Reverse design method and device for additive manufacturing nickel-based superalloy components, medium and product

Abstract

The invention discloses a reverse design method and device for additive manufacturing nickel-based superalloy components, a medium and a product, and relates to the technical field of laser additive manufacturing alloy optimizing.The method comprises the steps that a performance prediction model is constructed; setting performance constraint conditions; constructing an alloy component generation model; obtaining a random vector; inputting the random vectors and the performance constraint conditions into an alloy component generation model to obtain a plurality of groups of candidate alloy components, and inputting the random vectors and the performance constraint conditions into a performance prediction model to obtain phase contents of the candidate alloy components; screening out candidate alloy components of which the phase contents do not meet performance constraint conditions to obtain primary samples; taking a multi-dimensional crack sensitivity index value as an optimization target, and constructing a Pareto frontier solution set based on the primarily selected sample; candidate alloy components corresponding to the primary samples meeting set conditions in the Pareto frontier solution set are used as optimal alloy components; and obtaining the optimized target alloy based on the optimal alloy components by adopting an additive manufacturing technology. According to the method, the alloy component design efficiency and precision can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of laser additive manufacturing alloy optimization, and in particular to a reverse design method, equipment, medium and product for additive manufacturing nickel-based high-temperature alloy composition. Background Art

[0002] Laser Additive Manufacturing (LAM) technology has received widespread attention in the fields of aviation, aerospace, and energy in recent years. It is capable of producing complex and lightweight structural components. However, during the LAM process, due to rapid temperature cycles and high temperature gradients, nickel-based high-temperature alloys often face problems such as crack formation and degradation of high-temperature performance, which seriously restricts their reliability in high-end applications. Traditional alloy design methods mainly rely on a large number of experiments and limited thermodynamic simulations, which are not only costly, but also difficult to fully explore the composition space of complex multiphase alloys, making it impossible to effectively balance crack control and high-temperature performance optimization. Existing alloy design methods usually rely on traditional experiments or limited thermodynamic calculations, lack of effective model optimization, resulting in large uncertainties in the alloy design process and high costs of experimental verification. Summary of the Invention

[0003] The purpose of this application is to provide a reverse design method, equipment, medium and product for additive manufacturing of nickel-based high-temperature alloy compositions, which can improve the efficiency and accuracy of alloy composition design and ensure that the designed alloy composition meets the set performance indicators.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a reverse design method for additively manufacturing nickel-based high-temperature alloy compositions, comprising:

[0006] Obtaining multi-group composition data of a target nickel-based high-temperature alloy at a set temperature and phase content data corresponding to the composition data to generate a training data set;

[0007] Constructing an initial performance prediction model, training the initial performance prediction model based on the training data set until the output of the initial performance prediction model meets the set requirements, obtaining a trained initial performance prediction model, and using it as the performance prediction model;

[0008] Setting performance constraints;

[0009] Using a conditional variational autoencoder, constructing an alloy composition generation model based on the training data set and the performance constraints;

[0010] Obtain a random vector; the random vector satisfies a normal distribution;

[0011] Inputting the random vector and the performance constraint condition into the alloy composition generation model to obtain multiple groups of candidate alloy compositions;

[0012] Inputting multiple groups of candidate alloy components into the performance prediction model to obtain the phase contents of the candidate alloy components;

[0013] Screening out candidate alloy components whose phase contents do not satisfy the performance constraints from multiple groups of candidate alloy components to obtain a preliminary sample;

[0014] Taking the multidimensional crack sensitivity index value as the optimization target, constructing a Pareto front solution set based on the preliminary selected samples; taking the candidate alloy composition corresponding to the preliminary selected samples that meet the set conditions in the Pareto front solution set as the optimal alloy composition;

[0015] An optimized nickel-based high-temperature alloy is obtained based on the optimal alloy composition using additive manufacturing technology.

[0016] Optionally, constructing an initial performance prediction model, training the initial performance prediction model based on the training data set until the output of the initial performance prediction model meets set requirements, and obtaining the trained initial performance prediction model includes:

[0017] Build an initial performance prediction model based on the machine learning model;

[0018] training the initial performance prediction model based on the training data set;

[0019] The initial performance prediction model is evaluated using a first loss function until the loss function value of the prediction result output by the initial performance prediction model meets the set requirements, thereby obtaining a trained initial performance prediction model.

