A method and apparatus for reverse design of target property nickel-based superalloy compositions

Abstract

The application discloses a method and equipment for reverse design of a target performance nickel-based high-temperature alloy composition, and relates to the technical field of nickel-based high-temperature alloy composition design. The method comprises the following steps: obtaining formed nickel-based high-temperature alloy composition data and a performance prediction model; performing clustering on the formed nickel-based high-temperature alloy composition data by using a K-means algorithm to obtain a plurality of nickel-based high-temperature alloy data sets; training a variational autoencoder by using each of the nickel-based high-temperature alloy data sets to obtain a nickel-based high-temperature alloy composition generation model; and reverse designing a target performance nickel-based high-temperature alloy composition based on the nickel-based high-temperature alloy composition generation model and the performance prediction model to obtain a nickel-based high-temperature alloy composition design value. The application can realize reverse design of a nickel-based high-temperature alloy composition with a target performance.

Description

Technical Field

[0001] This invention relates to the field of nickel-based superalloy composition design technology, and in particular to a method and apparatus for reverse design of nickel-based superalloy compositions with target properties. Background Technology

[0002] In the data-driven design of nickel-based superalloys, most methods use alloy composition-performance datasets to train forward design models, i.e., predicting performance based on the target alloy composition. However, it is difficult to design alloy compositions in reverse order based on the target performance. This is because the dimension of the composition vector is much larger than the dimension of the predicted performance vector. The same performance parameter corresponds to multiple alloy compositions. Therefore, traditional discriminative models are difficult to use for reverse design of alloy compositions. Generative methods must be used to train a model that fits the composition distribution of nickel-based superalloys. Summary of the Invention

[0003] The purpose of this invention is to provide a method and apparatus for reverse designing nickel-based superalloy compositions with target properties, so as to achieve reverse design of nickel-based superalloy compositions with target properties.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for reverse engineering the composition of a nickel-based superalloy with target performance includes:

[0006] Acquire pre-formed nickel-based superalloy composition data and performance prediction models;

[0007] The K-means algorithm is used to cluster the formed nickel-based superalloy composition data to obtain several nickel-based superalloy class datasets; each nickel-based superalloy class dataset includes several nickel-based superalloy composition samples.

[0008] A variational autoencoder is trained using each of the aforementioned nickel-based superalloy datasets to obtain a nickel-based superalloy composition generation model. The variational autoencoder includes an encoder, a transformed normal distribution layer, and a decoder. The encoder generates a mean vector and a standard deviation vector based on the mixed feature vector of the nickel-based superalloy composition samples. The mixed feature vector includes an alloy composition feature vector and an elemental physical feature vector. The transformed normal distribution layer generates a random vector based on the mean vector and the standard deviation vector. The decoder generates a virtual nickel-based superalloy composition based on the random vector.

[0009] Based on the nickel-based superalloy composition generation model and the performance prediction model, the target performance nickel-based superalloy composition is designed in reverse to obtain the design value of the nickel-based superalloy composition.

[0010] Optionally, the target performance nickel-based superalloy composition is designed in reverse based on the nickel-based superalloy composition generation model and the performance prediction model to obtain the design value of the nickel-based superalloy composition, specifically including:

[0011] The nickel-based superalloy composition generation model is used to generate candidate nickel-based superalloy composition data;

[0012] The performance prediction model is used to predict the performance of the candidate nickel-based superalloy composition data to obtain the predicted performance;

[0013] From the candidate nickel-based superalloy composition data, virtual nickel-based superalloy compositions whose predicted performance reaches the target performance are selected to obtain the design value of the nickel-based superalloy composition.

[0014] Optionally, a variational autoencoder is trained using each of the aforementioned nickel-based superalloy datasets to obtain a nickel-based superalloy composition generation model, specifically including:

[0015] Calculate the elemental physical feature vector based on the alloy composition feature vector of the nickel-based high-temperature alloy composition sample, and then concatenate the alloy composition feature vector with the elemental physical feature vector to obtain a mixed feature vector;

[0016] The mixed feature vector of the nickel-based superalloy composition sample is input into the variational autoencoder to generate a virtual nickel-based superalloy composition.

