Prediction method of nickel-boron-aluminum master alloy based on machine learning
Through the prediction method based on machine learning, the problem of slag composition and content control in the preparation of nickel-bor-aluminum intermediate alloy is solved, the cost and cycle are reduced, and the efficient design of nickel-bor-aluminum intermediate alloy is achieved, ensuring the alloy quality.
Patent Information
- Application Number
- CN202411025538.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, there is a lack of effective methods for slag composition and content control of nickel-boron-aluminum intermediate alloys, which leads to high production costs, long cycles, and lack of direct design tasks for multiple goals, affecting the quality of the alloy.
Using machine learning-based prediction method, the design space for preparing nickel-boron-aluminum intermediate alloy is constructed, sample points are randomly extracted, training sets are constructed, samples that do not meet the requirements of components are eliminated, optimization expectations of candidate samples are calculated, training sets are updated, and model is evaluated. Accuracy, the slag-based component composition is designed using the optimal predictive alloy model to obtain a controllable nickel-boron-aluminum intermediate alloy with the target Al element and B element.
It avoids unnecessary trial and error steps, reduces experimental costs, shortens the experimental cycle, and explores the impact of raw material composition, slag ratio and reaction heat on Al and B elements in the alloy, and develops a meeting of the preset goals. nickel-boron-aluminum intermediate alloy.
Smart Images

Figure CN118553361B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of alloys, and in particular relates to a prediction method for nickel-boron-aluminum master alloys based on machine learning. Background Art
[0002] Adding boron to nickel-based high-temperature alloys for aviation can help purify grain boundaries, improve high-temperature properties and increase high-temperature fracture strength. Adding aluminum is beneficial to regulating the composition and morphology of the precipitated phase and improving the precipitation strengthening effect. However, elemental boron has a low density and is volatile, making it difficult to add directly to the melt. The commonly used solution at present is to add nickel-boron-aluminum intermediate alloys to introduce boron and aluminum elements into the high-temperature alloy. The existing intermediate alloy production processes mainly include self-propagating aluminothermic reduction, vacuum melting and powder metallurgy. Among them, the self-propagating aluminothermic reduction method uses low raw material cost, can be mass-produced, and has simple equipment. It is the most commonly used method for preparing intermediate alloys. The key to preparing intermediate alloys by aluminothermic reduction is raw material composition design, slag system selection and heat control. The combination of slag systems with different components affects the viscosity and density of the slag, the gold-slag separation effect and the alloy composition; the unit mass reaction heat affects the macroscopic quality of the alloy ingot generated by the reaction. Low reaction heat is prone to insufficient reaction and Al in the alloy ingot. 2 O 3 Slag inclusions and high reaction heat will cause the reaction to be too violent, resulting in defects such as pores and shrinkage in the alloy ingot.
[0003] Patents CN102011004B, CN116024460A, etc. disclose methods for preparing various nickel-based master alloys, mainly for master alloys such as nickel-boron, nickel-phosphorus-boron, etc., and propose specific slag composition ratio schemes, and successfully prepare high-quality master alloys by combining aluminothermic reduction experiments, two-step refining, etc. However, for nickel-boron-aluminum master alloys, the slag composition and content control are still to be explored, and there is a lack of new and efficient design methods to break through the bottleneck of high cost and long cycle of preparing master alloys by traditional experimental trial and error methods. The machine learning methods disclosed in patents CN113870957A, CN117637053A, etc. are expected to solve the above problems, but these methods are all aimed at the service performance of alloys and do not involve direct design tasks for multiple objectives. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a prediction method for nickel-boron-aluminum master alloy based on machine learning in view of the deficiencies in the above-mentioned prior art, adopt machine learning to guide the experimental process, avoid unnecessary trial and error steps, reduce the experimental cost, shorten the experimental cycle, and explore the influence of raw material composition, slag system ratio and reaction heat on Al and B elements in the alloy, and then develop a nickel-boron-aluminum master alloy that meets the preset goals.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a prediction method of nickel-boron-aluminum master alloy based on machine learning, characterized in that the method comprises the following steps:
[0006] Step 1: construct a design space for the components required for preparing nickel-boron-aluminum master alloy;
[0007] Step 2: randomly extract sample points in the design space, prepare nickel-boron-aluminum master alloy and construct a training set;
[0008] Step 3: Build a model group, calculate the uncertainty and mean of the samples in the design space based on the model group, and eliminate the samples in the design space that do not meet the component requirements;
[0009] Step 4: Calculate the optimal expectation of candidate samples for the Pareto front in the training set and sort the candidate samples;
[0010] Step 5: Update the training set and evaluate the model accuracy: Experimentally verify the content of Al and B in the top two candidate samples ranked in step 4, and update the training set to evaluate the prediction accuracy of the model. If the determination coefficient of the model is If it is greater than 0.9, the optimal prediction alloy model is obtained and step 6 is executed; otherwise, step 3 is executed;
[0011] Step 6: Use the optimal prediction alloy model to predict the influence of the slag component content on the Al and B element contents of the nickel-boron-aluminum master alloy, design the slag component composition according to the predicted influence rules, and obtain the nickel-boron-aluminum master alloy with controllable target Al and B elements.
