Alloy component optimization method based on neural network model
Through the alloy composition optimization method based on neural network model, the problem of low component screening efficiency in the development of nickel-based high-temperature alloys is solved. The neural network model is used to predict alloy performance and optimize components, and rapid and efficient alloy development is achieved, reducing resource waste and Manpower and material resources investment.
Patent Information
- Application Number
- CN202411064545.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-15
AI Technical Summary
During the development of traditional nickel-based high-temperature alloys, the alloy composition screening and design efficiency are low, and relying on trial and error methods leads to a large amount of resource waste. The existing efficient methods show limitations in complex nickel-based high-temperature alloy systems.
Using the alloy composition optimization method based on neural network model, the neural network model is trained, the input alloy composition is used to predict its performance, and the alloy composition is optimized based on the prediction results. The samples are prepared using thermal isostatic micro-synthesis technology to perform performance testing, and the formation of Dataset, feedback training model for improved accuracy.
It has achieved rapid optimization of alloy components, reduced manpower and material investment, and can machine learning training on the components and performance of hundreds of alloys, achieving rapid prediction of target alloy performance, and improving R&D efficiency.
Smart Images

Figure BDA0004978978500000031 
Figure HDA0004978978510000011 
Figure HDA0004978978510000012
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of material metallurgy, and in particular relates to an alloy composition optimization method based on a neural network model. Background Art
[0002] Nickel-based high-temperature alloys have a wide range of composition, novel design concepts, unique properties, and excellent freedom of performance control, but this also makes it difficult for engineers and technicians to quickly and efficiently screen out the required nickel-based high-temperature alloy system. The traditional new alloy development process is mainly based on trial and error, which involves preliminary manual screening of alloy components, and then using arc melting to prepare samples for actual measurement and verification. This method requires a large amount of trial and error costs, which is very unfavorable for the screening and design of target alloys.
[0003] With the development of technology, some new and more efficient alloy composition optimization methods have been proposed. For example, the low alloy steel alloy composition optimization method proposed in the document with patent number CN202311225177.7 has shown certain effectiveness in low alloy systems; however, it still exposes great limitations when facing the complex and changeable material system of nickel-based high temperature alloys. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides an alloy composition optimization method based on a neural network model to solve the problem of composition optimization efficiency of nickel-based high-temperature alloys in the research and development process.
[0005] The present invention achieves the above technical objectives through the following technical means.
[0006] A method for optimizing alloy composition based on a neural network model: for the target alloy to be prepared, a data set is prepared to train a neural network model, the input of the model is the alloy composition, and the output of the model is the alloy performance; the trained model is used to predict the alloy performance, and the alloy composition is optimized and adjusted according to the model prediction results.
[0007] Furthermore, the dataset is made as:
[0008] According to the phase diagram of the target alloy, several groups of basic alloy components are selected;
[0009] For each basic alloy component, other elements of different contents are added to obtain multiple alloy composition ratios as samples;
[0010] Samples were made according to the sample alloy composition ratio, and performance tests were carried out to obtain a data set.
[0011] Furthermore, in the preparation of the data set: based on the phase diagram of the target alloy, a group of alloy components is selected as a benchmark with the requirement of being able to form a face-centered cubic single-phase solid solution; then 10 to 20 groups of basic alloy components are expanded outward with the benchmark alloy components as the center.
[0012] Furthermore, in the preparation of the data set: a metal raw material with a purity of 99.9at% was used, and the alloy sample was prepared under vacuum state by using hot isostatic pressing micro-synthesis manufacturing technology.
[0013] Furthermore, in the preparation of the data set: performance tests include hardness, lattice constant, and grain size, wherein the hardness test uses a micron mechanical hardness tester and takes the average value through a 5x5 matrix; the grain size is observed and calculated through SEM.
[0014] Furthermore, the neural network model is a linear model.
[0015] Furthermore, the neural network model is an elastic network model.
[0016] Furthermore, it is used to optimize the alloy composition of nickel-based high-temperature alloys.
[0017] Furthermore, samples are prepared based on the optimized target alloy composition for performance measurement, and the measured results are fed back to the neural network model to further train the model.
