Nickel-based single crystal superalloy heat treatment process optimization method based on high-throughput experiment and machine learning

Abstract

The invention relates to a nickel-based single-crystal high-temperature alloy heat treatment process optimization method based on high-throughput experiment and machine learning, which comprises the following steps: constructing a multi-section solid solution temperature-time gradient window by taking as-cast nickel-based single-crystal high-temperature alloy as a starting point; batch vacuum-packaged samples in the window are synchronously fed into the furnace, are heated section by section, are immediately quenched after heat preservation, and are quickly frozen to corresponding tissues; measuring the Re element segregation ratio, the shrinkage porosity and porosity ratio and the eutectic content ratio of each section of the sample, and establishing a temperature-time-tissue data set; and inputting the optimal heat treatment parameters into a machine learning model for training, carrying out cross validation and fusion, reversely predicting the optimal heat treatment parameters capable of simultaneously meeting target tissue indexes, and finally verifying an optimization effect through a high-temperature mechanical test. Compared with the prior art, the method has the advantages that the element segregation improvement and microstructure morphology of the solid solution product are quickly learned by utilizing machine learning, so that the optimal solid solution process or heat treatment process of the single-crystal high-temperature alloy can be quickly predicted.

Description

Technical Field

[0001] This invention relates to the field of nickel-based single-crystal superalloy preparation, and in particular to an optimization method for the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning. Background Technology

[0002] Nickel-based single-crystal superalloys are widely used in high-pressure turbine blades for aero-engines due to their excellent high-temperature mechanical properties. Over the past few decades, engineers have significantly improved the mechanical properties of nickel-based single-crystal superalloys through compositional design and optimized heat treatment processes, effectively supporting the rapid development of modern aero-engines. Currently, nickel-based single-crystal superalloys have reached their fifth generation (first generation without Re; second generation with 2-3% Re; third generation with 4-5% Re; fourth and fifth generations incorporate 2-3% and 5-6% ruthenium, respectively). However, due to limitations in technological maturity and cost, second-generation single-crystal superalloys remain dominant in industrial applications, with the application of third-generation superalloys being gradually promoted. Therefore, with the alloy composition essentially fixed, optimizing heat treatment processes to further improve and stabilize the high-temperature mechanical properties of nickel-based single-crystal superalloys is the main focus of current research.

[0003] The heat treatment process for nickel-based single-crystal superalloys includes solution heat treatment and aging heat treatment. Because nickel-based single-crystal superalloys contain a significant amount of refractory elements, castings prepared using directional solidification processes exhibit significant elemental segregation between dendrites and dendrite trunks. The presence of abundant γ / γ' eutectic structures between dendrites makes these localized areas weak points in the alloy, and they are more prone to precipitating TCP phases during service, leading to premature material failure. Therefore, solution heat treatment needs to fully dissolve the carbides, γ / γ' eutectic, and coarse γ' phase microstructure in the single-crystal castings, and improve elemental segregation without initial melting, providing a foundation for the subsequent aging treatment to precipitate uniform strengthening phases.

[0004] Northwestern Polytechnical University (Patent CN 115613141 A) proposed a continuous heating heat treatment process for second-generation nickel-based single-crystal superalloys. By employing multi-stage stepped continuous heating in the low-temperature range and continuous heating in the high-temperature range, the traditional industrial heat treatment regime was optimized, and the peak heat treatment temperature was increased. This effectively reduced the segregation of refractory elements such as Re and W, improved the heat-treated microstructure, and enhanced mechanical properties. Zhejiang University (Patent CN 115747687 A), based on continuous stepped heating, effectively improved the creep rupture life of a second-generation nickel-based single-crystal superalloy by applying a high-temperature solution treatment approximately 15°C below the γ-phase solidus line. However, both of these methods are improvements on existing industrial heat treatment processes and may not be effective for other second- and third-generation nickel-based single-crystal superalloys, especially higher-generation alloys. To address the aforementioned drawbacks, Northwestern Polytechnical University (Patent CN 117252030 A) proposed a method and system for designing solution heat treatment regimes based on phase-field simulation software, and Tsinghua University (Patent CN 111640472A) proposed a solution heat treatment method for nickel-based single-crystal superalloys based on numerical simulation. However, due to limitations in the accuracy and feasibility of theoretical models, these methods still face many problems in practical operation. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning. By accurately measuring the thermophysical properties of the cast single-crystal superalloy to determine the range of process parameters, high-throughput experimental design is used to obtain samples under different heat treatment process parameters, and machine learning is used to quickly learn the improvement of elemental segregation and microstructure morphology of the solution product, thereby quickly predicting the optimal solution treatment process or heat treatment process for the single-crystal superalloy.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] This invention provides a method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning, comprising the following steps:

[0008] S1. Samples were taken from the as-cast nickel-based single-crystal superalloy, and the initial dissolution temperature of the γ / γ′ eutectic phase, γ′ phase, and γ phase, as well as the complete dissolution temperature of the γ phase, were determined by differential thermal analysis or thermogravimetric analysis.

[0009] S2. Based on the temperature range obtained in S1, set the controllable range of the solution heat treatment temperature steps and their corresponding times to form a multi-segment gradient process window.

[0010] S3. Prepare samples in batches within the gradient process window and simultaneously perform high-throughput heat treatment. Quench and sample immediately after each temperature step is completed.

[0011] S4. Measure the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain the microstructure feature dataset corresponding to the heat treatment process parameters in S3.

[0012] S5. Use the tissue feature dataset obtained in S4 as training data to train the machine learning model. Use the trained machine learning model and the target tissue feature data as input data to obtain optimized heat treatment process parameters.

[0013] Furthermore, in S1, the specific process of taking samples from the as-cast nickel-based single-crystal superalloy and determining the initial dissolution temperature of the γ / γ′ eutectic phase, the γ′ phase, and the γ phase, as well as the complete dissolution temperature of the γ phase, through differential thermal analysis or thermogravimetric analysis includes:

[0014] Taking cast nickel-based single-crystal superalloys as the object, first cut out a block sample that can match the crucible, and after surface cleaning, place it in a differential scanning calorimeter or a synchronous thermal analyzer.

[0015] Under an inert atmosphere, the temperature is continuously heated at a constant heating rate. By recording the heat flow or mass change curves, the onset points of the endothermic peaks of the γ / γ′ eutectic phase, γ′ phase, and γ phase are captured sequentially as the initial dissolution temperatures, and the temperature at which the endothermic peak of the γ phase returns to the baseline is taken as the complete dissolution temperature.

[0016] Furthermore, in S2, based on the temperature range obtained in S1, the specific process of setting the solution heat treatment temperature steps and their corresponding controllable time ranges to form a multi-segment gradient process window includes:

[0017] Using the initial dissolution temperature of the γ′ phase measured by S1 as the lower limit and the complete dissolution temperature of the γ phase as the upper limit, several continuous temperature steps are divided, and the holding time interval is set for each step according to the principle of gradient decrease.

[0018] By combining temperature steps and holding time into a multi-segment gradient process window, the temperature difference and time difference between adjacent segments are kept at a fixed preset step size, thereby allowing refractory elements to diffuse fully while avoiding initial melting.

[0019] Furthermore, in S3, the specific process of batch preparing samples and simultaneously performing high-throughput heat treatment within the gradient process window, with immediate quenching and sampling after each temperature step, includes:

[0020] Based on the gradient process window generated by S2, the same batch of cast nickel-based single crystal superalloys were cut into block samples of equal size and numbered sequentially.

[0021] Each sample was individually placed into a quartz glass tube and then vacuum-sealed to ensure that the heat treatment atmosphere was consistent with industrial vacuum conditions.

[0022] All sealed tube samples were simultaneously placed into a gradient temperature controlled muffle furnace, and the furnace temperature was strictly increased step by step according to the temperature steps set in the window.

[0023] After each solid solution temperature is reached and the holding time is completed, the corresponding numbered quartz tube is immediately removed and quenched in flowing argon or ice water to freeze the tissue at that temperature-time point.

[0024] Start the next step of heating, repeat heating, holding, sampling, quenching until the end of the window, thereby obtaining the process parameters for each segment within the S2 window.

[0025] Furthermore, in S4, the specific process of determining the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain the microstructure characteristic dataset corresponding to the heat treatment process parameters in S3 includes:

[0026] Each sample after S3 quenching was sequentially mounted, ground, polished, and electrolytically etched to expose the dendritic structure.

