Method, device and equipment for determining reference interval of tungsten alloy parameters

Abstract

The application provides a method, device and equipment for determining a parameter reference interval of a tungsten alloy, and relates to the technical field of refractory metal material design and intelligent prediction. The method comprises the following steps: obtaining tungsten alloy data; the tungsten alloy data comprises chemical composition information, sintering process parameters and second phase attribute information. Physical mechanism features of the tungsten alloy data are extracted; based on the physical mechanism features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed. The tungsten alloy data, the physical mechanism features, the dispersion strengthening channel codes and the solid solution strengthening channel codes are all input into a preset back propagation neural network model for performance prediction, and predicted mechanical property data is output. Based on the predicted mechanical property data, back analysis is performed to generate a tungsten alloy parameter reference interval. The problems of insufficient mechanism expression, weak model interpretability and difficulty in simultaneously describing the coupling effect of dispersion strengthening and solid solution strengthening in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of refractory metal material design and intelligent prediction technology, and more specifically, to a method, apparatus and equipment for determining the reference range of tungsten alloy parameters. Background Technology

[0002] Currently, tungsten and tungsten alloys possess significant application value in fields such as plasma-facing materials for fusion devices, aerospace thermal protection components, and high-temperature molds due to their high melting point, high strength, high thermal conductivity, and excellent high-temperature service capability. However, single tungsten materials still have limitations in terms of room-temperature brittleness, insufficient toughness, and microstructural stability. Introducing second phases such as carbides and oxides to achieve dispersion strengthening, or using elements such as Re, Ta, Mo, and Nb to achieve solid solution strengthening, are important technical pathways to improve the overall performance of tungsten alloys.

[0003] Existing methods for predicting material properties typically involve directly inputting composition and process parameters into neural networks or other machine learning models. While these methods can yield some fitting results, they fail to adequately characterize the physical differences between dispersion strengthening and solid solution strengthening mechanisms. Consequently, the interpretability of these models is limited, and they cannot distinguish between the dispersion strengthening contribution caused by second-phase particles and the lattice distortion strengthening contribution caused by solid solution elements. This makes it impossible to determine the impact of each strengthening mechanism on the predicted mechanical property data. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, and equipment for determining the reference range of tungsten alloy parameters, in order to solve the problems of insufficient mechanism expression, weak model interpretability, and difficulty in simultaneously characterizing the coupling effect of dispersion strengthening and solid solution strengthening in existing material performance prediction methods.

[0005] Firstly, a method for determining a reference range for tungsten alloy parameters is provided, which may include: Acquire tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase property information. The physical mechanism features of the tungsten alloy data are extracted; the physical mechanism features include chemical bond features, interface mismatch features, and thermodynamic stability features; based on the physical mechanism features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed. The tungsten alloy data, the physical mechanism characteristics, the dispersion strengthening channel encoding, and the solid solution strengthening channel encoding are all input into a preset backpropagation neural network model for performance prediction, and the predicted mechanical property data are output. Based on the predicted mechanical property data, a back-calculation analysis is performed to generate a reference range for tungsten alloy parameters.

[0006] Secondly, a device for determining a reference range for tungsten alloy parameters is provided, the device comprising: The acquisition module is used to acquire tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second-phase property information. An extraction module is used to extract the physical mechanism features of the tungsten alloy data; the physical mechanism features include chemical bond features, interface mismatch features, and thermodynamic stability features; based on the physical mechanism features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed; The prediction module is used to input the tungsten alloy data, the physical mechanism characteristics, the dispersion strengthening channel code, and the solid solution strengthening channel code into a preset backpropagation neural network model for performance prediction, and output predicted mechanical property data. The reverse calculation module is used to perform reverse analysis based on the predicted mechanical property data to generate the reference range of the tungsten alloy parameters.

[0007] Thirdly, an electronic device is provided, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements any of the steps described in the first aspect above.

[0008] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements any of the steps of the method described in the first aspect above.

[0009] This application provides a method, apparatus, and equipment for determining the reference range of tungsten alloy parameters. The method involves acquiring tungsten alloy data, including chemical composition information, sintering process parameters, and second-phase property information. Physical mechanism features of the tungsten alloy data are extracted, including chemical bond characteristics, interface mismatch characteristics, and thermodynamic stability characteristics. Based on these physical mechanism features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed. The tungsten alloy data, physical mechanism features, dispersion strengthening channel codes, and solid solution strengthening channel codes are all input into a preset backpropagation neural network model for performance prediction, outputting predicted mechanical property data. Back-propagation analysis is performed based on the predicted mechanical property data to generate the tungsten alloy parameter reference range. This solution can simultaneously adapt to dispersion-strengthened tungsten alloys, solid solution-strengthened tungsten alloys, and composite-strengthened tungsten alloy systems. Through back-propagation analysis such as interpretability analysis, key parameter windows can be back-propagated, providing a basis for alloy composition design and sintering process optimization. This solves the problems of insufficient mechanism expression, weak model interpretability, and difficulty in simultaneously characterizing the coupling effect of dispersion strengthening and solid solution strengthening in existing technologies. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a method for determining a reference range for tungsten alloy parameters provided in this application embodiment; Figure 2 A flowchart illustrating another method for determining the reference range of tungsten alloy parameters provided in this application; Figure 3 A flowchart illustrating a model training process provided in this application; Figure 4 A flowchart illustrating another method for determining the reference range of tungsten alloy parameters provided in this application; Figure 5 A flowchart of target performance-driven reverse design and process recommendation is provided for embodiments of this application; Figure 6 This is a schematic diagram of the output of a reverse design module provided in an embodiment of this application; Figure 7 A schematic diagram of a device for determining a reference range of tungsten alloy parameters provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The words "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The words "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but do not exclude other elements or objects. The words "connected," "coupled," or "connected," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0013] The symbols and terms defined in the embodiments of this application are as follows:

[0014] The method for determining the reference range of tungsten alloy parameters provided in this application can be applied to electronic devices, terminal devices, devices or apparatuses for determining the reference range of tungsten alloy parameters, or other devices or apparatuses capable of executing this embodiment, and there are no limitations on this application. This embodiment describes the execution subject as an electronic device. This application particularly relates to material mechanism characteristics, physical constraints, and backpropagation neural network models.

[0015] The terminal can be a user equipment (UE) such as a mobile phone, smartphone, laptop computer, digital broadcast receiver, personal digital assistant (PDA), or tablet computer (PAD), handheld device, in-vehicle device, wearable device, computing device, or other processing device connected to a wireless modem, mobile station (MS), or mobile terminal. This terminal has the ability to communicate with one or more core networks via a radio access network (RAN).

