but also requires all R&D personnel to have extensive knowledge, capabilities, and practical experience, which poses a challenge to quickly finding the composition regions of high-performance magnetostrictive materials.

[0003] In response to the problems existing in the field of rare earth magnetostrictive materials mentioned above, it is urgently necessary to propose a method that can predict the material composition region and its boundaries with excellent magnetostrictive properties, so as to greatly reduce the research and development cost of magnetostrictive materials and significantly shorten the magnetostrictive material mining cycle, thereby strongly supporting the rapid development of magnetostrictive devices. Summary of the Invention

[0004] The present invention aims to provide a machine learning-based method for screening rare earth giant magnetostrictive materials. Based on magnetostriction coefficient data for multi-component rare earth magnetic materials, a machine learning model is developed to predict the relationship between each component's composition and magnetostrictive performance. This method identifies the compositional range and boundaries of materials with excellent magnetostrictive properties, thereby avoiding magnetostrictive performance degradation caused by component segregation during the preparation process. This method also effectively reduces R&D costs and shortens the development cycle.

[0005] The technical solution adopted by the present invention is: a method for screening rare earth giant magnetostrictive materials based on machine learning, comprising:

[0006] Step 1: Establish the composition of multi-component Laves phase RM2-based rare earth magnetostrictive materials and the corresponding saturation magnetostriction coefficient λ s dataset;

[0007] Step 2: Divide the dataset into a training set and a test set, build a machine learning model based on the training set, and verify it with the test set to ultimately obtain the optimal model that can be used to predict the magnetostriction coefficient of the material;

[0008] Step 3: Pre-design the composition ratio of each element in the multi-component Laves phase RM2-based rare earth magnetostrictive material, establish a virtual rare earth magnetostrictive material library, predict the saturation magnetostriction coefficient of the materials in the virtual material library based on the optimized machine learning model, and determine the high magnetostriction coefficient material composition area and its boundary.

[0009] In step 1, it can be specifically divided into the following steps:

[0010] Step 1-1: Obtain data on the composition and corresponding saturation magnetostriction coefficient of a multi-component Laves phase RM2-based rare earth magnetostrictive material;

[0011] Step 1-2: Build a database of Laves phase RM2-based rare earth magnetostrictive material compositions and corresponding saturation magnetostriction coefficients;

[0012] Step 1-3: Use the material elements in the database as characteristic attributes, and use the composition percentage of the characteristic element in the material as the characteristic value. If there is no characteristic element, the characteristic value is zero. At the same time, the saturation magnetostriction coefficient under different compositions is used as the target attribute of the material.

[0013] The step 2 can be specifically divided into the following steps:

[0014] Step 2-1: Divide the dataset into a training set and a test set. First, select 80% of the data in the dataset as the training set for building and training the machine learning model. Then, use the remaining 20% ​​of the data in the dataset as the test set to evaluate the constructed machine learning model.

[0015] Step 2-2: Determine the model performance evaluation index, that is, use the determination coefficient R 2 And mean absolute error MAE are used to evaluate the accuracy of the model; the determination coefficient R 2 for:

[0016]

[0017]

[0018] SST=SSE+SSR

[0019] Where y i represents the true value of the data, represents the mean of the true values, f iRepresents the predicted value of the model, SST is the total deviation sum of squares, SSE is the residual sum of squares, SSR is the regression sum of squares; the mean absolute error MAE is:

[0020]

[0021] Step 2-3: Use the training set to train the model, and adjust the hyperparameters of the machine learning model through grid partitioning and five-fold cross validation. 2 And mean absolute error MAE to evaluate the model performance under different hyperparameters;

[0022] Step 2-4: Comprehensively consider the coefficient of determination R of the model under different hyperparameters 2 And mean absolute error MAE, select the best model hyperparameters;

[0023] Step 2-5: Evaluate the model selected in step 2-4 using the test set data, comparing the performance of the model on the training set and the test set;

[0024] Steps 2-6: Develop solutions for different performance scenarios. First, determine hyperparameters such as the number of decision trees, the depth of a single decision tree, the minimum number of leaf node samples, and the learning rate. If the model overfits, adjust the hyperparameters, especially reducing the number and depth of decision trees, to reduce the degree of model fit and thus improve the model's generalization ability. If the model underfits, increase the number and depth of decision trees to improve the model's fit.