[0012] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 1, the components required for preparing the nickel-boron-aluminum master alloy include raw material components, slag system components and reaction heat; the raw material components include boric anhydride and aluminum powder, and the amount of boric anhydride added and the amount of aluminum powder added are both 100% to 130% of the theoretical value calculated according to the law of conservation of mass; the slag system components include CaO and CaF 2 The required CaO value range is 0g~200g, CaF 2 The value range of is 0g~300g; the value range of the reaction heat is 3000J / g~3600J / g.
[0013] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 2, the training set includes the components required for preparing the nickel-boron-aluminum master alloy, and the contents of Al and B elements in the nickel-boron-aluminum master alloy; wherein the contents of Al and B elements in the nickel-boron-aluminum master alloy are obtained by randomly extracting sample points in the design space, conducting aluminothermic reduction experiments, preparing the nickel-boron-aluminum master alloy, and testing the nickel-boron-aluminum master alloy.
[0014] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 3, the input of the selected machine learning model is the amount of boric anhydride added, the amount of aluminum powder added, the CaO content, the CaF 2 The output of the selected machine learning model is the content of Al and B in the nickel-boron-aluminum master alloy.
[0015] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step three, by sampling the training set multiple times, a training group is constructed, and the training group is respectively input into the selected machine learning model for training to construct a model group for predicting the element content of the nickel-boron-aluminum master alloy, and the model group is used to calculate the uncertainty and mean of the content of Al element and B element in the nickel-boron-aluminum master alloy output in the design space sample, and the samples that do not meet the composition requirements in the design space are eliminated.
[0016] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step three, the samples that do not meet the composition requirements are samples with Al element content > 15% and B element content < 10%.
[0017] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 4, the candidate samples are the samples remaining after eliminating the samples that do not meet the composition requirements in the design space.
[0018] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 4, all candidate samples are sorted from large to small according to the optimization expectation value, and the calculation process of the optimization expectation of each group of candidate samples for the Pareto front in the training set is as follows:
[0019] The formula for calculating the optimal probability of candidate samples for the Pareto front in the training set is as follows:
[0020]
[0021] in, is the optimization area of the candidate sample for the current Pareto frontier, and There are two optimization targets, namely the Al element and the B element content. yes and The uncorrelated Gaussian probability density function is, ;
[0022] The calculation formula for the optimal expectation of each group of candidate samples for the Pareto frontier in the training set is as follows:
[0023] ,but ,in, , and is the centroid of all samples on the Pareto front.
[0024] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step five, an aluminum thermal reduction experiment is carried out on the top two candidate samples to prepare nickel-boron-aluminum master alloy samples, and the content of Al and B elements in the candidate samples is tested; the top two candidate samples are included in the training set to update the training set, the selected machine learning model is retrained using the updated training set, and the prediction accuracy of the retrained machine learning model is evaluated.
[0025] The above-mentioned prediction method of nickel-boron-aluminum master alloy based on machine learning is characterized in that: in step 5, in step 5, the determination coefficient The calculation formula is ;in, True value of the sample; is the model prediction value; is the model predicted mean.
[0026] The beneficial effect of the present invention is that machine learning is used to guide the experimental process, unnecessary trial and error steps are avoided, the experimental cost is reduced, the experimental cycle is shortened, the influence of raw material composition, slag system ratio and reaction heat on Al and B elements in the alloy is explored, and then a nickel-boron-aluminum intermediate alloy that meets the preset goals is developed.