[0016] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0017] Figure 1 This is a flowchart illustrating a method for determining a reference range for tungsten alloy parameters, provided in an embodiment of this application. Figure 1 As shown, the method may include: Step S101: Obtain tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase property information.

[0018] For example, tungsten alloy data is acquired; the tungsten alloy data includes chemical composition information, sintering process parameters, and second-phase attribute information. The chemical composition information includes the tungsten matrix (W matrix) and the content of alloying elements. The W matrix represents a matrix material with metallic tungsten as the main component. The alloying element content includes the type of solid solution element and the content of that solid solution element. The solid solution elements include at least one of the following: Re, Ta, Mo, and Nb. The second-phase attribute information includes at least one of the following: second-phase type, initial particle size of the second phase, volume fraction of the second phase, melting point of the second phase, and elastic modulus of the second phase. The second-phase attribute information mainly refers to information used to achieve dispersion strengthening. The second-phase type includes carbide type and oxide type. The second-phase type includes at least one of the following: TiC, ZrC, HfC, La2O3, and Y2O3. Sintering process parameters refer to the core thermal process parameters, pressure and atmosphere parameters, and raw material and product characteristic parameters in the actual sintering process. Core thermal process parameters include sintering temperature, sintering time, and heating / cooling rate. Pressure and atmosphere parameters include sintering pressure and sintering atmosphere. Raw material and product characteristic parameters include raw material (green) properties and sintered body microstructure.

[0019] In this step, at least one set of tungsten alloy data is acquired; each set of tungsten alloy data includes chemical composition information, sintering process parameters, and second-phase property information. Specifically, multiple sets of tungsten alloy data can be acquired based on tungsten matrices with different compositions, or multiple sets of tungsten alloy data can be acquired based on tungsten matrices with the same composition. For example, for tungsten matrices with the same composition, chemical composition information, and second-phase property information, multiple sets of tungsten alloy data are generated based on changes in sintering process parameters. Multiple sets of tungsten alloy data include: First set of tungsten alloy data: chemical composition information, second phase property information, and sintering temperature of 1500℃; The second set of tungsten alloy data includes: chemical composition information, second phase property information, and sintering temperature of 1600℃. The third set of tungsten alloy data includes: chemical composition information, second phase property information, and sintering temperature of 1700℃.

[0020] Step S102: Extract the physical mechanism characteristics of tungsten alloy data; the physical mechanism characteristics include chemical bond characteristics, interface mismatch characteristics and thermodynamic stability characteristics; based on the physical mechanism characteristics, construct dispersion strengthening channel codes and solid solution strengthening channel codes.

[0021] For example, physical mechanism features are extracted from tungsten alloy data, including chemical bond features, interfacial mismatch features, and thermodynamic stability features. Then, based on these features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed, each corresponding to a dispersion strengthening channel. The tungsten alloy data, chemical bond features, interfacial mismatch features, thermodynamic stability features, dispersion strengthening channel codes, and solid solution strengthening channel codes are then fused to obtain the fusion result.

[0022] Step S103: Input the tungsten alloy data, physical mechanism characteristics, dispersion strengthening channel encoding and solid solution strengthening channel encoding into the preset backpropagation neural network model for performance prediction, and output the predicted mechanical property data.

[0023] For example, the preset backpropagation neural network model is trained based on a tungsten alloy dataset, which includes at least multi-component chemical composition information, sintering process parameters, second-phase property information, and mechanical property data. In this step, the fusion result is input into the preset backpropagation neural network model for prediction to obtain predicted mechanical property data.

[0024] Among them, the chemical bond characteristics are composed of at least one of the following: average bond energy, element electronegativity difference, space group encoding, number of d electrons in transition metals, and point defect formation energy; the interface mismatch characteristics are established based on the difference in lattice constants between the reinforcing phase and the tungsten matrix; and the thermodynamic stability characteristics are established based on the formation enthalpy, mixing free energy, interface energy, or phase stability-related parameters.

[0025] Step S104: Perform reverse analysis based on the predicted mechanical property data to generate a reference range for tungsten alloy parameters.

[0026] For example, by combining dimensionality reduction analysis and interpretability analysis, a back-analysis is performed on the predicted mechanical property data to generate a reference range for tungsten alloy data parameters. This reference range includes a performance improvement range and a performance-sensitive range. The performance improvement range is the range of characteristic values ​​that cause the predicted mechanical property data to improve relative to the baseline performance and exceed a first preset threshold. The performance-sensitive range is the range of characteristic values ​​that cause the absolute value of the change in the predicted mechanical property data to exceed a second preset threshold. Both the first and second preset thresholds can be fixed values ​​or user-defined dynamically adjustable values.

[0027] The method provided in this application acquires tungsten alloy data, including chemical composition information, sintering process parameters, and second-phase property information. It extracts the physical mechanism characteristics of the tungsten alloy data, including chemical bond characteristics, interface mismatch characteristics, and thermodynamic stability characteristics. Based on these physical mechanism characteristics, it constructs dispersion strengthening channel codes and solid solution strengthening channel codes. The tungsten alloy data, physical mechanism characteristics, dispersion strengthening channel codes, and solid solution strengthening channel codes are all input into a preset backpropagation neural network model for performance prediction, outputting predicted mechanical property data. Based on the predicted mechanical property data, a backpropagation analysis is performed to generate a reference range for tungsten alloy parameters. This solution can simultaneously adapt to dispersion-strengthened tungsten alloys, solid solution-strengthened tungsten alloys, and composite-strengthened tungsten alloy systems. Through backpropagation analysis such as interpretability analysis, key parameter windows can be backpropagated, providing a basis for alloy composition design and sintering process optimization. This solves the problems of insufficient mechanism expression, weak model interpretability, and difficulty in simultaneously characterizing the coupling effect of dispersion strengthening and solid solution strengthening in existing technologies.

[0028] Figure 2 A flowchart illustrating another method for determining a reference range for tungsten alloy parameters provided in this application is shown below. Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the method is described in detail below, and the method includes: Step S201: Obtain tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase property information.

[0029] For example, this step is the same as step S101, and will not be described again.

[0030] Step S202: Extract the physical mechanism features of the tungsten alloy data; the physical mechanism features include chemical bond features, interface mismatch features and thermodynamic stability features; based on the physical mechanism features, construct dispersion strengthening channel codes and solid solution strengthening channel codes.

[0031] For example, this step is the same as step S102, and will not be described again.

