CN118352002A - A method for predicting the yield strength of γ' phase strengthened cobalt-based superalloys - Google Patents

Abstract

The present invention discloses a method for predicting the yield strength of a γ' phase strengthened cobalt-based high-temperature alloy, and belongs to the field of cobalt-based high-temperature alloys. The method is based on a γ' phase strengthened cobalt-based high-temperature alloy composition-heat treatment process-yield strength data set, and uses machine learning to construct a prediction model based on a gradient boosting algorithm to achieve yield strength predictions for this series of alloys at different temperatures. The input of the prediction model is alloy composition, heat treatment process parameters, and test temperature; the output is yield strength. The yield strength prediction method can make a fast and accurate prediction of the yield strength of a γ' phase strengthened cobalt-based high-temperature alloy within a large composition and temperature range, and has a strong application value in the composition design and optimization of this series of alloys.

Description

A method for predicting the yield strength of γ' phase strengthened cobalt-based superalloys

Technical Field

The invention belongs to the field of cobalt-based high-temperature alloys, and in particular relates to a method for predicting the yield strength of a γ' phase-strengthened cobalt-based high-temperature alloy.

Background technique

Cobalt-based superalloys occupy an indispensable position in the aviation and energy industries due to their excellent mechanical properties and thermal stability. They are widely used in key hot-end components such as aircraft engines and gas turbines. For this type of alloy, maintaining stable mechanical properties in a high-temperature working environment is crucial to improving the performance, efficiency and reliability of the engine. γ' phase-strengthened cobalt-based superalloys are a type of cobalt-based superalloy that has begun to develop in recent years. They have better high-temperature mechanical properties than traditional cobalt-based superalloys and have the potential to become the next generation of high-temperature structural materials. Yield strength is a key parameter for evaluating the high-temperature mechanical properties of this type of alloy. However, as a new alloy system, the prediction method for the yield strength of γ' phase-strengthened cobalt-based superalloys has not yet been established, which is not conducive to the rapid development and engineering application of this type of alloy.

Traditional methods for predicting the yield strength of high-temperature alloys mainly rely on classical theories and empirical formulas. These methods often face many challenges, including complex calculations of material behavior, estimation of multiple unknown parameters, and the need for a large amount of experimental data, which not only increases the cost but also affects the accuracy and efficiency of the prediction. In recent years, with the rapid development of artificial intelligence and computer science, machine learning methods have been introduced into materials science research. For example, machine learning has been shown to effectively process data sets and predict material properties through algorithmic learning. This approach can significantly reduce the need for experiments, reduce R&D costs, and accelerate the development of new materials.

At present, machine learning has been successfully applied in predicting the high-temperature mechanical properties of high-temperature alloys. CN114818481A proposes a method for predicting the creep rupture life of single-crystal high-temperature alloys. Based on the machine learning algorithm, a single-crystal high-temperature alloy creep rupture life prediction model that can quickly evaluate the creep performance of the alloy is obtained. For structural materials γ' phase strengthened cobalt-based high-temperature alloys, yield strength is the most important indicator. However, machine learning methods have high requirements for the amount of data, and the experimental data on yield strength provided in existing literature is very limited. Therefore, there is currently no existing technology that uses machine learning methods to predict yield strength.

Summary of the invention

In response to the above-mentioned technical problem that there is currently no machine learning method for predicting yield strength, the present invention provides a method for predicting the yield strength of γ' phase strengthened cobalt-based high-temperature alloy based on machine learning, which can quickly predict the yield strength of such alloys at different temperatures and is suitable for a larger alloy composition and test temperature range.

