High-temperature Ni-Ti shape memory alloy component optimization method based on deep learning

Abstract

Aiming at the problems of low phase transition temperature and fast function degradation of the traditional Ni-Ti shape memory alloy under the high-temperature working condition, the invention provides the high-temperature Ni-Ti shape memory alloy component optimization method based on deep learning, according to the method, a Ni-Ti-X ternary system is used as a search space, then an automatic encoder of a multi-head self-attention mechanism is combined, and the high-temperature Ni-Ti shape memory alloy component optimization method based on deep learning is provided. A Ni-Ti-X element screening model is established, precise reverse design of components of the Ni-Ti shape memory alloy within the phase transition temperature (150-400 DEG C) interval is achieved, and the error lt is verified through experiments; + / -10 DEG C; compared with a traditional trial and error method, the method has the advantages that the development period of alloy components is shortened from several months to several days, and the obtained Ni-Ti-15Zr (at.%) alloy has the phase transition temperature of about 170 DEG C and is obviously superior to that of existing Ni-Ti alloy (less than or equal to 100 DEG C). The invention provides an efficient and low-cost component design approach for high-temperature driving and sealing components of aero-engines, oil and gas wells and the like.

Description

Technical Field

[0001] This invention relates to the field of metallic material composition design, specifically to a method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning. Background Technology

[0002] Shape memory alloys (SMAs), due to their macroscopic shape recovery effect resulting from reversible martensitic phase transformation, have been regarded as key functional materials in adjustable bypass valves for aero-engines, packers for deep well oil and gas completions, and high-temperature sensing actuators. However, the martensitic reverse transformation temperature (Af) of traditional near-equal atomic ratio Ni-Ti systems is generally below 100°C, and under martensitic-austenitic cyclic loading, microscopic damage such as dislocation annihilation, Ti2Ni precipitation, and interface decoherence accumulates rapidly, leading to significant functional degradation and making it difficult to meet the long-life requirements of the ≥150°C service temperature range.

[0003] Previous studies have attempted to introduce tertiary components such as Hf, Zr, Pd, Pt, and Au into Ni-Ti binary matrices to improve phase transition temperature and enhance thermal stability through lattice distortion and electron concentration modulation. However, the composition-structure-property mapping relationship of multi-component alloys exhibits highly nonlinear, strongly coupled, and multi-extremum characteristics; the traditional "trial and error-characterization" experimental paradigm has extremely low search efficiency in high-dimensional composition space, and a single high-temperature cycle-fatigue experiment can take months, resulting in exponentially increasing costs, sparse data, and significant noise.

[0004] With the rise of materials genome engineering and data-driven paradigms, how to use machine learning models to uncover the physical laws hidden in limited experimental data under the constraints of "small sample size, high noise, and high dimensionality" and then achieve Ni-Ti-X reverse design with synergistic optimization of target phase transition temperature (150–400℃) and cycle stability has become an urgent problem to be solved in the field of high-temperature shape memory alloys. Summary of the Invention

[0005] In view of this, the present invention proposes a deep learning-based method for optimizing the composition of high-temperature Ni-Ti shape memory alloys, which can accurately construct the mapping relationship between Ni-Ti alloy composition and phase transformation temperature, thereby improving the high-temperature service performance of Ni-Ti shape memory alloys in a low-cost and high-efficiency manner.

[0006] A deep learning-based method for optimizing the composition of high-temperature Ni-Ti shape memory alloys, the method comprising the following: Step 1: Collect alloy composition and phase transformation temperature data from relevant literature experiments on Ni-Ti-X shape memory alloys, construct an initial sample dataset, with alloy composition as input, phase transformation temperature as output, and X as the third component composition; Step 2: Construct a Ni-Ti-X alloy composition dataset from all collected initial samples and preprocess it. The preprocessing includes data cleaning and data normalization of the Ni-Ti-X alloy composition dataset to obtain the preprocessed Ni-Ti-X alloy composition dataset, which will be used as the model input data. Step 3: Design the Ni-Ti-X element screening model. The Ni-Ti-X element screening model is a deep learning model that integrates an autoencoder and a multi-head self-attention mechanism. The Ni-Ti-X element screening model integrates the advantages of the two mechanisms and can efficiently learn the mapping relationship between alloy composition and phase transformation temperature on a small sample Ni-Ti-X alloy composition dataset, which can be used to achieve the optimized design of alloy composition. Step 4: Divide the preprocessed Ni-Ti-X alloy composition dataset into a training set and a test set; use the training set to train the Ni-Ti-X element screening model and iteratively optimize the model hyperparameters to improve the predictive ability and stability of the Ni-Ti-X element screening model. Step 5: Evaluate and validate the trained Ni-Ti-X element screening model using the test set; Step 6: Use the trained and optimized Ni-Ti-X element screening model to predict and screen Ni-Ti-X alloy composition combinations with high phase transformation temperatures, which can be used to guide the composition optimization design and preparation of high-temperature Ni-Ti-X shape memory alloys.