[0032] Step S203: Input the tungsten alloy data, physical mechanism characteristics, dispersion strengthening channel encoding and solid solution strengthening channel encoding into the preset backpropagation neural network model for performance prediction, and output the predicted mechanical property data.

[0033] In one example, the training process of the backpropagation neural network model includes: acquiring a tungsten alloy dataset; the tungsten alloy dataset includes at least multi-component chemical composition information, sintering process parameters, second-phase property information, and predicted mechanical property data; performing Pearson correlation coefficient pre-screening and recursive feature elimination screening on the tungsten alloy dataset sequentially to obtain a target feature subset for multi-objective performance prediction; extracting chemical bond features, interface mismatch features, and thermodynamic stability features from the target feature subset; constructing dispersion strengthening channel encoding and solid solution strengthening channel encoding based on the chemical bond features, interface mismatch features, and thermodynamic stability features; fusing the target feature subset, chemical bond features, interface mismatch features, thermodynamic stability features, dispersion strengthening channel encoding, and solid solution strengthening channel encoding to obtain a fusion result; constructing a candidate neural network model based on the fusion result; and training, validating, and optimizing the hyperparameters of the candidate neural network model using the backpropagation algorithm and a total loss function including a physical constraint loss term to obtain a converged backpropagation neural network model.

[0034] In one example, the dispersion strengthening channel encoding corresponds to the particle size, particle spacing, second phase volume fraction and interfacial modulus mismatch, lattice mismatch degree and Orowan correlation factor, while the solid solution strengthening channel encoding corresponds to the solid solution element concentration, atomic radius mismatch, shear modulus mismatch, valence electron concentration difference, mixing enthalpy or solid solution stability parameter.

[0035] In one example, the total loss function includes a data fitting loss term and a physical constraint loss term, wherein the physical constraint loss term includes at least one or more of the following: non-negative constraint of dispersion strengthening contribution, non-negative constraint of solid solution strengthening contribution, limit degradation consistency constraint, and dimensional consistency constraint.

[0036] In one example, the extreme degradation consistency constraint is used to control the output constraint of the backpropagation neural network model to be solid solution reinforcement dominant if the diffusion enhancement channel encoding is lower than a preset first threshold; and to control the output constraint of the backpropagation neural network model to be diffusion enhancement dominant if the solid solution enhancement channel encoding is lower than a preset second threshold.

[0037] For example, tungsten alloy data, physical mechanism characteristics, dispersion strengthening channel encoding, and solid solution strengthening channel encoding are all input into a preset backpropagation neural network model for performance prediction, and the predicted mechanical property data is output. The predicted mechanical property data includes at least one of the following: tensile strength, hardness, and elongation.

[0038] Optional, Figure 3 A flowchart illustrating the model training process provided in this application is shown below. Figure 3 As shown, the training process for obtaining the backpropagation neural network model includes: collecting and integrating historical and experimental data of dispersion-strengthened tungsten alloys and solid solution-strengthened tungsten alloys, constructing a tungsten alloy dataset in a unified format, namely the initial feature set F0 (i.e., Figure 3 The tungsten alloy dataset includes at least four types of information: first, multi-component chemical composition information, such as the content of W matrix, second-phase elements, and solid solution elements; second, sintering process parameters, such as heating rate, maximum sintering temperature, sintering pressure, and holding time; third, second-phase property information, such as second-phase type, initial particle size, volume fraction, melting point, and elastic modulus; and fourth, predicted mechanical property data, such as tensile strength, elongation, and hardness. Then, the tungsten alloy dataset undergoes coarse screening and fine selection. First, for the initial feature set F0, the Pearson correlation coefficient (PCC) between each feature and each target performance variable (tensile strength, elongation, hardness) is calculated. Features that are weakly correlated with all target performance variables or are not significant are deleted, i.e., features whose absolute values ​​of correlation coefficients with all target performance variables are below a preset threshold are deleted, resulting in the selected feature set F_PCC. Subsequently, a recursive feature elimination (RFE) screening process is performed with tensile strength, elongation, and hardness as target performance variables. The selected feature sets F_PCC (which includes multiple candidate key features) corresponding to each target performance variable are then merged to obtain the final optimal feature subset F*, i.e., the target feature subset, to balance the consistency and effectiveness of multi-output predictions. In this application, key features such as the original particle size of the second phase, the volume fraction of the second phase, the type of the second phase, the melting point of the second phase, the type of solid solution element, the concentration of solid solution element, the maximum sintering temperature, the sintering pressure, and the sintering holding time are preferably retained.

[0039] Figure 4 A flowchart illustrating another method for determining a reference range for tungsten alloy parameters provided in this application is shown below. Figure 4As shown, physical mechanism features are further constructed based on the target feature subset. Specifically, chemical bond features (phi_bond), interface misfit features (phi_misfit), and thermodynamic stability features (phi_thermo) are extracted from the target feature subset. Chemical bond features (phi_bond) and crystal structure features include average bond energy, space group encoding, Pauling electronegativity difference, crystal structure type encoding, number of d electrons in transition elements, and point defect formation energy. Chemical bond features are used to distinguish the differences in the chemical bond nature between carbide-reinforced phases and oxide-reinforced phases. Pauling electronegativity difference is used to characterize the differences in bonding tendency between elements, crystal structure type encoding is used to characterize the crystal structure type of the phase, and space group encoding is used to characterize the differences in crystal symmetry between different phases. Interface misfit features (phi_misfit) are used to characterize the difference in lattice constants and interfacial compatibility between the reinforcing phase and the tungsten matrix. The thermodynamic stability feature (phi_thermo) is used to characterize the phase stability of the second phase under sintering and service conditions. Interface misfit features and thermodynamic stability features (i.e....) Figure 4 The interface and thermodynamic characteristics (abbreviated as "interface and thermodynamic features") include the second phase / matrix lattice mismatch. Formation energy coefficient / Interfacial compatibility, thermodynamic stability parameters .

[0040] Furthermore, such as Figure 4 As shown, this application employs a mechanism discrimination and dual-channel enhancement mechanism coding method. Based on chemical bond characteristics, interfacial mismatch characteristics, and thermodynamic stability characteristics, dispersion enhancement channel code A and solid solution enhancement channel code B are constructed. The dispersion enhancement channel code corresponds to the dispersion enhancement channel z_d, and the solid solution enhancement channel code corresponds to the solid solution enhancement channel z_s.

[0041] The dispersion strengthening channel z_d primarily encodes variables related to the dispersion strengthening factor. These channels include: second-phase volume fraction, particle size, Orowan spacing (i.e., Orowan strengthening factor), interfacial modulus mismatch, interfacial effects, and second-phase / matrix lattice mismatch. The solid solution strengthening channel z_s primarily encodes variables related to the solid solution strengthening factor, and a mechanism discrimination function m is constructed based on these variables. These channels may include: solid solution element concentration, atomic radius mismatch, shear modulus mismatch, valence electron concentration difference, mixing enthalpy, or solid solution stability parameters.