The technical solution adopted by the present invention is specifically as follows:

A method for predicting the yield strength of a γ' phase strengthened cobalt-based high-temperature alloy is provided. A data set is first established, and then a prediction model for the yield strength of a γ' phase strengthened cobalt-based high-temperature alloy is established based on machine learning. The method specifically comprises the following steps:

S1. Establish a data set; collect relevant data on the yield strength of γ' phase strengthened cobalt-based high-temperature alloy at different temperatures to construct a data set, wherein the data includes alloy composition, heat treatment process parameters, test temperature and corresponding yield strength; the data set is randomly divided into a training set and a test set in a ratio of 6~8:2~4, wherein the data of the training set is used to train the prediction model, and the data of the test set is used to test the model accuracy;

S2. Establish a prediction model; for the training set constructed in S1, use different machine learning algorithms to train the prediction model, and select the gradient boosting algorithm with the best comprehensive prediction performance to build the optimal model;

S3. Yield strength prediction: The alloy composition, heat treatment process parameters and test temperature of the γ' phase strengthened cobalt-based high-temperature alloy to be predicted are input into the optimal model obtained in S2 to obtain the predicted value of the yield strength.

Furthermore, in S1, the specific process of data collection is: based on the existing literature (the data obtained from different databases will have some differences, but because there are more data samples, there will not be significant differences in the data, and thus there will not be obvious differences in the results due to differences in database sources. Therefore, no specific requirements are required for the database), the literature is obtained by searching keywords and other search methods (keywords such as Co, Ni, yield strength, yield stress, etc.), and the yield strength data of different alloys at different temperatures are extracted from the curve chart of the obtained literature using Origin software. Finally, all the data include alloy composition, heat treatment process parameters, test temperature and corresponding yield strength.

Furthermore, the data set contains more than 400 data, covering the contents of 8 to 12 alloy elements, more than 4 heat treatment process parameters and yield strength test temperatures.

Furthermore, the data are preferably 400 to 600, and the alloying elements are preferably 10, including Co, Ni, Al, W, Ti, Ta, Cr, Mo, V, and B; the heat treatment process parameters include solution temperature, solution time, aging temperature, and aging time.

Furthermore, the content of alloy elements ranges from Co 14-88at.%, Ni 0-54at.%, Al 0-10at.%, W0-11at.%, Ti 0-12at.%, Ta 0-2.8at.%, Cr 0-18at.%, Mo 0-9at.%, V 0-12at.%, B 0-0.02at.%; the solution temperature is 1100-1350℃, the solution time is 2-100 h, the aging temperature is 760-1100℃, the aging time is 4-400 h, the yield strength test temperature is 25-1100℃, and the yield strength is 94-888MPa.

Furthermore, in S2, different machine learning algorithms preferably select different classic machine learning algorithms, including nearest neighbor algorithm, random forest algorithm, gradient boosting algorithm and support vector machine algorithm.

Furthermore, in S2, the number of trees in the gradient boosting algorithm parameters is optimized in the range of 160-200, and the depth of the tree is 4-5.

Furthermore, in S3, the input parameters of the model are alloy composition, solution temperature, solution time, aging temperature, aging time and test temperature; and the output parameter is yield strength.

In order to solve the problem of the data volume of yield strength, the inventors not only used the experimental data directly presented in the existing literature, but also thought of extracting the yield strength data of different alloys at different temperatures from the curve diagrams of the existing literature, thereby significantly expanding the data range of yield strength and being able to meet the data volume requirements of the machine learning method. However, the data obtained in this way do have a certain degree of error. Therefore, in the initial process of constructing the yield strength prediction model, the inventors used the alloy composition and test temperature, the parameters most relevant to yield strength, as input data. The results showed that the error was large. Through continuous attempts and analysis, the inventors added heat treatment process parameters on the basis of alloy composition and test temperature, and finally obtained a very ideal error rate and accuracy value, which can effectively predict the yield strength of γ' phase strengthened cobalt-based high-temperature alloy.