[0007] Furthermore, in step 1, the Ni-Ti-X alloy composition dataset is obtained, specifically including the Ni-Ti-X alloy composition and phase transformation temperature, where X can be one of Pt, Pd, Zr, Hf, Co, Cu, Cr, Mo, V, Ta, Nb, Fe, Y, Al, W, Nd, La, B, and Re.

[0008] Furthermore, in step 2, the preprocessing methods employed include data cleaning and normalization based on linear regression, K-nearest neighbor model and rules, and domain knowledge. Data cleaning includes noise removal, outlier removal, and missing value imputation.

[0009] Furthermore, in step 3, the Ni-Ti-X element screening model clearly defines two core tasks: classification and regression. The classification task aims to determine whether the high-temperature Ni-Ti-X shape memory alloy has phase transformation behavior; the regression task is used to accurately predict the phase transformation temperature of the Ni-Ti-X alloy. Thus, the Ni-Ti-X element screening model is divided into a regressor and a classifier.

[0010] Furthermore, in step 4, the ratio of the number of samples in the training set to the number of samples in the test set is 80%:20%.

[0011] Furthermore, in step 4, the Bayesian optimization algorithm is used for hyperparameter selection and fine-tuning when iteratively optimizing the model hyperparameters.

[0012] Furthermore, in step 4, the hyperparameters of the Ni-Ti-X element screening model include batch size, number of features k_features, and number of hidden layer nodes hidden_layer.

[0013] Furthermore, in step 5, when evaluating and validating the trained Ni-Ti-X element screening model, the regressor uses the coefficient of determination R. 2 The root mean square error (RMSE), mean absolute error (MAE), and explanatory error (EV) were used to comprehensively analyze the predictive performance of the Ni-Ti-X element screening model.

[0014] Furthermore, in step 6, the Ni-Ti-X element screening model is used to construct the mapping value of Ni-Ti-X alloy composition-phase transformation temperature, and a composition optimization diagram is constructed to guide the composition optimization of Ni-Ti-X shape memory alloy.

[0015] The beneficial effects of this invention are: This invention eliminates the need for traditional trial-and-error characterization experiments to optimize the third component composition of Ni-Ti shape memory alloys, effectively increasing the phase transition temperature while reducing R&D costs, thus accelerating the development and design of high-temperature Ni-Ti shape memory alloy materials. Furthermore, the Ni-Ti-X screening model holds promise for targeted development of Ni-Ti alloy materials in various scenarios, including solid-state refrigeration, vibration reduction and noise reduction, and extreme cold flexible actuation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the prior art or embodiments of the present invention, the accompanying drawings used in the prior art or embodiments will be briefly described below. Obviously, the drawings described below are merely embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without any creative effort.