[0027] The Re element segregation ratio was quantitatively determined along the dendrite trunk-interdendritic direction using scanning electron microscopy-energy dispersive spectroscopy.

[0028] The shrinkage porosity ratio was statistically analyzed by X-ray tomography or optical microscopy, and the eutectic content ratio was calculated by image grayscale thresholding.

[0029] By binding the Re element segregation ratio, shrinkage and porosity ratio, and eutectic content ratio with the temperature-time parameters corresponding to the same numbered sample, a microstructure feature dataset covering the entire gradient window is constructed.

[0030] Furthermore, in S5, the process of using the tissue feature dataset obtained in S4 as training data to train a machine learning model, and then using the trained machine learning model with the target tissue features as input data to obtain optimized heat treatment process parameters includes:

[0031] Based on the microstructure feature dataset obtained from S4, the heat treatment process parameters contained therein are used as input features, and the corresponding Re element segregation ratio, shrinkage porosity ratio and eutectic content ratio are used as output labels to construct a supervised learning training set.

[0032] The supervised learning training set is used to train multiple different types of machine learning models in parallel, and the prediction accuracy of each model is evaluated through cross-validation during the training process.

[0033] The machine learning model with the smallest prediction error is selected as the preferred model. When the prediction errors of multiple models are similar, a multi-model prediction result fusion mechanism is established.

[0034] The desired target Re element segregation ratio, target shrinkage and porosity ratio, and target eutectic content ratio are input into the preferred model or multi-model fusion system.

[0035] The optimal combination of heat treatment process parameters that can simultaneously achieve the target Re element segregation ratio, target shrinkage porosity ratio, and target eutectic content ratio is output through the reverse prediction function of the preferred model or multi-model fusion system.

[0036] Furthermore, in S5, the machine learning model includes at least two of the following: Ridge Regression Model, XGBoost Model, and LightGBM Model;

[0037] The machine learning model is configured to establish a nonlinear mapping relationship between heat treatment process parameters and Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio.

[0038] Furthermore, in S5, 90% of the total dataset obtained in S4 is used as the training set, and the remaining 10% is used as the test set. The training set is used to train the parameters of the machine learning model, and the test set is used to verify the prediction accuracy of the trained model.

[0039] Furthermore, evaluating the prediction accuracy of each model through cross-validation specifically includes:

[0040] The prediction errors of each model for Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio were calculated using the ten-fold cross-validation method. The prediction errors were quantified using at least one of the mean absolute error and root mean square error.

[0041] Furthermore, in S5, after obtaining the optimized heat treatment process parameters, a process verification step is also included:

[0042] The optimized heat treatment process parameters were used to heat treat the new nickel-based single-crystal superalloy sample. The high-temperature tensile properties and high-temperature creep properties of the heat-treated sample were then tested to verify the effectiveness of the optimized heat treatment process parameters.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1) This invention uses high-throughput experiments to quickly obtain a large number of machine learning samples, and uses three key indicators, namely Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio, as machine learning targets. Machine learning is used to quickly predict the microstructure and state of single-crystal high-temperature alloys after heat treatment. This invention can quickly and effectively develop heat treatment processes for single-crystal high-temperature alloys.

[0045] 2) This method not only avoids the inefficiency of traditional experimental methods, but also significantly improves the efficiency of developing heat treatment processes for single-crystal high-temperature alloys. Furthermore, this method expands the range of heat treatment process parameters for single-crystal high-temperature alloys, avoids blind spots in the heat treatment process, significantly improves the efficiency of the heat treatment process, shortens the heat treatment time, and enhances the high-temperature mechanical properties of single-crystal high-temperature alloys. Attached Figure Description

[0046] Figure 1 Range of high-throughput heat treatment process parameters for single-crystal superalloys;

[0047] Figure 2 Flowchart of machine learning prediction for heat treatment process of single crystal superalloys based on high-throughput experiments;

[0048] Figure 3 Comparison of the high-temperature yield strength of the third-generation single-crystal superalloy obtained by this method with the performance of the original process;

[0049] Figure 4 Comparison of the high-temperature tensile strength of the third-generation single-crystal superalloy obtained by this method with the performance of the original process;

[0050] Figure 5 Comparison of the high-temperature creep performance of the third-generation single-crystal superalloy obtained by this method with that of the original process. Detailed Implementation

[0051] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, circuit structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0052] Example 1

[0053] This embodiment provides a method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning, including the following steps:

[0054] S1. Samples were taken from the as-cast nickel-based single-crystal superalloy, and the initial dissolution temperatures of the γ / γ′ eutectic phase, γ′ phase, and γ phase, as well as the complete dissolution temperature of the γ phase, were determined by differential thermal analysis or thermogravimetric analysis.