The beneficial effects of the present invention are:

The present invention obtains a yield strength prediction method for γ' phase strengthened cobalt-based high-temperature alloys based on a gradient boosting model by screening different machine learning algorithms. The method can quickly predict the yield strength of the series of alloys at different temperatures under different compositions and process conditions. It not only has high prediction accuracy, but also has a large applicable composition and temperature range. The specific implementation effect of the optimal model on the training set and the test set is shown in Figure 2. The results show that the average relative error of the test set of the optimal model is 6.4%, and the accuracy value ^R2 reaches 0.91, that is, it has excellent prediction performance. Therefore, the prediction method of the present invention has significant engineering application value in the composition design and process optimization of γ' phase strengthened cobalt-based high-temperature alloys.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the present invention;

Figures 2 to 5 are the result graphs of the average error rate of the test set under different parameters of the nearest neighbor algorithm, the random forest algorithm, the gradient boosting algorithm and the support vector machine algorithm;

FIG6 is a diagram showing the specific implementation effect of the prediction method of the present invention on a training set and a test set;

FIG. 7 is a diagram showing the implementation effect of the prediction method of the present invention on alloys outside the training set.

Detailed ways

The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments, but the present invention is not limited thereto.

Example 1

With Co, Ni, yield strength and yield stress as keywords, the existing literature was retrieved from the Wanfang database. Through denoising, the literature that obviously did not meet the conditions was excluded. 402 relevant data on the yield strength of γ' phase strengthened cobalt-based high-temperature alloys at different temperatures were collected to construct a data set with the data structure of alloy composition-heat treatment process parameters-test temperature-yield strength. The data set was randomly divided into a training set and a test set in a ratio of 8:2. The data in the training set was used to train the prediction model, and the data in the test set was used to test the model accuracy.

Based on the above training set and test set, alloy composition, solution temperature, solution time, aging temperature, aging time and test temperature are used as input parameters of the model; the output parameter is yield strength. The nearest neighbor algorithm, random forest algorithm, gradient boosting algorithm and support vector machine algorithm are used to train the prediction model. From Figures 2 to 5, it can be seen that the nearest neighbor algorithm has a minimum average error rate of 16% when the K value is 2; the random forest algorithm shows a rule that the average error rate is small at the top and large at the bottom. When the number of trees is 180 and the depth is 20, the average error rate is 11%; the gradient boosting algorithm has a minimum average error rate of 6% when the number of trees is 200 and the depth is 5; when the regularization parameter of the support vector machine algorithm is around 100 and the kernel function parameter is around 0.01, the average error rate of the test set is in a small range of 15%, so the gradient boosting algorithm with the best comprehensive prediction performance is selected to build the optimal model.

Based on the gradient boosting algorithm, the number of trees in the algorithm parameters is optimized in the range of 160-200, and the depth of the tree is 4-5. The average error rate of the test set of the optimal model is 6.4%, and the accuracy value ^R2 reaches 0.91. The specific implementation effect of the optimal model on the training set and the test set is shown in Figure 6.

Five γ′ phase strengthened cobalt-based high-temperature alloy data are selected from the test set. These alloys cover ternary, quaternary, quinary, hexavalent and heptad system alloys. The test temperature ranges from 25°C to 1000°C. The five different alloy compositions and process parameters are shown in Table 1.

Table 1 Five different alloy compositions and process parameters

The alloy composition, solution temperature, solution time, aging temperature, aging time, and test temperature of the above five alloys were input into the gradient lifting model according to the data in Table 1, and the predicted values of yield strength were obtained respectively. The prediction results of the five alloys are shown in FIG7 , and the error rates between the predicted yield strength values and the experimental yield strength values of the five alloys obtained by the method of the present invention are 5.25%, 6.57%, 4.67%, 5.00%, and 0.96%, respectively, which shows that the prediction method of the present invention has high prediction reliability.

Two γ′ phase strengthened cobalt-based high-temperature alloy data were selected from the test set. The input data were only the alloy composition and the test temperature. They were substituted into the optimal prediction model. The two different alloy compositions and prediction results are shown in Table 2.

