A Method and System for Shape Optimization of a Superalloy Disk-Type Preform

Abstract

The present invention belongs to the technical field of materials science and discloses a method for optimizing the shape of a high-temperature alloy disk pre-forging. The present invention combines finite element numerical simulation, support vector machine and genetic algorithm to establish a GA-SVR prediction model between input variables and objective function, predicts the optimal value of the objective function, and obtains the best combination of pre-forging geometric parameters. First, the shape of the pre-forging is preliminarily designed through experience, then the results of finite element simulation are extracted to establish a data set, and the parameters of the support vector machine model are used as the objective function of the genetic algorithm by using the data set. The genetic algorithm is used to optimize the parameters of the support vector machine model to obtain the best prediction model. For the combination of size parameters of the pre-forging when the deformation uniformity evaluation index is the best, the global optimization of the genetic algorithm effectively improves the deformation uniformity of the final forging, thereby improving the final performance of the forging.

Description

Technical Field

[0001] The present invention belongs to the technical field of materials science and technology, and particularly relates to a method and system for optimizing the shape of a high-temperature alloy disk-type pre-forging. Background Art

[0002] As a key hot-end component of an aero-engine, high-temperature alloy disk forgings need to work under extreme high temperature, high pressure, high rotational speed and complex load environments, and the requirements for their performance and dimensional accuracy are extremely strict. At present, the die forging process of turbine disks faces problems such as difficult filling, uneven deformation and coarse grains, making it difficult to meet the quality requirements of forgings. Therefore, it is necessary to optimize the tissue compactness and uniformity of forgings to improve their reliability and stability during service. Due to the complex structure of disk forgings, the forging process usually requires multiple processes such as upsetting, pre-forging and final forging. The shape of the pre-forging has an important influence on the flow and distribution of materials in the die during the final forging stage, thus determining the dimensional accuracy, deformation uniformity and microstructure state of the formed forging.

[0003] Existing pre-forging design methods usually design based on experience or analogy with existing die forging drawings, and physical experiments are used to verify the feasibility of the pre-forging scheme. Although it is practical, the efficiency is low. With the application of finite element numerical simulation technology, the costs of scheme verification, structure optimization and process parameter calibration in forging development have been greatly reduced, and simulation methods have been widely used in the optimization process. However, for the geometric design of highly complex pre-forgings, it is obviously difficult to meet the multi-objective optimization requirements only relying on empirical design and finite element simulation.

[0004] In the current technology, in order to optimize the performance of forgings and effectively control the grain size and deformation uniformity, global optimization methods such as response surface method, neural network and sensitivity analysis are generally adopted. By establishing the relationship between design variables and objective functions in quantitative analysis through these methods, multi-objective optimization design becomes more efficient and meets the process requirements. However, the construction of the response surface model in the existing methods is relatively cumbersome. For example, both PB experimental design and rotational composite design require manual statistics and variance calculation, which are time-consuming and laborious. In addition, in order to obtain the experimental response value, each group of parameters needs to be re-modeled and imported into the simulation, and the process is cumbersome; and although the second-order response surface model is suitable for simulating curve relationships, it has poor robustness when dealing with highly nonlinear problems and is easily interfered by outliers. The insufficient number of experimental schemes results in the final model being difficult to accurately capture the complex interaction and nonlinear characteristics between multiple variables, affecting the optimization efficiency.

[0005] Therefore, the problems and deficiencies in the prior art are as follows: The traditional forging process cannot meet the requirements of high precision and high performance of the turbine disk, and the existing response surface method lacks an accurate design method for the optimization of the pre-forging die design. There is an urgent need to develop a new method based on the efficient optimization of the pre-forging shape, combined with parametric geometric design, improved experimental design, and optimization models, to improve the forging performance, shorten the production cycle, and reduce the production cost, so as to meet the stringent service requirements of the high-temperature alloy disk forgings in aeroengines. Summary of the Invention

[0006] In view of the problems existing in the prior art, the present invention provides a method for optimizing the shape of a pre-forging of a high-temperature alloy disk.

[0007] The present invention is implemented as follows. A method for optimizing the shape of a pre-forging of a high-temperature alloy disk includes:

[0008] Step 1: Simulate the disk part without the pre-forging process. According to the overall equivalent strain of the disk part after final forging, use the inverse simulation technology to quickly design the initial geometric shape of the pre-forging, and design the pre-forging die shape according to the pre-forging shape;

[0009] Step 2: Combine the experimental design method and the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forging in Step 1 as the key optimization parameters, and determine the value range of the optimization parameters;

[0010] Step 3: Use the Latin hypercube sampling method to generate approximate random samples within the multi-parameter value range, set the number of samples extracted by the Latin hypercube sampling, and use the finite element simulation method to automatically batch simulate all random samples;

[0011] Step 4: Establish an objective function describing the forming quality, and extract the data required in the simulation results of all samples in Step 3 as the data set for establishing the prediction model;

[0012] Step 5: Establish a GA-SVR prediction model between the input variables and the objective function through the data set in Step 4, predict the optimal value of the objective function, and obtain the best combination of pre-forging geometric parameters.

