A reverse design method and system for alloy composition based on knowledge distillation

Abstract

This invention discloses a method and system for reverse alloy composition design based on knowledge distillation, relating to the fields of materials genome and alloy reverse design. The method includes: collecting alloy composition data and corresponding performance data to obtain a dataset; calculating the atomic characteristics of each alloy in the dataset; constructing and training a supervised learning model based on the atomic characteristics and performance data to establish a mapping relationship between alloy atomic characteristics and performance; training a deep neural network model based on the composition data and performance data to construct a mapping relationship between alloy composition and performance; adjusting network parameters using a backpropagation algorithm during the deep neural network model training process to minimize the prediction difference between the deep neural network model and the supervised learning model; performing sampling optimization and optimizing the composition using the gradient information of the deep neural network model to find the optimal alloy composition. This invention combines the composition and atomic characteristics of alloys to analyze the composition of high-performance alloys, realizing customized material design and optimization.

Description

Technical Field

[0001] This invention relates to the fields of materials genes and alloy reverse design, and more specifically to a method and system for reverse design of alloy composition based on knowledge distillation. Background Technology

[0002] Compared to single metals, alloys often exhibit superior physical, chemical, and mechanical properties, such as high hardness, heat resistance, and corrosion resistance. These properties make alloy materials widely used in various industries, including aerospace, energy, automotive manufacturing, and electronic equipment.

[0003] However, traditional alloy design relies on experience and trial-and-error methods, which are not only time-consuming and costly but also struggle to fully explore the complex relationship between material properties and composition. In contrast, machine learning algorithms can analyze data, learn and predict the intrinsic correlation between alloy composition and its properties, thereby efficiently screening alloy materials with desired performance. However, a significant challenge in alloy design is how to reverse-engineer the composition based on the alloy's design target performance—the so-called "reverse design." Faced with the vast material space of multi-element alloys, resolving the compositional composition of potential high-performance alloys through reverse design methods is crucial for achieving customized material design and optimization.

[0004] Therefore, how to develop a reverse design method for alloy composition is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a reverse design method and system for alloy composition based on knowledge distillation, aiming to build an efficient method for predicting alloy performance and reverse designing composition. By combining knowledge distillation technology with machine learning models, it provides strong technical support for the design of alloy materials and realizes the accurate design of high-performance alloy composition.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention discloses a reverse design method for alloy composition based on knowledge distillation, the specific steps of which are as follows:

[0008] Step 1: Collect chemical composition data and corresponding performance data of multi-element alloys, organize them to obtain a multi-element alloy dataset, and calculate the atomic characteristics of each alloy in the multi-element alloy dataset.

[0009] Step 2: Based on the atomic characteristics and performance data of each alloy, construct and train a supervised learning model to establish the mapping relationship between alloy atomic characteristics and performance.

[0010] Step 3: Based on the chemical composition data and performance data of each alloy, train a deep neural network model to construct the mapping relationship from alloy composition to performance; during the training process of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters to minimize the prediction difference between the deep neural network model and the supervised learning model;

[0011] Step 4: Use Langevin Dynamics for sampling optimization and use the gradient information of the deep neural network model for composition optimization to find the optimal alloy composition.

[0012] Furthermore, each alloy in the multi-element alloy dataset is represented as X. i = [x1,x2,……,x m ], satisfying ∑x j =1; the corresponding performance index is represented by Y. i ; where X i Let X be a component vector representing the i-th alloy component in the dataset, where i = 1, 2, ..., N, and N is the total number of alloys in the dataset. i Composed of m elements, x j This represents the molar percentage of the j-th element.

[0013] Furthermore, the atomic characteristics include: average electronegativity, average ionization energy, and atomic size difference.

[0014] Furthermore, the mapping relationship between the atomic characteristics of the alloy and its properties is expressed as follows: in Let f(x) represent the predicted value of the supervised learning model, and let f(x) represent the supervised learning model.

[0015] The mapping relationship between alloy composition and performance is expressed as follows: in Let g(x) represent the predicted value of the deep neural network model; the deep neural network model is a four-layer fully connected network, with the input dimension being the number of alloy elements in the dataset, and the output being the target performance.