[0018] The beneficial effects of the present invention are:
[0019] The present invention provides an alloy composition optimization method based on a neural network model, which can achieve rapid optimization of alloy composition and reduce the investment of a large amount of manpower and material resources. The composition, organization and performance data of hundreds of alloys can be trained by machine learning, and a model can be established and optimized to achieve rapid prediction of the performance of the target alloy. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 It is a curve chart showing the change of RMSE value relative to the feature number of the two selected neural network models;
[0021] Figure 2 The distribution diagram of the predicted values of the two selected neural network models relative to the actual values. DETAILED DESCRIPTION
[0022] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.
[0023] In view of the composition optimization problem of nickel-based high-temperature alloys, the present invention provides the following alloy composition optimization method:
[0024] Step 1: Sampling
[0025] Step 1.1, phase diagram analysis: Based on the phase diagram of the target alloy, a group of alloy components is selected as a benchmark with the ability to form a face-centered cubic single-phase solid solution as the basic requirement; then 10 to 20 groups of basic alloy components are expanded outward with the benchmark alloy components as the center; finally, other elements of different contents are added to the above-selected groups of basic alloy components to form multiple alloy composition ratios as samples.
[0026] For example, taking NiTiCr alloy as an example, 14 groups of basic alloy components are selected based on the phase diagram; then different contents of Fe are added to the above 14 groups of basic alloy components to obtain 105 alloy components as samples; wherein the Ni content is 10-30, the Cr content is 10-20, and the Ti content is 0.1-5 (percentage).
[0027] Step 1.2, sample preparation: using metal raw materials with a purity of 99.9at%, according to the sample composition ratio selected in step 1.1, the alloy samples were prepared in one go under vacuum using hot isostatic pressing micro-synthesis manufacturing technology to ensure a uniform experimental environment.
[0028] Step 1.3, performance test: The prepared alloy samples were subjected to performance tests, including hardness, lattice constant, grain size, etc. The hardness test was performed using a micron mechanical hardness tester, and the average value was taken through a 5x5 matrix; the grain size was observed and calculated by SEM.
[0029] Step 1.4, create a data set: summarize the measured performance data corresponding to the above samples to form a data set for training the neural network model.
[0030] Step 2: Model training
[0031] Based on the data set formed by sampling in step 1, the neural network model is trained, where 80% of the data in the data set is used as a training set and 20% as a test set. To improve the prediction accuracy of the model, the Bayesian optimization method is used to optimize the model's hyperparameters before training.
[0032] Exemplarily, this embodiment selects a linear model (Linear) and an elastic network model (ElasticNet_CrossValidation, EN_CV) for testing.
[0033] After the model training is completed, the cross-validation method can be used to test the model performance. Specifically, the stability and accuracy of the model are evaluated by 10-fold cross-validation and repeated random splits 100 times:
[0034]
[0035] Where n is the number of samples, y i represents the actual performance value of the i-th sample alloy, μ i It represents the corresponding model prediction value. The smaller the RMSE, the higher the prediction accuracy of the model.
[0036] like Figure 1 As shown in the figure, the errors of the two models gradually decrease as the number of random combinations of features increases; (a) is for the Linear model, and (b) is for the EN_CV model. Figure 2 The following is an intuitive distribution diagram of the predicted values and actual values of the two models, where (a) and (b) are for the Linear model, and (c) and (d) are for the EN_CV model. After comprehensive testing, the RMSE value of the EN_CV model is 7.31, which is better than the Linear model. Therefore, EN_CV is preferred as the final prediction model.
[0037] Step 3: Alloy composition optimization
[0038] Use the trained neural network model to predict the properties of the target alloy to be prepared; based on the model prediction results, adjust the alloy composition according to the required alloy performance requirements to determine the final target alloy composition.
[0039] Step 4: Feedback and Optimization Loop
[0040] Samples are prepared based on the optimized target alloy composition, and performance measurements are carried out. The measured results are fed back to the neural network model to further train the model and continuously improve the accuracy of the model.
[0041] Finally, for example, based on the above alloy composition optimization method, this embodiment selects the following nickel-based high-temperature alloy composition: C0.05%Ni15%Cr15%Mo2.2%Mn2.0%Si0.8%Ti0.3%FeBal.
