Prediction method and system for phase change of beta-type titanium alloy microstructure

Abstract

The invention discloses a beta-type titanium alloy microstructure phase change prediction method and system, and the method comprises the steps: obtaining the composition proportion information of a to-be-predicted beta-type titanium alloy sample, and obtaining corresponding electronic structure parameters, lattice distortion parameters, thermodynamic parameters and other physical characteristic data based on the analysis; and inputting the physical characteristic data into a constructed phase change prediction model to obtain a phase change prediction result which is used for guiding and adjusting the composition proportion of the titanium alloy. Wherein the phase change prediction model constructs an initial prediction model through a machine learning mechanism fusing uncertainty quantization, robust mode recognition and decision boundary learning, adopts a layered five-fold cross validation strategy for training, and determines an integrated weight based on prediction accuracy and diversity indexes of each mechanism on a validation set; and finally carrying out weighted integration to obtain a high-precision prediction model. According to the method, the accuracy and reliability of phase change prediction of the beta-type titanium alloy are effectively improved.

Description

Technical Field

[0001] This invention relates to the fields of material property prediction and machine learning technology, and in particular to a method and system for predicting the microstructure phase transformation of β-type titanium alloys. Background Technology

[0002] Compared to common low-temperature metallic structural materials, titanium alloys possess higher specific strength and corrosion resistance, along with low density, non-magnetic properties, and low thermal conductivity and coefficient of thermal expansion at low temperatures, making them ideal structural materials for aerospace cryogenic engineering. In the field of metallic materials design and development, accurately predicting the microstructure evolution of alloys is a crucial prerequisite for achieving targeted control of material properties. Therefore, establishing effective prediction models for twinning and stress-induced martensitic phase transformation that may occur in metastable β-type titanium alloys during deformation is essential for developing novel alloys with excellent strength-ductility matching.

[0003] However, the limited experimental sample size makes it difficult to support the training of complex models, while simple models cannot capture complex nonlinear correlation mechanisms such as the synergistic effect between electronic structure parameters and lattice distortion parameters, and the non-monotonic influence of key physical parameters on phase transformation behavior. This contradiction makes the existing titanium alloy phase transformation prediction methods generally have the defects of insufficient generalization ability and large fluctuations in prediction results, which seriously restricts their practical application value in alloy composition optimization design. Summary of the Invention

[0004] This invention provides a method and system for predicting the microstructure phase transformation of β-type titanium alloys, in order to solve the technical problems of insufficient accuracy and poor generalization ability of existing technologies in predicting twin / martensitic phase transformation of β-type titanium alloys. The goal is to achieve rapid prediction of phase transformation state based on alloy composition, reduce experimental costs, and improve prediction accuracy under limited sample conditions.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for predicting the microstructure phase transformation of β-type titanium alloys, comprising:

[0006] Obtain the composition ratio information of the β-type titanium alloy sample to be predicted;

[0007] Based on the analysis results of the component ratio information, physical characteristic data corresponding to the β-type titanium alloy sample are obtained;

[0008] The physical characteristic data is input into the pre-constructed phase transformation prediction model to obtain the phase transformation prediction result, which is used to guide the adjustment of the composition ratio information of the β-type titanium alloy.

[0009] The construction process of the phase transition prediction model includes:

[0010] Construct a training sample dataset;

[0011] Integrate at least one type of machine learning mechanism to construct an initial prediction model;

[0012] A hierarchical K-fold cross-validation strategy is adopted, and the initial prediction model is trained using the training sample dataset. Based on the training results, the machine learning mechanism in the initial prediction model is optimized to determine the final weights.

[0013] Based on the final weights, the initial prediction model is weighted and integrated to obtain the phase transition prediction model.

[0014] As one preferred embodiment, the method of employing a hierarchical K-fold cross-validation strategy to train the initial prediction model using the training sample dataset includes:

[0015] The training sample dataset is divided into a basic training set, a validation set, and a test set.

[0016] In each round of training, the initial prediction model is trained using the basic training set, the validation set, and the test set.

[0017] As one preferred embodiment, training the initial prediction model using the basic training set, the validation set, and the test set includes:

[0018] The basic training set is input into the initial prediction model to train the initial prediction model;

[0019] The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity indicators are obtained based on the prediction results.

[0020] Based on the prediction accuracy and the diversity index, the ensemble weights of the initial prediction model are determined;

[0021] The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

[0022] As one preferred embodiment, the step of weighted integration of the initial prediction model based on the final weights to obtain the phase transition prediction model includes:

[0023] By integrating the ensemble weights of the initial prediction model after each round of training, the final weights of each machine learning mechanism in the initial prediction model are determined.