[0017] The loss function is determined based on the virtual nickel-based superalloy composition and the nickel-based superalloy composition sample.

[0018] With the goal of minimizing the loss function, variational autoencoders are trained using each of the nickel-based superalloy datasets to obtain several trained variational autoencoders.

[0019] The decoders of all trained variational autoencoders are collectively determined as the nickel-based superalloy composition generation model.

[0020] Optionally, before clustering the formed nickel-based superalloy composition data using the K-means algorithm, the method further includes:

[0021] The preformed nickel-based superalloy composition data is transformed to [0,1] using maximum-minimum normalization.

[0022] Optionally, the K-means algorithm is used to cluster the formed nickel-based superalloy composition data to obtain several nickel-based superalloy class datasets, specifically including:

[0023] Set the number of target categories to K;

[0024] K nickel-based superalloy composition samples were randomly selected as the initial cluster centers;

[0025] Calculate the Euclidean distance from each nickel-based superalloy composition sample to the K cluster centers, and assign it to the category corresponding to the cluster center with the smallest Euclidean distance;

[0026] The cluster centers are updated based on the mean of all nickel-based superalloy composition samples in each category, and it is determined whether the set number of iterations has been reached.

[0027] If the set number of iterations is not reached, return to the step of "calculate the Euclidean distance from each nickel-based superalloy composition sample to the K cluster centers and assign it to the category corresponding to the cluster center with the smallest Euclidean distance";

[0028] If the set number of iterations is reached, each cluster will be identified as a separate dataset for nickel-based superalloys, resulting in K datasets for nickel-based superalloys.

[0029] Optionally, the performance prediction model is a simulation model, a machine learning model, or an experimental measurement model; the type of the target performance includes at least one of the following: high-temperature creep performance, high-temperature rupture life, high-temperature tensile strength, room-temperature plasticity, γˋ solution temperature, γˋ phase volume fraction, high-temperature yield strength, and hardness.

[0030] Optionally, the encoder includes an encoder input layer, a first fully connected layer, a first activation function ReLU layer, and an encoder output layer connected in sequence; the decoder includes a decoder input layer, a second fully connected layer, a second activation function ReLU layer, and a decoder output layer connected in sequence.

[0031] Optionally, the elemental physical characteristic vector includes: the mean, maximum, minimum, and range values ​​of the element's inherent characteristics; the elemental inherent characteristics include at least one of: atomic number, Mendeleev number, relative atomic mass, number of rows and columns in the periodic table, atomic radius, number of orbitals occupied by electrons in each shell, number of unpaired electrons, and melting point of the element.

[0032] Optionally, the expression for the loss function is:

[0033] Loss = Loss KL +Loss penalty ;

[0034]

[0035]

[0036] Where Loss is the total loss, Loss KL For KL divergence loss, Loss penaltyThe penalty is for missing components, where N is the total number of alloy components, and x is the total number of components. i The content of alloy component i in the nickel-based superalloy composition sample, z i The content of alloy component i in the virtual nickel-based superalloy composition.

[0037] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to achieve the steps of the method described above for reverse engineering a nickel-based superalloy composition with target performance.

[0038] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0039] The method for reverse-engineering nickel-based superalloy compositions with target properties provided by this invention employs the K-means algorithm to address the difficulty of encoding sample points at edge positions in the high-dimensional nickel-based superalloy composition space into effective feature vectors. Furthermore, it combines the K-means algorithm with a variational autoencoder embedding elemental physical information to achieve efficient alloy composition generation. By specifying target properties and generating and filtering nickel-based superalloy compositions, it solves the problem of reverse-engineering alloy compositions that is difficult to address in traditional discrimination models. Therefore, this invention can achieve reverse-engineering of nickel-based superalloy compositions with target properties. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating the method for reverse engineering the composition of nickel-based superalloys with target performance, provided by this invention.