[0013] Further, the simulation of the disk part without the pre-forging process, according to the overall equivalent strain of the disk part after final forging, uses the inverse simulation technology to quickly design the initial geometric shape of the pre-forging, and designs the pre-forging die shape according to the pre-forging shape:

[0014] Directly conduct finite element simulation on the cake blank, reverse track the positions of the small deformation area and the large deformation area in the original blank based on the strain results of the forgings obtained from the finite element simulation, and preliminarily design the shape of the pre-forging. This method can eliminate the small deformation area and the large deformation area, which can significantly improve the deformation uniformity of the forgings. The specific design method is as follows: First, divide the cake blank into N rectangular areas with equal volume along the radial direction according to the deformation after final forging. The more complex the shape of the final forging, the more the number of partitions for the final forging. Then, divide the cake blank into N rectangular areas with the same volume according to the partitions of the final forging; write a script to extract the strain value ∈i of each grid after final forging, and calculate the average effective strain value of the forging after final forging Then analyze the deformation conditions of the large deformation area and the small deformation area of the final forging, mark the small deformation area smaller than the forging and the large deformation area, and use the reverse tracking method to track the positions of the small deformation area and the large deformation area in the original blank. Select the points to be tracked after final forging through point tracking in the post-processing process of the finite element simulation software, reverse simulate and track the positions of the selected points in the cake blank, then reduce the material at the corresponding small deformation area of the cake blank, and add the reduced volume along the radial direction to the large deformation area. Finally, smooth the sharp parts of each different rectangular area to obtain the initial pre-forging shape;

[0015] Average effective strain value

[0016] ∈i is the effective strain of the i-th grid, and n is the total number of grids;

[0017] Small deformation area

[0018] Large deformation area

[0019] α and β are parameters for selecting the overall deformation amount, and their value ranges are 0.1 to 0.5.

[0020] Furthermore, combine the experimental design method with the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forging in step 1 as the key optimization parameters, and determine the value ranges of the optimization parameters;

[0021] Select the geometric dimensions that mainly affect the effective strain of the forging during the pre-forging process as the key geometric dimensions. The selection method is as follows: extract all the geometric parameters of the preliminarily designed pre-forging die in step 1, then set all the parameters as the design variables of the finite element simulation. Combine the experimental design method with the finite element method and conduct a screening experimental design to evaluate the influence degree of the geometric dimensions of the die on the effective strain of the forging. Obtain the relationship diagram of the influence degree of all geometric parameters on the effective strain in the post-processing of the finite element simulation. Use the Spearman correlation coefficient (rs) to evaluate the influence degree of each geometric parameter variable on the effective strain. When the Spearman correlation coefficient is greater than q, select this parameter as the characteristic geometric parameter. q is the threshold for the geometric parameter to have a greater influence on the effective strain, and its range is 0.1 - 0.3. Filter out the geometric dimensions with less influence on the effective strain through this method to reduce the complexity and diversity of the data, and at the same time reduce the number of samples required for establishing the GA-SVR prediction model between the input variables and the objective function, improving the prediction accuracy and generalization ability of the final model; and select the value range of the required optimization parameters according to the Spearman correlation coefficient of each characteristic geometric dimension;

[0022] Spearman correlation coefficient

[0023] Among them, dj is the range difference between the variable-effective strain relationship, j is the number of key geometric dimensions after screening, and k is the number of data points;

[0024] r s The value range of is from -1 to +1;

[0025] Positive correlation means that as one variable increases, the other variable also tends to increase;

[0026] Negative correlation indicates that as one variable increases, the other variable tends to decrease;

[0027] r s Close to 0 means there is no significant monotonic relationship between the two, r s Values close to -1 or +1 indicate a strong correlation between the two;

[0028] Value range of the optimization parameter: L j ×(1 + rs) > L , j > L j ×(1 - rs);

[0029] L j Is the original size of the jth key geometric dimension, L , j Is the jth key geometric dimension.

[0030] Further, the Latin hypercube sampling method is used to generate approximate random samples within the multi-parameter value range, and the number of samples extracted by the Latin hypercube sampling is set, and the finite element simulation method is used to automatically batch simulate and simulate all random samples;

[0031] Latin hypercube sampling is adopted to generate approximate random samples from the multi-characteristic geometric parameter distribution determined in step 2. In this sampling method, the value range of each parameter is divided into several probability intervals, and a sample point is randomly selected within each interval. These randomly selected sample points are combined to form the final sample set. By ensuring the uniform distribution of samples in the multi-dimensional space, it guarantees the diversity and representativeness of the samples and is applicable to complex problems that require a large number of simulations and analyses; the required number of samples depends on multiple factors, including the complexity of the problem, the diversity of the data, the number of features, and the generalization ability of the model. If the complexity of the data set is high, or the data is widely distributed in the input space, more samples are needed to ensure that the model can capture all important features; after setting the required number of samples, the simulation samples can be submitted in the corresponding finite element simulation software for automatic batch simulation and simulation.