[0025] Step 2-7: Observe the model's performance on the training and test sets, and apply the solution developed in step 2-6 to further adjust the hyperparameters to ensure that the model not only has high accuracy but also has good generalization ability. Ultimately, the optimal machine learning model for predicting the material's saturation magnetostriction coefficient is obtained.

[0026] The step 3 can be specifically divided into the following steps:

[0027] Step 3-1: Pre-design the composition ratio of each element in the multi-component Laves phase RM2-based rare earth magnetostrictive material and establish a virtual magnetostrictive material library;

[0028] Step 3-2: Use the optimal machine learning model for predicting the saturation magnetostriction coefficient of the material established in step 2-7 to predict the magnetostrictive properties of each component material in the virtual material library in step 3-1;

[0029] Step 3-3: Perform data analysis and visualization on the results predicted in step 3-2 to find the material composition region with high magnetostriction coefficient and its boundary.

[0030] Beneficial effects of the present invention:

[0031] The present invention provides a method for screening rare earth giant magnetostrictive materials based on machine learning. It can accurately and quickly predict the saturation magnetostriction coefficient of each component of a rare earth magnetostrictive material and determine the composition range and boundaries of materials with excellent magnetostrictive properties.

[0032] In the embodiment of the present invention, an optimal machine learning model that can predict the correlation between each component composition and magnetostrictive performance is obtained. The model determination coefficient is above 0.9 on the training set and the test set, and the model shows high accuracy and good generalization ability. At the same time, the model is used to predict the saturation magnetostriction coefficient of each component material in the RFe2 (R = Tb, Dy, Pr and other rare earth elements) virtual material library. It is determined that when the Tb composition range is 0.26-0.53 and the Fe composition range is 1.92-1.97, Tb x Dy 1-x Fe y λ S In particular, when the Tb content is in the range of 0.28 to 0.34 and the Fe content is in the range of 1.94 to 1.97, the Tb x Dy 1-x Fe y λ S All are greater than 1700ppm, showing excellent magnetostrictive properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a method for screening rare earth giant magnetostrictive materials based on machine learning according to an embodiment of the present invention;

[0034] Figure 2 The material composition range and boundary diagram with excellent magnetostrictive properties determined for the embodiments of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form.

[0036] like Figure 1 As shown, in this embodiment, the prediction method is specifically used to predict the saturation magnetostriction coefficient of RFe2 (R = rare earth elements such as Tb, Dy, Pr) based rare earth magnetostrictive materials. By completing steps 1-3 below, the saturation magnetostriction coefficient of the RFe2 based rare earth magnetostrictive material is predicted. The specific steps are as follows:

[0037] Step 1: Establish the composition of multi-component RFe2-based rare earth magnetostrictive materials and the corresponding saturation magnetostriction coefficient λ s dataset;

[0038] The specific steps are as follows:

[0039] Step 1-1: Obtain data on the composition and corresponding saturation magnetostriction coefficient of a multi-component RFe2-based rare earth magnetostrictive material;

[0040] Step 1-2: Build a database of RFe2-based rare earth magnetostrictive material compositions and corresponding saturation magnetostriction coefficients;

[0041] Step 1-3: Use the material elements in the database as characteristic attributes, and use the composition percentage of the characteristic element in the material as the characteristic value. If there is no characteristic element, the characteristic value is zero. At the same time, the saturation magnetostriction coefficient under different compositions is used as the target attribute of the material.