[0042] Figure 2 A detailed step diagram illustrating the reverse engineering of the composition of a nickel-based superalloy with target performance, provided by this invention.

[0043] Figure 3 This is a schematic diagram of the variational autoencoder provided by the present invention. Detailed Implementation

[0044] 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.

[0045] The purpose of this invention is to provide a method and apparatus for reverse designing nickel-based superalloy compositions with target properties, so as to achieve reverse design of nickel-based superalloy compositions with target properties.

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] like Figure 1 and Figure 2 As shown, this invention provides a method for reverse engineering the composition of a nickel-based superalloy with target performance, the method comprising:

[0048] Step S1: Obtain the composition data and performance prediction model of the formed nickel-based superalloy.

[0049] The pre-formed nickel-based superalloy composition data should have nickel as the main element, meaning that nickel has the highest content. The dataset should include all alloy components that are common in nickel-based superalloys.

[0050] The performance prediction model, including simulation models, machine learning models, and experimental measurement models, can predict performance based on existing alloy composition data. These three models are briefly introduced below.

[0051] 1. Simulation model: See the website https: / / jmatpro.cntech.com. This website provides the Jmatpro simulation software, which uses thermodynamic methods to calculate the properties of nickel-based superalloys. It can calculate the corresponding properties based on the input nickel-based superalloy composition.

[0052] 2. Machine learning model: See https: / / github.com / CompRhys / aviary. This website releases the Roost pre-trained model package, which can predict performance based on material composition. To apply this machine learning performance prediction model, it is necessary to obtain the composition data and corresponding performance data of nickel-based superalloys, and complete the training using the pre-training module of the Roost model. The pre-training process is as follows: Set the weights of the last fully connected layer in the Roost model to be trainable, freeze the weight parameters of other layers and set them to be non-trainable, and use the obtained composition data and corresponding performance data of the nickel-based superalloys for training, thereby obtaining a machine learning model for predicting the performance of nickel-based superalloys.

[0053] 3. Experimental measurement model: This involves purchasing the raw materials that make up the components of the nickel-based superalloy and mixing them together using experimental methods. The experiment is typically conducted under high temperature and high pressure to ensure the mixture is homogeneous. The measurement method is selected based on the performance to be measured; for example, a tensile testing machine is used to measure elastic properties.

[0054] Step S2: The K-means algorithm is used to cluster the formed nickel-based superalloy composition data to obtain several nickel-based superalloy class datasets; the nickel-based superalloy class datasets include several nickel-based superalloy composition samples.

[0055] First, the pre-formed nickel-based superalloy composition data is transformed to [0,1] using maximum-minimum normalization, so that the sum of the contents of all elements in a set of nickel-based superalloy composition data is equal to 1.

[0056] Then, the K-means algorithm is used to cluster the formed nickel-based superalloy composition data. Each cluster represents the same type of alloy in the high-dimensional space, resulting in several nickel-based superalloy class datasets. This ensures that when the encoder extracts the alloy composition, it can effectively encode each alloy composition sample in the high-dimensional space into a feature vector.

[0057] The specific implementation process includes: 1. First, manually set the value K, where K represents the number of classes to be divided into for the nickel-based superalloy samples, i.e., the target number of classes. Then, randomly select K alloy samples as initial cluster centers; 2. Calculate the Euclidean distance of each sample to the K cluster centers and assign it to the class corresponding to the cluster center with the smallest distance; 3. For each class, recalculate its cluster center, which is the average vector of all alloy composition vectors in that class; repeat steps 2 and 3 above until the manually set number of iterations is reached; finally, each class obtained represents the same type of alloy in the high-dimensional space, ensuring that the encoder can effectively encode each alloy composition sample in the high-dimensional space into a feature vector when extracting alloy composition later.

