Search

CN-121983202-A - High-temperature Ni-Ti shape memory alloy component optimization method based on deep learning

CN121983202ACN 121983202 ACN121983202 ACN 121983202ACN-121983202-A

Abstract

Aiming at the problems of low phase transition temperature and rapid function degradation of the traditional Ni-Ti shape memory alloy under the high-temperature working condition, the invention provides a deep learning-based high-temperature Ni-Ti shape memory alloy component optimization method which takes a Ni-Ti-X ternary system as a search space, and then, establishing a Ni-Ti-X element screening model by combining an automatic encoder of a multi-head self-attention mechanism, and realizing accurate reverse design of Ni-Ti shape memory alloy components in a phase transition temperature (150-400 ℃) interval, wherein experimental verification errors are < +/-10 ℃. Compared with the traditional trial-and-error method, the method shortens the development period of the alloy components from a few months to a few days, and the obtained Ni-Ti-15Zr (at%) alloy has a phase transition temperature of about 170 ℃ and is obviously superior to the existing Ni-Ti alloy (less than or equal to 100 ℃). The invention provides a high-efficiency and low-cost component design way for high-temperature driving and sealing components of aeroengines, oil-gas wells and the like.

Inventors

  • Yang Kongyuan
  • WANG TONGJIAN
  • YANG KONGHUA
  • LIU CHUNBAO
  • ZHAI LU
  • SUN HAIKUAN

Assignees

  • 吉林大学

Dates

Publication Date
20260505
Application Date
20260327

Claims (9)

  1. 1. The high-temperature Ni-Ti shape memory alloy component optimization method based on deep learning is characterized by comprising the following steps of: Step 1, collecting alloy components and phase transition temperature data from a Ni-Ti-X shape memory alloy related literature experiment, constructing an initial sample data set, wherein the alloy components are used as input, the phase transition temperature is used as output, and X is a third component; Step 2, constructing all collected initial samples into a Ni-Ti-X alloy component data set, and preprocessing the Ni-Ti-X alloy component data set, wherein the preprocessing comprises data cleaning and data normalization of the Ni-Ti-X alloy component data set to obtain a preprocessed Ni-Ti-X alloy component data set as model input data; Step 3, designing a Ni-Ti-X element screening model, wherein the Ni-Ti-X element screening model is a deep learning model integrating an automatic encoder and a multi-head self-attention mechanism, integrates the advantages of the two mechanisms, can efficiently learn the mapping relation between alloy components and phase transition temperature on a small sample Ni-Ti-X alloy component data set, and is used for realizing the optimal design of alloy components; Step 4, dividing the preprocessed Ni-Ti-X alloy component data set into a training set and a testing set, training the Ni-Ti-X element screening model by using the training set, and iteratively optimizing the super parameters of the model so as to improve the prediction capability and stability of the Ni-Ti-X element screening model; Step 5, evaluating and verifying the trained Ni-Ti-X element screening model by using a test set; And 6, predicting and screening the Ni-Ti-X alloy component combination with high phase transition temperature by using the trained and optimized Ni-Ti-X element screening model, and guiding the component optimization design and preparation of the high-temperature Ni-Ti-X shape memory alloy.
  2. 2. The deep learning-based high temperature Ni-Ti shape memory alloy composition optimization method of claim 1, wherein the Ni-Ti-X alloy composition dataset obtained in step 1 specifically comprises one of Ni-Ti-X alloy composition, phase transition temperature, X may be Pt, pd, zr, hf, co, cu, cr, mo, V, ta, nb, fe, Y, al, W, nd, la, B, re.
  3. 3. The deep learning-based high-temperature Ni-Ti shape memory alloy composition optimization method according to claim 1, wherein in the step 2, the pretreatment method adopted includes performing data cleaning and normalization by using domain knowledge based on linear regression, K nearest neighbor model and rule, and the data cleaning includes noise removal, outlier rejection and missing value filling.
  4. 4. The deep learning-based high-temperature Ni-Ti shape memory alloy composition optimization method according to claim 1, wherein in the step 3, the Ni-Ti-X element screening model is used for clearly classifying and regressing two core tasks, namely a classification task is used for judging whether the high-temperature Ni-Ti-X shape memory alloy has phase transition behavior or not, and a regression task is used for accurately predicting the phase transition temperature of the Ni-Ti-X alloy, so that the Ni-Ti-X element screening model is divided into a regressor and a classifier.
  5. 5. The deep learning-based high temperature Ni-Ti shape memory alloy composition optimization method of claim 1, wherein in step 4, the number of samples divided by training set and test set is 80% to 20%.
  6. 6. The deep learning-based high-temperature Ni-Ti shape memory alloy composition optimization method according to claim 1, wherein in the step 4, a bayesian optimization algorithm is adopted to perform super-parameter selection and fine tuning when the model super-parameters are optimized iteratively.
  7. 7. The method for optimizing the composition of the high-temperature Ni-Ti shape memory alloy based on deep learning according to claim 1, wherein the super parameters of the Ni-Ti-X element screening model in the step 4 include batch size, feature number k_features, hidden layer node number hidden_layer.
  8. 8. The method for optimizing the composition of the high-temperature Ni-Ti shape memory alloy based on deep learning according to claim 1, wherein in the step 5, when the trained Ni-Ti-X element screening model is evaluated and verified, a regressor adopts a decision coefficient R 2 , a root mean square error RMSE, an average absolute error MAE and an interpretation error EV, and a classifier adopts an Accuracy and an Accuracy to comprehensively analyze the prediction performance of the Ni-Ti-X element screening model.
  9. 9. The deep learning-based high-temperature Ni-Ti shape memory alloy composition optimization method according to claim 1, wherein in the step 6, a Ni-Ti-X element screening model is used to construct a map of Ni-Ti-X alloy composition-phase transition temperature, and a composition optimization graph is constructed to guide the optimization of Ni-Ti-X shape memory alloy composition.