[0032] Further, the objective function describing the forming quality is established, and the required data in all sample simulation results in step 3 is extracted as the data set for establishing the prediction model;

[0033] The input variables required for the above-mentioned step data set are the key geometric parameters to be optimized. The objective function of the data set is mainly the sub-objectives for evaluating the forging forming quality, including but not limited to: the average effective strain y1 describing the deformation degree of the final forging, the standard deviation of the equivalent effective strain y2 describing the deformation uniformity, the average grain size y3 describing the tissue uniformity, the standard deviation of the average grain size y4, and the recrystallized volume fraction y5. This involves the problem of multi-objective optimization. Now each sub-objective is normalized to obtain y , i ∈(0, 1), and then the normalized sub-objectives are integrated by linear weighting to obtain the objective function M is the number of selected sub-objectives, and the weight w of each sub-objective l The distribution needs to consider the influence of the average effective strain, the standard deviation of the equivalent effective strain, the average grain size, the standard deviation of the average grain size, and the recrystallized volume fraction of each target on the final forming quality of the forging. The greater the influence on the final forming quality of the forging, the greater the weight value. The weights of each sub-objective satisfy Since the smaller the standard deviation of equivalent strain, the average grain size, and the standard deviation of average grain size are, the better, they are taken as positive numbers. Since the larger the average equivalent strain and the recrystallized volume fraction are, the better, they are taken as negative numbers. The smaller the final objective function Y is, the better the forging forming quality is. Through a Python script, corresponding data is automatically and batch extracted from the file storing the simulation results through the corresponding keywords. The extracted data includes the geometric dimension parameters of each sample and the equivalent strain and grain size of all grids at the end of the simulation. After processing the extracted data according to the required objectives, the finally obtained data is used as the dataset for establishing the prediction model. This dataset is a tabular file with multiple inputs and a single output.

[0034] Average equivalent strain value

[0035] Standard deviation of equivalent strain

[0036] Average grain size

[0037] Standard deviation of average grain size

[0038] Recrystallized volume fraction

[0039] ∈i is the equivalent strain of the i-th grid, is the average equivalent strain value, d i is the grain size of the i-th grid, is the average grain size, is the recrystallized volume fraction of the i-th grid, and n is the total number of grids;

[0040] Min-max normalization

[0041] Among them, y l is each sub-objective, y , l is the normalized data. This method scales all data into the interval (0, 1).

[0042] Furthermore, the GA-SVR prediction model between the input variables and the objective function is established through the dataset in step 4 to predict the optimal value of the objective function and obtain the best combination of pre-forging geometric parameters;

[0043] Call the dataset established in step 4, combine the support vector regression machine with the genetic algorithm, and construct a prediction and optimization model for the deformation uniformity of the superalloy disk forging;

[0044] Establish the sample set of the model: (a1, b1), (a2, b2)...(a p , bp )...(a t ,b t ), a p ,b p ∈R

[0045] The objective function of SVR can be expressed as:

[0046] Subject to the following conditions:

[0047] The kernel function of SVR:

[0048] where a p is the input variable value of the support vector regression machine, b p is the corresponding output variable value, φ(a p ) is the kernel function that maps the data to a high-dimensional space, ω is the weight vector, μ is the bias term, ξ p and are slack variables used to handle points outside the insensitive interval; C is the regularization parameter that controls the penalty intensity for interval violations; the kernel function parameters γ and the error tolerance ε;

[0049] Search for the optimal solution of the problem through the selection, crossover, and mutation operations in the genetic algorithm iteration process; when optimizing the SVR model, use the genetic algorithm to optimize the parameters of SVR, including the regularization parameter C, the kernel function parameter γ, and the error tolerance ε, and the variation range of each parameter is set to (0, 100), (0, 1000), (0, 10). The optimization process is divided into the following steps

[0050] Encoding: Encode the parameters of SVR, the regularization parameter C, the kernel function parameter γ, and the error tolerance ε into chromosomes;

[0051] Initializing the population: Randomly generate a set of solutions as the initial population;

[0052] Fitness calculation: Calculate the fitness of each individual using the cross-validation method, usually the prediction accuracy;

[0053] Selection, crossover, mutation: The chromosomes in the population evolve towards the optimal values of C, γ, and ε through selection, crossover, and mutation, and generate a new population;

[0054] Termination condition: Repeat the above process until the termination condition is met, reaching the maximum number of iterations or the fitness reaches a certain threshold;

[0055] After using the genetic algorithm to iteratively optimize the parameters of the support vector regression model to obtain the optimal model, predict the optimal value of the objective function to obtain the best combination of preform geometric parameters.

[0056] Another object of the present invention is to provide a high-temperature alloy disk pre-forging part shape optimization system, comprising:

[0057] A design module, which is used to simulate the disk part without the pre-forging process. According to the overall effective strain condition after the final forging of the disk part, the reverse simulation technology is used to quickly design the initial geometric shape of the pre-forging part, and the shape of the pre-forging die is designed according to the shape of the pre-forging part;

[0058] A screening module, which is used to combine the experimental design method and the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forging part as the key optimization parameters, and determine the value range of the optimization parameters;

[0059] A simulation and emulation module, which is used for the number of samples extracted by hypercube sampling, and automatically batch-simulates and emulates all random samples by using the finite element simulation method;

[0060] An extraction module, which is used to establish an objective function describing the forming quality, and extract the required data from all sample simulation results as the data set for establishing the prediction model;

[0061] A prediction module, which is used to establish a GA-SVR prediction model between the input variables and the objective function through the data set, predict the optimal value of the objective function, and obtain the best combination of pre-forging part geometric parameters.

[0062] Another object of the present invention is to provide a computer device, which includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the high-temperature alloy disk pre-forging part shape optimization method.