[0058] Step S3: Train a variational autoencoder using each of the aforementioned nickel-based superalloy datasets to obtain a nickel-based superalloy composition generation model. The variational autoencoder includes an encoder, a transformed normal distribution layer, and a decoder. The encoder generates a mean vector and a standard deviation vector based on the mixed feature vector of the nickel-based superalloy composition samples. The mixed feature vector includes an alloy composition feature vector and elemental physical feature vectors. The transformed normal distribution layer generates a random vector based on the mean vector and the standard deviation vector. The decoder generates a virtual nickel-based superalloy composition based on the random vector.

[0059] Furthermore, such as Figure 3 As shown, the encoder includes an encoder input layer, a first fully connected layer, a first activation function ReLU layer, and an encoder output layer connected in sequence; the decoder includes a decoder input layer, a second fully connected layer, a second activation function ReLU layer, and a decoder output layer connected in sequence. The variational autoencoder adds a transform normal distribution layer between the encoder output layer and the decoder input layer. In this layer, two fully connected operations are used to calculate the mean and standard deviation of the normal distribution, thereby generating a normally distributed random vector as the input to the decoder. The two fully connected layers output the mean vector mean = (m1, m2, ..., m...). N ) and the standard deviation vector standard=(s1,s2,..,s N Two vectors have the same dimension N, and for the k-th element m... k and s k Generate a normal distribution X' ~ N(m) k ,s k 2 ), and extract a number r k As the k-th element of the random vector, the random vector input to the decoder is R = (r1, r2, ..., r...). N ) is a normally distributed random vector.

[0060] The specific process for determining the composition generation model of nickel-based superalloys is as follows:

[0061] 1. Calculate the elemental physical feature vector based on the alloy composition feature vector of the nickel-based high-temperature alloy composition sample, and concatenate the alloy composition feature vector with the elemental physical feature vector to obtain a mixed feature vector.

[0062] Preferably, the elemental physical characteristic vector includes: the mean, maximum, minimum, and range values ​​of the element's inherent characteristics; the elemental inherent characteristics include at least one of: atomic number, Mendeleev number, relative atomic mass, number of rows and columns in the periodic table, atomic radius, number of orbitals occupied by electrons in each shell, number of unpaired electrons, and melting point of the element.

[0063] Specifically, based on known individual elemental properties, electronic structure, and elemental information, such as atomic number, Mendeleev number, relative atomic mass, number of rows and columns in the periodic table, atomic radius, number of orbitals occupied by electrons in each shell, number of unpaired electrons, and melting point of the element, a set of alloy composition feature vectors X = (x1, x2, ..., x...) is calculated. N The element physical characteristic vector P = (p1, p2, ..., p) is... M N represents the total number of alloy components, M represents the total number of inherent characteristics of elements, and x i This represents the content of each alloy component, and x i ∈[0,1], i represents the various elements that make up the nickel-based superalloy, i∈{Ni,Co,Al…}, p j This represents the inherent characteristics of a certain element obtained through calculation, where j represents the physical feature vector of the j-th calculated element. Calculation methods include mean, maximum, minimum, and range values. The mean is calculated as follows: The maximum value is calculated as follows: The minimum value is calculated as follows: The calculation method for the range value is as follows The hybrid feature vector V = (X, P) obtained by concatenating the alloy composition feature vector X and the element physical feature vector P, which embeds the element physical information, is used as the input of the variational autoencoder.

[0064] For example, in the nickel-based superalloy Ni3Al, the Ni content is 75% and the Al content is 25%. Given that the relative atomic mass of Ni is 58.7 and the relative atomic mass of Al is 27, the calculated results for the mean, maximum, minimum, and range values ​​of the relative atomic mass characteristics of the nickel-based superalloy Ni3Al are as follows:

[0065] Mean: 0.75 * 58.7 + 0.25 * 27 = 50.775;

[0066] Maximum value: max(58.7,27)=58.7;

[0067] Minimum value: min(58.7,27)=27;

[0068] Range value: 58.7 - 27 = 31.7.