Description

High-temperature Ni-Ti shape memory alloy component optimization method based on deep learning Technical Field The invention relates to the field of metal material component design, in particular to a high-temperature Ni-Ti shape memory alloy component optimization method based on deep learning. Background Shape Memory Alloys (SMA) have been regarded as key functional materials in aero-engine adjustable bypass valves, deep well completion packers, and high temperature sensing actuators due to their macroscopic shape recovery effect from reversible martensitic transformation. However, the martensite reverse transformation temperature (Af) of the traditional near-equiatomic-ratio Ni-Ti system is generally lower than 100 ℃, and under the action of martensite-austenite cyclic load, microscopic damages such as dislocation annihilation, ti2Ni precipitation and interfacial non-co-graining are rapidly accumulated, so that the function attenuation is remarkable, and the long-life requirement of a service temperature zone of more than or equal to 150 ℃ is difficult to meet. Researchers in the past research try to introduce Hf, zr, pd, pt, au and other third components into a Ni-Ti binary matrix so as to improve the phase transition temperature and enhance the thermal stability through lattice distortion and electron concentration regulation. However, the multi-element alloy composition-structure-performance mapping relation presents the characteristics of high nonlinearity, strong coupling and multipole value, the traditional trial-error-representation experimental paradigm has extremely low searching efficiency in a high-dimensional composition space, the single high-temperature cycle-fatigue experimental period is as long as several months, the cost is exponentially increased, and the data is sparse and the noise is obvious. With the rise of material genetic engineering and data driving paradigm, how to utilize a machine learning model to mine the physical rule hidden in limited experimental data under the constraint of 'small sample, high noise and high dimensionality', and further realize the Ni-Ti-X reverse design with the cooperative optimization of target phase transition temperature (150-400 ℃) and cycle stability, has become a breakthrough problem in the field of high-temperature shape memory alloy. Disclosure of Invention In view of the above, the invention provides a deep learning-based high-temperature Ni-Ti shape memory alloy component optimization method, which can accurately construct the mapping relation between Ni-Ti alloy components and phase transition temperature, thereby improving the high-temperature service performance of the Ni-Ti shape memory alloy by adopting a low-cost and high-efficiency method. A method for optimizing the composition of a high-temperature Ni-Ti shape memory alloy based on deep learning, the method comprising the following: Step 1, collecting alloy components and phase transition temperature data from a Ni-Ti-X shape memory alloy related literature experiment, constructing an initial sample data set, wherein the alloy components are used as input, the phase transition temperature is used as output, and X is a third component; Step 2, constructing all collected initial samples into a Ni-Ti-X alloy component data set, and preprocessing the Ni-Ti-X alloy component data set, wherein the preprocessing comprises data cleaning and data normalization of the Ni-Ti-X alloy component data set to obtain a preprocessed Ni-Ti-X alloy component data set as model input data; Step 3, designing a Ni-Ti-X element screening model, wherein the Ni-Ti-X element screening model is a deep learning model integrating an automatic encoder and a multi-head self-attention mechanism, integrates the advantages of the two mechanisms, can efficiently learn the mapping relation between alloy components and phase transition temperature on a small sample Ni-Ti-X alloy component data set, and is used for realizing the optimal design of alloy components; Step 4, dividing the preprocessed Ni-Ti-X alloy component data set into a training set and a testing set, training the Ni-Ti-X element screening model by using the training set, and iteratively optimizing the super parameters of the model so as to improve the prediction capability and stability of the Ni-Ti-X element screening model; Step 5, evaluating and verifying the trained Ni-Ti-X element screening model by using a test set; And 6, predicting and screening the Ni-Ti-X alloy component combination with high phase transition temperature by using the trained and optimized Ni-Ti-X element screening model, and guiding the component optimization design and preparation of the high-temperature Ni-Ti-X shape memory alloy. Further, in the step 1, a Ni-Ti-X alloy composition dataset is obtained, specifically including a Ni-Ti-X alloy composition, a phase transition temperature, and X may be one of Pt, pd, zr, hf, co, cu, cr, mo, V, ta, nb, fe, Y, al, W, nd, la, B, re.