[0063] Another object of the present invention is to provide a computer-readable storage medium, storing a computer program, which when executed by a processor, causes the processor to execute the steps of the high-temperature alloy disk pre-forging part shape optimization method.

[0064] Another object of the present invention is to provide an information data processing terminal, which is used to implement the high-temperature alloy disk pre-forging part shape optimization system.

[0065] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by the present invention are:

[0066] First, the present invention combines finite element numerical simulation, support vector machine, and genetic algorithm to establish a GA-SVR prediction model between input variables and objective function, predict the optimal value of the objective function, and obtain the best combination of preform geometric parameters. First, the preform shape is preliminarily designed through experience, and then the results of finite element simulation are extracted to establish a data set. The parameters of the support vector machine model are used as the objective function of the genetic algorithm, and the genetic algorithm is used to optimize the parameters of the support vector machine model to obtain the best prediction model. For the combination of preform size parameters when the deformation uniformity evaluation index is the best, the global optimization of the genetic algorithm effectively improves the deformation uniformity of the final forging, thereby improving the final performance of the forging. Compared with the traditional preform design method, this method has the advantages of strong generality, less manual operation, and good deformation uniformity, shortening the design cycle, improving production efficiency, and reducing production costs.

[0067] Second, the technical solution of the present invention fills the technical gaps in the domestic and international industries: provides an efficient method and system for the shape optimization of high-temperature alloy disk preforms. The latest finite element simulation software is used to automatically select and batch simulate samples, greatly improving the efficiency in experimental scheme design and simulation result acquisition. Then, the simulation results are extracted through scripts to generate a data set, and an intelligent algorithm model is established to optimize the preform shape. This method is not only applicable to high-temperature alloy disk parts, but also suitable for parts that cannot be mass-produced in actual production and need to continuously seek the optimal design through simulation. Only need to use the same method to obtain a batch of samples, then extract the simulation results to establish a data set, and then establish an intelligent algorithm model of design variables and objective function according to the situation, and then the intelligent algorithm model automatically optimizes the design variables.

[0068] The technical solution of the present invention overcomes technical biases: an intelligent algorithm model is established through a data set and then optimized. This method combines the advantages of machine learning. The intelligent algorithm can automate the optimization process, reduce the cost and time of manual trial and error, and improve the optimization efficiency; the model established based on the data set can provide data-driven decision support, increasing the objectivity and accuracy of decision-making; many intelligent optimization algorithms, such as genetic algorithm, particle swarm optimization, etc., have global optimization capabilities, can avoid local optimal solutions, and increase the probability of finding the global optimal solution; optimizing through the intelligent algorithm model can reduce the need for physical experiments, thereby reducing costs and risks; the optimization process of the intelligent algorithm model can be automated and can be flexibly adjusted according to different business needs. This method can be extended to large-scale data sets and high-dimensional data and is applicable to problems of various scales.

[0069] Thirdly, the pre-forging part shape optimization method for superalloy discs provided by the present invention significantly solves many problems in the prior art and achieves technological progress in multiple aspects in industrial applications.

[0070] 1. Solving the problems of uneven deformation and coarse-grained structure of forgings in the prior art: Turbine disc forgings under traditional forging processes are difficult to meet the requirements of high precision and high performance, especially problems such as uneven deformation and coarse-grained structure of forgings. Through the inverse simulation method of the present invention, the shape of the pre-forging part can be designed more scientifically, making the equivalent strain distribution of the forging after final forging more uniform, significantly improving the deformation uniformity of the forging, thereby enhancing the microstructure quality of the forging, refining the grain structure, and meeting the stringent requirements of key components of aeroengines.

[0071] 2. Improving the optimization efficiency, reducing production costs and time: Traditional empirical analogy design and physical experiment verification methods are time-consuming and laborious and are difficult to cope with complex forging designs. By using the experimental design method of the present invention in combination with finite element simulation, the geometric parameters that have the greatest impact on the forging performance are screened out, and simulation samples are generated through Latin hypercube sampling, reducing the number of physical experiments and the amount of data required, thereby accelerating the development process. This method reduces the need for repeated experiments, shortens the R & D cycle, and reduces production costs, and is especially suitable for the design requirements of pre-forging parts with multiple parameters and complex structures.

[0072] 3. Providing accurate multi-objective optimization design: Existing response surface models are insufficient in dealing with complex non-linear relationships and are difficult to effectively capture the deep relationships between various design variables. Through the multi-objective optimization method of the present invention, multiple sub-objectives (such as equivalent strain, grain size, and recrystallized volume fraction) that affect the forming quality of the forging are normalized, and an overall objective function is constructed based on the linear weighting method, providing a more reliable evaluation criterion for the forming quality of the forging, thereby ensuring the accuracy of the optimization design and the forming quality of the forging.

[0073] 4. Improvement in intelligence and automation: The present invention uses a support vector regression machine (SVR) combined with a genetic algorithm (GA) for multi-objective optimization modeling, realizing the intelligence and automation of the optimization process. The parameters of the SVR model are automatically optimized through the genetic algorithm, making the model have higher prediction accuracy and generalization ability. The optimized model can efficiently predict the optimal combination of pre-forging part geometric parameters, improve the forming quality of the forging, avoid frequent manual intervention in traditional optimization methods, and improve the optimization efficiency.