[0020] Optionally, the conditional variational autoencoder includes an encoder and a decoder; using the conditional variational autoencoder, constructing an alloy composition generation model based on the training data set and the performance constraint conditions, including:

[0021] training a conditional variational autoencoder based on the training dataset and the performance constraints;

[0022] Evaluating the conditional variational autoencoder using a second loss function until a loss function value of output data of the conditional variational autoencoder meets a set requirement, thereby obtaining a trained conditional variational autoencoder; the second loss function includes a KL divergence loss function term and a reconstruction loss function term;

[0023] The decoder in the trained conditional variational autoencoder is used as the alloy composition generation model.

[0024] Optionally, setting a target performance based on the performance constraint;

[0025] Performing feature splicing on the composition data of the target alloy and the target performance in the training data set, and inputting the features into the encoder to generate a mean vector and a standard deviation vector;

[0026] Obtain a normally distributed random vector based on the mean vector and the standard deviation vector;

[0027] The normally distributed random vector is input into the decoder to obtain the output data.

[0028] Optionally, the phase content of the candidate alloy composition includes the phase content of the matrix phase and the phase content of the strengthening phase;

[0029] The performance constraint condition includes that the phase content of the reinforcement phase is greater than a first set threshold value and the sum of the phase content of the matrix phase and the phase content of the reinforcement phase is greater than a second set threshold value at the set temperature;

[0030] Eliminate candidate alloy components whose phase contents do not satisfy the performance constraints from multiple groups of candidate alloy components to obtain a preliminary sample, including:

[0031] screening the phase contents of the candidate alloy components using the performance constraints;

[0032] determining corresponding candidate alloy components according to the phase contents of the screened candidate alloy components;

[0033] The preliminary selected sample is obtained based on the phase contents of the screened candidate alloy components and the corresponding candidate alloy components.

[0034] Optionally, taking the multi-dimensional crack sensitivity index as the optimization target, constructing the Pareto front solution set based on the preliminary selected samples includes:

[0035] Determining the multi-dimensional crack sensitivity index value corresponding to the primary selected sample;

[0036] A non-dominated sorting genetic algorithm is adopted to construct the Pareto front solution set based on the preliminary selected samples and the multi-dimensional crack sensitivity index values corresponding to the preliminary selected samples.

[0037] Optionally, an optimized target alloy is obtained based on the optimal alloy composition using additive manufacturing technology, comprising:

[0038] Using vacuum induction melting technology, the alloy elements in the optimal alloy composition are preliminarily melted to obtain an ingot;

[0039] treating the ingot until the impurity content of the ingot reaches a set content, and using the treated ingot as a master alloy;

[0040] atomizing the master alloy to obtain alloy powder;

[0041] A laser metal deposition process is adopted to set the laser power, scanning speed and layer thickness, and the alloy powder is deposited and formed to obtain an optimized target alloy.

[0042] In a second aspect, the present application provides a computer device comprising: 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 implement the steps of the inverse design method for additively manufactured nickel-based high-temperature alloy composition described in any one of the above.

[0043] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the inverse design method for additively manufacturing nickel-based high-temperature alloy composition as described in any one of the above.

[0044] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the inverse design method for additively manufactured nickel-based high-temperature alloy composition described in any one of the above.

[0045] According to the specific embodiments provided in this application, this application has the following technical effects:

[0046] This application provides a method, apparatus, medium, and product for reverse design of nickel-based superalloy compositions for additive manufacturing. The method generates a training dataset by acquiring multiple sets of compositional data for a target nickel-based superalloy at a set temperature, along with phase content data corresponding to the compositional data. Performance constraints are set, and an initial performance prediction model is trained based on the training dataset to obtain a performance prediction model. A conditional automatic variational encoder is then used to construct an alloy composition generation model based on the training dataset and the performance constraints, ensuring that the designed alloy composition meets the set performance criteria. A random vector satisfying a normal distribution is input into the alloy composition generation model to obtain multiple sets of candidate alloy compositions. The performance prediction model is used to determine the phase content of each set of candidate alloy compositions. Preliminary samples are screened based on the performance constraints, and a multi-dimensional crack sensitivity index evaluation is performed to construct a Pareto front solution set. From these samples, candidate alloy compositions that meet the set criteria are selected as the optimal alloy composition. An optimized nickel-based superalloy based on the optimal alloy composition is then obtained using additive manufacturing technology, thereby improving the efficiency and accuracy of optimized alloy composition design for nickel-based superalloys manufactured using additive manufacturing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 This is a flow chart of a reverse design method for additively manufacturing nickel-based high-temperature alloy composition in one embodiment of the present application;

[0049] Figure 2 A schematic diagram of a conditional variational autoencoder provided in one embodiment of the present application;

[0050] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0053] With the rapid development of high-throughput thermodynamic computation, machine learning, and generative artificial intelligence technologies, multidisciplinary integrated design methods are emerging as a new approach to solving challenging materials design challenges. By constructing large-scale thermodynamic databases, using machine learning to establish predictive models, and employing conditional variational autoencoders for inverse generation, it is possible to efficiently explore the intrinsic connections between alloy composition and microstructure and properties, thereby reducing repeated experiments and improving design accuracy and optimization efficiency.