[0042] Step 2: Divide the dataset into a training set and a test set, build a machine learning model based on the training set, and verify it with the test set to ultimately obtain the optimal model that can be used to predict the magnetostriction coefficient of the material;

[0043] The specific steps are as follows:

[0044] Step 2-1: Divide the dataset into a training set and a test set. First, select 80% of the data in the dataset as the training set for building and training the machine learning model. Then, use the remaining 20% ​​of the data in the dataset as the test set to evaluate the constructed machine learning model.

[0045] Step 2-2: Determine the model performance evaluation index, that is, use the determination coefficient R 2 And mean absolute error MAE are used to evaluate the accuracy of the model; the determination coefficient R 2 for:

[0046]

[0047]

[0048] SST=SSE+SSR

[0049] Where y i represents the true value of the data, f i Represents the predicted value of the model, SST is the total deviation sum of squares, SSE is the residual sum of squares, SSR is the regression sum of squares; the mean absolute error MAE is:

[0050]

[0051] Step 2-3: Use the training set to train the model, and adjust the hyperparameters of the machine learning model through grid partitioning and five-fold cross validation. 2 And mean absolute error MAE to evaluate the model performance under different hyperparameters;

[0052] Step 2-4: Comprehensively consider the coefficient of determination R of the model under different hyperparameters 2 And mean absolute error MAE, select the best model hyperparameters;

[0053] Step 2-5: Evaluate the model selected in step 2-4 using the test set data, comparing the performance of the model on the training set and the test set;

[0054] Steps 2-6: Develop solutions for different performance scenarios. First, determine hyperparameters such as the number of decision trees, the depth of a single decision tree, the minimum number of leaf node samples, and the learning rate. If the model overfits, adjust the hyperparameters, especially reducing the number and depth of decision trees, to reduce the degree of model fit and thus improve the model's generalization ability. If the model underfits, increase the number and depth of decision trees to improve the model's fit.

[0055] Step 2-7: Observe the model's performance on the training and test sets, and apply the solution developed in step 2-6 to further adjust the hyperparameters to ensure that the model not only has high accuracy but also has good generalization ability. Ultimately, the optimal machine learning model for predicting the material's saturation magnetostriction coefficient is obtained.

[0056] Step 3: Pre-design the composition ratio of each element in the multi-component RFe2-based rare earth magnetostrictive material, establish a virtual magnetostrictive material library, predict the saturation magnetostriction coefficient of the materials in the virtual material library based on the optimized machine learning model, and determine the high magnetostriction coefficient material composition area and its boundary.

[0057] The specific steps are as follows:

[0058] Step 3-1: Pre-design the composition ratio of each element in the multi-component RFe2-based rare earth magnetostrictive material and establish a virtual magnetostrictive material library;

[0059] Step 3-2: Use the optimal machine learning model for predicting the saturation magnetostriction coefficient of the material established in step 2-7 to predict the magnetostrictive properties of each component material in the virtual material library in step 3-1;

[0060] Step 3-3: Perform data analysis and visualization on the results predicted in step 3-2 to find the material composition region with high magnetostriction coefficient and its boundary.

[0061] Depend on Figure 2 It can be seen that the present invention predicts that the embodiment Tb x Dy 1-x Fe y The saturation magnetostriction coefficient of each component material in the virtual material library is determined to be 0.26-0.53 when the Tb component ranges from 0.26 to 0.53 and the Fe component ranges from 1.92 to 1.97. x Dy 1-x Fe y λ S In particular, when the Tb content is in the range of 0.28 to 0.34 and the Fe content is in the range of 1.94 to 1.97, the Tb content in the embodiment x Dy 1-x Fe y λ S All are greater than 1700ppm, showing excellent magnetostrictive properties.