[0055] In specific implementation, in S1, the process of taking samples from the as-cast nickel-based single-crystal superalloy and determining the initial dissolution temperature of the γ / γ′ eutectic phase, γ′ phase, and γ phase, and the complete dissolution temperature of the γ phase through differential thermal or thermogravimetric analysis includes:

[0056] Taking cast nickel-based single-crystal superalloys as the object, first cut out a block sample that can match the crucible, and after surface cleaning, place it in a differential scanning calorimeter or a synchronous thermal analyzer.

[0057] Under an inert atmosphere, the temperature is continuously heated at a constant heating rate. By recording the heat flow or mass change curves, the onset points of the endothermic peaks of the γ / γ′ eutectic phase, γ′ phase, and γ phase are captured sequentially as the initial dissolution temperatures, and the temperature at which the endothermic peak of the γ phase returns to the baseline is taken as the complete dissolution temperature.

[0058] S2. Based on the temperature range obtained in S1, set the controllable range of the solution heat treatment temperature steps and their corresponding times to form a multi-segment gradient process window.

[0059] In specific implementation, in S2, based on the temperature range obtained in S1, the process of setting the solution heat treatment temperature steps and their corresponding controllable time ranges to form a multi-segment gradient process window includes:

[0060] Using the initial dissolution temperature of the γ′ phase measured by S1 as the lower limit and the complete dissolution temperature of the γ phase as the upper limit, several continuous temperature steps are divided, and the holding time interval is set for each step according to the principle of gradient decrease.

[0061] By combining temperature steps and holding time into a multi-segment gradient process window, the temperature difference and time difference between adjacent segments are kept at a fixed preset step size, thereby allowing refractory elements to diffuse fully while avoiding initial melting.

[0062] S3. Prepare samples in batches within the gradient process window and simultaneously perform high-throughput heat treatment. Quench and sample immediately after each temperature step is completed.

[0063] In specific implementation, S3 involves batch preparation of samples within the gradient process window and simultaneous high-throughput heat treatment, with quenching and sampling immediately after each temperature step. The specific process includes:

[0064] Based on the gradient process window generated by S2, the same batch of cast nickel-based single crystal superalloys were cut into block samples of equal size and numbered sequentially.

[0065] Each sample was individually placed into a quartz glass tube and then vacuum-sealed to ensure that the heat treatment atmosphere was consistent with industrial vacuum conditions.

[0066] All sealed tube samples were simultaneously placed into a gradient temperature controlled muffle furnace, and the furnace temperature was strictly increased step by step according to the temperature steps set in the window.

[0067] After each solid solution temperature is reached and the holding time is completed, the corresponding numbered quartz tube is immediately removed and quenched in flowing argon or ice water to freeze the tissue at that temperature-time point.

[0068] Start the next step of heating, repeat heating, holding, sampling, quenching until the end of the window, thereby obtaining the process parameters for each segment within the S2 window.

[0069] The core idea of ​​S1 to S3 described above is to first measure temperature, then define the temperature window, then sample, and finally freeze. The first step involves continuously scanning the as-cast nickel-based single-crystal superalloy in an inert atmosphere using differential thermal analysis (DTA) or thermogravimetric analysis (TGA). Based on heat flow or mass mutations, the initial melting points of the γ / γ′ eutectic, γ′ phase, and γ phase are precisely captured. The γ phase return point is used as the upper limit of solid solution, defining a safe and sufficient temperature band for subsequent processes. The second step involves designing a multi-stage solid solution ladder with progressively increasing temperature and constant time difference, using this band as the boundary. This prevents initial melting while providing sufficient kinetic space for the diffusion of high-melting-point elements, forming an executable high-throughput process window. The third step involves cutting the same batch of alloy into samples of equal size, vacuum-sealing them, and placing them into the furnace as a whole. The temperature is then strictly increased and held according to the window sequence. After each set temperature is reached and the holding period ends, the corresponding numbered sample is immediately removed and rapidly quenched, instantly freezing the microstructure at that temperature-time node. Thus, each sample uniquely corresponds to a specific set of process parameters, providing a one-to-one temperature-time-microstructure dataset for subsequent machine learning.