[0016] Furthermore, step 4 specifically includes:

[0017] Step 4.1: Determine the sampling starting point based on the alloy composition to be optimized;

[0018] Step 4.2: Input the composition vector of the alloy to be optimized into the deep neural network model to obtain the predicted value;

[0019] Step 4.3: Calculate the gradient information of the current predicted value, and calculate the optimized alloy composition sampling points based on the gradient information. If the number of samplings is less than a preset threshold, return to step 4.2; otherwise, proceed to step 4.4.

[0020] Step 4.4: Calculate the error between the predicted values ​​of the supervised learning model and the predicted values ​​of the deep neural network model for all sampling points; determine whether the stopping condition is met. If it is met, select the optimal alloy composition from all sampling points; otherwise, add the chemical composition data of all sampling points and the corresponding predicted values ​​of the supervised learning model as new samples to the multivariate alloy dataset, and return to step 3.

[0021] Furthermore, the sampling point is represented as X. i,k = [x1,x2,…x j …,x m ], where X i,k This represents the composition vector of the i-th alloy sampled in the dataset at the k-th sampling time, composed of m elements, x j This represents the molar percentage of the j-th element.

[0022] The formula for calculating and optimizing the alloy composition sampling points is: X i,k+1 =X i,k +λ*grad; where λ>0 is the learning rate, and grad represents the gradient information. express For x j The partial derivatives of .

[0023] Furthermore, the error is expressed as in This represents the predicted value of the supervised learning model. This represents the predicted value of the deep neural network model;

[0024] The stopping condition is: there exists a sampling point X. i,k+1 ,satisfy and The target performance requirement is achieved or the maximum number of iterations is reached, where ε represents the error factor of the supervised learning model and the deep neural network model.

[0025] This invention also discloses a reverse design system for alloy composition based on knowledge distillation, comprising:

[0026] Preprocessing module: Collects chemical composition data and corresponding performance data of multi-element alloys, organizes them to obtain multi-element alloy dataset, and calculates the atomic characteristics of each alloy in the multi-element alloy dataset;

[0027] Supervised learning module: Based on the atomic characteristics and performance data of each alloy, a supervised learning model is constructed and trained to establish the mapping relationship between alloy atomic characteristics and performance;

[0028] Deep learning module: Based on the chemical composition data and performance data of each alloy, a deep neural network model is trained to construct the mapping relationship between alloy composition and performance; during the training process of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters to minimize the prediction difference between the deep neural network model and the supervised learning model;

[0029] Optimization module: Sampling optimization is performed using Langevin Dynamics, and composition optimization is performed using the gradient information of the deep neural network model to find the optimal alloy composition.

[0030] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and system for reverse design of alloy composition based on knowledge distillation. Utilizing deep neural networks and gradient optimization algorithms, it performs rapid search and optimization in a high-dimensional alloy composition space, thereby quickly finding alloy compositions that meet the target performance, improving search efficiency compared to high-throughput forward search. Secondly, during the optimization process, by continuously adding newly generated alloy compositions and their predicted performance to the training dataset, knowledge distillation and data augmentation are achieved. This not only improves the prediction accuracy and generalization ability of the DNN model but also allows the composition feature model to gradually approach the performance of the atomic parameter feature model, further optimizing the reverse design effect. This invention also combines the composition and atomic parameter features of the alloy, using a DNN to approximate the composition feature model with the atomic parameter feature model, resolving the compositional composition of potential high-performance alloys, and realizing customized material design and optimization. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] This invention discloses a reverse design method for alloy composition based on knowledge distillation, such as... Figure 1 As shown, the specific steps are as follows:

[0035] Step 1: Collect chemical composition data and corresponding performance data of multi-element alloys, organize them into a multi-element alloy dataset, and calculate the atomic characteristics of each alloy in the multi-element alloy dataset.

[0036] Step 2: Based on the atomic characteristics and performance data of each alloy, construct and train a supervised learning model to establish the mapping relationship between the atomic characteristics and performance of the alloy.

[0037] Step 3: Based on the chemical composition and performance data of each alloy, train a deep neural network model to construct the mapping relationship between alloy composition and performance; during the training of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters to minimize the prediction difference between the deep neural network model and the supervised learning model.

[0038] Step 4: Use Langevin Dynamics for sampling optimization and use the gradient information of the deep neural network model for composition optimization to find the optimal alloy composition.