[0042] Computable expressions include: The matrix lattice mismatch δ = |a_p-a_m| / a_m × 100%; Atomic radius mismatch η = |r_s-r_W| / r_W × 100%; Shear modulus mismatch μ = |G_s-G_W| / G_W × 100%; Where a_p represents the lattice constant of the second phase particles or the reinforcing phase; a_m represents the lattice constant of the tungsten matrix or the W-based solid solution matrix; r_s represents the atomic radius of the solid solution element; r_W represents the atomic radius of the W element; G_s represents the shear modulus of the solid solution element or the corresponding solid solution; and G_W represents the shear modulus of the W matrix.

[0043] The target feature subset, chemical bond features, interface mismatch features, thermodynamic stability features, dispersion strengthening channel encoding and solid solution strengthening channel encoding, etc., are spliced ​​and fused to obtain the fusion result. A candidate neural network model is then established based on the fusion result.

[0044] The target feature subset is concatenated with the physical mechanism features phi_bond, phi_misfit, phi_thermo, and the dual-channel mechanism encodings z_d and z_s to form a joint feature space X_fuse. A candidate neural network model is then established. Finally, the candidate neural network model is trained, validated, and its hyperparameters optimized using the backpropagation algorithm, a total loss function including a physical constraint loss term, and a mechanism discriminant function m, resulting in a converged backpropagation neural network model. Specifically, a backpropagation neural network (BPNN) is used to construct a multi-output regression model. Preferably, the regression model includes one input layer, at least two hidden layers, and one output layer. The number of neurons in the hidden layers can be set to 128, 64, or 32, etc., and the number of neurons in the output layer is consistent with the number of target performance values ​​to be predicted. The hidden layers preferably use ReLU, Softplus activation functions, or other continuously differentiable activation functions, and the output layer adopts a multi-output regression structure to balance nonlinear expressiveness and gradient propagation stability. During training, the tungsten alloy dataset is divided into training, validation, and test sets, or K-fold cross-validation is used to evaluate model performance. At least one of the following metrics—coefficient of determination R², mean absolute error (MAE), and root mean square error (RMSE)—is used as the evaluation metric to obtain a converged backpropagation neural network model. The network parameters are updated using the Adam or AdamW optimizer, and the learning rate, number of neurons in the hidden layers, batch size, and regularization parameters are determined through grid search, random search, or Bayesian optimization. The objective is to minimize the total loss function L_total. To prevent overfitting, L2 regularization, Dropout, and early stopping strategies can be introduced.

[0045] The training model framework includes both a mechanism discrimination function and a total loss function. The total loss function is the objective function used to optimize model parameters during neural network training. It constrains the error between the model's predicted mechanical properties and the actual mechanical properties, and ensures the model output conforms to the material strengthening laws through a physical constraint loss term. The total loss function is expressed as L_total = L_data + λ * L_phy, where λ is the weight coefficient of the physical constraint loss term, L_data is the data fitting loss term used to constrain the error between the model's predicted mechanical properties and the actual mechanical properties, and L_phy represents the physical constraint loss term used to constrain the model output to conform to the physical laws of dispersion strengthening and solid solution strengthening. The physical constraints L_phy include at least the following: First, a non-negative constraint on the contribution of dispersion enhancement; second, a non-negative constraint on the contribution of solid solution enhancement; third, a limit degradation consistency constraint, i.e., when z_d is lower than a preset first threshold and approaches zero, the model output degenerates into a solid solution enhancement-dominated result, and when z_s is lower than a preset second threshold and approaches zero, the model output degenerates into a dispersion enhancement-dominated result; and fourth, a dimension consistency constraint, i.e., performing dimensionless or standardized processing on the input features.

[0046] The mechanism discrimination function is mainly used to determine whether the current sample belongs to a state dominated by diffusion enhancement, solid solution enhancement, or a state of coupling dominance based on the diffusion enhancement channel code z_d and the solid solution enhancement channel code z_s. It is a mechanism classification or mechanism weight discrimination function.

[0047] The relationship between the mechanism discriminant function and the total loss function is as follows: the mechanism discriminant function can be used as part of the physical constraint loss term, or used to generate mechanism weights and participate in the calculation of the physical constraint loss term, but the mechanism discriminant function itself is not equivalent to the total loss function. When training a neural network, the total loss function can include a data fitting loss term and a physical constraint loss term; among them, the physical constraint loss term can use the results of the mechanism discriminant function (e.g., mechanism weights) to constrain the contributions of diffusion enhancement, solid solution enhancement, and their coupling state.

[0048] Therefore, the mechanism discrimination function m is used to determine "which mechanism this sample is mainly reinforced by"; the total loss function is used to train the model and make the model predict more accurately; for example, if z_d is high and z_s is low, m judges it to be dominated by diffusion reinforcement; if z_d is low and z_s is high, m judges it to be dominated by solid solution reinforcement; if both z_d and z_s are high, m judges it to be dominated by composite reinforcement.

[0049] Step S204: Perform reverse analysis based on the predicted mechanical property data to generate a reference range for tungsten alloy parameters.

[0050] In one example, the tungsten alloy parameter reference range includes a performance improvement range and a performance sensitive range; wherein, the performance improvement range is the range of characteristic values ​​that make the predicted mechanical performance data improve relative to the baseline performance and exceed a preset first threshold, and the performance sensitive range is the range of characteristic values ​​that make the absolute value of the change in the predicted mechanical performance data exceed a second preset threshold.

[0051] In one example, step S204 includes several implementations: The first implementation of step S204 is as follows: Principal component analysis is used to project the input tungsten alloy data into a two-dimensional or three-dimensional space to generate principal component coordinate data; based on t-distribution random neighborhood embedding, the tungsten alloy data is nonlinearly reduced to a two-dimensional or three-dimensional embedding space to obtain embedded coordinate data; based on the embedded coordinate data, the clustering trend is determined; a partial dependency graph is used to determine the influence trend of a single feature on the predicted mechanical property data, and the PDP curve data of that feature is generated according to the influence trend; the features are information on each chemical composition, each sintering process parameter, and each second phase attribute information; the PDP curve data is generated by calling a backpropagation neural network model to predict different feature values; the coupling effect between pairs of features is analyzed using a feature interaction graph to obtain the mechanical property change trend under the joint change of the two features; based on the principal component coordinate data, clustering trend, PDP curve data, and mechanical property change trend, a reference interval for tungsten alloy parameters and a reference interval for tungsten alloy data are generated.