[0070] S4. Measure the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain a dataset of microstructure characteristics corresponding to the heat treatment process parameters in S3.

[0071] In specific implementation, in S4, the process of determining the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain the microstructure characteristic dataset corresponding to the heat treatment process parameters in S3 includes:

[0072] Each sample after S3 quenching was sequentially mounted, ground, polished, and electrolytically etched to expose the dendritic structure.

[0073] The Re element segregation ratio was quantitatively determined along the dendrite trunk-interdendritic direction using scanning electron microscopy-energy dispersive spectroscopy.

[0074] The shrinkage porosity ratio was statistically analyzed by X-ray tomography or optical microscopy, and the eutectic content ratio was calculated by image grayscale thresholding.

[0075] By binding the Re element segregation ratio, shrinkage and porosity ratio, and eutectic content ratio with the temperature-time parameters corresponding to the same numbered sample, a microstructure feature dataset covering the entire gradient window is constructed.

[0076] S5. Use the tissue feature dataset obtained in S4 as training data to train the machine learning model. Use the trained machine learning model and the target tissue feature data as input data to obtain optimized heat treatment process parameters.

[0077] In specific implementation, in S5, the tissue feature dataset obtained in S4 is used as training data to train a machine learning model. The specific process of using the trained machine learning model with the target tissue features as input data to obtain optimized heat treatment process parameters includes:

[0078] Based on the microstructure feature dataset obtained from S4, the heat treatment process parameters contained therein are used as input features, and the corresponding Re element segregation ratio, shrinkage porosity ratio and eutectic content ratio are used as output labels to construct a supervised learning training set.

[0079] The supervised learning training set is used to train multiple different types of machine learning models in parallel, and the prediction accuracy of each model is evaluated through cross-validation during the training process.

[0080] The machine learning model with the smallest prediction error is selected as the preferred model. When the prediction errors of multiple models are similar, a multi-model prediction result fusion mechanism is established.

[0081] The desired target Re element segregation ratio, target shrinkage and porosity ratio, and target eutectic content ratio are input into the preferred model or multi-model fusion system.

[0082] The optimal combination of heat treatment process parameters that can simultaneously achieve the target Re element segregation ratio, target shrinkage porosity ratio, and target eutectic content ratio is output through the reverse prediction function of the preferred model or multi-model fusion system.

[0083] In specific implementation, in S5, the machine learning model includes at least two of the following: Ridge Regression Model, XGBoost Model, and LightGBM Model;

[0084] The machine learning model is configured to establish a nonlinear mapping relationship between heat treatment process parameters and Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio.

[0085] In specific implementation, in S5, 90% of the total dataset obtained in S4 is used as the training set, and the remaining 10% is used as the test set. The training set is used for parameter training of the machine learning model, and the test set is used to verify the prediction accuracy of the trained model.

[0086] In practice, evaluating the prediction accuracy of each model through cross-validation includes:

[0087] The prediction errors of each model for Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio were calculated using the ten-fold cross-validation method. The prediction errors were quantified using at least one of the mean absolute error and root mean square error.

[0088] In practice, S5, after obtaining the optimized heat treatment process parameters, also includes a process verification step:

[0089] The optimized heat treatment process parameters were used to heat treat the new nickel-based single-crystal superalloy sample. The high-temperature tensile properties and high-temperature creep properties of the heat-treated sample were then tested to verify the effectiveness of the optimized heat treatment process parameters.