[0062] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for screening rare earth giant magnetostrictive materials based on machine learning, characterized in that: include: Step 1: Establish the composition of multi-component Laves phase RM2-based rare earth magnetostrictive materials and the corresponding saturation magnetostriction coefficient λ s dataset; Step 2: Divide the dataset into a training set and a test set, build a machine learning model based on the training set, and verify it with the test set to ultimately obtain the optimal model that can be used to predict the magnetostriction coefficient of the material; Step 3: Pre-design the composition ratio of each element in the multi-component Laves phase RM2-based rare earth magnetostrictive material, establish a virtual rare earth magnetostrictive material library, predict the saturation magnetostriction coefficient of the materials in the virtual material library based on the optimized machine learning model, and determine the high magnetostriction coefficient material composition region and its boundary; In step 2, it is specifically divided into the following steps: Step 2-1: Divide the established dataset into a training set and a test set. First, select 80% of the data in the dataset as the training set for building and training the machine learning model; then use the remaining 20% ​​of the data in the dataset as the test set to evaluate the constructed machine learning model. Step 2-2: Determine the model performance evaluation index, that is, use the determination coefficient R 2 And mean absolute error MAE are used to evaluate the accuracy of the model; the determination coefficient R 2 for: SST=SSE+SSR Where y i represents the true value of the data, represents the mean of the true values, f i Represents the predicted value of the model, SST is the total deviation sum of squares, SSE is the residual sum of squares, SSR is the regression sum of squares; the mean absolute error MAE is: Step 2-3: Use the training set to train the model, and adjust the hyperparameters of the machine learning model through grid partitioning and five-fold cross validation. 2 And mean absolute error MAE to evaluate the model performance under different hyperparameters; Step 2-4: Comprehensively consider the coefficient of determination R of the model under different hyperparameters 2 And mean absolute error MAE, select the best model hyperparameters; Step 2-5: Evaluate the model selected in step 2-4 using the test set data, comparing the performance of the model on the training set and the test set; Steps 2-6: Develop solutions for different performance scenarios. First, determine hyperparameters such as the number of decision trees, the depth of a single decision tree, the minimum number of leaf samples, and the learning rate. If the model overfits, adjust the hyperparameters, especially reducing the number and depth of decision trees, to reduce the degree of model fit and thus improve the model's generalization ability. If the model underfits, increase the number and depth of decision trees to improve the model's fit. Step 2-7: Observe the model's performance on the training and test sets, and apply the solution developed in step 2-6 to further adjust the hyperparameters to ensure that the model not only has high accuracy but also has good generalization ability. Ultimately, the optimal machine learning model for predicting the material's saturation magnetostriction coefficient is obtained.

2. The method for screening rare earth giant magnetostrictive materials based on machine learning according to claim 1, characterized in that: In step 1, it is specifically divided into the following steps: Step 1-1: Obtain data on the composition and corresponding saturation magnetostriction coefficient of a multi-component Laves phase RM2-based rare earth magnetostrictive material; Step 1-2: Build a database of Laves phase RM2-based rare earth magnetostrictive material compositions and corresponding saturation magnetostriction coefficients; Steps 1-3: Use the elements of the material in the database as characteristic attributes, and use the percentage of the element in the material as the characteristic value. If the element is not present, the characteristic value is zero. At the same time, the saturation magnetostriction coefficient under different compositions is used as the target attribute of the material.

3. The method for screening rare earth giant magnetostrictive materials based on machine learning according to claim 1, characterized in that: In step 3, it is specifically divided into the following steps: Step 3-1: Pre-design the composition ratio of each element in the multi-component Laves phase RM2-based rare earth magnetostrictive material and establish a virtual magnetostrictive material library; Step 3-2: Use the optimal machine learning model for predicting the saturation magnetostriction coefficient of the material established in step 2-7 to predict the magnetostrictive properties of each component material in the virtual material library in step 3-1; Step 3-3: Perform data analysis and visualization on the results predicted in step 3-2 to find the material composition region with high magnetostriction coefficient and its boundary.

Images