[0024] The phase transition prediction model is obtained by integrating the various machine learning mechanisms using the final weights.

[0025] As a preferred embodiment, the physical characteristic data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters. The physical characteristic data corresponding to the β-type titanium alloy sample, obtained based on the analysis results of the component ratio information, includes:

[0026] Based on the composition ratio information, the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters are calculated.

[0027] By integrating the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters, the physical characteristic data corresponding to the β-type titanium alloy sample are obtained.

[0028] Another embodiment of the present invention provides a prediction system for the microstructure phase transformation of β-type titanium alloys, comprising:

[0029] The acquisition module is used to acquire the composition ratio information of the β-type titanium alloy sample to be predicted;

[0030] The feature extraction module is used to obtain physical feature data corresponding to the β-type titanium alloy sample based on the analysis results of the component ratio information;

[0031] The phase transformation prediction module is used to input the physical characteristic data into the pre-constructed phase transformation prediction model to obtain the phase transformation prediction result. The phase transformation prediction result is used to guide the adjustment of the composition ratio information of the β-type titanium alloy.

[0032] The phase transition prediction module further includes:

[0033] Building units are used to construct training sample datasets;

[0034] Fusion unit, used to fuse at least one type of machine learning mechanism to build an initial prediction model;

[0035] The training unit is used to train the initial prediction model with the training sample dataset using a hierarchical K-fold cross-validation strategy, and optimize the machine learning mechanism in the initial prediction model based on the training results to determine the final weights.

[0036] An integration unit is used to perform weighted integration of the initial prediction model based on the final weights to obtain the phase transition prediction model.

[0037] As one preferred embodiment, the training unit is further used for:

[0038] The training sample dataset is divided into a basic training set, a validation set, and a test set.

[0039] In each round of training, the initial prediction model is trained using the basic training set, the validation set, and the test set.

[0040] As one preferred embodiment, the training unit is further used for:

[0041] The basic training set is input into the initial prediction model to train the initial prediction model;

[0042] The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity indicators are obtained based on the prediction results.

[0043] Based on the prediction accuracy and the diversity index, the ensemble weights of the initial prediction model are determined;

[0044] The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

[0045] As one preferred embodiment, the integrated unit is further configured to:

[0046] By integrating the ensemble weights of the initial prediction model after each round of training, the final weights of each machine learning mechanism in the initial prediction model are determined.

[0047] The phase transition prediction model is obtained by integrating the various machine learning mechanisms using the final weights.

[0048] As one preferred embodiment, the physical feature data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters, and the feature extraction module is further used for:

[0049] Based on the composition ratio information, the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters are calculated.

[0050] By integrating the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters, the physical characteristic data corresponding to the β-type titanium alloy sample are obtained.

[0051] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0052] (1) This invention calculates physically interpretable characteristic parameters such as Md and Bo based on the composition of β-type titanium alloy, and then constructs a diversified machine learning mechanism. Combining "accuracy + diversity" weight allocation, it avoids the problem of weak generalization ability of traditional single model and solves the pain point of unreasonable weight allocation in small sample scenarios. It realizes accurate prediction of stress-induced twinning and strain-induced martensitic phase transformation of β-type titanium alloy. Compared with traditional methods that rely on a large number of experiments, it greatly improves prediction efficiency and reduces R&D costs.

[0053] (2) In the model construction, this invention introduces feature correlation analysis and hierarchical five-fold cross-validation. First, strongly correlated features are removed to reduce redundant interference. Then, hierarchical segmentation is used to ensure that the class ratio of the basic training set and the validation set is consistent. At the same time, weight mode is set for the difference in sample size so that the model can adapt to different data scales. Compared with the existing ensemble models that mostly use fixed voting mechanism, the adaptive weight strategy of this invention further improves the prediction stability. The whole process forms a closed loop of "data preprocessing - model training - result verification", which is convenient for engineering application and repeated implementation. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating a method for predicting the microstructure phase transformation of β-type titanium alloys in one embodiment of the present invention.

[0055] Figure 2 This is a structural diagram of a prediction system for the microstructure phase transformation of β-type titanium alloy in one embodiment of the present invention;

[0056] Figure 3 This is a structural diagram of a phase transformation prediction model for the microstructure phase transformation of a β-type titanium alloy in one embodiment of the present invention.