[0090] Application Example 1

[0091] To achieve the above objectives, the present invention has established the following implementation path:

[0092] 1) Samples were taken from the as-cast single-crystal superalloy, and the phase transformation points of the samples were analyzed using differential thermal analysis (DSC) or thermogravimetric analysis (TGA) to obtain the initial dissolution temperatures of the γ / γ' eutectic phase, γ' phase, and γ phase, as well as the complete dissolution temperature of the γ phase in the single-crystal superalloy. These temperatures were denoted as T. γ / γ'共晶 T γ' T s and T l ;

[0093] 2) Define the solution heat treatment process range for single-crystal superalloys. Assuming the solution treatment is divided into n segments, each segment will be labeled as P1, P2, P3, P4…P n The corresponding solution treatment temperatures are T1, T2, T3, T4…T n The corresponding solution times are t1, t2, t3, t4…t n , Figure 1 This refers to the range of process parameters for high-throughput heat treatment of single-crystal superalloys.

[0094] In practice, the main process parameters for selecting high-throughput experimental parameters should follow this strategy: T1 = T γ' -15,T n =T s +15, T n -T n-1 =5; t n -t n-1 =4~8;

[0095] 3) Develop a high-throughput heat treatment process experiment for single-crystal superalloys based on the heat treatment process parameter range obtained in step two. Prepare n samples, and then place all samples into the heat treatment furnace to begin heat treatment. During the heat treatment of the single-crystal superalloys, when the solution temperature T... n After the heat treatment is completed, the sample at the solution treatment temperature is taken out and quenched.

[0096] In practice, to facilitate high-throughput and rapid implementation, the heat-treated samples were vacuum-sealed in quartz glass tubes, and the heat treatment process was carried out in a muffle furnace.

[0097] In practice, in order to simulate the quenching speed of industrial vacuum heat treatment, the size of the heat treatment sample is 10mm*10mm*5mm.

[0098] 4) The microstructure of samples at different solution temperatures was observed, and the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of the samples corresponding to the solution temperature of the parameter were analyzed and statistically analyzed.

[0099] 5) A machine learning model is used to learn the microstructure obtained from high-throughput heat treatment experiments and to verify its error band. A suitable machine learning model is used to solve the process parameters of the target microstructure, so as to quickly obtain the heat treatment process of single crystal high-temperature alloy.

[0100] Preferably, 90% of the heat-treated samples obtained from high-throughput experiments are used as samples for machine learning, and 10% are used as test samples to test the reliability of prediction results.

[0101] Preferably, in order to improve the reliability of machine learning, it is recommended to use multiple models to learn from the samples and to verify the deviation of the models; the model with the smallest deviation should be selected first. If the deviations of the models are similar, multiple models can be used for prediction, and the final prediction results should be normalized.

[0102] 6) The single-crystal superalloy sample was subjected to vacuum heat treatment using the heat treatment process obtained by this method, and the sample was tested for high-temperature tensile and high-temperature creep properties.

[0103] In practical applications:

[0104] The composition range of the third-generation single-crystal high-temperature alloy used is as follows: C≤0.04, Cr:4.1~5.0, Co:10.0~15.0, W:5.0~6.5, Mo:0.5~2.0, Al:5.0~6.25, Ta:7.0~9.25, Re:5.1~5.6, Hf:0.1~0.2, B:0.003~0.006, Y≤0.03, and the remainder is Ni.

[0105] 1) Samples were taken from the as-cast single-crystal superalloy, and the phase transformation points in the alloy were analyzed using DSC to obtain the Ti values. γ' =1255℃, T s =1361℃ and T l =1400℃;

[0106] 2) Define the solution heat treatment process range for single-crystal high-temperature alloys: T1 = 1240℃, T2 = 1250℃, T3 = 1260℃, T4 = 1270℃, T5 = 1280℃, T6 = 1290℃, T7 = 1300℃, T8 = 1310℃, T9 = 1320℃, T... 10=1330℃, T 11 =1340℃, T 12 =1350℃, T 13 =1360℃, T 14 =1370℃; the heat treatment holding time for Pn in each stage is 4 hours; such as Figure 2 As shown;

[0107] 3) Prepare 14 single-crystal high-temperature alloy samples with dimensions of 10mm*10mm*5mm. Vacuum seal them in quartz glass tubes and then place them in a muffle furnace for heat treatment experiments. During the heat treatment process, take out one sample after each heat treatment stage.

[0108] 4) The microstructure of the samples heat-treated at solution temperature of 1240-1370℃ was observed, and the Re element segregation ratio, shrinkage porosity ratio and eutectic content ratio corresponding to the solution temperature were analyzed and statistically analyzed.