[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 The figure is a flowchart of the method of the present invention.
[0029] Figure 2 Schematic diagram of the effect of the amount of boric anhydride added on the contents of Al and B in the alloy.
[0030] Figure 3 Schematic diagram of the effect of the amount of aluminum powder added on the content of Al and B elements in the alloy.
[0031] Figure 4 Schematic diagram of the effect of the amount of CaO added on the contents of Al and B in the alloy according to the present invention.
[0032] Figure 5 The CaF 2 Schematic diagram of the effect of addition amount on the content of Al and B elements in the alloy.
[0033] Figure 6Schematic diagram of the effect of reaction heat on the content of Al and B elements in the alloy according to the present invention.
[0034] Figure 7 This is a microstructure scan of the nickel-boron-aluminum master alloy prepared by the present invention.
[0035] Figure 8 This is the distribution diagram of Al element in the nickel-boron-aluminum master alloy prepared by the present invention.
[0036] Fig. 9 This is the distribution diagram of the B element in the nickel-boron-aluminum master alloy prepared by the present invention. DETAILED DESCRIPTION
[0037] like Figures 1 to 9 The prediction method of nickel-boron-aluminum master alloy based on machine learning is shown, and the method comprises the following steps:
[0038] Step 1: construct a design space for the components required for preparing nickel-boron-aluminum master alloy;
[0039] Step 2: randomly extract sample points in the design space, prepare nickel-boron-aluminum master alloy and construct a training set;
[0040] Step 3: Build a model group, calculate the uncertainty and mean of the samples in the design space based on the model group, and eliminate the samples in the design space that do not meet the component requirements;
[0041] Step 4: Calculate the optimal expectation of candidate samples for the Pareto front in the training set and sort the candidate samples;
[0042] Step 5: Update the training set and evaluate the model accuracy: Experimentally verify the content of Al and B in the top two candidate samples ranked in step 4, and update the training set to evaluate the prediction accuracy of the model. If the determination coefficient of the model is If it is greater than 0.9, the optimal prediction alloy model is obtained and step 6 is executed; otherwise, step 3 is executed;
[0043] Step 6: Use the optimal prediction alloy model to predict the influence of the slag component content on the Al and B element contents of the nickel-boron-aluminum master alloy, design the slag component composition according to the predicted influence rules, and obtain the nickel-boron-aluminum master alloy with controllable target Al and B elements.
[0044] The present invention adopts machine learning to guide the experimental process, avoids unnecessary trial and error steps, reduces experimental costs, shortens the experimental cycle, explores the influence of raw material composition, slag system ratio and reaction heat on Al and B elements in the alloy, and then develops a nickel-boron-aluminum master alloy that meets the preset goals.
[0045] In step 1, based on the consumption of 1000 g of boric anhydride, the nickelous oxide required in the reactants is calculated according to the law of conservation of mass, and the constructed design space is shown in Table 1 below:
[0046] Table 1 Design space constructed in this example
[0047]
[0048] As can be seen from Table 1, the design space contains a total of 191,394 sample points.
[0049] In step 2, the constructed training set is shown in Table 2 below:
[0050] Table 2 Training set constructed in this example
[0051]
[0052] In step 3, the model hyperparameter regularization coefficient C, penalty coefficient γ, and model prediction accuracy are combined with cross-validation to obtain a machine learning model for predicting the element content of nickel-boron-aluminum master alloy. The selected machine learning model is a support vector regression model, and the leave-one-out cross-validation method is used to determine the optimal model hyperparameters and the correlation coefficient between the predicted value and the measured value on the training set. As shown in Table 3;
[0053] Table 3 Hyperparameters and model accuracy of the support vector regression model constructed in this embodiment
[0054]
[0055] All samples in the design space are predicted through the predictive machine learning model until all samples that do not meet the ingredient requirements are eliminated.