[0074] In summary, through precise pre-forging shape optimization and intelligent, multi-objective optimization modeling, the method of the present invention effectively solves problems such as uneven forging deformation and coarse grains in industrial applications, improves the forming quality and production efficiency of forgings, realizes technological progress in forging shape optimization, and provides a strong guarantee for the reliability and safety of superalloy disk forgings. Description of the Drawings

[0075] Figure 1 It is a flowchart of the method for optimizing the shape of a superalloy disk pre-forging provided by an embodiment of the present invention.

[0076] Figure 2 It is a block diagram of the structure of the system for optimizing the shape of a superalloy disk pre-forging provided by an embodiment of the present invention.

[0077] Figure 3 It is a process diagram of point tracking through post-processing of finite element simulation software provided by an embodiment of the present invention.

[0078] Figure 4 It is a diagram of the initial pre-forging shape and key geometric parameters provided by an embodiment of the present invention.

[0079] Figure 5 It is a diagram showing the relationship between the influence degree of key geometric parameters on the equivalent strain provided by an embodiment of the present invention.

[0080] Figure 6 It is a schematic diagram of SVR provided by an embodiment of the present invention.

[0081] Figure 7 They are the equivalent strain distribution nephograms, grain size distribution nephograms, and recrystallized volume fraction nephograms before and after optimization provided by an embodiment of the present invention;

[0082] Figure 8 It is the fitting error between the predicted value and the actual value of the training samples of the GA-SVR model provided by an embodiment of the present invention;

[0083] Figure 9 It is the fitting error between the predicted value and the actual value of the test samples of the GA-SVR model provided by an embodiment of the present invention. Detailed Embodiments

[0084] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0085] The following are two specific embodiments of the application of the present invention in actual industries:

[0086] Embodiment 1: Optimization Design of the Pre-forging of an Aero-engine Turbine Disk

[0087] In the manufacturing of aero-engines, as a key forged disk made of superalloy for high temperature applications, the properties of the turbine disk directly affect the reliability of the engine. By using the optimization method of the present invention, the optimized design of the pre-forged part of the turbine disk can be achieved through the following steps:

[0088] 1. Reverse simulation of the initial design: First, perform the final forging simulation of the turbine disk without the pre-forging process to obtain its deformation distribution and equivalent strain distribution, and design the initial geometric shape of the pre-forged part through reverse simulation technology.

[0089] 2. Screening and optimization of key parameters: Using the experimental design method combined with finite element simulation, extract the key geometric parameters that affect the uniformity of final forging deformation and grain refinement, and obtain the pre-forging geometric dimensions most suitable for the final forging of the turbine disk.

[0090] 3. Multi-objective optimization: Use the Latin hypercube sampling method to generate samples and conduct simulations, calculate indicators such as the equivalent strain, grain size distribution, and recrystallized volume fraction of the forging, establish an optimization model based on GA-SVR, predict the optimal value of the objective function, and obtain the final pre-forging shape.

[0091] Through this optimization process, the final forging of the turbine disk achieves a uniform distribution of dense structure and refined grains, significantly improves the mechanical properties, and meets the strict quality standards of key components of aero-engines. At the same time, compared with the traditional method, the R & D cycle is shortened by about 30%, and the production cost is reduced by more than 20%.

[0092] Example 2: Precision forging optimization of the industrial gas turbine disk

[0093] In an industrial gas turbine, the disk needs to withstand the extreme environment of high temperature and high stress for a long time, and the quality of its forging directly determines the service life and performance stability of the turbine. The specific process of using the optimization method of the present invention for the design of the pre-forged part of the disk is as follows:

[0094] 1. Final forging simulation and pre-forged part design: Perform the finite element simulation of the final forging of the disk to obtain the results of equivalent strain and grain distribution. Using the reverse simulation method, quickly design the initial pre-forged part shape with a reasonable volume distribution to improve the deformation uniformity after final forging.

[0095] 2. Experimental design for multi-parameter optimization: Determine the key geometric parameters and use the Latin hypercube sampling to generate samples of different parameter combinations. Conduct simulation calculations for each group of samples, analyze key quality indicators such as deformation uniformity and recrystallized volume fraction, and establish a GA-SVR prediction model including key geometric parameters and forming quality objectives.

[0096] 3. Optimal shape optimization: Through iterative optimization using genetic algorithms, an optimal combination of geometric parameters is obtained to ensure that the mechanical properties of the disk forging meet the design requirements, reducing subsequent machining and heat treatment adjustments.

[0097] After adopting this optimized design, the forming accuracy of the final forging of the gas turbine disk is increased by about 15%, and the grain uniformity is greatly improved, meeting the service requirements under complex working conditions.

[0098] As Figure 1 shown, a method for optimizing the shape of a superalloy disk pre-forging provided by an embodiment of the present invention includes the following steps:

[0099] S101: Simulate the disk part without the pre-forging process. According to the overall equivalent strain situation of the disk part after final forging, use reverse simulation technology to quickly design the initial geometric shape of the pre-forging, and design the shape of the pre-forging die according to the shape of the pre-forging.