[0039] In a specific embodiment, each alloy in the multi-element alloy dataset is represented as X. i = [x1,x2,……,x m ], satisfying ∑x j =1; the corresponding performance index is represented by Y. i ; where X i Let X be a component vector representing the i-th alloy component in the dataset, where i = 1, 2, ..., N, and N is the total number of alloys in the dataset. i Composed of m elements, x j This represents the molar percentage of the j-th element.

[0040] In one specific embodiment, atomic features include: average electronegativity, average ionization energy, and atomic size variation. Atomic features, due to the incorporation of prior physicochemical knowledge, often exhibit higher reliability compared to models based solely on composition. However, the composition cannot be derived in reverse from atomic features. This problem is addressed by introducing knowledge distillation techniques.

[0041] In a specific embodiment, the mapping relationship from alloy atomic characteristics to properties is expressed as follows: in The value represents the prediction of the supervised learning model, and f(x) represents the supervised learning model, which serves as the teacher model for the subsequent knowledge distillation model. The supervised learning model is built based on supervised learning algorithms, such as support vector machines, random forests, or gradient boosting trees.

[0042] The mapping relationship between alloy composition and properties is expressed as follows: in Let f(x) represent the predicted value of the deep neural network model. g(x) serves as the student model for knowledge distillation, aiming to mimic the teacher model's ability to predict alloy properties. During training, the backpropagation algorithm is used to adjust the network parameters and minimize the prediction difference between f(x) and g(x). The deep neural network model is a four-layer fully connected network with the input dimension being the number of alloy elements in the dataset and the output being the target performance.

[0043] In one specific embodiment, step 4 specifically includes:

[0044] Step 4.1: Based on the alloy composition to be optimized, determine the sampling starting point. The sampling starting point composition vector is represented as X. i,0 = [x1,x2,……,x m The number of samples is n;

[0045] Step 4.2: Input the composition vector of the alloy to be optimized into the deep neural network model to obtain the predicted value;

[0046] Step 4.3: Calculate the gradient information of the current predicted value, and calculate the optimized alloy composition sampling points based on the gradient information. If the number of samplings is less than the preset threshold n, return to step 4.2; otherwise, proceed to step 4.4.

[0047] Step 4.4: Calculate the error between the predicted values ​​of the supervised learning model and the predicted values ​​of the deep neural network model for all sampling points; determine whether the stopping condition is met. If it is met, select the optimal alloy composition from all sampling points; otherwise, add the chemical composition data of all sampling points and the corresponding predicted values ​​of the supervised learning model as new samples to the multivariate alloy dataset, and return to step 3.

[0048] Specifically, to find the optimal alloy composition in a high-dimensional composition space, Langevin Dynamics is used for composition optimization. Initially, an alloy composition X from the dataset is selected. i As the sampling starting point, it is input into the DNN model g(x) to obtain the prediction performance. Then, the gradient of the predicted performance with respect to the mole percentage of each component is calculated, i.e., the partial derivative of the predicted performance value with respect to each component. Based on the gradient information, the component vector is updated. This process is iterated continuously, and the predicted performance is checked after each iteration. If the difference between the current performance and the previous performance is less than a preset threshold ε, and the target performance is met or the preset maximum number of iterations is reached, then a potential optimization component is considered to have been found; otherwise, optimization continues.

[0049] During the optimization process, the newly generated alloy composition and its predicted properties will be considered. The data is dynamically expanded by adding it to the original dataset D. The newly added data helps the DNN model g(x) better simulate the behavior of f(x), gradually improving the model's prediction accuracy and generalization ability. This process is called knowledge distillation, which allows the DNN model not only to approximate the performance of the base model but also to be further optimized, achieving more efficient reverse design.

[0050] In one specific embodiment, the sampling point is represented as X. i,k = [x1,x2,…x j …,x m ], where X i,k This represents the composition vector of the i-th alloy sampled in the dataset at the k-th sampling time, composed of m elements, x j This represents the molar percentage of the j-th element.

[0051] The formula for calculating the optimal alloy composition sampling points is: X i,k+1 =X i,k +λ*grad; where λ>0 is the learning rate, and grad represents the gradient information. express For x j The partial derivatives of .

[0052] After n samplings, an optimized trajectory with a total of n component sampling points is obtained.