[0052] In one example, a reference interval for tungsten alloy parameters is generated based on the overall distribution of principal component coordinate data, clustering trends, PDP curve data, and mechanical property variation trends. This includes: determining candidate reference intervals for each feature based on PDP curve data and mechanical property variation trends; determining the overall distribution of tungsten alloy data in low-dimensional space based on principal component coordinate data, and determining target sample regions that satisfy preset sample rules based on multiple sample regions in the overall distribution; determining the spatial distribution characteristic parameters of the target sample regions; determining target sample clusters that satisfy preset priority labels based on the clustering labels of each sample cluster characterized by clustering trends; taking the intersection of the target sample regions and target sample clusters based on the spatial distribution characteristic parameters; and restricting the candidate reference intervals to the intersection to obtain the final reference interval for tungsten alloy parameters.

[0053] In one example, based on PDP curve data and mechanical performance change trends, candidate reference intervals for each feature are determined, including: calculating the absolute value of the change in performance relative to the baseline at each point on the curve based on the PDP curve data, extracting feature value sub-intervals whose absolute value of change is greater than a preset threshold, and using them as candidate reference intervals for that feature; baseline performance refers to the baseline performance data predicted by the backpropagation neural network model when a single feature is minimized and other features are fixed at preset reference values; for two features that meet preset coupling conditions, based on the mechanical performance change trends, identifying the region where the absolute value of the total performance change predicted by the backpropagation neural network model exceeds a preset threshold when the two features change jointly from the baseline performance to the current value, and using this region as a joint constraint region; if the candidate reference interval for that feature is inconsistent with the joint constraint region, then the candidate reference interval for that feature is corrected according to the joint constraint region.

[0054] The second implementation of step S204 is as follows: taking at least two of the tensile strength, elongation and hardness in the predicted mechanical property data as optimization objectives, and using chemical composition information, sintering process parameters and second phase attribute information as decision variables, a multi-objective Bayesian optimization algorithm is used to search for the Pareto front; for continuous variables in the decision variables, the minimum and maximum values ​​of each variable are extracted from the searched Pareto front, and the numerical interval defined by the minimum and maximum values ​​is used as the reference interval; for discrete variables in the decision variables, the set of categories appearing in the Pareto front is used as the reference interval for tungsten alloy parameters.

[0055] For example, in the first implementation of step S204, principal component analysis (PCA) and t-distributed random neighborhood embedding (t-SNE) are used to perform dimensionality reduction visualization of the samples (i.e., tungsten alloy data) and features. Partial Dependence Plot (PDP), feature importance ranking plot, or feature interaction plot are used to analyze the influence of composition, process, and dual-channel mechanism encoding on the predicted mechanical property data. Here, features refer to each data point included in each chemical composition information, each data point included in each sintering process parameter, and each data point included in each second-phase attribute information.

[0056] Specifically, principal component analysis (PCA) is used to project the input tungsten alloy data into a two-dimensional or three-dimensional space, generating principal component coordinate data. Based on the principal component coordinate data, the overall distribution of the tungsten alloy data in the low-dimensional space is determined. Furthermore, the predicted mechanical property data (tensile strength, elongation, hardness) can be divided to determine the mechanical property levels, thereby determining the overall distribution of tungsten alloy data of different mechanical property levels in the low-dimensional space; the division methods include, but are not limited to, equal-width binning or equal-frequency binning. Based on t-distribution random neighborhood embedding, the tungsten alloy data is nonlinearly reduced to a two-dimensional or three-dimensional embedding space to obtain embedding coordinate data. Based on the embedding coordinate data, the clustering trend is determined. A partial dependency graph is used to determine the influence trend of a single feature on the predicted mechanical property data, and PDP curve data for that feature is generated based on the influence trend; the PDP curve data is generated by calling a backpropagation neural network model to predict different feature values. Using a feature interaction graph, other features are fixed, and all combinations of values ​​for two features are taken to analyze the coupling effect between the two features, obtaining the mechanical property change trend under the joint change of the two features. The mechanical property change trend refers to whether the effect is positive or negative, i.e., whether the mechanical property change trend is increasing.

[0057] Finally, based on the PDP curve data and the trend of mechanical property changes, candidate reference intervals for each feature are determined. According to the principal component coordinate data, the overall distribution of tungsten alloy data in low-dimensional space is determined. Based on multiple sample regions within the overall distribution, target sample regions satisfying preset sample rules are identified, and the spatial distribution characteristic parameters of the target sample regions are determined. These spatial distribution characteristic parameters include density center, confidence ellipse, or convex hull boundary. The target sample regions are the high-performance sample clusters. Each sample cluster represented by the clustering trend corresponds to a clustering label. Based on the clustering labels of each sample cluster represented by the clustering trend, target sample clusters satisfying preset priority labels are determined; these target sample clusters are the high-performance sample clusters. Based on the spatial distribution characteristic parameters, the intersection of the target sample regions and the target sample clusters is taken. Then, the boundaries of the candidate reference intervals are updated based on the intersection, thereby restricting the candidate reference intervals to within the intersection. The restricted candidate reference intervals are then merged to obtain the final tungsten alloy parameter reference intervals. At this point, the final tungsten alloy parameter reference intervals are restricted to the original feature value range corresponding to the high-performance sample clusters. The tungsten alloy parameter reference range includes the composition range, sintering process range, and / or second phase property range. When merging, if the candidate reference range contains multiple discontinuous sub-ranges for the same feature, they can be merged. For multiple features, a multi-dimensional parameter window is formed, which is jointly defined by the reference ranges of each feature. When there are interactive constraints, the multi-dimensional parameter window is modified according to the joint constraint region. The joint constraint region is explained in detail below.

[0058] Optionally, based on PDP curve data and the trend of mechanical property changes, candidate reference intervals for each feature are determined, including: based on PDP curve data, calculating the absolute value of the change in performance at each point on the curve relative to the baseline, and extracting feature value sub-intervals with absolute values ​​greater than a preset threshold as candidate reference intervals for that feature. Baseline performance refers to the baseline performance data predicted by the backpropagation neural network model when a single feature is minimized and other features are fixed to preset reference values. To ensure the accuracy of prediction and backpropagation, multiple sets of tungsten alloy data (i.e., multiple samples) are acquired by default. For each value of a single feature, the value of that feature in all samples is replaced with that value, keeping other features unchanged. The backpropagation neural network model is then used to predict the predicted performance of each sample. The average predicted performance of all samples is then averaged to obtain the average predicted performance corresponding to that value. The baseline average predicted performance is usually based on the average prediction of each sample when the feature is minimized. The absolute value of the change refers to the absolute value of the average predicted performance of all samples corresponding to a point on the curve minus the absolute value of the baseline average predicted performance.