[0054] In an exemplary embodiment, Figure 1 As shown, a reverse design method for the composition of a nickel-based high-temperature alloy for additive manufacturing is provided. The method is executed by a computer device, specifically, a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the composition of the target alloy is reversely designed by using the method, and the optimized target alloy is manufactured by additive manufacturing technology as an example, including:

[0055] Step 100: Acquire multiple groups of composition data of a target nickel-based high-temperature alloy at a set temperature and phase content data corresponding to the composition data to generate a training data set.

[0056] Step 200: construct an initial performance prediction model. The initial performance prediction model is trained based on a training data set until the output of the initial performance prediction model meets the set requirements, thereby obtaining a trained initial performance prediction model and using it as the performance prediction model.

[0057] Step 300: Setting performance constraints. Using a conditional variational autoencoder, an alloy composition generation model is constructed based on the training dataset and the performance constraints.

[0058] Step 400: Obtain a random vector that satisfies a normal distribution. Input the random vector and performance constraints into an alloy composition generation model to obtain multiple groups of candidate alloy compositions.

[0059] Step 500: Input multiple groups of candidate alloy components into a performance prediction model to obtain phase contents of the candidate alloy components. Candidate alloy components whose phase contents do not meet the performance constraint conditions are screened out from the multiple groups of candidate alloy components to obtain a preliminary selection sample.

[0060] Step 600: Taking the multi-dimensional crack sensitivity index as the optimization target, a Pareto front solution set is constructed based on the preliminary selected samples. The candidate alloy composition corresponding to the preliminary selected samples that meet the set conditions in the Pareto front solution set is taken as the optimal alloy composition.

[0061] Step 700: Using additive manufacturing technology to obtain an optimized nickel-based high-temperature alloy based on the optimal alloy composition.

[0062] As an optional embodiment, to enhance the flexibility of the inverse design method for additively manufactured nickel-based superalloy compositions, in step 100, thermodynamic calculation software is used to obtain multiple sets of phase content data (i.e., phase content data corresponding to the composition data) based on the multiple sets of composition data of the target nickel-based superalloy. The multiple sets of composition data of the target nickel-based superalloy are obtained by sampling at multiple composition points of the target nickel-based superalloy. For example, the multiple sets of composition data of the target nickel-based superalloy at a set temperature (e.g., 900°C) are obtained. Thermo-Calc thermodynamic calculation simulation software is used to calculate the multiple sets of phase content data corresponding to the composition data based on the multiple sets of composition data of the target nickel-based superalloy. Thus, the multiple sets of composition data of the target nickel-based superalloy at the set temperature and the phase content data corresponding to the composition data are obtained to generate a training dataset. The phase content data includes the phase content of the matrix phase (e.g., γ phase) and the strengthening phase (e.g., γ' phase) in the alloy. This step provides a comprehensive understanding of the phase stability and phase ratio of different alloy compositions at the set temperature, thereby providing a theoretical basis for subsequent nickel-based superalloy composition design and optimization.

[0063] As an optional embodiment, in order to improve the reliability of the reverse design method for additively manufacturing nickel-based high-temperature alloy composition, step 200 includes:

[0064] Step 210: construct an initial performance prediction model based on a machine learning model. The machine learning model may use common regression algorithms in the scikit-learn library, such as linear regression, support vector machine regression (SVR), random forest regression, etc.

[0065] Step 220: Train an initial performance prediction model based on the training data set.

[0066] Step 230 , using a first loss function to evaluate the initial performance prediction model until the loss function value of the prediction result output by the initial performance prediction model meets the set requirements, thereby obtaining a trained initial performance prediction model and using it as the performance prediction model.