[0100] Directly perform finite element simulation on the cake blank for final forging. Reverse track the positions of the small deformation area and the large deformation area in the original blank through the equivalent strain results of the forging obtained by finite element simulation, and preliminarily design the shape of the pre-forging. This method can eliminate the small deformation area and the large deformation area, which can significantly improve the deformation uniformity of the forging. The specific design method is as follows: First, divide the cake blank into N rectangular areas with equal volume along the radius direction according to the deformation situation after final forging. The more complex the shape of the final forging, the more the number of partitions for final forging. Then, divide the cake blank into N rectangular areas with the same volume according to the partitions of the final forging. Write a script to extract the strain value ∈i of each grid after final forging, and calculate the average equivalent strain value of the forging after final forging Then analyze the deformation situations of the large deformation area and the small deformation area of the final forging, mark the small deformation area smaller than the forging and the large deformation area, and use the reverse tracking method to track the positions of the small deformation area and the large deformation area in the original blank. Select the points to be tracked after final forging through point tracking in the post-processing process of the finite element simulation software, reverse simulate and track the positions of the selected points in the cake blank. Then reduce the material at the corresponding small deformation area of the cake blank, and add the reduced volume along the radius direction to the large deformation area. Finally, smooth the sharp parts of each different rectangular area to obtain the initial pre-forging shape.

[0101] Average equivalent strain value

[0102] ∈i is the equivalent strain of the i-th grid, and n is the total number of grids;

[0103] Small deformation area

[0104] Large deformation area

[0105] α and β are parameters for selecting the overall deformation amount, and their value ranges are from 0.1 to 0.5.

[0106] S102: Combine the experimental design method with the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the S101 pre-forged part as the key optimization parameters, and determine the value ranges of the optimization parameters.

[0107] Select the geometric dimensions that mainly affect the equivalent strain of the forging during the pre-forging process as the key geometric dimensions. The selection method is to extract all the geometric parameters of the pre-forging die preliminarily designed in step 1, then set all the parameters as the design variables of the finite element simulation, combine the experimental design method with the finite element method, and conduct screening experimental design to evaluate the influence degree of the geometric dimensions of the die on the equivalent strain of the forging. Obtain the relationship diagram of the influence degree of all geometric parameters on the equivalent strain in the post-processing of the finite element simulation. Use the Spearman correlation coefficient (rs) to evaluate the influence degree of each geometric parameter variable on the equivalent strain. When the Spearman correlation coefficient is greater than q, select this parameter as the characteristic geometric parameter. q is the threshold for the geometric parameter to have a greater influence on the equivalent strain, and its range is from 0.1 to 0.3. Filter out the geometric dimensions with less influence on the equivalent strain through this method to reduce the complexity and diversity of the data, and at the same time reduce the number of samples required for establishing the GA-SVR prediction model between the input variables and the objective function, and improve the prediction accuracy and generalization ability of the final model. And select the value ranges of the required optimization parameters according to the Spearman correlation coefficient of each characteristic geometric dimension.

[0108] Spearman correlation coefficient

[0109] Among them, dj is the rank difference between the variable - equivalent strain relationship, j is the number of key geometric dimensions after screening, and k is the number of data points;

[0110] r s The value range is from -1 to +1;

[0111] Positive correlation means that as one variable increases, the other variable also tends to increase;

[0112] Negative correlation indicates that as one variable increases, the other variable tends to decrease;

[0113] r s Close to 0 indicates that there is no significant monotonic relationship between the two, and r s Values close to -1 or +1 indicate a strong correlation between the two;

[0114] Value range of the optimization parameter: L j ×(1 + rs) > L , j>L j ×(1 - rs);

[0115] L j is the original dimension of the j-th critical geometric dimension, L , j is the j-th critical geometric dimension.

[0116] S103: Generate approximate random samples within the multi-parameter value range using the Latin Hypercube Sampling method, set the number of samples drawn by the Latin Hypercube Sampling, and use the finite element simulation method to automatically batch simulate and simulate all random samples.

[0117] Take Latin Hypercube Sampling to generate approximate random samples from the multi-characteristic geometric parameter distribution determined in S102. In this sampling method, the value range of each parameter is divided into several intervals with equal probabilities, and a sample point is randomly selected within each interval. These randomly selected sample points are combined to form the final sample set. It ensures the uniform distribution of samples in the multi-dimensional space, thus guaranteeing the diversity and representativeness of the samples, and is suitable for complex problems that require a large number of simulations and analyses. The required number of samples depends on multiple factors, including the complexity of the problem, the diversity of the data, the number of features, and the generalization ability of the model. If the complexity of the data set is high, or the data is widely distributed in the input space, more samples are needed to ensure that the model can capture all important features. After setting the required number of samples, the simulation samples can be submitted and automatically batch simulated in the corresponding finite element simulation software.

[0118] S104: Establish an objective function describing the forming quality, and extract the data required in all sample simulation results in S103 as the data set for establishing the prediction model.