[0057] Figure label:

[0058] The module includes: acquisition module 11, feature extraction module 12, phase transition prediction module 13, construction unit 21, fusion unit 22, training unit 23, and integration unit 24. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0060] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0061] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0062] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0063] One embodiment of the present invention provides a method for predicting the microstructure phase transformation of β-type titanium alloys. For details, please refer to [link to documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a method for predicting the microstructure phase transformation of β-type titanium alloys according to one embodiment of the present invention, including steps S1-S3:

[0064] S1: Obtain the composition ratio information of the β-type titanium alloy sample to be predicted;

[0065] Among them, the β-type titanium alloy sample to be predicted refers to the metastable β-type titanium alloy. Compared with α-type or (α+β)-type titanium alloys, β-type titanium alloys usually contain enough β-stabilizing elements to allow their metastable β phase to be retained to room temperature after rapid cooling. This is a prerequisite and core research object for subsequent twinning or stress-induced martensitic phase transformation (Twin / SIM phase transformation).

[0066] Composition information is the most basic and fundamental input data driving the entire phase transformation prediction model. It refers to the composition of each alloying element and its atomic percentage (at.%) in the titanium alloy sample. For example, for the sample Ti-8.5Cr-1.5Sn (atomic percentage), its composition information is as follows: main β-stabilizing element: Cr = 8.5 at.%; alloying / neutral element: Sn = 1.5 at.%; matrix element: Ti = 90.0 at.%. The obtained elemental composition and atomic percentage data are validated to remove invalid data with incorrect element types or abnormal atomic percentage values, and the valid data is retained as the composition information for the β-type titanium alloy sample to be processed.

[0067] S2: Based on the analysis results of the component ratio information, physical characteristic data corresponding to the β-type titanium alloy sample are obtained;

[0068] Among them, physical characteristic data refers to a set of parameters that quantitatively describe the key influencing factors of the material's phase transformation behavior, obtained from the compositional information of the β-type titanium alloy sample through a series of calculation formulas based on materials physics and thermodynamics. Because the compositional information of the β-type titanium alloy sample to be processed only provides raw data on chemical composition and cannot directly reflect the phase transformation tendency of the β-type titanium alloy material under stress, it is necessary to convert qualitative compositional information into quantitative physical characteristics through calculation. These characteristics are directly related to the electronic level mechanism, lattice strain state, and thermodynamic driving force of the phase transformation.

[0069] Preferably, in one embodiment of the present invention, the physical characteristic data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters. Based on the analysis results of the composition ratio information, physical characteristic data corresponding to the β-type titanium alloy sample is obtained, including:

[0070] Based on the composition ratio information, the electronic structure parameters, lattice distortion parameters and thermodynamic parameters are calculated;

[0071] By integrating electronic structure parameters, lattice distortion parameters, and thermodynamic parameters, physical characteristic data corresponding to β-type titanium alloy samples were obtained.

[0072] Specifically, based on the composition information of the β-type titanium alloy sample, three types of core parameters were calculated through a series of physicochemical formulas: electronic structure parameters, lattice distortion parameters, and thermodynamic parameters.

[0073] Electronic structure parameters: d orbital binding energies are calculated based on the atomic percentage of each element. Average electron orbital parameters and average valence electron concentration These parameters characterize the electronic structure of the alloy from a quantum mechanical perspective. Md reflects the position of the d-electron energy level, Bo represents the interatomic bonding strength, and e / a determines the position of the Fermi level. Together, they influence the electronic driving force of the phase transition.

[0074] d-orbital binding energy The calculation formula is:

[0075]

[0076] In the formula c i It represents the atomic percentage of the i-th element, (Md). i Let be the d-orbital binding energy parameter of component i, and n be the number of alloying elements.

[0077] Average electron orbital parameters The calculation formula is:

[0078]

[0079] In the formula c i It represents the atomic percentage of the i-th element, (Bo). i Let i be the average electron orbital parameter of component i, and n be the number of alloying elements.

[0080] Average valence electron concentration The calculation formula is:

[0081]

[0082] In the formula c i It is the atomic percentage of the i-th element, e i is the number of valence electrons of the i-th element, and n is the number of alloying elements.

[0083] Lattice distortion parameter: The degree of lattice distortion is characterized by calculating the standard deviation of atomic size (ΔR). Based on the differences in atomic radii of each constituent element, this parameter uses a standard deviation algorithm to quantify the distribution characteristics of the lattice strain field, which can accurately reflect the lattice strain energy during the phase transition process.

[0084] The formula for calculating the standard deviation of atomic size (ΔR) is:

[0085]

[0086] In the formula c i It is the atomic percentage of the i-th element, n is the number of alloying elements, and r i Let represent the atomic radius of the i-th type of atom. This represents the average atomic radius of the alloy.