[0067] For example, a suitable machine learning model is selected to construct an initial performance prediction model. The initial performance prediction model is trained based on the training data set. The first loss function includes the mean absolute error (MAE) and the coefficient of determination (R 2 ), use the first loss function to evaluate the initial performance prediction model until the loss function value (including mean absolute error and determination coefficient) of the prediction result output by the initial performance prediction model meets the set requirements, and obtain the trained initial performance prediction model, which is used as the performance prediction model. The initial performance prediction model outputs the MAE value and R 2 The values are expressed as:

[0068] Where n represents the number of samples of the target alloy in the training data set (i.e., the number of groups of composition data), y i Represents the true value in the training dataset, y′ i Represents the i-th prediction result. MAE is an indicator to measure the closeness between the predicted γ' phase dissolution temperature and the true value. The smaller its value, the more accurate the prediction result of the performance prediction model. 2 It is used to quantify the ability of independent variables to explain the proportion of predictable variance in dependent variables. The closer its value is to 1, the more accurate the prediction results of the performance prediction model are.

[0069] As an optional embodiment, to ensure that the designed alloy composition meets the set performance indicators, the conditional variational autoencoder in step 300 includes an encoder and a decoder. The process of using the conditional variational autoencoder to construct an alloy composition generation model based on the training data set and performance constraints includes:

[0070] Step 310: training a conditional variational autoencoder based on the training dataset and performance constraints.

[0071] Step 320 : Evaluate the conditional variational autoencoder using a second loss function until the loss function value of the conditional variational autoencoder's output data meets a set requirement, thereby obtaining a trained conditional variational autoencoder. The second loss function includes a KL divergence loss function term and a reconstruction loss function term.

[0072] The processing of the output data from the conditional variational autoencoder involves setting target performance based on performance constraints. The target alloy composition data and the corresponding phase content data from the training dataset are concatenated with the target performance data, and the resultant mean and standard deviation vectors are fed into the encoder. A normally distributed random vector is generated based on the mean and standard deviation vectors. This normally distributed random vector and the performance constraints are then fed into the decoder to generate the output data.

[0073] Step 330: Use the decoder in the trained conditional variational autoencoder as the alloy composition generation model.

[0074] For example, the composition and principle of Conditional Variational Autoencoder (CVAE) are as follows: Figure 2 As shown in the figure, by combining variational inference and conditional generation models, CVAE can generate new material composition combinations given the desired performance indicators. The multi-group composition data in the training dataset (i.e., input composition database), the phase content data corresponding to the composition data, and the set target performance are feature-concatenated to form the input data of CVAE. The encoder maps the input data to the latent space and captures the potential relationship between the composition data and the phase content data through the latent vector. The latent vector is sampled from the latent space, and the latent vector and the set performance constraints are input into the decoder to generate a new alloy composition. Through this design, the decoder not only generates new samples based on the alloy composition characteristics (i.e., latent vectors) in the latent space, but also can directly generate alloy compositions that meet specific performance targets (i.e., performance constraints, such as the desired phase content).

[0075] During the training process of the conditional variational autoencoder, optimization is performed through the second loss function, which includes the KL divergence loss function term and the reconstruction loss function term. The reconstruction loss function term ensures that the generated alloy composition is close to the real composition and that the output alloy composition is reasonable, that is, the sum of the content of each element is 100%. The KL divergence loss function term is used to maintain the good structure of the latent space and train the conditional variational autoencoder to learn the composition distribution of the real nickel-based high-temperature alloy, thereby improving the generation ability of CVAE. In this way, CVAE can efficiently generate alloy compositions that meet specific performance requirements. The second loss function is expressed as

[0076] in,

[0077] Where μ i is the mean of the i-th dimension in the latent space. is the variance of the i-th dimension in the latent space. d is the dimensionality of the latent space. It is the reconstruction loss function term, usually using MSE or cross entropy to measure the difference between the input and reconstructed data. is the KL divergence loss function term, which measures the difference between the latent space distribution and the standard normal distribution. β is a hyperparameter that is usually used to balance the two parts of the loss. If β is large, the model will be more inclined to normalize the latent space; if β is small, the model will pay more attention to the quality of the reconstructed data. i is the i-th input sample. x recon,i is the i-th reconstructed sample obtained by the decoder. N is the total number of samples.

[0078] As an optional embodiment, to improve the efficiency and accuracy of the alloy composition inverse design method, the phase contents of the candidate alloy compositions include the phase contents of the matrix phase and the phase contents of the strengthening phase. Performance constraints include the phase contents of the strengthening phase being greater than a first set threshold at a set temperature, and the sum of the phase contents of the matrix phase and the strengthening phase being greater than a second set threshold.

[0079] In all the embodiments provided in this application, the performance constraints are set as the phase content of the reinforcement phase being greater than 30% and the sum of the phase content of the matrix phase and the phase content of the reinforcement phase being greater than 90% at a set temperature.