[0069] For example, Ni has an atomic number of 28, and Al has an atomic number of 13. In the nickel-based superalloy Ni3Al, Ni accounts for 75% of the total content, and Al accounts for 25%. The calculated results for the atomic number characteristics of the nickel-based superalloy Ni3Al, including the mean, maximum, minimum, and range values, are as follows:

[0070] Mean: 0.75*28 + 0.25*13 = 24.25;

[0071] Maximum value: max(28,13)=28;

[0072] Minimum value: min(28,13)=13;

[0073] Range value: 28-13=15.

[0074] 2. Input the mixed feature vector of the nickel-based superalloy composition sample into the variational autoencoder to generate a virtual nickel-based superalloy composition.

[0075] 3. Determine the loss function based on the virtual nickel-based superalloy composition and the nickel-based superalloy composition sample.

[0076] The loss function for training the entire variational autoencoder consists of the sum of a KL divergence loss term and a missing composition penalty term. The KL divergence loss term is used to train the variational autoencoder to learn the true composition distribution of nickel-based superalloys, while the missing composition penalty term ensures that the output alloy composition is reasonable, i.e., the sum of the contents of each element is 100%. The specific formula for the loss function is as follows:

[0077] The formula for the KL divergence loss term is:

[0078]

[0079] The formula for the component missing penalty term is:

[0080]

[0081] The overall loss function is:

[0082] Loss = Loss KL +Loss penalty ;

[0083] Where Loss is the total loss, Loss KL For KL divergence loss, Loss penalty The penalty is for missing components, where N is the total number of alloy components, and x is the total number of components. i The content of alloy component i in the nickel-based superalloy composition sample, z i The content of alloy component i in the virtual nickel-based superalloy composition.

[0084] 4. With the goal of minimizing the loss function, variational autoencoders are trained using each of the nickel-based superalloy datasets to obtain several trained variational autoencoders.

[0085] Specifically, the nickel-based superalloy composition data is used as the input training set. First, features embedding elemental physical information are calculated based on the composition data and concatenated with the composition features as the input vector of the encoder. In the encoder part, after being processed by a fully connected layer and the ReLU activation function, the output is sent to the encoder output layer to generate a mean vector and a standard deviation vector. A normally distributed random vector is generated by calculating the mean vector and the standard deviation vector and used as the input of the decoder. In the decoder, after being processed by a fully connected layer and the ReLU activation function, the output is sent to the decoder output layer. The KL loss function is calculated by the composition vector of the nickel-based superalloy and the output of the decoder output layer, and a composition missing penalty function is used to ensure the rationality of the output alloy composition.

[0086] 5. The decoders of all trained variational autoencoders are collectively determined as the nickel-based superalloy composition generation model.

[0087] Specifically, based on the class labels obtained by the K-means algorithm, nickel-based superalloys are divided into multiple classes; in each class, the encoder and decoder parts of the variational autoencoder are trained using hybrid features that embed the physical characteristics of the elements; finally, the trained decoder part is taken out separately as a generator to generate various nickel-based superalloy compositions.

[0088] It should be noted that since the nickel-based superalloy composition generation model includes multiple generators, the generator is selected probabilistically based on the number of samples in each category during the generation of virtual nickel-based superalloy compositions. For example, if there are three categories with 100, 200, and 300 samples each, and three generators are trained, the probabilities of selecting these three generators in subsequent applications would be 100 / 600, 200 / 600, and 300 / 600, respectively. That is, the more samples a category has, the higher the probability of selecting the generator for that category; conversely, the fewer samples a category has, the lower the probability of selecting the generator for that category.

[0089] For each class of generator, a random vector following a standard normal distribution is input. The generator neural network maps this random vector to a potential alloy composition that does not exist in the existing nickel-based superalloy dataset but whose data distribution is the same as that of real nickel-based superalloy compositions. The generator then outputs a set of nickel-based superalloy composition parameters consistent with the composition distribution of that class. The generated virtual nickel-based superalloys exhibit complex relationships between their elemental components, providing valuable candidate samples for subsequent screening.