[0017] Figure 1 This is a flowchart of the high-temperature Ni-Ti shape memory alloy composition optimization method based on deep learning disclosed in this invention; Figure 2 This is a schematic diagram illustrating the modeling principle of the Ni-Ti-X element screening model in this invention. Figure 3 The evaluation metric for the Ni-Ti-X element screening model in this invention is to iterate 50 times on the training set. Figure 4 This invention evaluates and verifies the prediction accuracy of the Ni-Ti-X element screening model on the test set. Figure 5 The composition optimization design of Ni-Ti-Zr shape memory alloy in the embodiments of the present invention is described, wherein "×" indicates the Ni-Ti-Zr alloy composition without phase transformation; the asterisk represents the Ni-Ti-Zr alloy composition selected by the model. Figure 6 This invention relates to the phase transformation temperature test of Ni-Ti-Zr shape memory alloy in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Furthermore, the embodiments described herein are only a part, not all, of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0019] Reference Figure 1 As shown, this invention discloses a method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning, comprising the following steps: Step 1: Collect alloy composition and phase transformation temperature data from relevant literature experiments on Ni-Ti-X shape memory alloys to construct an initial sample dataset, where the alloy composition is used as input, the phase transformation temperature is used as output, and X is the third component. The alloy composition includes Ni and its content, Pt and its content, Pd and its content, Zr and its content, Hf and its content, Co and its content, Cu and its content, Cr and its content, Mo and its content, V and its content, Ta and its content, Nb and its content, Fe and its content, Y and its content, Al and its content, W and its content, Nd and its content, La and its content, B and its content, and Re and their respective contents; the phase transformation temperatures include the martensite initiation temperature M. s Martensite termination temperature M f Austenite initiation temperature A s and austenite termination temperature A f ; Step 2: Construct a Ni-Ti-X alloy composition dataset from all the collected initial samples and perform systematic preprocessing on it. Preprocessing includes data cleaning and data normalization of the Ni-Ti-X alloy composition dataset to obtain the preprocessed Ni-Ti-X alloy composition dataset, which is used as the model input data. Data cleaning includes noise removal, outlier removal, and missing value imputation. Preprocessing was performed on the Ni-Ti-X alloy composition dataset to remove noise and outliers. This process identified and addressed data points caused by random fluctuations, minor measurement errors, or non-systematic interference resulting from variations in Ni-Ti-X shape memory alloy preparation and experimentation, which could interfere with the model's learning of the true underlying patterns. Based on the model / rules, a simple linear regression model was established using domain knowledge (the known empirical relationship between alloy composition and phase transformation temperature) to fit the data points. Data points significantly deviating from the fitted curve / surface were considered noise and corrected or removed. A linear relationship was established between the phase transformation temperature and the content of a certain element, with the fitting formula as follows: (1); in, The phase transformation temperature (M) of Ni-Ti-X shape memory alloys s M f A s A f ), and These are the fitted values. Alloy element composition.

[0020] Missing value imputation is performed on the Ni-Ti-X alloy composition dataset to address the issue of missing data in certain samples, preventing the entire dataset from being discarded or the model from becoming unprocessable due to missing data. The main method utilizes similarity sample imputation (K-nearest neighbor algorithm). For a missing sample A, the K most similar samples in the dataset are found to it in terms of other features. The weighted mean of these K samples on the missing feature (weighted by similarity) is used to impute the missing value of sample A. Similarity is typically measured using Euclidean distance, as shown in the following formula: (2); in, a is the Euclidean distance metric. i and b i This is a sample dataset of Ni-Ti-X alloy composition.

[0021] The formula for data normalization is as follows: (3); in, Representing data points The original value, This is its scaled value. Additionally, and These are the mean and standard deviation of each feature, respectively.

[0022] Step 3: Design the Ni-Ti-X element screening model. The Ni-Ti-X element screening model is a deep learning model that integrates an autoencoder and a multi-head self-attention mechanism (hereinafter referred to as the Ni-Ti-X element screening model). Figure 2 As shown, this model integrates the advantages of both mechanisms, enabling efficient learning of the mapping relationship between alloy composition and phase transformation temperature on small-sample Ni-Ti-X alloy composition datasets, which can be used to achieve optimized design of alloy composition.

[0023] The Ni-Ti-X element screening model has two core tasks: classification and regression. The classification task aims to determine whether high-temperature Ni-Ti-X shape memory alloys exhibit phase transformation behavior; the regression task is used to accurately predict the phase transformation temperature of Ni-Ti-X alloys. Therefore, the Ni-Ti-X element screening model consists of a regressor and a classifier.

[0024] The Ni-Ti-X element screening model captures different features of the input through a multi-head self-attention mechanism. Specifically, the query Q, key K, and value V are subjected to different linear transformations and projected onto the head. i Different subspaces (i.e., different heads) i For each head i Calculate using formula 4: (4); in, , , yes The projection matrix, query Q, key K, and value V are generated from the input through a linear transformation, where X is the input sequence. , and These are the weight matrices for Q, K, and V, as shown in Formula 5: (5); All The outputs of each attention point are concatenated to form a large vector. As shown in Formula 6: (6); Therefore, an 8-head attention mechanism is introduced into the Ni-Ti-X element screening model, enabling each element in the input sequence to establish a relationship with other elements in the sequence. Specifically, the attention mechanism allows each position to focus on information from other positions, thereby effectively capturing the interdependencies between alloy components. This mechanism allows the model to automatically focus on key features of the alloy composition during the learning process, improving prediction accuracy and model expressive power. Furthermore, parameters are employed... encoder The alloy composition is used as input to learn a low-dimensional representation, providing training data for the MLP prediction task. The encoder transforms the alloy composition information into a latent space by compressing it. This space contains important characteristic information about the alloy composition. Decoder Using the learned potential space and decoder parameters This generates new alloy compositions and provides candidate Ni-Ti-X alloy designs.