[0053] In a specific embodiment, the error is expressed as: in This represents the predicted value of the supervised learning model. This represents the predicted value of the deep neural network model;

[0054] The stopping condition is: there exists a sampling point X. i,k satisfy and The target performance requirement or the maximum number of iterations is achieved, where ε represents the error factor of the supervised learning model and the deep neural network model. The target performance requirement includes, but is not limited to, alloy predictability and the specified range of alloy composition molar percentage.

[0055] If the stopping condition is not met, the alloy composition data of the sampling points in this iteration and the corresponding... The values ​​are added to the dataset as new sample data to retrain the deep neural network model. The gradually added data enables the deep neural network model g(x) to learn the knowledge in the supervised learning model f(x), gradually becoming its surrogate model, and more accurately calculating the gradient of alloy properties with respect to composition.

[0056] Based on reliable literature data and modeling methods, this invention establishes an alloy melting point prediction model that is simple, fast, low-cost, and pollution-free. Through knowledge distillation, the DNN model can flexibly explore the alloy composition space based on gradient information. By learning the performance advantages and disadvantages of existing alloys, the DNN model can propose novel composition combinations, promoting the rapid development of alloy materials.

[0057] In a specific embodiment, an alloy composition with a design requirement of melting point <7 is used as an example for illustration. The data in step 1 comes from 150 low-melting-point alloys with different compositions reported in the literature, some of which are shown in Table 1:

[0058] Table 1. Data on Low Melting Point Alloys

[0059] Chemical formula Melting point Ga71.9In17.9Sn9.0Zn0.13 8.000 Ga78.3In14.9Sn6.8 10.000 In60.1Sn18.8Bi21.1 63.000 Sn56Bi44 139.060 In34.3Sn22.8Bi43.0 80.700 In60.7Sn17.8Bi21.5 60.420 Sn33.1Bi42.0 101.130 Sn 33.1Bi 42.7Pb 24.1Cd 0.1 97.060 Sn33.2Bi38.5Ag7.8 96.840 Bi 48.0 Pb 37.8 Ag 14.2 92.550 Ga54.5Ge11.1Tl34.4 92.970 In 30.2Sn 16.7Bi 38.7Pb 14.4 144.990 Sn 19.7Bi 42.0Pb 22.6Sb 15.6 59.730

[0060] In step 4, Ga71.9In17.9Sn9.0Zn0.13 is selected as the starting point for Langevin Dynamics sampling optimization. Table 2 shows an example of the trajectory of the sampling optimization process for this sample. The design sample obtained in this optimization is the sample from the eighth iteration.

[0061] Table 2. Sampling optimization trajectory of Ga-In-Sn-Zn reverse design

[0062]

[0063]

[0064] This embodiment demonstrates a novel data-driven materials genome approach based on knowledge distillation for reverse alloy composition design, guiding the composition design of multi-element alloys. This embodiment illustrates that this technology can rapidly obtain the composition of a designed low-melting-point alloy based on a model.

[0065] In one specific embodiment, the process is essentially the same as the embodiments described above, except that the starting point for optimization is a pentagonal alloy, and the Pb content is set to not exceed 0.15%. Table 3 shows an example of the trajectory of the sampling optimization process for this sample.

[0066] Table 3. Sampling optimization trajectory of Ga-In-Sn-Zn-Pb reverse design

[0067]

[0068] This embodiment demonstrates that the reverse design method for alloy composition based on knowledge distillation can flexibly set the composition gradient and add composition optimization constraints to meet the design requirements of different alloy materials.

[0069] This invention discloses a reverse design system for alloy composition based on knowledge distillation, comprising:

[0070] Preprocessing module: Collects chemical composition data and corresponding performance data of multi-element alloys, organizes them into a multi-element alloy dataset, and calculates the atomic characteristics of each alloy in the multi-element alloy dataset;

[0071] Supervised learning module: Based on the atomic characteristics and performance data of each alloy, a supervised learning model is constructed and trained to establish the mapping relationship between the atomic characteristics of the alloy and its performance.

[0072] Deep learning module: Based on the chemical composition and performance data of each alloy, a deep neural network model is trained to construct the mapping relationship between alloy composition and performance; during the training process of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters and minimize the prediction difference between the deep neural network model and the supervised learning model;

[0073] Optimization module: Langevin Dynamics is used for sampling optimization, and gradient information from a deep neural network model is used for composition optimization to find the optimal alloy composition.