[0109] 5) A machine learning model was used to learn the tissue obtained from high-throughput heat treatment experiments at 1240–1360℃. The specific process is as follows: Figure 3 As shown. To improve the reliability of machine learning, we sequentially used Ridge Regression, XGBoost, and LightGBM models to learn from the samples and validated the model deviations. The error ranges for predicting Re element segregation were 0.025, 0.028, and 0.023, respectively; the error ranges for predicting shrinkage and porosity ratio were 0.018, 0.042, and 0.025, respectively; and the error ranges for predicting eutectic content ratio were 0.052, 0.035, and 0.041, respectively. Clearly, the three models did not exhibit particularly strong superiority or inferiority. Therefore, in practical operation, we normalized the prediction results of the three models.

[0110] The heat treatment process results obtained by machine learning are: 1100℃ / holding for 2h → 1240℃ / holding for 4h → 1300℃ / holding for 4h → 1320℃ / holding for 4h → 1330℃ / holding for 4h → 1340℃ / holding for 4h → 1350℃ / holding for 4h, followed by argon blowing to cool to room temperature;

[0111] 6) The single-crystal superalloy samples were subjected to vacuum heat treatment using the heat treatment process obtained by this method, and high-temperature tensile and high-temperature creep rupture performance tests were performed on the samples. To demonstrate the superiority of this invention, we performed a solution treatment + primary aging + secondary aging heat treatment on the third-generation single-crystal superalloy using the original process parameters, and compared its performance with that obtained by this invention. The primary and secondary aging process parameters used in this invention are the same as those in the original process.

[0112] from Figure 4, Figure 5 It can be seen that the high-temperature yield strength, tensile strength and high-temperature creep life of the single-crystal superalloys treated by the heat treatment process obtained by this method are significantly improved. Specifically, the yield strength at 1100℃ is increased by 26% and the creep life at 1120℃ is increased by more than 30%.

[0113] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning, characterized in that, Includes the following steps: S1. Samples were taken from the as-cast nickel-based single-crystal superalloy, and the initial dissolution temperature of the γ / γ′ eutectic phase, γ′ phase, and γ phase, as well as the complete dissolution temperature of the γ phase, were determined by differential thermal analysis or thermogravimetric analysis. S2. Based on the temperature range obtained in S1, set the controllable range of the solution heat treatment temperature steps and their corresponding times to form a multi-segment gradient process window. S3. Prepare samples in batches within the gradient process window and simultaneously perform high-throughput heat treatment. Quench and sample immediately after each temperature step is completed. S4. Measure the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain the microstructure feature dataset corresponding to the heat treatment process parameters in S3. S5. Use the tissue feature dataset obtained in S4 as training data to train the machine learning model. Use the trained machine learning model and the target tissue feature data as input data to obtain optimized heat treatment process parameters.

2. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S1, the specific process of taking samples from the as-cast nickel-based single-crystal superalloy and determining the initial dissolution temperature of the γ / γ′ eutectic phase, γ′ phase, and γ phase, and the complete dissolution temperature of the γ phase through differential thermal or thermogravimetric analysis includes: Taking cast nickel-based single-crystal superalloys as the object, first cut out a block sample that can match the crucible, and after surface cleaning, place it in a differential scanning calorimeter or a synchronous thermal analyzer. Under an inert atmosphere, the temperature is continuously heated at a constant heating rate. By recording the heat flow or mass change curves, the onset points of the endothermic peaks of the γ / γ′ eutectic phase, γ′ phase, and γ phase are captured sequentially as the initial dissolution temperatures, and the temperature at which the endothermic peak of the γ phase returns to the baseline is taken as the complete dissolution temperature.

3. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S2, based on the temperature range obtained in S1, the specific process of setting the solution heat treatment temperature steps and their corresponding controllable time ranges to form a multi-segment gradient process window includes: Using the initial dissolution temperature of the γ′ phase measured by S1 as the lower limit and the complete dissolution temperature of the γ phase as the upper limit, several continuous temperature steps are divided, and the holding time interval is set for each step according to the principle of gradient decrease. By combining temperature steps and holding time into a multi-segment gradient process window, the temperature difference and time difference between adjacent segments are kept at a fixed preset step size, thereby allowing refractory elements to diffuse fully while avoiding initial melting.

4. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S3, the specific process of batch preparing samples and simultaneously performing high-throughput heat treatment within the gradient process window, with immediate quenching and sampling after each temperature step, includes: Based on the gradient process window generated by S2, the same batch of cast nickel-based single crystal superalloys were cut into block samples of equal size and numbered sequentially. Each sample was individually placed into a quartz glass tube and then vacuum-sealed to ensure that the heat treatment atmosphere was consistent with industrial vacuum conditions. All sealed tube samples were simultaneously placed into a gradient temperature controlled muffle furnace, and the furnace temperature was strictly increased step by step according to the temperature steps set in the window. After each solid solution temperature is reached and the holding time is completed, the corresponding numbered quartz tube is immediately removed and quenched in flowing argon gas or ice water to instantly freeze the tissue at that temperature-time point. Start the next step of heating, repeat heating, holding, sampling, quenching until the end of the window, thereby obtaining the process parameters for each segment within the S2 window.

5. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S4, the specific process of determining the Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio of each sample segment to obtain the microstructure characteristic dataset corresponding to the heat treatment process parameters in S3 includes: Each sample after S3 quenching was sequentially mounted, ground, polished, and electrolytically etched to expose the dendritic structure. The Re element segregation ratio was quantitatively determined along the dendrite trunk-interdendritic direction using scanning electron microscopy-energy dispersive spectroscopy. The shrinkage porosity ratio was statistically analyzed by X-ray tomography or optical microscopy, and the eutectic content ratio was calculated by image grayscale thresholding. By binding the Re element segregation ratio, shrinkage and porosity ratio, and eutectic content ratio with the temperature-time parameters corresponding to the same numbered sample, a microstructure feature dataset covering the entire gradient window is constructed.

6. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S5, the process of using the tissue feature dataset obtained in S4 as training data to train a machine learning model, and then using the trained machine learning model with the target tissue features as input data to obtain optimized heat treatment process parameters includes: Based on the microstructure feature dataset obtained from S4, the heat treatment process parameters contained therein are used as input features, and the corresponding Re element segregation ratio, shrinkage porosity ratio and eutectic content ratio are used as output labels to construct a supervised learning training set. The supervised learning training set is used to train multiple different types of machine learning models in parallel, and the prediction accuracy of each model is evaluated through cross-validation during the training process. The machine learning model with the smallest prediction error is selected as the preferred model. When the prediction errors of multiple models are similar, a multi-model prediction result fusion mechanism is established. The desired target Re element segregation ratio, target shrinkage and porosity ratio, and target eutectic content ratio are input into the preferred model or multi-model fusion system. The optimal combination of heat treatment process parameters that can simultaneously achieve the target Re element segregation ratio, target shrinkage porosity ratio, and target eutectic content ratio is output through the reverse prediction function of the preferred model or multi-model fusion system.

7. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S5, the machine learning model includes at least two of the following: Ridge Regression Model, XGBoost Model, and LightGBM Model; The machine learning model is configured to establish a nonlinear mapping relationship between heat treatment process parameters and Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio.

8. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S5, 90% of the total dataset obtained in S4 is used as the training set, and the remaining 10% is used as the test set. The training set is used to train the parameters of the machine learning model, and the test set is used to verify the prediction accuracy of the trained model.

9. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 6, characterized in that, The evaluation of the prediction accuracy of each model through cross-validation specifically includes: The prediction errors of each model for Re element segregation ratio, shrinkage porosity ratio, and eutectic content ratio were calculated using the ten-fold cross-validation method. The prediction errors were quantified using at least one of the mean absolute error and root mean square error.

10. The method for optimizing the heat treatment process of nickel-based single-crystal superalloys based on high-throughput experiments and machine learning according to claim 1, characterized in that, In S5, after obtaining the optimized heat treatment process parameters, a process verification step is also included: The optimized heat treatment process parameters were used to heat treat the new nickel-based single-crystal superalloy sample. The high-temperature tensile properties and high-temperature creep properties of the heat-treated sample were then tested to verify the effectiveness of the optimized heat treatment process parameters.

Images