[0087] Thermodynamic parameters: The thermodynamic state of the alloy is characterized by calculating the enthalpy of mixing (ΔHmix) and the entropy of mixing (ΔSmix). The enthalpy of mixing reflects the strength of interatomic interactions, while the entropy of mixing characterizes the degree of disorder in the configuration. Together, they determine the thermodynamic driving force and direction of the phase transition.

[0088] The formula for calculating the enthalpy of mixing (ΔHmix) is:

[0089]

[0090] In the formula c i c j These represent the atomic percentages of the i-th and j-th elements, respectively, where n is the number of alloying elements, and Ω... ij for in The enthalpy of mixing of AB binary alloys under specific compositions is simplified to a weighted sum of all components constituting the binary system.

[0091] The formula for calculating the mixing entropy (ΔSmix) is:

[0092]

[0093] In the formula c i c j where are the atomic percentages of the i-th and j-th elements, respectively, n is the number of alloying elements, and R is the gas constant.

[0094] After eliminating dimensional differences through feature standardization, electronic structure parameters, lattice distortion parameters, and thermodynamic parameters are combined to form a unified physical feature dataset. This dataset not only preserves the independent physical meaning of each parameter but also constructs a complete descriptive system for phase transition behavior through the synergistic effects between the parameters.

[0095] S3: Input the physical characteristic data into the pre-constructed phase transformation prediction model to obtain the phase transformation prediction results. The phase transformation prediction results are used to guide the adjustment of the composition ratio information of the β-type titanium alloy.

[0096] Specifically, the physical characteristic data calculated in step S2 is used as input and imported into the trained phase transformation prediction model. This model intelligently infers the tendency of the β-type titanium alloy sample to undergo twin / stress-induced martensite (Twin / SIM) phase transformation. The model's output contains two key pieces of information: a classification judgment, which provides a clear classification label, i.e., "Twin / SIM tendency exists" or "Twin / SIM tendency does not exist"; and a probability confidence level, which outputs a probability value between 0 and 1 to quantify the reliability of the classification judgment. Predictions with high confidence levels are more instructive.

[0097] For β-type titanium alloy samples predicted to have a Twin / SIM tendency, they can be included in the subsequent experimental verification sequence for further evaluation through actual melting, processing, and mechanical property testing. At the same time, this composition can be used as a reference to conduct exploratory fine-tuning in its compositional neighborhood (e.g., adjusting the content of key elements by ±0.5 at.%), and then predicted again using the phase transformation prediction model to explore a compositional range with better performance.

[0098] For β-type titanium alloy samples that are predicted to lack Twin / SIM tendency, it indicates that the current composition of the β-type titanium alloy may be insufficient to induce the required phase transformation strengthening effect. It is necessary to re-examine or adjust the current composition scheme, such as by introducing new alloying elements or adjusting the proportion of the main components, in order to explore other possible compositional spaces and help avoid potential R&D risks.

[0099] Through the above process, the phase transformation prediction results can be transformed into specific and actionable R&D instructions, which significantly improves the R&D efficiency and success rate of new high-performance β-type titanium alloys.

[0100] In one embodiment of the present invention, the process of constructing the phase transition prediction model includes steps 1-4:

[0101] Step 1: Construct the training sample dataset;

[0102] Specifically, alloy data that has been experimentally characterized can be extracted from academic literature on metastable β-titanium alloy twins and stress-induced martensite, or relevant computational or experimental data on titanium alloys can be obtained from public materials database platforms. Each sample in the training dataset contains input features (physical feature data) and a target variable, where the target variable refers to the phase transformation type label. When a sample contains twins or stress-induced martensite (SIM), it is marked as positive; when a sample is confirmed to contain neither twins nor SIM, it is marked as negative.

[0103] To ensure data quality and consistency, the collected raw data was processed as follows: Methods such as `dropna()` from the pandas library were used to directly delete any rows containing missing values, ensuring data integrity; text-based phase transition type labels were uniformly mapped to numeric labels (1 for positive, 0 for negative) to meet the requirements of machine learning algorithms; and all physical feature data were standardized to eliminate differences in units and orders of magnitude between different feature parameters. After processing the raw data, a standardized training dataset with a unified format was obtained, providing solid data support for subsequent model training.