Table 2 Two different alloy compositions and prediction results

The alloy composition and test temperature of the above two alloys are input into the gradient lifting model according to the data in Table 2, and the predicted values of yield strength are obtained respectively. It can be seen from Table 2 that the error rate between the predicted yield strength value of alloy 1 and the experimental yield strength value is 54%, and the error rate between the predicted yield strength value of alloy 2 and the experimental yield strength value is 63%. It can be seen that only using the alloy composition and test temperature that are strongly related to yield strength as input data will result in too large an error rate, so that stable and accurate prediction results cannot be obtained. It can be seen from Table 1 above that introducing heat treatment process parameters as output on the basis of alloy composition and test temperature can significantly reduce the error rate of the prediction result, thereby ensuring the effectiveness and reliability of the prediction method of the present invention.

Claims (8)

1. The method for predicting the yield strength of the gamma' -phase reinforced cobalt-based superalloy is characterized by comprising the following steps of:

S1, establishing a data set; collecting related data of yield strength of the gamma' -phase reinforced cobalt-based superalloy at different temperatures to construct a data set, wherein the data comprises alloy components, heat treatment process parameters, test temperatures and corresponding yield strengths; the data set is randomly divided into a training set and a testing set according to the proportion of 6-8:2-4, wherein the data of the training set is used for training a prediction model, and the data of the testing set is used for testing model precision;

S2, establishing a prediction model; for the training set constructed in the step S1, training a prediction model by using different machine learning algorithms, and selecting a gradient lifting algorithm with the best comprehensive prediction performance to construct an optimal model;

s3, predicting yield strength; and (2) inputting the alloy components, the heat treatment process parameters and the test temperature of the gamma' -phase strengthened cobalt-based superalloy to be predicted into the optimal model obtained in the step (S2) to obtain a predicted value of the yield strength.

2. The method for predicting yield strength of gamma prime phase strengthened cobalt-based superalloy according to claim 1, wherein in S1, the specific data collection process is: the method comprises the steps of obtaining a document by searching keywords on the basis of the existing document, extracting yield strength data of different alloys at different temperatures from a graph of the obtained document by using Origin software, and finally obtaining all data including alloy components, heat treatment process parameters, test temperatures and corresponding yield strength data.

3. The method for predicting yield strength of a gamma prime phase strengthened cobalt-based superalloy according to claim 1, wherein the dataset comprises 400 or more data including 8-12 alloy element contents, 4 or more heat treatment process parameters, and a yield strength test temperature.

4. A method of predicting yield strength of a gamma prime phase strengthened cobalt-based superalloy as in claim 3, wherein the number of alloying elements is 10, including Co, ni, al, W, ti, ta, cr, mo, V, B; the heat treatment process parameters comprise a solid solution temperature, a solid solution time, an aging temperature and an aging time.

5. The method for predicting the yield strength of a gamma prime phase strengthened cobalt-based superalloy according to claim 4, wherein the content of the alloying elements is in the range of Co 14-88at.%,Ni 0-54at.%,Al 0-10at.%,W 0-11at.%,Ti 0-12at.%,Ta 0-2.8at.%,Cr 0-18at.%,Mo 0-9at.%,V 0-12at.%,B 0-0.02at.%; at a solution temperature of 1100-1350 ℃, a solution time of 2-100 h, an aging temperature of 760-1100 ℃, an aging time of 4-400 h, a yield strength test temperature of 25-1100 ℃ and a yield strength of 94-888MPa.

6. The method for predicting yield strength of a gamma prime phase strengthened cobalt-based superalloy according to claim 1, wherein in S2, the different machine learning algorithms include a neighbor algorithm, a random forest algorithm, a gradient lifting algorithm, and a support vector machine algorithm.

7. The method for predicting yield strength of a gamma prime phase strengthened cobalt-based superalloy according to claim1, wherein in S2, the number of trees in the gradient lifting algorithm parameters is optimized in the range of 160-200, and the depth of the tree is 4-5.

8. The method for predicting the yield strength of a gamma prime phase strengthened cobalt-based superalloy according to claim 1, wherein in S3, the input parameters of the model are alloy composition, solution temperature, solution time, aging temperature, aging time and test temperature; the output parameter is the yield strength.