[0119] The input variables required for the above-mentioned step data set are the critical geometric parameters to be optimized. The objective function of the data set is mainly the sub-objectives for evaluating the forging forming quality, including but not limited to: the average equivalent strain y1 describing the deformation degree of the final forging, the standard deviation of the equivalent strain y2 describing the deformation uniformity, the average grain size y3 describing the tissue uniformity, the standard deviation of the average grain size y4, the recrystallized volume fraction y5, etc. It involves the problem of multi-objective optimization. Now, each sub-objective is normalized to obtain y , i ∈(0, 1), and then the normalized sub-objectives are integrated using linear weighting to obtain the objective function M is the number of selected sub-objectives, and the weight w of each sub-objective lThe distribution needs to consider the influence of the average equivalent strain value, standard deviation of equivalent strain, average grain size, standard deviation of average grain size, recrystallized volume fraction, etc. on the final forming quality of the forging. The greater the influence on the final forming quality of the forging, the greater the weight value. The weights of each sub-objective satisfy Since the standard deviation of equivalent strain, average grain size, and standard deviation of average grain size are better when they are smaller, positive numbers are taken. The average equivalent strain value and recrystallized volume fraction are better when they are larger, so negative numbers are taken. The smaller the final objective function Y, the better the forming quality of the forging. Through the python script, the corresponding data is automatically extracted in batches from the file storing the simulation results through the corresponding keywords. The extracted data includes the geometric size parameters of each sample and the equivalent strain and grain size of all grids at the end of the simulation. After processing the extracted data according to the required objectives, the finally obtained data is used as the data set for establishing the prediction model. This data set is a multi-input and single-output table file.

[0120] Average equivalent strain value

[0121] Standard deviation of equivalent strain

[0122] Average grain size

[0123] Standard deviation of average grain size

[0124] Recrystallized volume fraction

[0125] ∈i is the equivalent strain of the i-th grid, is the average equivalent strain value, d i is the grain size of the i-th grid, is the average grain size, is the recrystallized volume fraction of the i-th grid, and n is the total number of grids;

[0126] Minimum-maximum normalization

[0127] Among them, y l is each sub-objective, y , l is the normalized data. This method scales all data into the interval (0, 1).

[0128] S105: Establish a GA-SVR prediction model between the input variables and the objective function through the data set in S104, predict the optimal value of the objective function, and obtain the best combination of pre-forging geometric parameters.

[0129] Call the data set established in S104, combine the support vector regression machine with the genetic algorithm, and construct a prediction and optimization model for the deformation uniformity of superalloy disk forgings.

[0130] Establish the sample set of the model: (a1, b1), (a2, b2)...(a p , b p )...(a t , b t ), a p , b p ∈R

[0131] The objective function of SVR can be expressed as:

[0132] Subject to the following conditions:

[0133] The kernel function of SVR:

[0134] Among them, a p is the input variable value of the support vector regression machine, b p is the corresponding output variable value, φ(a p ) is the kernel function that maps the data to a high-dimensional space, ω is the weight vector, μ is the bias term, ξ p and are slack variables used to handle points outside the insensitive interval; C is the regularization parameter that controls the penalty intensity of interval violations; the kernel function parameter γ and the error tolerance ε.

[0135] Search for the optimal solution of the problem through the selection, crossover, and mutation operations in the genetic algorithm iteration process. When optimizing the SVR model, use the genetic algorithm to optimize the parameters of SVR, including the regularization parameter C, the kernel function parameter γ, and the error tolerance ε. The variation range of each parameter is set to (0, 100), (0, 1000), (0, 10). The optimization process is divided into the following steps

[0136] Encoding: Encode the parameters of SVR, the regularization parameter C, the kernel function parameter γ, and the error tolerance ε into chromosomes;

[0137] Initializing the population: Randomly generate a set of solutions as the initial population;

[0138] Fitness calculation: Calculate the fitness of each individual using methods such as cross-validation, usually the prediction accuracy;

[0139] Selection, crossover, mutation: The chromosomes in the population evolve towards the optimal values of C, γ, and ε through selection, crossover, and mutation, and generate a new population;

[0140] Termination condition: Repeat the above process until the termination condition is met, i.e., the maximum number of iterations is reached or the fitness reaches a certain threshold;

[0141] After using the genetic algorithm to iteratively optimize the parameters of the support vector regression model to obtain the optimal model, predict the optimal value of the objective function to obtain the best combination of preform geometric parameters.

[0142] As Figure 2 shown, the high-temperature alloy disk-shaped preform shape optimization system provided by the embodiment of the present invention includes:

[0143] A design module, which is used to simulate the disk-shaped parts without the pre-forging process, and according to the overall effective strain situation after the final forging of the disk-shaped parts, use the inverse simulation technology to quickly design the initial geometric shape of the preform, and design the pre-forging die shape according to the preform shape;

[0144] A screening module, which is used to combine the experimental design method and the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the preform as the key optimization parameters, and determine the value range of the optimization parameters;

[0145] A simulation and simulation module, which is used for the number of samples extracted by hypercube sampling, and uses the finite element simulation method to automatically batch simulate all random samples;

[0146] An extraction module, which is used to establish an objective function describing the forming quality, and extract the required data from all sample simulation results as the data set for establishing the prediction model;

[0147] A prediction module, which is used to establish a GA-SVR prediction model between the input variables and the objective function through the data set, predict the optimal value of the objective function, and obtain the best combination of preform geometric parameters.

[0148] Another object of the present invention is to provide a computer device, which includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the high-temperature alloy disk-shaped preform shape optimization method.

[0149] Another object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the processor executes the steps of the high-temperature alloy disk-shaped preform shape optimization method.

[0150] Another object of the present invention is to provide an information data processing terminal, which is used to implement the high-temperature alloy disk-shaped preform shape optimization system.