[0104] Step 2: Integrate at least one type of machine learning mechanism to build an initial prediction model;

[0105] Specifically, by integrating multiple machine learning mechanisms with different core capabilities, an initial prediction model with high accuracy, robustness, and reliability is ultimately constructed. Here, "mechanism" refers to a machine learning algorithm or model with a specific learning paradigm and capabilities. In one embodiment of this invention, it is necessary to integrate machine learning mechanisms including at least uncertainty quantification, robust pattern recognition, and decision boundary learning to construct the initial prediction model. That is, these different machine learning mechanisms are placed together within a unified framework or set to construct the initial prediction model. The current initial prediction model possesses basic predictive capabilities, but the optimal combination of its internal mechanisms is still unknown.

[0106] In one embodiment of the present invention, the uncertainty quantification mechanism is implemented by a Gaussian Process Classifier. This mechanism not only provides a "yes" or "no" classification prediction, but also outputs a probabilistic prediction result and its uncertainty range. The Gaussian Process Classifier is a nonparametric model based on Bayesian probability theory. Its core value lies in its ability to not only provide classification labels for samples, but also to quantify the uncertainty of the prediction result. When the phase transition prediction model faces alloys with complex compositions or those at the prediction boundary (decision ambiguity region), it can provide a large prediction variance, equivalent to issuing a warning that "this prediction result has low confidence and should be treated with caution," thereby guiding the prioritization of experimental verification and improving the reliability of decisions.

[0107] In one embodiment of the invention, the robust pattern recognition mechanism is implemented using a Random Forest Classifier, which excels at capturing complex, nonlinear feature interactions from high-dimensional features. The Random Forest Classifier is an efficient ensemble learning algorithm that makes predictions by constructing a large number of decision trees and synthesizing their conclusions. The inherent feature importance ranking capability of Random Forests can reveal which physical features contribute most to predicting the Twin / SIM phase transition, greatly enhancing the interpretability of the model and making the prediction results more transparent and reliable for materials scientists.

[0108] In one embodiment of the present invention, the decision boundary learning mechanism is implemented by a Support Vector Machine (SVC) classifier. The core objective of this mechanism is to construct a clear and optimal decision boundary in a high-dimensional feature space. The SVC classifier is a powerful classifier based on the principle of structural risk minimization in statistical learning theory. Its core idea is not simply to fit the data, but to find a decision hyperplane that can optimally distinguish between different categories of samples. Specifically, the SVC classifier employs various kernel functions, such as the Radial Basis Function (RBF kernel) and the polynomial kernel (degree = n). Different kernel functions enable it to learn different boundary morphologies, thereby precisely defining the critical separation conditions in the physical feature space for alloys that exhibit "Twin / SIM" versus "Twin / SIM" non-existence.

[0109] Step 3: Employ a hierarchical K-fold cross-validation strategy, train the initial prediction model using the training sample dataset, and optimize the machine learning mechanism in the initial prediction model based on the training results to determine the final weights;

[0110] Hierarchical K-fold cross-validation is a common variation of cross-validation. Its core purpose is to maximize the use of existing data with limited samples, while obtaining a reliable estimate of the model's generalization ability and optimizing the model parameters accordingly. Specifically, the training dataset is randomly and uniformly divided into K roughly equal parts, called K "folds." The partitioning process ensures that the proportion of samples from each category in each fold is consistent with the original dataset, especially suitable for imbalanced datasets (i.e., large differences in the number of samples from different categories). This avoids imbalanced category distribution in some folds due to random partitioning, which could affect the accuracy of model evaluation. The decision to optimize the three complementary machine learning mechanisms within the initial prediction model is not pre-set but depends on the actual performance of the subsequent model on the "validation set" of cross-validation. The final output of the optimization is a set of values, namely the "final weights" for each machine learning mechanism. These weights satisfy the following conditions: all weights are positive and their sum is 1; the magnitude of the weight directly reflects the "voice" of that mechanism in the ensemble decision.

[0111] Preferably, in one embodiment of the present invention, the step of using a hierarchical K-fold cross-validation strategy to train the initial prediction model with the training sample dataset includes:

[0112] The training sample dataset is divided into a basic training set, a validation set, and a test set.

[0113] In each round of training, the initial prediction model is trained using the basic training set, the validation set, and the test set.

[0114] Specifically, the complete training sample dataset is randomly and uniformly divided into K mutually exclusive subsets. During the partitioning process, stratified sampling is used to ensure that the distribution ratio of phase transition type labels within each fold remains consistent with the original distribution of the entire dataset, thus guaranteeing that each fold is a representative sample of the entire set. The basic training set consists of K-2 folds of data and serves as the primary training data for learning the internal parameters of each machine learning mechanism in the initial prediction model. The validation set consists of a single fold of data and is not used in the initial prediction model training; it is specifically used to evaluate the performance of each trained machine learning mechanism. The test set consists of the remaining single folds of data and is completely independent of the basic training and validation sets in each round; it is specifically used to objectively evaluate the final performance of the model formed after the weights are determined by the validation set.