[0059] For two features that satisfy the preset coupling condition, based on the trend of mechanical performance change, the region where the absolute value of the total performance change predicted by the backpropagation neural network model exceeds a preset threshold when the two features change jointly from the baseline performance to the current value is identified as the joint constraint region. It is then determined whether the candidate reference interval of the feature is consistent with this joint constraint region. If the candidate reference interval of the feature is inconsistent with the joint constraint region, the candidate reference interval of the feature is corrected according to the joint constraint region. The preset coupling condition refers to the difference between the absolute value of the total performance change predicted by the model when the two features change jointly and the sum of the absolute values ​​of the performance changes when the two features change independently, and the difference is greater than the preset interaction effect threshold. The absolute value of the performance change when feature 1 changes independently refers to the absolute value of the performance change predicted by the model when feature 1 changes from the baseline value to the current value, with feature 2 fixed as the baseline value. The calculation method for the absolute value of the performance change when feature 2 changes independently is similar. The preset interaction effect threshold can be user-defined and is not limited.

[0060] Optionally, the candidate reference interval is adjusted based on the overall distribution and clustering trend. This includes: based on the data clustering regions represented by the overall distribution and the principal component coordinate data, if it is determined that the feature value corresponding to the corrected candidate reference interval does not fall within the data clustering region in the principal component coordinate data, the corrected candidate reference interval is shrunk to the range overlapping with the data clustering region. Based on the main clustering regions represented by the clustering trend, feature values ​​in the shrunk candidate reference interval that deviate from the main clustering regions are identified, and these feature values ​​are removed from the shrunk candidate reference interval to obtain the adjusted candidate reference interval.

[0061] In the second implementation of step S204, at least two of the tensile strength, elongation, and hardness in the predicted mechanical property data are used as optimization objectives, and chemical composition information, sintering process parameters, and second phase attribute information are used as decision variables. A multi-objective Bayesian optimization algorithm is used to search for the Pareto front. For continuous variables in the decision variables, the minimum and maximum values ​​of each variable are extracted from the searched Pareto front, and the numerical range defined by the minimum and maximum values ​​is used as a reference range. For discrete variables in the decision variables, the set of categories appearing in the Pareto front is used as a reference range for tungsten alloy parameters.

[0062] It should be noted that any of the above implementation methods for back-analysis can be used, or a combination thereof, without limitation.

[0063] Step S205: In response to the output operation, obtain the target mechanical property data input by the user; using the target mechanical property data as a constraint, and based on the tungsten alloy parameter reference range, reverse-engineer the recommended combination of candidate alloy composition and sintering process parameters; generate and output the recommended scheme based on the combination of candidate alloy composition and sintering process parameters.

[0064] For example, the tungsten alloy parameter reference range can identify composition-process-mechanism variables that significantly affect tensile strength, elongation, and hardness. To shorten the R&D cycle and improve experimental screening efficiency, a target performance-driven reverse design module is established based on the tungsten alloy parameter reference range obtained in step S204. Figure 5 A target performance-driven reverse design and process recommendation flowchart is provided for embodiments of this application, such as... Figure 5 As shown, the user inputs the target mechanical property vector y_target (i.e., the target performance in the figure), such as a target tensile strength of 1000 MPa, and can also attach lower limits for elongation, hardness, or mechanism proportion. The reverse design module calls the trained forward prediction model f(x,p) to obtain the predicted performance vector y_pred, where x represents the composition and second phase correlation parameters, and p represents the sintering process parameters.

[0065] Furthermore, based on the target mechanical performance data, an optimized target J is constructed: J = J = min { ‖y_pred y_target‖ + αR_mech + βR_proc}, where R_mech represents the mechanism consistency constraint, used to ensure that the contribution ratio of dispersion strengthening channels and solid solution strengthening channels falls within a set range; R_proc represents the process feasibility constraint, used to ensure that the sintering temperature, pressure, and holding time are within the preset process boundaries; J represents the comprehensive optimization objective function in the reverse design process; y_pred represents the predicted mechanical property vector output by the trained forward prediction model; R_mech represents the mechanism consistency constraint, used to ensure that the contribution ratio of dispersion strengthening channels and solid solution strengthening channels or the mechanism dominance state falls within a set range; y_target represents the target mechanical property vector input by the user or preset; α represents the weight coefficient of the mechanism consistency constraint; β represents the weight coefficient of the process feasibility constraint; R_proc represents the process feasibility constraint, used to ensure that process parameters such as sintering temperature, sintering pressure, and holding time are within the manufacturable range. The element range / total amount is constrained based on the candidate composition generator; the temperature / pressure / time window is generated based on the candidate process generator; based on the optimization objective and the tungsten alloy parameter reference range, the recommended combination of candidate alloy composition and sintering process parameters is calculated in reverse, and each combination of candidate alloy composition and sintering process parameters is a candidate scheme.

[0066] For multiple candidate solutions that satisfy the target mechanical property vector, a reverse optimizer is used to further rank them according to manufacturability, prediction confidence, and material cost, outputting multiple sets of candidate solutions. Each candidate solution includes at least: second phase type, second phase volume fraction, types and contents of solid solution elements, sintering temperature, sintering pressure, holding time, and corresponding predicted tensile strength, elongation, and hardness. The search algorithm of the reverse optimizer can employ genetic algorithms, particle swarm optimization, Bayesian optimization, or a combination thereof.

[0067] Based on physical and manufacturability constraints, the module performs target deviation calculations, obtains interpretable outputs, generates recommended solutions (i.e., recommended results), and finally performs experimental verification / re-entry into the database, completing a closed-loop iterative update. Therefore, this module forms a closed-loop iterative process of "forward prediction - reverse design - experimental verification - data feedback".

[0068] Figure 6 This is a schematic diagram of the output of a reverse design module provided in an embodiment of this application, such as... Figure 6 As shown, it includes: user-defined objectives (i.e., user-input target mechanical performance data); design space constraints; model return mechanism preferences; visual feedback; and multiple candidate solutions A, B, and C.