[0042] 1) The corresponding model prediction performance is:
[0043] Room temperature mechanical properties: Rm = 732MPa, Rp0.2 = 622MPa, Rm represents tensile strength, Rp0.2 represents yield strength, room temperature micro-Vickers hardness HV0.5 = 271.
[0044] High temperature mechanical properties: at 100℃, Rm=653MPa, Rp0.2=562MPa, at 200℃, Rm=625MPa, Rp0.2=551MPa, at 300℃, Rm=619MPa, Rp0.2=544MPa, at 400℃, Rm=598MPa, Rp0.2=507MPa, at 500℃, Rm=591MPa, Rp0.2=495MPa, at 600℃, Rm=534MPa, Rp0.2=444MPa, at 700℃, Rm=437MPa, Rp0.2=278MPa, at 800℃, Rm=330MPa, Rp0.2=176MPa.
[0045] 2) The measured performance data is:
[0046] Room temperature mechanical properties: Rm = 730MPa, Rp0.2 = 620MPa, A = 29.0%, Rm represents tensile strength, Rp0.2 represents yield strength, and A represents elongation after fracture; room temperature micro-Vickers hardness HV0.5 = 272, 269, 268;
[0047] High temperature mechanical properties: at 100℃, Rm=650MPa, Rp0.2=560MPa, A=17.0%; at 200℃, Rm=623MPa, Rp0.2=549MPa, A=12.0%; at 300℃, Rm=617MPa, Rp0.2=541MPa, A=11.0%; at 400℃, Rm=598MPa, Rp0.2=509MPa, A=13 .0%, at 500℃, Rm=589MPa, Rp0.2=494MPa, A=14.0%, at 600℃, Rm=535MPa, Rp0.2=442MPa, A=10.0%, at 700℃, Rm=439MPa, Rp0.2=278MPa, A=27.0%, at 800℃, Rm=331MPa, Rp0.2=178MPa, A=45.0%.
[0048] The overall error does not exceed five percent.
[0049] The present invention is not limited to the above-mentioned embodiments. Any obvious improvement, substitution or deformation that can be made by those skilled in the art without departing from the essential content of the present invention belongs to the protection scope of the present invention.
Claims
1. A method for optimizing alloy composition based on a neural network model, characterized in that: For the target alloy to be prepared, a data set is prepared to train a neural network model, the input of the model is the alloy composition, and the output of the model is the alloy performance; the trained model is used to predict the alloy performance, and the alloy composition is optimized and adjusted based on the model prediction results.
2. The alloy composition optimization method based on the neural network model according to claim 1, characterized in that: The dataset is made as: According to the phase diagram of the target alloy, several groups of basic alloy components are selected; For each basic alloy component, other elements of different contents are added to obtain multiple alloy composition ratios as samples; Samples were made according to the sample alloy composition ratio, and performance tests were carried out to obtain a data set.
3. The alloy composition optimization method based on the neural network model according to claim 2, characterized in that: In the preparation of the data set: based on the phase diagram of the target alloy, a group of alloy components is selected as a benchmark with the requirement of being able to form a face-centered cubic single-phase solid solution; then 10 to 20 groups of basic alloy components are expanded outward with the benchmark alloy components as the center.
4. The alloy composition optimization method based on the neural network model according to claim 2, characterized in that: In the preparation of the data set: a metal raw material with a purity of 99.9at% was used, and the alloy sample was prepared under vacuum by using hot isostatic pressing micro-synthesis manufacturing technology.
5. The alloy composition optimization method based on the neural network model according to claim 2, characterized in that: The data set is being prepared: performance tests include hardness, lattice constant, and grain size, wherein the hardness test uses a micron mechanical hardness tester and takes the average value through a 5x5 matrix; the grain size is observed and calculated through SEM.
6. The alloy composition optimization method based on the neural network model according to claim 1, characterized in that: The neural network model is a linear model.
7. The alloy composition optimization method based on the neural network model according to claim 1, characterized in that: The neural network model is an elastic network model.
8. The alloy composition optimization method based on the neural network model according to claim 1, characterized in that: Used to optimize the alloy composition of nickel-based high-temperature alloys.
9. The alloy composition optimization method based on the neural network model according to claim 1, characterized in that: Samples are prepared based on the optimized target alloy composition for performance measurement, and the measured results are fed back to the neural network model for further model training.