[0080] In step 500, the process of screening out candidate alloy components whose phase contents do not satisfy the performance constraints from multiple groups of candidate alloy components to obtain a preliminary sample includes: screening the phase contents of the candidate alloy components using the performance constraints; determining corresponding candidate alloy components based on the phase contents of the screened candidate alloy components; and obtaining a preliminary sample based on the phase contents of the screened candidate alloy components and the corresponding candidate alloy components.

[0081] In step 600, the process of constructing a Pareto front solution set based on the preliminary samples using the multidimensional crack sensitivity index as the optimization objective includes determining the multidimensional crack sensitivity index values corresponding to the preliminary samples. A non-dominated sorting genetic algorithm is employed to construct the Pareto front solution set based on the preliminary samples and the multidimensional crack sensitivity index values corresponding to the preliminary samples.

[0082] For example, the multidimensional crack sensitivity index includes: FR = T l -T s , M sac =W Al +0.5 W Ti +0.3 W Nb +0.15 W Ta , Wherein, FR represents the freezing range (FR), CSC represents the crack sensitivity coefficient (CSC), HSC represents the hot-tearing susceptibility criterion (HSC), SCI represents the solidification cracking index (SCI), M sac Indicates strain aging cracking in the alloy, SAC stands for strain aging cracking index, which is used to evaluate the sensitivity of the alloy to stress aging cracking. l Indicates the liquidus temperature, T s represents the solid phase temperature, t v is the time period of stress release process in the vulnerable area, t R is the time period of stress release process in stress relaxation zone, Indicates the solid content as f s Time, T v The temperature range of the stress release process, T R Indicates the temperature range of the stress relaxation process in the stress relaxation zone, Indicates the solid content as f s Temperature at time t = temperature at time - solid fraction, represents the volume fraction of the γ phase (i.e., matrix phase), Indicates the critical temperature when the γ phase content reaches 0.8, W 元素 Indicates the mass fraction of an element.

[0083] As an optional embodiment, in order to perform additive manufacturing on the obtained optimal alloy composition to obtain an optimized nickel-based high-temperature alloy, step 700 includes:

[0084] Step 710 , using vacuum induction melting technology, preliminarily melts the alloy elements in the optimal alloy composition to obtain an ingot.

[0085] Step 720 , processing the ingot until the impurity content of the ingot reaches a set content, and using the processed ingot as a master alloy.

[0086] Step 730: atomize the master alloy to obtain alloy powder.

[0087] Step 740 , using a laser metal deposition process, setting the laser power, scanning speed and layer thickness, and depositing the alloy powder to form an optimized target alloy.

[0088] In an exemplary embodiment, in combination with the steps in the above embodiments, the reverse design method of the additive manufacturing nickel-based high-temperature alloy composition is used to reverse design the alloy composition of the nickel-based high-temperature alloy, and the additive manufacturing technology is used to manufacture the nickel-based high-temperature alloy sample as an example.

[0089] Step S1, within the predetermined composition range of the nickel-based high-temperature alloy, the main alloying elements (such as Ni, Cr, Co, Ti, Al, etc.) are clearly defined according to the alloy design requirements to determine the sampling area and sampling point distribution of the composition space. By adopting methods such as random sampling, grid sampling or Latin hypercube sampling to sample the composition points, a uniform and representative discretization of the multi-dimensional composition space is achieved. The composition data of each sampling point is obtained, and the composition data of each sampling point is calculated using thermodynamic calculation software (for example) to obtain the phase content data corresponding to the multi-group composition data of the nickel-based high-temperature alloy at a set temperature, including the phase content of the matrix phase (such as γ phase) and the strengthening phase (such as γ' phase) in the nickel-based high-temperature alloy. In this calculation process, not only the solidification curve, liquidus temperature, solidus temperature and phase transformation behavior of the alloy are taken into account, but also the composition of other phases closely related to performance (such as residual liquid phase, carbide, nitride, boride, etc.) can be predicted.

[0090] In addition, in practical applications, historical experimental data on nickel-based superalloys (such as data collected from literature and real-world laboratory data) can also be obtained. This historical experimental data includes composition data and corresponding phase content data to increase the training data sample. A training dataset is constructed based on the sampled composition data, corresponding phase content data, and historical experimental data. During the construction of the training dataset, the sampled data and historical experimental data are preprocessed, feature extracted, and normalized to ensure the accuracy and consistency of the training data.