[0115] Preferably, in one embodiment of the present invention, training an initial prediction model using a basic training set, a validation set, and a test set includes:

[0116] The basic training set is input into the initial prediction model to train the initial prediction model;

[0117] The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity metrics are obtained based on the prediction results.

[0118] Based on prediction accuracy and diversity metrics, the ensemble weights of the initial prediction model are determined.

[0119] The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

[0120] Specifically, the basic training set of the current round is input into the initial prediction model, and the three machine learning mechanisms constituting the model are trained independently to optimize their uncertainty quantification ability, robust pattern recognition ability, and decision boundary learning ability. Then, the validation set of the current round is input into the trained initial prediction model, and prediction accuracy and diversity indicators are obtained based on the prediction results. Prediction accuracy is a basic indicator that measures the classification performance of a single machine learning mechanism, representing the proportion of samples in which the mechanism makes correct predictions on the validation set. The diversity indicator is a quantitative standard that measures the degree of difference in predictions between different machine learning mechanisms, and its calculation process is as follows.

[0121] First, it is necessary to calculate the predictive discrepancy (D) between each pair of mechanisms. ij The calculation formula is as follows:

[0122]

[0123] In the formula, Dij This represents the difference in predictions between mechanism i and mechanism j, where N is the number of samples in the validation set. and Let represent the prediction results of the two mechanisms for the k-th sample, respectively. This is an indicator function; its value is 1 when the condition is true, and 0 otherwise.

[0124] Then calculate the diversity score (div) for each mechanism. i The calculation formula is as follows:

[0125]

[0126] In the formula, div i Let M represent the diversity score of mechanism i, where M is the total number of mechanisms (M = 3 here). This score is the average of the predictive differences between mechanism i and other mechanisms.

[0127] Based on the prediction accuracy and the diversity index, the ensemble weights (w) of the initial prediction model are determined using the softmax function. i The calculation formula is as follows:

[0128]

[0129] In the formula, acc i For the prediction accuracy of mechanism i, div i Let α represent the diversity score of mechanism i, β represent the accuracy incentive factor, β represent the diversity incentive factor, and M represent the total number of mechanisms (here, M = 3). In the model, the accuracy incentive factor α is set to 1.0, and the diversity incentive factor β is set to 2.0, which can effectively balance the accuracy and diversity of the model.

[0130] Repeat the above process K times, ensuring that each subset is used as a test set once, to complete the full cross-validation training process. Finally, input the test set into the initial prediction model, and perform weighted fusion of the prediction results based on the ensemble weights to complete one round of training of the initial prediction model.

[0131] Step 4: Based on the final weights, perform weighted integration on the initial prediction model to obtain the phase transition prediction model.

[0132] The final weights are fixed weight coefficients determined for each machine learning mechanism in the initial prediction model after the complete training process described above. The phase transformation prediction model obtained based on these final weights can receive physical characteristic data of β-type titanium alloys and output predictions of their tendency to undergo twinning / stress-induced martensitic phase transformation.

[0133] Preferably, in one embodiment of the present invention, the initial prediction model is weighted and integrated based on the final weights to obtain a phase transition prediction model, including:

[0134] By integrating the ensemble weights of the initial prediction model after each round of training, the final weights of each machine learning mechanism in the initial prediction model are determined.

[0135] The phase transition prediction model is obtained by integrating the various machine learning mechanisms using the final weights.

[0136] Specifically, the ensemble weights calculated for each machine learning mechanism (i.e., Gaussian process classifier, random forest classifier, support vector machine classifier) ​​in the initial prediction model are collected after each round of training. For each machine learning mechanism, the arithmetic mean of its K weights obtained in five rounds of training is taken to obtain the final weight. The final weight is a fixed weight coefficient determined for each machine learning mechanism in the initial prediction model after the complete training process described above. The final weights are then used to perform a weighted ensemble of the various machine learning mechanisms that have been trained in the initial prediction model to obtain the final phase transition prediction model.

