A method for designing a high-temperature alloy composition and process

Abstract

The application discloses a high-temperature alloy component and process design method, comprising the following steps: collecting data of a certain brand high-temperature alloy manual, experiment and literature, and constructing a database; using a machine learning algorithm to construct a corresponding performance prediction model, using a Bayesian optimization technology, taking the required optimized alloy component range as a search range, and taking a comprehensive performance index as an optimization target to realize alloy component optimization to obtain excellent comprehensive performance; through the machine learning algorithm, process optimization of small sample data comprehensive performance is realized; and the optimized alloy component and process casting alloy are used to verify the optimized comprehensive performance. Through collecting data as a training set to train the model, the application analyzes the influence law of alloy components and processing technology on the structure performance, designs the optimized alloy components and process, reduces the cost of manpower and material resources, and reduces the research and development cycle of cast nickel-based high-temperature alloy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of high-temperature alloy preparation, and particularly relates to a high-temperature alloy composition and process design method. BACKGROUND

[0002] Nickel-based high-temperature alloy occupies a special important position in the whole high-temperature alloy field, and is widely used to manufacture the hottest end parts of aviation jet engines and various industrial gas turbines, and has excellent comprehensive mechanical properties, including excellent yield strength, tensile strength, creep rupture performance, fatigue resistance and corrosion resistance.

[0003] With the development of computers and the improvement of computing power, computational materials science has rapidly emerged, promoting the transformation of material research and development from the "experience + trial and error" mode to the computationally driven research and development mode. Machine learning algorithms have played a great role in the research and development of nickel-based high-temperature alloys. Bayesian optimization technology, as a very effective global optimization algorithm, has been widely used in design problems in recent years. By designing appropriate probability proxy models and acquisition functions, the Bayesian optimization framework can obtain ideal solutions with only a few evaluations of the objective function. However, high-temperature alloys for high-temperature use have high high-temperature strength, but often have poor room temperature plasticity and toughness, and the processing performance of the alloy is poor. The addition of multiple alloying elements in the high-temperature alloy often leads to a contradictory relationship between high-temperature strength and room temperature plasticity, and existing research has focused on optimizing single properties of the alloy such as endurance life, making it difficult to meet the design requirements of high-temperature alloys with excellent comprehensive performance. SUMMARY

[0004] The main purpose of the present application is to provide a high-temperature alloy composition and process design method based on machine learning, which trains the model by collecting data as a training set, analyzes the influence of alloy composition and processing technology on the structure and performance, designs the composition and process of the optimized alloy, reduces the cost of manpower and material resources, and reduces the research and development cycle of cast nickel-based high-temperature alloys.

[0005] To this end, the high-temperature alloy composition and process design method provided by the present application comprises the following steps:

[0006] Step one, collect the data of a certain brand of high-temperature alloy manual, experiment, literature, and construct a material composition-process-performance database, wherein the performance includes tensile strength, elongation, reduction of area, endurance time, yield strength and endurance life;

[0007] Step two, based on the established database, use machine learning algorithms to establish corresponding prediction models for each performance;

[0008] Step three, use a multi-objective Bayesian optimization algorithm for multi-objective comprehensive optimization

[0009] First, the alloy composition is optimized, Y=tensile property elongation + yield strength + lambda * endurance life as the comprehensive optimization target, within the selected grade of nickel-based superalloy composition range, the maximum value is searched by using the Bayesian optimization algorithm, and then the optimal composition point is determined;

[0010] Wherein, lambda is 100 to keep three variables in the same order of magnitude;

[0011] Step four, using small sample data set for process optimization, first using various machine learning algorithms for modeling, and comparing the fitting effect to determine the optimal fitting model;

[0012] Step five, using the selected optimal fitting model to model each performance, Z = lambda1 * tensile strength + lambda2 * elongation + lambda3 * reduction of area + lambda4 * endurance time as the comprehensive optimization target, the shell preheating temperature, pouring temperature, air cooling condition as the input variable, the tensile strength, elongation, reduction of area, endurance time as the output result, the maximum value is searched by using the Bayesian optimization algorithm, and then the optimal process parameters are determined; wherein,

[0013] In order to maintain the consistency of dimension, lambda2, lambda3, lambda4 are respectively 20000, 10000, 25, in addition, because the tensile strength fitting effect is not good, lambda1 is 0.1, to weaken its influence on the optimization target;

[0014] Step six, using the optimized alloy composition, process casting alloy, the optimized comprehensive performance is verified.