[0091] Step S2, constructing an initial performance prediction model based on the machine learning model. Using a weighted machine learning algorithm, the weight of the historical experimental data in the training data set is set to 0.6, and the weight of the phase content data corresponding to the sampled composition data and the composition data obtained by thermodynamic calculation is set to 0.4, and the initial performance prediction model is trained. The initial performance prediction model is continuously optimized by minimizing the first loss function until the loss function value of the prediction result output by the initial performance prediction model meets the set requirements, and a trained initial performance prediction model is obtained and used as the performance prediction model. This enables the performance prediction model to accurately predict the volume fraction of the γ matrix and γ' phase and related performance indicators under unknown composition.

[0092] When training the initial performance prediction model, historical experimental data with a weight of 0.6 is considered more important as a reference during model training. Its errors are amplified (multiplied by 0.6) in the first loss function, while the errors of the sampled data are weakened (multiplied by 0.4). This makes the model more likely to fit the historical experimental data during training. In practical applications, the weights of historical experimental data and sampled data can be adjusted appropriately, with the weight of historical experimental data being greater than that of sampled data.

[0093] Step S3, set the performance constraints: the volume fraction of the γ' phase is greater than 30% and the sum of the γ phase and γ' phase contents is greater than 95%. Based on the performance constraints and the training data set, the conditional variational autoencoder is trained. The CVAE consists of two parts: the encoder and the decoder. Figure 2 As shown. Based on the performance constraint conditions, the target performance is set (the target performance is the specific γ phase content and γ' phase content), and the component data in the training data set, the phase content data corresponding to the component data, and the target performance are feature spliced to obtain a mixed feature vector. The mixed feature vector (corresponding to Figure 2The input composition database (input composition database) and performance constraints are input to the encoder. After processing through the first fully connected layer, ReLU activation function and multi-layer nonlinear projection, the mean and standard deviation of the latent vector are output, defining the distribution of the latent space. The mixed feature vector is first encoded into mean (μ) and standard deviation (σ) by the encoder. Then, the standard deviation σ is combined with a normal distribution random function to obtain a noise vector ∈~N(0,I). The latent vector is calculated by z=μ+σ*∈, and then the latent vector is passed as input to the decoder. Based on the latent vector and the performance constraints, the decoder reconstructs the alloy composition through the second fully connected layer and ReLU activation function, and outputs the predicted results of the target alloy composition (including multiple sets of alloy compositions), ensuring that the content of γ phase and γ' phase of the alloy meets the set requirements. Through this design, the decoder not only generates new candidate alloy compositions based on the alloy composition characteristics in the latent space, but also can directly generate alloy compositions that meet specific performance targets (such as the desired phase content).

[0094] The conditional variational autoencoder is evaluated using a second loss function. By minimizing the reconstruction loss function and iteratively optimizing the KL divergence loss function, the conditional variational autoencoder output data reaches the set loss function value, resulting in a trained conditional variational autoencoder. The decoder in the trained conditional variational autoencoder is then used as the alloy composition generation model.

[0095] Step S4: Obtain a random vector that satisfies a normal distribution. Input the random vector and the performance constraints into the alloy composition generation model to obtain multiple sets of candidate alloy compositions. Input the candidate alloy compositions into the performance prediction model to obtain the corresponding phase contents. From the multiple sets of candidate alloy compositions, candidate alloy compositions whose phase contents do not meet the performance constraints are screened out to obtain a preliminary sample.

[0096] The liquidus temperature (T l ), solidus temperature (T s ) and solid-liquid coexistence solidification interval data. Based on these data, the multi-dimensional crack sensitivity index calculation formula in the above embodiment is used, such as the crack sensitivity coefficient FR=T l -T s , and other CSC, HSC, M reflected by changes in temperature and time parameters sac, SAC, and other indicators were used to calculate the crack sensitivity coefficients of each group of candidate alloy compositions. Using multidimensional crack sensitivity as the optimization target, a non-dominated sorting genetic algorithm was employed to construct a Pareto front solution set based on the initial sample. The optimal alloy composition was selected as the one that achieved the best balance between crack resistance and high-temperature performance. This method comprehensively considers various performance requirements, ensuring that the final composition selected not only meets the high-temperature stability and mechanical property requirements of the laser additive manufacturing process, but also significantly reduces the likelihood of cracking.

[0097] In this embodiment, the hot crack criterion FR value of the additively manufactured nickel-based high-temperature alloy is 214K, and the γ' phase content at the set temperature (900°C) is 39%. The chemical composition of this group of nickel-based high-temperature alloys is shown in Table 1.