[0137] Another embodiment of the present invention provides a prediction system for the microstructure phase transformation of β-type titanium alloys. For details, please refer to [link to documentation]. Figure 2 , Figure 2 The diagram shown illustrates a prediction system for the microstructure phase transformation of a β-type titanium alloy in one embodiment of the present invention, comprising:

[0138] The acquisition module 11 is used to acquire the composition ratio information of the β-type titanium alloy sample to be predicted;

[0139] The feature extraction module 12 is used to obtain physical feature data corresponding to the β-type titanium alloy sample based on the analysis results of the composition ratio information.

[0140] The phase transformation prediction module 13 is used to input physical characteristic data into the pre-constructed phase transformation prediction model to obtain phase transformation prediction results. The phase transformation prediction results are used to guide the adjustment of the composition ratio information of the β-type titanium alloy.

[0141] Preferably, in one embodiment of the present invention, the physical feature data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters, and the feature extraction module is further used for:

[0142] Based on the composition ratio information, the electronic structure parameters, lattice distortion parameters and thermodynamic parameters are calculated;

[0143] By integrating electronic structure parameters, lattice distortion parameters, and thermodynamic parameters, physical characteristic data corresponding to β-type titanium alloy samples were obtained.

[0144] Another embodiment of the present invention provides a phase transformation prediction model for predicting the microstructure phase transformation of β-type titanium alloys. For details, please refer to [link to relevant documentation]. Figure 3 , Figure 3 The diagram shown is a structural diagram of a phase transformation prediction model for the microstructure phase transformation of a β-type titanium alloy according to one embodiment of the present invention, which includes:

[0145] Building unit 21 is used to build the training sample dataset;

[0146] Fusion unit 22 is used to fuse at least one type of machine learning mechanism to build an initial prediction model;

[0147] Training unit 23 is used to train the initial prediction model with the training sample dataset using a hierarchical K-fold cross-validation strategy, and optimize the machine learning mechanism in the initial prediction model based on the training results to determine the final weights.

[0148] Integration unit 24 is used to perform weighted integration of the initial prediction model based on the final weights to obtain the phase transition prediction model.

[0149] Preferably, in one embodiment of the present invention, the training unit is further configured to:

[0150] The training sample dataset is divided into corresponding basic training set, validation set, and test set;

[0151] In each round of training, the initial prediction model is trained using the basic training set, validation set, and test set.

[0152] Preferably, in one embodiment of the present invention, the training unit is further configured to:

[0153] The basic training set is input into the initial prediction model to train the initial prediction model;

[0154] The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity metrics are obtained based on the prediction results.

[0155] Based on prediction accuracy and diversity metrics, the ensemble weights of the initial prediction model are determined.

[0156] The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

[0157] Preferably, in one embodiment of the present invention, the integration unit is further configured to:

[0158] Integrate the ensemble weights of the initial prediction model after each round of training to determine the final weights of each machine learning mechanism in the initial prediction model.

[0159] By integrating the various machine learning mechanisms using the final weights, a phase transition prediction model is obtained.

[0160] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0161] (1) This invention calculates physically interpretable characteristic parameters such as Md and Bo based on the composition of β-type titanium alloy, and then constructs a diversified machine learning mechanism. Combining "accuracy + diversity" weight allocation, it avoids the problem of weak generalization ability of traditional single model and solves the pain point of unreasonable weight allocation in small sample scenarios. It realizes accurate prediction of stress-induced twinning and strain-induced martensitic phase transformation of β-type titanium alloy. Compared with traditional methods that rely on a large number of experiments, it greatly improves prediction efficiency and reduces R&D costs.

[0162] (2) In the model construction, this invention introduces feature correlation analysis and hierarchical five-fold cross-validation. First, strongly correlated features are removed to reduce redundant interference. Then, hierarchical segmentation is used to ensure that the class ratio of the basic training set and the validation set is consistent. At the same time, weight mode is set for the difference in sample size so that the model can adapt to different data scales. Compared with the existing ensemble models that mostly use fixed voting mechanism, the adaptive weight strategy of this invention further improves the prediction stability. The whole process forms a closed loop of "data preprocessing - model training - result verification", which is convenient for engineering application and repeated implementation.

[0163] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for predicting the microstructure phase transformation of β-type titanium alloys, characterized in that, include: Obtain the composition ratio information of the β-type titanium alloy sample to be predicted; Based on the analysis results of the component ratio information, physical characteristic data corresponding to the β-type titanium alloy sample are obtained; The physical characteristic data is input into the pre-constructed phase transformation prediction model to obtain the phase transformation prediction result, which is used to guide the adjustment of the composition ratio information of the β-type titanium alloy. The construction process of the phase transition prediction model includes: Construct a training sample dataset; Integrate at least one type of machine learning mechanism to construct an initial prediction model; A hierarchical K-fold cross-validation strategy is adopted, and the initial prediction model is trained using the training sample dataset. Based on the training results, the machine learning mechanism in the initial prediction model is optimized to determine the final weights. Based on the final weights, the initial prediction model is weighted and integrated to obtain the phase transition prediction model.