[0015] The present application has the following beneficial effects:

[0016] 1. The method for realizing the comprehensive performance improvement of high-temperature alloy material by using the Bayesian optimization technology realizes the efficient composition design of high-temperature alloy in a complex composition space.

[0017] 2. The small sample data alloy process optimization method is developed, the deviation risk caused by a single prediction model is reduced through the performance screening of multiple machine learning models, and the alloy process is successfully optimized.

[0018] 3. The alloy elongation, yield strength, endurance life and tensile strength are better than the level before optimization. DETAILED DESCRIPTION

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 is a flow chart of an embodiment of the present application;

[0021] Figure 2 is an optimal composition point found by the Bayesian optimization technique involved in an embodiment of the present application;

[0022] Figure 3 is a comparison chart of modeling and fitting effects of various machine learning algorithms involved in an embodiment of the present application;

[0023] Figure 4 is a feature importance chart of various performances involved in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] The specific process of the present application will be described in detail below with the cast nickel-based superalloy with the trade name K403 as an embodiment:

[0026] The embodiment alloy is the cast nickel-based superalloy K403, and the alloy composition range is (mass fraction / %) C: 0.11-0.18, Cr: 10.00-12.00, Co: 4.50-6.00, W: 4.80-5.50, Mo: 3.80-4.50, Al: 5.30-5.90, Ti: 2.30-2.90, Fe: ≤2.00, Ce: 0.010, B: 0.012-0.022, Zr: 0.030-0.080, Mn ≤0.50, Si ≤0.50, Ni: balance

[0027] The specific implementation steps are as follows:

[0028] Step one, collect the related properties, compositions, and process data of the cast nickel-based superalloy in the superalloy manual, experiments, and literature, and build a nickel-based superalloy material database after data standardization, missing value supplement, and other data processing. The database contains various alloy element compositions, process parameters, tensile properties, and other mechanical properties.

[0029] Step two, use machine learning algorithms to establish corresponding prediction models for each performance, such as tensile strength, elongation, reduction of area, endurance time, yield strength, and endurance life. Select the model with better fitting effect for each performance, and the multi-output neural network model has better fitting effect on elongation, yield strength, and endurance life.

[0030] Step three, multi-objective comprehensive optimization by Bayesian optimization technology, first optimize alloy composition, Y = tensile properties elongation + yield strength + lambda * endurance life as the comprehensive optimization target, where lambda takes 100 to keep three variables in the same order of magnitude, within the composition range of K403 cast nickel-based superalloy, search for the maximum value by using Bayesian optimization technology, the possible results are shown in Figure 2 The following possible composition points are searched out:

[0031]

[0032]

[0033] Step four, the data collected in the database is mainly the relationship between composition and performance, and the amount of process performance related data is small. Considering the high cost of one test, process optimization is carried out using small sample data set to find the process conditions under which tensile strength, elongation, reduction of area and endurance time are comprehensively optimal. Since the corresponding data in the collected data is less, various machine learning algorithms are first tried to model and fit the comparison effect to reduce the deviation caused by a single prediction model. The fitting effect is shown in Figure 3 It can be seen that GradientBoosting has better fitting effect for each performance. GradientBoosting is selected as the regressor for fitting.

[0034] Step five, after modeling each performance, the process parameters are optimized by using Bayesian optimization technology. Y = lambda1 * tensile strength + lambda2 * elongation + lambda3 * reduction of area + lambda4 * endurance time as the comprehensive optimization target. Similarly, in order to maintain the consistency of the dimension, lambda2, lambda3, lambda4 are respectively taken as 20000, 10000 and 25. In addition, because the fitting effect of tensile strength is not good, lambda1 is taken as 0.1 to weaken its influence on the optimization target. Input: shell preheating temperature, pouring temperature, air cooling condition; output: tensile strength, elongation, reduction of area, endurance time.