[0098] Table 1 Chemical composition

[0099] Ni(%) Cr(％) Co(%) W(％) Ti(%) Al(％) Ta(%) Nb(%) C(％) B(％) margin 24.6 17 2.15 3.74 1.9 2.83 4.45 0.097 0.007

[0100] Step S5: performing additive manufacturing on the obtained optimal alloy composition to obtain a nickel-based high-temperature alloy additive manufacturing sample.

[0101] Based on the optimal alloy composition, nickel and various alloying elements (including but not limited to Cr, Co, Ti, Al, Mo, and W) are initially melted using vacuum induction melting under an inert atmosphere to produce an ingot. The ingot is then heated to 1500-1600°C for 5-8 minutes in a vacuum environment to ensure alloy homogeneity and chemical stability. The resulting ingot undergoes multiple mechanical and chemical surface treatments (impurity content is controlled below 10 ppm) to produce a master alloy. The master alloy is then treated using an improved gas atomization process that utilizes both conventional argon atomization and mixed shielding gas atomization. By controlling the spray pressure (e.g., 6-10 MPa), temperature (1500-1600°C), and flow pattern, a nearly spherical alloy powder with a wide particle size distribution (target particle size range of 50-100 μm) is obtained. Finally, the alloy powder is deposited and formed using laser metal deposition, direct energy deposition, or other process platforms suitable for additive manufacturing, and printed into additively manufactured samples. The controlled process parameters—including laser or energy source power (controlled within the range of 100 to 300 W), scanning speed (set at 1000 to 2000 mm / s), filling spacing (1.2-1.6 mm) and layer thickness (20 to 80 μm)—can all be flexibly adjusted according to actual process requirements.

[0102] After printing, the nickel-based superalloy additively manufactured samples undergo comprehensive performance testing to verify their material properties. These tests primarily include mechanical property evaluation and crack sensitivity determination. Mechanical property testing comprehensively assesses the structural strength and toughness of the samples through tensile, compression, and hardness testing. In crack sensitivity testing, the samples undergo multiple cycles of thermal cycling at high temperatures to determine their resistance to thermal cracking, solidification cracking, and liquefaction cracking.

[0103] Based on the steps in the above embodiments, the reverse design method for additive manufacturing of nickel-based high-temperature alloy composition provided by this application has the following beneficial effects:

[0104] 1. Efficiency and Accuracy: By combining thermodynamic calculations, machine learning, and alloy composition generation models, this approach can rapidly generate alloy compositions that meet specific performance requirements, significantly improving the efficiency and accuracy of alloy design. Compared to traditional empirical methods or experimental verification processes, this approach can generate multiple candidate alloy compositions in a shorter time, avoiding lengthy and expensive experimental costs.

[0105] 2. Performance target orientation: By introducing target performance (such as the γ-phase and γ'-phase content at a specific temperature) into the conditional variational autoencoder, the alloy composition can be reverse-engineered directly based on preset performance indicators (i.e., performance constraints), ensuring that the generated alloy has the desired functional properties and meets specific engineering application requirements.

[0106] 3. Innovation and Flexibility: This application breaks through the limitations of traditional alloy design by automatically exploring and generating high-potential material combinations from a large number of possible alloy compositions. By learning the complex relationship between alloy composition and properties, CVAE provides a design space that is more flexible and innovative than traditional methods, making the alloy design optimization process more systematic and intelligent.

[0107] 4. Comprehensiveness and Scalability: This application allows for the modification of target alloy types, obtaining training datasets for different alloy types, and applying them to reverse design optimization of other alloys or materials, thereby increasing the flexibility of alloy composition design methods. By combining thermodynamic calculations with machine learning, this approach can adapt to different material systems and target performance requirements, providing a universal optimization tool for the materials science field.

[0108] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data related to the reverse design method of the composition of the additively manufactured nickel-based high-temperature alloy. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a reverse design method of the composition of the additively manufactured nickel-based high-temperature alloy is implemented.

[0109] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps of the above-mentioned method embodiments when executing the computer program.

[0110] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0111] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0113] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may 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 may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0114] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0115] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, 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.