[0025] Finally, the degree of phase transformation temperature improvement used to evaluate the new Ni-Ti-X alloy composition was calculated using a score, as follows: (7); in, Phase transition temperature (M) s M f A s、 A f The weight matrix of ) is the phase transformation temperature of the Ni-Ti-X alloy.

[0026] Step 4: Divide the preprocessed Ni-Ti-X alloy composition dataset into training and testing sets; use the training set to train the Ni-Ti-X element screening model, and apply Bayesian optimization algorithm for hyperparameter tuning. Bayesian optimization constructs a probabilistic model of the objective function (model accuracy), selects the most promising combination of hyperparameters, and thus efficiently iteratively optimizes the model, improving the prediction accuracy and generalization stability of the Ni-Ti-X element screening model. The hyperparameter optimization process is shown in Table 1.

[0027] Table 1: Top 5 hyperparameter optimization configurations: Model Time batch_size k_features hidden_layer MAE Regressor 1 24 12 842 -29.9 2 21 11 843 -33.1 3 21 10 840 -34.5 4 24 13 836 -35.0 5 222 9 843 -35.5 Classifier 1 92 20 335 0.95 2 125 20 258 0.94 3 21 19 497 0.93 4 115 20 405 0.93 5 162 17 477 0.93 In this step, the regressor in the Ni-Ti-X element screening model uses the coefficient of determination R. 2 The accuracy of the regressor is evaluated by the explanatory error (RV), mean absolute error (MAE), and root mean square error (RMSE), and the calculation formulas are as follows: (8); in, Indicates the number of data samples. Indicates the actual value. Indicates the predicted value. This represents the mean of the actual values. The model's ability to fit the data is related to its R-squared value. 2 The values ​​are directly proportional; the closer the value is to 1, the better the fit.

[0028] (9); Var() represents variance, which is a statistic that measures the degree of dispersion of the data distribution.

[0029] (10); In this step, the Ni-Ti-X element screening model uses both accuracy and precision to comprehensively analyze its predictive performance. The calculation formula is as follows: (11); (12); Among them, true positives (TP) represent the number of positive samples correctly predicted as positive by the model, false positives (FP) represent the number of negative samples incorrectly predicted as positive by the model, false negatives (FN) represent the number of positive samples incorrectly predicted as negative by the model, and true negatives (TN) represent the number of negative samples correctly predicted as negative by the model.

[0030] Figure 3 The evaluation metrics of the Ni-Ti-X element screening model after 50 iterations on the training set are shown. The evaluation metrics of the regressor (such as R0) are also presented. 2 The results (MAE, RMSE, and EV) show that the ES model has high stability and low volatility. Through multiple iterations, it can be seen that the model's prediction results in the regressor are highly stable, and the prediction error fluctuation is small. Specifically, the means of all evaluation metrics are within acceptable ranges: R0 2 = 0.71, EV = 0.72, MAE = 0.41 (×10 -2 ), RMSE = 0.65 (×10) -2 The results show that the ES model can maintain high prediction accuracy and small prediction error when facing different datasets. The evaluation metrics of the classifier (mean Accuracy = 93% and Precision = 96%) also demonstrate the robustness of the model, with both accuracy and precision maintained at a high level and with small fluctuations.

[0031] Step 5: As Figure 4As shown, the trained Ni-Ti-X elemental screening model was evaluated and validated using a test set. The regressor's prediction results on the test set are presented in the form of a marginal square plot. In the figure, the x-axis represents the actual phase transition temperature, and the y-axis represents the model's predicted value. By observing the distribution of data points, the relationship between the actual and predicted values ​​can be intuitively understood. Except for a few outliers, the data points are mainly distributed along the diagonal, indicating that the model has good predictive performance. In addition, the confusion matrix of the classifier's prediction results is shown, where "0" indicates that the prediction result is no phase transition, and "1" indicates that the prediction result is a phase transition. This verifies its accuracy in handling complex compositional changes and phase transition temperature prediction tasks.

[0032] Step 6: Use the trained and optimized Ni-Ti-X element screening model to predict and screen Ni-Ti-X alloy composition combinations with high phase transformation temperatures, which can be used to guide the composition optimization design and preparation of high-temperature Ni-Ti-X shape memory alloys.

[0033] The implementation of the deep learning-based high-temperature Ni-Ti shape memory alloy composition optimization method of this application embodiment is introduced with a specific example.