2. The method for predicting the microstructure phase transformation of β-type titanium alloys as described in claim 1, characterized in that, The step of employing a hierarchical K-fold cross-validation strategy to train the initial prediction model using the training sample dataset includes: The training sample dataset is divided into a basic training set, a validation set, and a test set. In each round of training, the initial prediction model is trained using the basic training set, the validation set, and the test set.

3. The method for predicting the microstructure phase transformation of β-type titanium alloys as described in claim 2, characterized in that, Training the initial prediction model using the basic training set, the validation set, and the test set includes: The basic training set is input into the initial prediction model to train the initial prediction model; The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity indicators are obtained based on the prediction results. Based on the prediction accuracy and the diversity index, the ensemble weights of the initial prediction model are determined; The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

4. A method for predicting the microstructure phase transformation of a β-type titanium alloy as described in claim 1 or claim 3, characterized in that, The step of weighted integration of the initial prediction model based on the final weights to obtain the phase transition prediction model includes: By integrating the ensemble weights of the initial prediction model after each round of training, the final weights of each machine learning mechanism in the initial prediction model are determined. The phase transition prediction model is obtained by integrating the various machine learning mechanisms using the final weights.

5. The method for predicting the microstructure phase transformation of β-type titanium alloys as described in claim 1, characterized in that, The physical characteristic data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters. The physical characteristic data corresponding to the β-type titanium alloy sample, obtained based on the analysis results of the component ratio information, includes: Based on the composition ratio information, the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters are calculated. By integrating the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters, the physical characteristic data corresponding to the β-type titanium alloy sample are obtained.

6. A prediction system for the microstructure phase transformation of β-type titanium alloys, characterized in that, include: The acquisition module is used to acquire the composition ratio information of the β-type titanium alloy sample to be predicted; The feature extraction module is used to obtain physical feature data corresponding to the β-type titanium alloy sample based on the analysis results of the component ratio information; The phase transformation prediction module is used to input the physical characteristic data into the pre-constructed phase transformation prediction model to obtain the phase transformation prediction result. The phase transformation prediction result is used to guide the adjustment of the composition ratio information of the β-type titanium alloy. The phase transition prediction module further includes: Building units are used to construct training sample datasets; Fusion unit, used to fuse at least one type of machine learning mechanism to build an initial prediction model; The training unit is used to train the initial prediction model with the training sample dataset using a hierarchical K-fold cross-validation strategy, and optimize the machine learning mechanism in the initial prediction model based on the training results to determine the final weights. An integration unit is used to perform weighted integration of the initial prediction model based on the final weights to obtain the phase transition prediction model.

7. The prediction system for the microstructure phase transformation of β-type titanium alloys as described in claim 6, characterized in that, The training unit is also used for: The training sample dataset is divided into a basic training set, a validation set, and a test set. In each round of training, the initial prediction model is trained using the basic training set, the validation set, and the test set.

8. The prediction system for the microstructure phase transformation of β-type titanium alloys as described in claim 7, characterized in that, The training unit is also used for: The basic training set is input into the initial prediction model to train the initial prediction model; The validation set is input into the trained initial prediction model, and the prediction accuracy and diversity indicators are obtained based on the prediction results. Based on the prediction accuracy and the diversity index, the ensemble weights of the initial prediction model are determined; The test set is input into the initial prediction model, and the prediction results are processed based on the ensemble weights to complete one round of training of the initial prediction model.

9. A prediction system for the microstructure phase transformation of a β-type titanium alloy as described in claim 6 or claim 8, characterized in that, The integration unit is also used for: By integrating the ensemble weights of the initial prediction model after each round of training, the final weights of each machine learning mechanism in the initial prediction model are determined. The phase transition prediction model is obtained by integrating the various machine learning mechanisms using the final weights.

10. The prediction system for the microstructure phase transformation of β-type titanium alloys as described in claim 6, characterized in that, The physical feature data includes at least electronic structure parameters, lattice distortion parameters, and thermodynamic parameters. The feature extraction module is also used for: Based on the composition ratio information, the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters are calculated. By integrating the electronic structure parameters, the lattice distortion parameters, and the thermodynamic parameters, the physical characteristic data corresponding to the β-type titanium alloy sample are obtained.

Images