[0116] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A reverse design method for additive manufacturing of nickel-based high-temperature alloy composition, characterized in that: include: Obtaining multi-group composition data of a target nickel-based high-temperature alloy at a set temperature and phase content data corresponding to the composition data to generate a training data set; Constructing an initial performance prediction model, training the initial performance prediction model based on the training data set until the output of the initial performance prediction model meets the set requirements, obtaining a trained initial performance prediction model, and using it as the performance prediction model; Setting performance constraints; Using a conditional variational autoencoder, constructing an alloy composition generation model based on the training data set and the performance constraints; Obtain a random vector; the random vector satisfies a normal distribution; Inputting the random vector and the performance constraint condition into the alloy composition generation model to obtain multiple groups of candidate alloy compositions; Inputting multiple groups of candidate alloy components into the performance prediction model to obtain the phase contents of the candidate alloy components; Screening out candidate alloy components whose phase contents do not satisfy the performance constraints from multiple groups of candidate alloy components to obtain a preliminary sample; Taking the multi-dimensional crack sensitivity index value as the optimization target, constructing a Pareto front solution set based on the preliminary selected samples; taking the candidate alloy composition corresponding to the preliminary selected samples that meet the set conditions in the Pareto front solution set as the optimal alloy composition; An optimized nickel-based high-temperature alloy is obtained based on the optimal alloy composition using additive manufacturing technology.

2. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 1, characterized in that: Constructing an initial performance prediction model, training the initial performance prediction model based on the training data set until the output of the initial performance prediction model meets the set requirements, and obtaining the trained initial performance prediction model includes: Build an initial performance prediction model based on the machine learning model; training the initial performance prediction model based on the training data set; The initial performance prediction model is evaluated using a first loss function until the loss function value of the prediction result output by the initial performance prediction model meets the set requirements, thereby obtaining a trained initial performance prediction model.

3. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 1, characterized in that: The conditional variational autoencoder includes an encoder and a decoder; A conditional variational autoencoder is used to construct an alloy composition generation model based on the training data set and the performance constraints, including: training a conditional variational autoencoder based on the training dataset and the performance constraints; Evaluating the conditional variational autoencoder using a second loss function until a loss function value of output data of the conditional variational autoencoder meets a set requirement, thereby obtaining a trained conditional variational autoencoder; the second loss function includes a KL divergence loss function term and a reconstruction loss function term; The decoder in the trained conditional variational autoencoder is used as the alloy composition generation model.

4. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 3, characterized in that: setting a target performance based on the performance constraints; Performing feature splicing on the composition data of the target alloy and the target performance in the training data set, and inputting the features into the encoder to generate a mean vector and a standard deviation vector; Obtain a normally distributed random vector based on the mean vector and the standard deviation vector; The normally distributed random vector and the performance constraint condition are input into the decoder to obtain the output data.

5. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 1, characterized in that: The phase content of the candidate alloy components includes the phase content of the matrix phase and the phase content of the strengthening phase; The performance constraint condition includes that the phase content of the reinforcement phase is greater than a first set threshold value and the sum of the phase content of the matrix phase and the phase content of the reinforcement phase is greater than a second set threshold value at the set temperature; Eliminate candidate alloy components whose phase contents do not satisfy the performance constraints from multiple groups of candidate alloy components to obtain a preliminary sample, including: screening the phase contents of the candidate alloy components using the performance constraints; determining corresponding candidate alloy components according to the phase contents of the screened candidate alloy components; The preliminary selected sample is obtained based on the phase contents of the screened candidate alloy components and the corresponding candidate alloy components.

6. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 1, characterized in that: Taking the multi-dimensional crack sensitivity index as the optimization target, the Pareto front solution set is constructed based on the preliminary selected samples, including: Determining the multi-dimensional crack sensitivity index value corresponding to the primary selected sample; A non-dominated sorting genetic algorithm is adopted to construct the Pareto front solution set based on the preliminary selected samples and the multi-dimensional crack sensitivity index values corresponding to the preliminary selected samples.

7. The reverse design method for additive manufacturing nickel-based high-temperature alloy composition according to claim 1, characterized in that: The optimized target alloy is obtained based on the optimal alloy composition by using additive manufacturing technology, including: Using vacuum induction melting technology, the alloy elements in the optimal alloy composition are preliminarily melted to obtain an ingot; treating the ingot until the impurity content of the ingot reaches a set content, and using the treated ingot as a master alloy; atomizing the master alloy to obtain alloy powder; A laser metal deposition process is adopted to set the laser power, scanning speed and layer thickness, and the alloy powder is deposited and formed to obtain an optimized target alloy.

8. A computer device comprising: 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 implement the inverse design method for additively manufactured nickel-based high-temperature alloy composition according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the reverse design method for the additively manufactured nickel-based high-temperature alloy composition according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the reverse design method for the additively manufactured nickel-based high-temperature alloy composition according to any one of claims 1 to 7 is implemented.

Images