[0069] Optionally, tensile strength, elongation, and hardness can be used as a joint target performance vector to establish a multi-output regression model. After completing database construction, feature selection, and model training, the user inputs a target tensile strength of 1000 MPa and can further input lower limits for elongation and hardness. The reverse design module generates candidate composition and process combinations within a search space that satisfies the W matrix, second phase type range, element content boundaries, and sintering parameter boundaries. Subsequently, it calls the forward BPNN model to simultaneously predict tensile strength, elongation, and hardness, and iteratively optimizes the model based on minimizing the target deviation. Finally, it outputs several candidate schemes that meet the target performance requirements and corresponding sintering parameter windows for priority experimental verification.

[0070] The method provided in this application embodiment acquires tungsten alloy data, including chemical composition information, sintering process parameters, and second-phase attribute information. It extracts physical mechanism features from the tungsten alloy data, including chemical bond features, interface mismatch features, and thermodynamic stability features. Based on these physical mechanism features, it constructs dispersion strengthening channel codes and solid solution strengthening channel codes. The tungsten alloy data, physical mechanism features, dispersion strengthening channel codes, and solid solution strengthening channel codes are all input into a preset backpropagation neural network model for performance prediction, outputting predicted mechanical property data. Based on the predicted mechanical property data, it performs back-calculation analysis to generate a tungsten alloy parameter reference range. In response to the output operation, it acquires the user-inputted target mechanical property data; using the target mechanical property data as constraints, and based on the tungsten alloy parameter reference range, it back-calculates recommended candidate alloy composition and sintering process parameter combinations; and based on the candidate alloy composition and sintering process parameter combinations, it generates and outputs a recommended scheme. This scheme follows a main line of "data construction - feature selection - mechanism encoding - physical constraint modeling - training optimization - design feedback," introducing physical mechanism features and physical constraints on the basis of data-driven modeling. Two-stage feature selection reduces redundant input, improving training efficiency and model robustness. Dual-channel mechanism encoding and physical constraint loss terms enhance the consistency between prediction results and material strengthening mechanisms, significantly improving the stability, generalization, and interpretability of the backpropagation neural network model. Therefore, this application can simultaneously adapt to dispersion-strengthened tungsten alloys, solid-solution-strengthened tungsten alloys, and composite-strengthened tungsten alloy systems. Through interpretability analysis and other back-calculation analyses, key parameter windows can be deduced, providing a basis for alloy composition design and sintering process optimization. This solves the problems of insufficient mechanism expression, weak model interpretability, and difficulty in simultaneously characterizing the coupling effect of dispersion strengthening and solid-solution strengthening in existing technologies.

[0071] In one embodiment, as an possible implementation, a tungsten alloy sample set covering different second-phase types, different solid solution element types, and different sintering regimes can be constructed, with tensile strength, elongation, and hardness as prediction targets. After cleaning, dimensionless transformation, and feature screening of the sample data, a physically constrained BPNN model is established for training, and the optimal model is determined based on R², MAE, and RMSE. Subsequently, the model results are analyzed using PCA, t-SNE, and PDP to identify parameter combinations that are beneficial to the synergistic improvement of strength, plasticity, and hardness, which are used to guide subsequent experimental verification.

[0072] It should be noted that the number of samples, network layers, number of neurons, learning rate, loss term weights, and evaluation index thresholds in the specific embodiments can be adjusted according to the actual data scale and research objectives. All equivalent substitutions or modifications made within the spirit and principles of this application should be included within the scope of protection of this application.

[0073] Corresponding to the above method, this application also provides a device for determining the reference range of tungsten alloy parameters, such as... Figure 7 As shown, the device includes: The acquisition module 41 is used to acquire tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase attribute information. Extraction module 42 is used to extract the physical mechanism features of the tungsten alloy data; the physical mechanism features include chemical bond features, interface mismatch features and thermodynamic stability features; based on the physical mechanism features, dispersion strengthening channel coding and solid solution strengthening channel coding are constructed; Prediction module 43 is used to input the tungsten alloy data, the physical mechanism characteristics, the dispersion strengthening channel code and the solid solution strengthening channel code into a preset backpropagation neural network model for performance prediction, and output predicted mechanical property data. The reverse calculation module 44 is used to perform reverse analysis based on the predicted mechanical property data to generate the reference range of the tungsten alloy parameters.

[0074] The functions of each functional unit in the tungsten alloy parameter reference range determination device provided in the above embodiments of this application can be implemented through the above method steps. Therefore, the specific working process and beneficial effects of each unit in the tungsten alloy parameter reference range determination device provided in the embodiments of this application will not be repeated here.

[0075] This application also provides an electronic device, such as... Figure 8 As shown, it includes a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540.

[0076] Memory 530 is used to store computer programs; The processor 510 performs the above steps when executing the program stored in the memory 530.

[0077] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0078] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0079] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0080] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0081] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments for solving the problem can be found in [reference needed]. Figure 2 The steps in the illustrated embodiments are used to implement the electronic device. Therefore, the specific working process and beneficial effects of the electronic device provided in this application will not be repeated here.

[0082] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform any of the 1111 methods described in the above embodiments.

[0083] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the 1111 methods described in the above embodiments.

[0084] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.

[0089] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims in this application and their equivalents, then this application also intends to include these modifications and variations.

Claims

1. A method for determining a reference range for tungsten alloy parameters, characterized in that, include: Acquire tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase property information. The physical mechanism characteristics of the tungsten alloy data are extracted; these physical mechanism characteristics include chemical bond characteristics, interfacial mismatch characteristics, and thermodynamic stability characteristics. Based on the aforementioned physical mechanism characteristics, dispersion enhancement channel coding and solid solution enhancement channel coding are constructed; The tungsten alloy data, the physical mechanism characteristics, the dispersion strengthening channel encoding, and the solid solution strengthening channel encoding are all input into a preset backpropagation neural network model for performance prediction, and the predicted mechanical property data are output. Based on the predicted mechanical property data, a back-calculation analysis is performed to generate a reference range for tungsten alloy parameters.

2. The method as described in claim 1, characterized in that, Based on the predicted mechanical property data, a back-calculation analysis is performed to generate a reference range for tungsten alloy parameters, including: Principal component analysis is used to project the input tungsten alloy data into a two-dimensional or three-dimensional space to generate principal component coordinate data. Based on t-distributed random neighborhood embedding, the tungsten alloy data is nonlinearly reduced to a two-dimensional or three-dimensional embedding space to obtain embedded coordinate data; based on the embedded coordinate data, the clustering trend is determined. A partial dependency graph is used to determine the influence trend of a single feature on the predicted mechanical property data, and PDP curve data of that feature is generated based on the influence trend; the feature includes information on each chemical composition, each sintering process parameter, and each second phase attribute; the PDP curve data is generated by calling the backpropagation neural network model to predict different feature values; By analyzing the coupling effect between pairs of features using feature interaction graphs, the trend of mechanical property changes under the joint changes of the two features can be obtained. Based on the principal component coordinate data, the clustering trend, the PDP curve data, and the mechanical property change trend, a reference range for tungsten alloy parameters is generated.