[0056] In step 5 of this embodiment, after four iterations, the optimal alloy raw materials, slag system ratios, and alloy Al and B element contents searched by the model are provided in Table 4;
[0057] Table 4 Optimal search and experimental results of the machine learning model in this embodiment
[0058]
[0059] In step 6, the model obtained in step 5 is used to predict all candidate samples in the design space, and the boric anhydride, aluminum powder, CaO, CaF 2 The influence of reaction heat on the Al and B element contents of nickel-boron-aluminum master alloy is shown in the figure. Figure 2-Figure 6 As shown. Using the optimal alloy raw material and slag system ratio scheme searched by the optimal prediction alloy model, a self-propagating aluminum thermal reduction experiment was carried out to obtain a nickel-boron-aluminum master alloy. The alloy microstructure scanning diagram is shown in Figure 7As shown, the distribution diagram of Al element in the alloy is as follows Figure 8 As shown, the distribution diagram of B element in the alloy is as follows Fig. 9 shown.
[0060] In this embodiment, in step 1, the components required for preparing the nickel-boron-aluminum master alloy include raw material components, slag system components and reaction heat; the raw material components include boric anhydride and aluminum powder, and the amount of boric anhydride added and the amount of aluminum powder added are both 100% to 130% of the theoretical value calculated according to the law of conservation of mass; the slag system components include CaO and CaF 2 The required CaO value range is 0g~200g, CaF 2 The value range of is 0g~300g; the value range of the reaction heat is 3000J / g~3600J / g.
[0061] In this embodiment, based on the consumption of 1000 g of boric anhydride, the theoretical values of other required components are calculated according to the law of conservation of mass, and the amount of boric anhydride and the amount of aluminum powder added are added at 100% to 130% of the calculated theoretical values.
[0062] In this embodiment, in step 2, the training set includes the components required for preparing the nickel-boron-aluminum master alloy, and the contents of Al and B in the nickel-boron-aluminum master alloy; wherein the contents of Al and B in the nickel-boron-aluminum master alloy are obtained by randomly extracting sample points in the design space, conducting aluminothermic reduction experiments, preparing the nickel-boron-aluminum master alloy, and testing the nickel-boron-aluminum master alloy.
[0063] In this embodiment, in step 3, the inputs of the selected machine learning model are the amount of boric anhydride added, the amount of aluminum powder added, the CaO content, the CaF 2 The output of the selected machine learning model is the content of Al and B in the nickel-boron-aluminum master alloy.
[0064] In this embodiment, in step three, a training group is constructed by sampling the training set multiple times, and the training groups are respectively input into the selected machine learning models for training to construct a model group for predicting the element content of the nickel-boron-aluminum intermediate alloy. The model group is used to calculate the uncertainty and mean of the content of Al and B elements in the nickel-boron-aluminum intermediate alloy output in the design space sample, and the samples that do not meet the composition requirements in the design space are eliminated.
[0065] In this embodiment, in step 3, the samples that do not meet the composition requirements are samples with an Al element content of >15% and a B element content of <10%.
[0066] In this embodiment, in step 4, the candidate samples are the samples remaining after removing the samples that do not meet the component requirements in the design space.
[0067] In this embodiment, in step 4, all candidate samples are sorted from large to small according to the optimization expectation value, and the calculation process of the optimization expectation of each group of candidate samples on the Pareto front in the training set is as follows:
[0068] The formula for calculating the optimal probability of candidate samples for the Pareto front in the training set is as follows:
[0069]
[0070] in, is the optimization area of the candidate sample for the current Pareto frontier, and There are two optimization targets, namely the Al element and the B element content. yes and The uncorrelated Gaussian probability density function is, ;
[0071] The calculation formula for the optimal expectation of each group of candidate samples for the Pareto frontier in the training set is as follows:
[0072] ,but ,in, , and is the centroid of all samples on the Pareto front.
[0073] In this embodiment, in step five, an aluminum thermal reduction experiment is carried out on the top two candidate samples to prepare nickel-boron-aluminum intermediate alloy samples, and the contents of Al and B elements in the candidate samples are tested; the top two candidate samples are included in the training set to update the training set, the selected machine learning model is retrained using the updated training set, and the prediction accuracy of the retrained machine learning model is evaluated.
[0074] In this embodiment, in step 5, the determination coefficient The calculation formula is ;in, True value of the sample; is the model prediction value; is the model predicted mean.