[0074] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those 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 invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A reverse design method for alloy composition based on knowledge distillation, characterized in that, The specific steps are as follows: Step 1: Collect chemical composition data and corresponding performance data of multi-element alloys, organize them to obtain a multi-element alloy dataset, and calculate the atomic characteristics of each alloy in the multi-element alloy dataset. Step 2: Based on the atomic characteristics and performance data of each alloy, construct and train a supervised learning model to establish the mapping relationship between alloy atomic characteristics and performance. Step 3: Based on the chemical composition data and performance data of each alloy, train a deep neural network model to construct the mapping relationship from alloy composition to performance; during the training process of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters to minimize the prediction difference between the deep neural network model and the supervised learning model; Step 4: Use Langevin Dynamics for sampling optimization and use the gradient information of the deep neural network model for composition optimization to find the optimal alloy composition; The atomic characteristics include: average electronegativity, average ionization energy, and atomic size variation; Step 4 specifically includes: Step 4.1: Determine the sampling starting point based on the alloy composition to be optimized; Step 4.2: Input the composition vector of the alloy to be optimized into the deep neural network model to obtain the predicted value; Step 4.3: Calculate the gradient information of the current predicted value, and calculate the optimized alloy composition sampling points based on the gradient information. If the number of samplings is less than a preset threshold, return to step 4.2; otherwise, proceed to step 4.

4. Step 4.4: Calculate the error between the predicted values ​​of the supervised learning model and the predicted values ​​of the deep neural network model for all sampling points; determine whether the stopping condition is met. If it is met, select the optimal alloy composition from all sampling points; otherwise, add the chemical composition data of all sampling points and the corresponding predicted values ​​of the supervised learning model as new samples to the multi-element alloy dataset, and return to step 3. The sampling points are represented as follows: ,in Represents the first in the dataset The first alloy The component vector of the second sampling is derived from Composition of elements; The formula for calculating and optimizing the alloy composition sampling points is as follows: ;in For learning rate, Represents gradient information, ( ) , express right The partial derivatives of .

2. The reverse design method for alloy composition based on knowledge distillation according to claim 1, characterized in that, Each alloy in the multi-element alloy dataset is represented as follows: ,satisfy =1; the corresponding performance metric is expressed as ;in, Let the component vector represent the first element in the dataset. alloy composition, i =1,2,..., N , N For the total number of alloys in the dataset, Depend on Composed of elements, Indicates the first The molar percentage of each element.

3. The reverse design method for alloy composition based on knowledge distillation according to claim 1, characterized in that, The mapping relationship between the atomic characteristics of the alloy and its properties is expressed as follows: ,in This represents the predicted value of the supervised learning model. This represents the supervised learning model; The mapping relationship between alloy composition and performance is expressed as follows: ,in This represents the predicted value of the deep neural network model. The deep neural network model is described above; the deep neural network model is a four-layer fully connected network, with the input dimension being the number of alloy elements in the dataset and the output being the target performance.

4. The reverse design method for alloy composition based on knowledge distillation according to claim 1, characterized in that, The error is expressed as ,in This represents the predicted value of the supervised learning model. This represents the predicted value of the deep neural network model; The stopping condition is: there is a sampling point. satisfy ,and To achieve the target performance requirements or reach the maximum number of iterations, where This represents the error factor of the supervised learning model and the deep neural network model.

5. A knowledge-based distillation-based alloy composition reverse design system, employing the knowledge-based distillation-based alloy composition reverse design method according to any one of claims 1 to 4, characterized in that, include: Preprocessing module: Collects chemical composition data and corresponding performance data of multi-element alloys, organizes them to obtain multi-element alloy dataset, and calculates the atomic characteristics of each alloy in the multi-element alloy dataset; Supervised learning module: Based on the atomic characteristics and performance data of each alloy, a supervised learning model is constructed and trained to establish the mapping relationship between alloy atomic characteristics and performance; Deep learning module: Based on the chemical composition data and performance data of each alloy, a deep neural network model is trained to construct the mapping relationship between alloy composition and performance; during the training process of the deep neural network model, the backpropagation algorithm is used to adjust the network parameters to minimize the prediction difference between the deep neural network model and the supervised learning model; Optimization module: Sampling optimization is performed using Langevin Dynamics, and composition optimization is performed using the gradient information of the deep neural network model to find the optimal alloy composition.

Images