[0034] The composition of Ni-Ti-Zr alloy was screened and optimized using the Ni-Ti-X elemental screening model. Figure 5 This paper demonstrates how the Ni-Ti-X elemental screening model guides the design of high-temperature NiTiZr alloys. In the figure, "×" indicates that the NiTiZr alloy composition exhibits no phase transformation behavior. As shown in the figure, with increasing Ni, Ti, and Zr content, some alloy components lose their phase transformation behavior; these components are all marked with "×". To ensure that the designed NiTiZr alloy possesses phase transformation behavior and has a high possible phase transformation temperature, the phase transformation temperature boundaries of different compositions were analyzed, and Ni was selected... 43.2 Ti 41.8 Zr 15 (at.%) represents the target alloy composition, indicated by an asterisk in the figure.

[0035] Ni optimized for vacuum casting 43.2 Ti 41.8 Zr 15 Phase transformation temperature tests were conducted on the alloy, and the experimental results were compared and analyzed with the predicted values ​​from the Ni-Ti-X elemental screening model. The experimental results show that the predicted phase transformation temperature is basically consistent with the actual measured value. It is worth noting that Ni... 43.2 Ti 41.8 Zr 15 The phase transformation temperature of the alloy was significantly increased (M s =99.6℃; M f =80.4℃; A s =126.6℃; A f=161.6℃).

[0036] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to the embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning, characterized in that, Includes the following steps: Step 1: Collect alloy composition and phase transformation temperature data from relevant literature experiments on Ni-Ti-X shape memory alloys, construct an initial sample dataset, with alloy composition as input, phase transformation temperature as output, and X as the third component composition; Step 2: Construct a Ni-Ti-X alloy composition dataset from all collected initial samples and preprocess it. The preprocessing includes data cleaning and data normalization of the Ni-Ti-X alloy composition dataset to obtain the preprocessed Ni-Ti-X alloy composition dataset, which will be used as the model input data. Step 3: Design the Ni-Ti-X element screening model. The Ni-Ti-X element screening model is a deep learning model that integrates an autoencoder and a multi-head self-attention mechanism. The Ni-Ti-X element screening model integrates the advantages of the two mechanisms and can efficiently learn the mapping relationship between alloy composition and phase transformation temperature on a small sample Ni-Ti-X alloy composition dataset, which can be used to achieve the optimized design of alloy composition. Step 4: Divide the preprocessed Ni-Ti-X alloy composition dataset into a training set and a test set; use the training set to train the Ni-Ti-X element screening model and iteratively optimize the model hyperparameters to improve the predictive ability and stability of the Ni-Ti-X element screening model. Step 5: Evaluate and validate the trained Ni-Ti-X element screening model using the test set; Step 6: Use the trained and optimized Ni-Ti-X element screening model to predict and screen Ni-Ti-X alloy composition combinations with high phase transformation temperatures, which can be used to guide the composition optimization design and preparation of high-temperature Ni-Ti-X shape memory alloys.

2. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, The Ni-Ti-X alloy composition dataset obtained in step 1 specifically includes the Ni-Ti-X alloy composition and phase transformation temperature, where X can be one of Pt, Pd, Zr, Hf, Co, Cu, Cr, Mo, V, Ta, Nb, Fe, Y, Al, W, Nd, La, B, and Re.

3. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 2, the preprocessing methods used include data cleaning and normalization based on linear regression, K-nearest neighbor model and rules, and domain knowledge. Data cleaning includes noise removal, outlier removal and missing value imputation.

4. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 3, the Ni-Ti-X element screening model has two core tasks: classification and regression. The classification task aims to determine whether the high-temperature Ni-Ti-X shape memory alloy has phase transformation behavior; the regression task is used to accurately predict the phase transformation temperature of the Ni-Ti-X alloy. Thus, the Ni-Ti-X element screening model is divided into a regressor and a classifier.

5. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 4, the ratio of the number of samples in the training set to the number of samples in the test set is 80%:20%.

6. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 4, the Bayesian optimization algorithm is used to select and fine-tune the hyperparameters when iteratively optimizing the model hyperparameters.

7. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 4, the hyperparameters of the Ni-Ti-X element screening model include batch size, number of features k_features, and number of hidden layer nodes hidden_layer.

8. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 5, when evaluating and validating the trained Ni-Ti-X element screening model, the regressor uses the coefficient of determination R. 2 The root mean square error (RMSE), mean absolute error (MAE), and explanatory error (EV) were used to comprehensively analyze the predictive performance of the Ni-Ti-X element screening model.

9. The method for optimizing the composition of high-temperature Ni-Ti shape memory alloys based on deep learning according to claim 1, characterized in that, In step 6, the Ni-Ti-X element screening model is used to construct the mapping value of Ni-Ti-X alloy composition-phase transformation temperature, and a composition optimization diagram is constructed to guide the composition optimization of Ni-Ti-X shape memory alloy.

Images