[0075] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. A prediction method for nickel-boron-aluminum master alloy based on machine learning, It is characterized in that The method comprises the following steps: Step 1: construct a design space for the components required for preparing nickel-boron-aluminum master alloy; Step 2: randomly extract sample points in the design space, prepare nickel-boron-aluminum master alloy and construct a training set; Step 3: Build a model group, calculate the uncertainty and mean of the samples in the design space based on the model group, and eliminate the samples in the design space that do not meet the component requirements; Step 4: Calculate the optimal expectation of candidate samples for the Pareto front in the training set and sort the candidate samples; Step 5: Update the training set and evaluate the model accuracy: Experimentally verify the content of Al and B in the top two candidate samples ranked in step 4, and update the training set to evaluate the prediction accuracy of the model. If the determination coefficient of the model is If it is greater than 0.9, the optimal prediction alloy model is obtained and step 6 is executed; otherwise, step 3 is executed; Step 6: Use the optimal prediction alloy model to predict the influence of the slag system component content on the Al and B element contents of the nickel-boron-aluminum master alloy, design the slag system component composition according to the predicted influence law, and obtain the nickel-boron-aluminum master alloy with controllable target Al and B elements; In step 1, the components required for preparing the nickel-boron-aluminum master alloy include raw material components, slag system components and reaction heat; the raw material components include boric anhydride and aluminum powder, and the amount of boric anhydride added and the amount of aluminum powder added are both 100% to 130% of the theoretical value calculated according to the law of conservation of mass; the slag system components include CaO and CaF 2 The required CaO value range is 0g~200g, CaF 2 The value range of is 0g~300g; the value range of the reaction heat is 3000J / g~3600J / g; In step 2, the training set includes the components required for preparing the nickel-boron-aluminum master alloy, and the contents of the Al element and the B element in the nickel-boron-aluminum master alloy; wherein the contents of the Al element and the B element in the nickel-boron-aluminum master alloy are obtained by randomly extracting sample points in the design space, conducting an aluminothermic reduction experiment, preparing the nickel-boron-aluminum master alloy, and testing the nickel-boron-aluminum master alloy; In step 3, the inputs of the selected machine learning model are the amount of boric anhydride added, the amount of aluminum powder added, the CaO content, and the CaF 2 content, and reaction heat. The output of the selected machine learning model is the content of Al and B elements in the nickel-boron-aluminum master alloy; In step three, training groups are constructed by sampling the training set multiple times, and the training groups are input into the selected machine learning models for training. A model group for predicting the element content of the nickel-boron-aluminum master alloy is constructed, and the model group is used to calculate the uncertainty and mean of the content of Al and B elements in the nickel-boron-aluminum master alloy output in the design space samples, and the samples that do not meet the composition requirements in the design space are eliminated.
2. The method for predicting nickel-boron-aluminum master alloy based on machine learning according to claim 1, Features: In step 3, the samples that do not meet the composition requirements are those with an Al content > 15% and a B content < 10%.
3. The prediction method of nickel-boron-aluminum master alloy based on machine learning according to claim 1, Features: In step 4, the candidate samples are the samples remaining in the design space after removing the samples that do not meet the component requirements.
4. The method for predicting nickel-boron-aluminum master alloy based on machine learning according to claim 1, Features: In step 4, all candidate samples are sorted from large to small according to the optimization expectation value. The calculation process of the optimization expectation of each group of candidate samples on the Pareto front in the training set is as follows: The formula for calculating the optimal probability of candidate samples for the Pareto front in the training set is as follows: in, is the optimization area of the candidate sample for the current Pareto frontier, and There are two optimization targets, namely the Al element and the B element content. yes and The uncorrelated Gaussian probability density function is, ; The calculation formula for the optimal expectation of each group of candidate samples for the Pareto frontier in the training set is as follows: but ,in, , and are the centroids of all samples on the Pareto front.
5. The method for predicting nickel-boron-aluminum master alloy based on machine learning according to claim 1, Features: In step five, aluminothermic reduction experiments are carried out on the top two candidate samples to prepare nickel-boron-aluminum intermediate alloy samples, and the contents of Al and B elements in the candidate samples are tested; the top two candidate samples are included in the training set to update the training set, and the selected machine learning model is retrained using the updated training set, and the prediction accuracy of the retrained machine learning model is evaluated.
6. The method for predicting nickel-boron-aluminum master alloy based on machine learning according to claim 1, Features: In step 5, the coefficient of determination The calculation formula is ;in, The true value of the sample; is the model prediction value; is the model predicted mean.