[0035] The inventors found that the change of alloy composition has a greater influence on the elongation and yield strength in tensile properties, and has a greater influence on the endurance life in endurance properties. Therefore, elongation, yield strength and endurance life are selected as the comprehensive optimization target in the composition optimization process. The change of process such as heat treatment temperature has a greater influence on the tensile strength, elongation and reduction of area in tensile properties, and has a greater influence on the endurance time in endurance properties. Therefore, the optimization target of process optimization is selected as tensile strength, elongation, reduction of area and endurance time. The characteristic importance of each performance can be as follows: Figure 4The comprehensive optimization target is selected to make the comprehensive optimization target change greatly when the alloy composition or the process changes, facilitate the screening process, and select the tensile property and the endurance property index for both the composition optimization and the process optimization, which is beneficial to the improvement of the comprehensive performance of the alloy.

[0036] Step six, prediction result: among the above corresponding composition points, the optimal value is obtained under the following process parameters: shell preheating temperature: 1037 DEG C, pouring temperature: 1432 DEG C, and air cooling condition, the tensile strength: 1040 MPa, the elongation: 7.2%, the reduction of area: 11.6%, and the endurance time: 43.89 h.

[0037] The values of λ2, λ3 and λ4 are based on the consideration that the four properties are inconsistent in the order of magnitude, for example, the endurance time is larger in the order of magnitude, so even a small change in the property will greatly affect the comprehensive optimization target relative to other properties, and therefore it is necessary to keep the order of magnitude of each property data consistent. Meanwhile, the importance of each property in the comprehensive optimization can be adjusted in the process of value taking, and the weight of the more important property can be made larger to have a larger proportion in the optimization process, and the weight of the property with lower importance or lower prediction accuracy such as the tensile strength can be weakened to obtain better optimization effect.

[0038] Step seven, according to the prediction result, the alloy melting experiment is carried out, the experimental result is similar to the prediction result, and is better than the level before optimization. If the experimental result does not meet the preset error requirement, the parameters of the established performance prediction model need to be adjusted, and the experimental verification data is further supplemented into the database, and the optimization and experimental verification steps are repeated until the experimental result and the preset result meet the requirements within a certain error range.

[0039] The above examples are only examples for clearly illustrating the present application, and are not limited to the embodiments. For those skilled in the art, other different forms of changes or variations can be made on the basis of the above description. Here, all the examples are not required to be exhausted, and the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A method of high temperature alloy composition and process design, characterized by, Comprising the following steps: Step one, collect the data of a certain brand of high-temperature alloy manual, experiment, literature, and after processing to build a material composition-process-performance database, including tensile strength, elongation, reduction of area, endurance time, yield strength and endurance life; Step two, based on the established database, use machine learning algorithm to establish the corresponding prediction model for each performance; Step three, use multi-objective Bayesian optimization algorithm for multi-objective comprehensive optimization First, optimize the alloy composition, take Y=tensile performance elongation+yield strength+λ*endurance life as the comprehensive optimization target, within the composition range of the selected brand of nickel-based high-temperature alloy, use Bayesian optimization algorithm to search for the maximum value, and then determine the optimal composition point; Where, λ is 100 to keep the three variables in the same order of magnitude; Step four, process optimization using small sample data set, first use various machine learning algorithms to model, and compare the fitting effect to determine the optimal fitting model; Step five, use the selected optimal fitting model to model each performance, take Z=λ1*tensile strength+λ2*elongation+λ3*reduction of area+λ4*endurance time as the comprehensive optimization target, take the shell preheating temperature, pouring temperature, air cooling conditions as the input variables, take the tensile strength, elongation, reduction of area, endurance time as the output results, use Bayesian optimization algorithm to search for the maximum value, and then determine the optimal process parameters; Where, In order to maintain the consistency of the dimension, λ2, λ3, λ4 take 20000, 10000, 25 respectively, in addition, because the tensile strength fitting effect is not good, λ1 take 0.1 to weaken its influence on the optimization target; Step six, use the optimized alloy composition and process to cast the alloy to verify the optimized comprehensive performance.

2. The method of claim 1, wherein: GradientBoosting is selected as the regressor for fitting in step four.

3. The method of claim 1, wherein: Various machine learning algorithms in step 4 include AdaBoost, Extratrees, GradientBoosting, SVR, XGB, Gaussian process regression, and neural network.

4. The method of claim 1, wherein: Data processing mainly includes data cleaning and data standardization.

5. The method of claim 1, wherein: The prediction model constructed by machine learning algorithm in step two is a multi-input multi-output neural network model.

Images