3. The method as described in claim 2, characterized in that, Based on the principal component coordinate data, the clustering trend, the PDP curve data, and the mechanical property variation trend, a reference range for tungsten alloy parameters is generated, including: Based on the PDP curve data and the trend of mechanical property changes, candidate reference intervals for each feature are determined; Based on the principal component coordinate data, the overall distribution of the tungsten alloy data in low-dimensional space is determined, and based on multiple sample regions in the overall distribution, a target sample region that meets the preset sample rules is determined; the spatial distribution characteristic parameters of the target sample region are determined. Based on the clustering labels of each sample cluster characterized by the clustering trend, target sample clusters that meet the preset priority labels are determined. Based on the spatial distribution characteristic parameters, the intersection of the target sample region and the target sample cluster is taken; By restricting the candidate reference range to the intersection, the final tungsten alloy parameter reference range is obtained.

4. The method as described in claim 3, characterized in that, Based on the PDP curve data and the trend of mechanical property changes, candidate reference intervals for each feature are determined, including: Based on the PDP curve data, the absolute value of the change in performance of each point on the curve relative to the baseline is calculated, and feature value sub-intervals with absolute values ​​of change greater than a preset threshold are extracted as candidate reference intervals for the feature; the baseline performance refers to the baseline performance data predicted by the backpropagation neural network model when a single feature is taken as the minimum value and other features are fixed at preset reference values. For two features that satisfy the preset coupling conditions, based on the trend of mechanical performance change, the region where the absolute value of the total performance change predicted by the backpropagation neural network model exceeds the preset threshold when the two features change together from the baseline performance to the current value is identified as the joint constraint region; if the candidate reference interval of the feature is inconsistent with the joint constraint region, the candidate reference interval of the feature is corrected according to the joint constraint region.

5. The method as described in claim 1, characterized in that, Based on the predicted mechanical property data, a back-calculation analysis is performed to generate a reference range for tungsten alloy parameters, including: Using at least two of the tensile strength, elongation and hardness in the predicted mechanical property data as optimization objectives, and the chemical composition information, the sintering process parameters and the second phase attribute information as decision variables, a multi-objective Bayesian optimization algorithm is used to search the Pareto front. For continuous variables among the decision variables, the minimum and maximum values ​​of each variable are extracted from the Pareto front, and the numerical interval defined by the minimum and maximum values ​​is used as the reference interval; for discrete variables among the decision variables, the set of categories appearing in the Pareto front is used as the reference interval for the tungsten alloy parameters.

6. The method according to any one of claims 1-5, characterized in that, The tungsten alloy parameter reference range includes a performance improvement range and a performance sensitive range; wherein, the performance improvement range is the range of characteristic values ​​that make the predicted mechanical performance data improve relative to the baseline performance and exceed a first preset threshold, and the performance sensitive range is the range of characteristic values ​​that make the absolute value of the change in the predicted mechanical performance data exceed a second preset threshold.

7. The method according to any one of claims 1-5, characterized in that, The method further includes: In response to output operations, acquire the target mechanical performance data input by the user; Using the target mechanical property data as constraints, and based on the tungsten alloy parameter reference range, the recommended combination of candidate alloy composition and sintering process parameters is calculated in reverse; based on the combination of candidate alloy composition and sintering process parameters, a recommended scheme is generated and output.

8. The method according to any one of claims 1-5, characterized in that, The training process of the backpropagation neural network model includes: Obtain a tungsten alloy dataset; the tungsten alloy dataset includes at least multi-component chemical composition information, sintering process parameters, second-phase property information, and predicted mechanical property data; The tungsten alloy dataset is subjected to Pearson correlation coefficient pre-screening and recursive feature elimination screening in sequence to obtain a subset of target features for multi-objective performance prediction; Extract the chemical bond features, interface mismatch features, and thermodynamic stability features of the target feature subset; construct dispersion enhancement channel coding and solid solution enhancement channel coding based on the chemical bond features, interface mismatch features, and thermodynamic stability features; The target feature subset, the chemical bond feature, the interface mismatch feature, the thermodynamic stability feature, the dispersion enhancement channel encoding, and the solid solution enhancement channel encoding are fused to obtain a fusion result; a candidate neural network model is constructed based on the fusion result. The candidate neural network model is trained, validated, and its hyperparameters optimized using the backpropagation algorithm and a total loss function that includes a physical constraint loss term, to obtain a converged backpropagation neural network model.

9. The method as described in claim 8, characterized in that, The dispersion strengthening channel encoding includes particle size, particle spacing, second phase volume fraction and interfacial modulus mismatch, lattice mismatch degree and Orowan correlation factor. The solid solution strengthening channel encoding includes solid solution element concentration, atomic radius mismatch, shear modulus mismatch, valence electron concentration difference, mixing enthalpy or solid solution stability parameter.

10. The method as described in claim 8, characterized in that, The total loss function includes a data fitting loss term and a physical constraint loss term. The physical constraint loss term includes at least one or more of the following: non-negative constraint of dispersion strengthening contribution, non-negative constraint of solid solution strengthening contribution, limit degradation consistency constraint, and dimensional consistency constraint.

11. The method as described in claim 10, characterized in that, The extreme degradation consistency constraint is used to control the output constraint of the backpropagation neural network model to be solid solution enhancement-dominated output if the diffusion enhancement channel encoding is lower than a preset first threshold. If the solid solution enhancement channel encoding is lower than a preset second threshold, then the output constraint of the backpropagation neural network model is controlled to be diffusion enhancement-dominated output.

12. A device for determining a reference range for tungsten alloy parameters, characterized in that, The device includes: The acquisition module is used to acquire tungsten alloy data; the tungsten alloy data includes chemical composition information, sintering process parameters, and second phase property information. An extraction module is used to extract the physical mechanism features of the tungsten alloy data; the physical mechanism features include chemical bond features, interface mismatch features, and thermodynamic stability features; based on the physical mechanism features, dispersion strengthening channel codes and solid solution strengthening channel codes are constructed; The prediction module is used to input the tungsten alloy data, the physical mechanism characteristics, the dispersion strengthening channel code, and the solid solution strengthening channel code into a preset backpropagation neural network model for performance prediction, and output predicted mechanical property data. The reverse calculation module is used to perform reverse analysis based on the predicted mechanical property data to generate the reference range of the tungsten alloy parameters.

13. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method of any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-11.

Images