[0090] Step S4: Based on the nickel-based superalloy composition generation model and the performance prediction model, reverse design the target performance nickel-based superalloy composition to obtain the nickel-based superalloy composition design value.

[0091] Preferably, the nickel-based superalloy composition generation model is used to generate candidate nickel-based superalloy composition data; the performance prediction model is used to predict the performance of the candidate nickel-based superalloy composition data to obtain the predicted performance; and virtual nickel-based superalloy compositions whose predicted performance reaches the target performance are selected from the candidate nickel-based superalloy composition data to obtain the design value of the nickel-based superalloy composition.

[0092] Specifically, an existing composition prediction performance model is used as a surrogate model. A target performance is input, and by comparing it with the performance of various generated alloy samples, the alloy composition sample with the closest performance is selected, thus achieving reverse design of the nickel-based superalloy composition with the target performance. The type of target performance can be high-temperature creep performance, high-temperature rupture life, high-temperature tensile strength, room-temperature plasticity, γˋ solution temperature, γˋ phase volume fraction, high-temperature yield strength, hardness, etc. By setting a relative error standard, the output is achieved when the relative error between the predicted performance and the target performance of the alloy to be screened is less than the set relative error standard.

[0093] In summary, the method for reverse engineering the composition of nickel-based superalloys with target performance provided by this invention has the following beneficial effects:

[0094] 1. This invention uses the k-means algorithm to solve the problem that sample points at the edge positions in the composition space of high-dimensional nickel-based superalloys are difficult to encode into effective feature vectors.

[0095] 2. This invention combines a variational autoencoder with a K-means algorithm, embedding elemental physical features to enhance the encoding capability of the variational autoencoder, thereby enabling efficient generation of the composition of various nickel-based high-temperature alloys.

[0096] 3. This invention uses existing models that predict performance based on composition as surrogate models. It inputs a target performance and compares it with the performance of various alloy samples that have already been generated. It then selects the alloy composition sample with the closest performance and achieves reverse design of the target performance nickel-based superalloy composition.

[0097] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0098] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps in the above method embodiments.

[0099] In one embodiment, a computer device is also provided, including a memory and a processor, and a computer program stored in the memory and executable on the processor, which, when executing the computer program, implements the steps in the above method embodiments.

[0100] Furthermore, the computer device also includes an input / output interface (I / O) and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores pending transactions. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0101] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0102] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0103] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for reverse engineering the composition of a nickel-based superalloy with target performance, characterized in that, include: Acquire pre-formed nickel-based superalloy composition data and performance prediction models; The K-means algorithm is used to cluster the formed nickel-based superalloy composition data to obtain several nickel-based superalloy class datasets; each nickel-based superalloy class dataset includes several nickel-based superalloy composition samples. A variational autoencoder is trained using each of the aforementioned nickel-based superalloy datasets to obtain a nickel-based superalloy composition generation model. The variational autoencoder includes an encoder, a transformed normal distribution layer, and a decoder. The encoder generates a mean vector and a standard deviation vector based on the mixed feature vector of the nickel-based superalloy composition samples. The mixed feature vector includes an alloy composition feature vector and an elemental physical feature vector. The transformed normal distribution layer generates a random vector based on the mean vector and the standard deviation vector. The decoder generates a virtual nickel-based superalloy composition based on the random vector. Based on the nickel-based superalloy composition generation model and the performance prediction model, a target performance nickel-based superalloy composition is designed in reverse to obtain the design value of the nickel-based superalloy composition. Specifically, this includes: generating candidate nickel-based superalloy composition data using the nickel-based superalloy composition generation model; predicting the performance of the candidate nickel-based superalloy composition data using the performance prediction model to obtain the predicted performance; and selecting virtual nickel-based superalloy compositions from the candidate nickel-based superalloy composition data whose predicted performance reaches the target performance to obtain the design value of the nickel-based superalloy composition. The K-means algorithm was used to cluster the pre-formed nickel-based superalloy composition data to obtain several nickel-based superalloy class datasets, specifically including: Set the number of target categories to K; K nickel-based superalloy composition samples were randomly selected as the initial cluster centers; Calculate the Euclidean distance from each nickel-based superalloy composition sample to the K cluster centers, and assign it to the category corresponding to the cluster center with the smallest Euclidean distance; The cluster centers are updated based on the mean of all nickel-based superalloy composition samples in each category, and it is determined whether the set number of iterations has been reached. If the set number of iterations is not reached, return to the step of "calculate the Euclidean distance from each nickel-based superalloy composition sample to the K cluster centers and assign it to the category corresponding to the cluster center with the smallest Euclidean distance"; If the set number of iterations is reached, each cluster will be identified as a separate dataset for nickel-based superalloys, resulting in K datasets for nickel-based superalloys. Since the nickel-based superalloy composition generation model includes multiple generators, the generator is selected probabilistically based on the number of samples for each category during the generation of virtual nickel-based superalloy compositions.

2. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 1, characterized in that, Variational autoencoders were trained using the datasets for each of the aforementioned nickel-based superalloy classes to obtain nickel-based superalloy composition generation models, specifically including: Calculate the elemental physical feature vector based on the alloy composition feature vector of the nickel-based high-temperature alloy composition sample, and then concatenate the alloy composition feature vector with the elemental physical feature vector to obtain a mixed feature vector; The mixed feature vector of the nickel-based superalloy composition sample is input into the variational autoencoder to generate a virtual nickel-based superalloy composition. The loss function is determined based on the virtual nickel-based superalloy composition and the nickel-based superalloy composition sample. With the goal of minimizing the loss function, variational autoencoders are trained using each of the nickel-based superalloy datasets to obtain several trained variational autoencoders. The decoders of all trained variational autoencoders are collectively determined as the nickel-based superalloy composition generation model.

3. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 1, characterized in that, Before clustering the formed nickel-based superalloy composition data using the K-means algorithm, the following steps are also included: The preformed nickel-based superalloy composition data is transformed to [0,1] using maximum-minimum normalization.

4. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 1, characterized in that, The performance prediction model is a simulation model, a machine learning model, or an experimental measurement model; the target performance type includes at least one of the following: high temperature creep performance, high temperature creep life, high temperature tensile strength, room temperature plasticity, γ' solution temperature, γ' phase volume fraction, high temperature yield strength, and hardness.

5. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 1, characterized in that, The encoder includes an encoder input layer, a first fully connected layer, a first activation function ReLU layer, and an encoder output layer connected in sequence; the decoder includes a decoder input layer, a second fully connected layer, a second activation function ReLU layer, and a decoder output layer connected in sequence.

6. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 1, characterized in that, The elemental physical characteristic vector includes: the mean, maximum, minimum, and range values ​​of the element's inherent characteristics; the elemental inherent characteristics include at least one of: atomic number, Mendeleev number, relative atomic mass, number of rows and columns in the periodic table, atomic radius, number of orbitals occupied by electrons in each shell, number of unpaired electrons, and melting point of the element.

7. The method for reverse engineering the composition of a nickel-based superalloy with target performance according to claim 2, characterized in that, The expression for the loss function is: Loss=Loss KL +Loss penalty ; ； ； Where Loss is the total loss, Loss KL For KL divergence loss, Loss penalty The penalty is for missing components, where N is the total number of alloy components, and x is the total number of components. i The content of alloy component i in the nickel-based superalloy composition sample, z i The content of alloy component i in the virtual nickel-based superalloy composition.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of a method for reverse-engineering a nickel-based superalloy composition with target performance as claimed in any one of claims 1-7.

Images