[0151] Specific implementation of the present invention:

[0152] (1) Simulate the disk-shaped parts without pre-forging process. According to the overall equivalent strain of the disk-shaped parts after final forging, use inverse simulation technology to quickly design the initial geometric shape of the pre-forged parts, and design the shape of the pre-forging die according to the shape of the pre-forged parts.

[0153] First, perform upsetting and final forging finite element simulations on the cylindrical billet through finite element simulation software. According to the overall equivalent strain after final forging, divide the cake billet before final forging into 5 regions with equal volume along the radial direction. Write a script to extract the strain value ∈i of each grid after final forging, and then calculate the average equivalent strain value of the forgings after final forging is 2.54, and α and β are taken as 0.3 and 0.5, to obtain a small deformation region ∈ small ≤ 1.78 and a large deformation region ∈ large ≥ 3.81. Then use the reverse tracking method to track the positions of the small and large deformation regions in the original billet. Select the points to be tracked after final forging through point tracking in the post-processing of the finite element simulation software, and then inversely simulate and track the positions of the selected points in the cake billet. The tracking process is as Figure 3 shown. P1 - P30 are the points in the small deformation region, and P31 - P50 are the points in the large deformation region. Then reduce the material at the corresponding small deformation region of the cake billet, and add the reduced volume along the radial direction to the large deformation region. Then smooth the sharp parts of each different rectangular region to obtain the initial pre-forged part shape as Figure 4 shown, and design the pre-forging die according to the shape of the pre-forged part.

[0154] (2) Combine the experimental design method with the finite element method to conduct screening experimental design to determine the key geometric parameters of the pre-forged parts in step 1 as the required optimization parameters, and determine the value range of the required optimization parameters.

[0155] Establish a multi-process full-flow simulation of a superalloy disk-shaped forging through finite element simulation software. Its technological process includes blanking - heating - upsetting - heating - pre-forging - heating - final forging. After the full-flow simulation is completed, extract all geometric parameters of the pre-forging process for screening experimental design, evaluate the influence degree of the change of these geometric dimensions on the equivalent strain, and obtain the relationship diagram of the influence degree of all geometric parameters on the equivalent strain in the post-processing of the finite element simulation as Figure 5 shown, and screen out the key geometric parameters through the Spearman correlation coefficient in the figure. When the Spearman correlation coefficient is greater than or equal to 0.2, select this parameter as the key geometric parameter. Select the key geometric dimension parameters of adding the pre-forging die during the pre-forging process as the input variables, and obtain three variables for each of the upper and lower dies through screening as Figure 4 shown. According to the calculation formula of the value range of the optimization parameters L j×(1±rs), and the obtained value ranges are as follows: the lower die H1 is 35 - 65 mm, the lower die H2 is 20 - 40 mm, the lower die D1 is 15 - 45 mm, the upper die H3 is 17 - 47 mm, the upper die H4 is 8 - 28 mm, and the upper die D2 is 40 - 80 mm.

[0156] (3) Use the Latin hypercube sampling method to generate approximate random samples within the multi - parameter value range, set the number of samples extracted by the Latin hypercube sampling, and use the finite element simulation software to automatically batch simulate all random samples.

[0157] Adopt Latin hypercube sampling to generate approximate random samples from the multi - variable key geometric parameter distribution determined in step 2. The input variables include six geometric dimensions. Set the number of samples to 50, where the number of training set samples is 40 and the number of test set samples is 10. After setting the required number of samples, the simulation samples can be submitted and automatically batch simulated in the corresponding finite element simulation software.

[0158] (4) Establish an objective function to describe the forming quality, and extract the data required from all sample simulation results in step 3 as the data set for establishing the prediction model.

[0159] The input variables of the data set are the key geometric parameters to be optimized. The objective function of the data set describes the average equivalent strain y1 of the final forging deformation degree, the standard deviation of equivalent strain y2 describing deformation uniformity, the average grain size y3 describing tissue uniformity, the standard deviation of average grain size y4, and the recrystallized volume fraction y5, which involves the problem of multi - objective optimization. Now, each sub - objective is normalized to obtain y , l ∈(0, 1), and then the normalized sub - objectives are integrated using linear weighting to obtain the objective function The weight w of each sub - objective l The distribution of weights needs to consider the influence of the average equivalent strain, standard deviation of equivalent strain, average grain size, standard deviation of average grain size, recrystallized volume fraction, etc. of each objective on the final forming quality of the forging. The greater the influence on the final forming quality of the forging, the greater the weight value. The weights of each sub - objective satisfy Since the smaller the standard deviation of equivalent strain, the average grain size, and the standard deviation of average grain size are, the better, they are taken as positive numbers. Since the larger the average value of equivalent strain and the recrystallized volume fraction are, the better, they are taken as negative numbers. The smaller the final objective function Y is, the better the forging forming quality is. Through a Python script, corresponding data is automatically and batch extracted from the files storing simulation results through corresponding keywords. The extracted data includes the geometric dimension parameters of each sample and the equivalent strain and grain size of all grids at the end of the simulation. After processing the extracted data according to the required objectives, the finally obtained data is used as the dataset for establishing the prediction model. This dataset is a tabular file with multiple inputs and a single output.

[0160] (5) Through the dataset in step 4, establish a GA-SVR prediction model between the input variables and the objective function, predict the optimal value of the objective function, and obtain the best combination of pre-forging geometric parameters.

[0161] Call the dataset established in step 4, combine the support vector regression machine with the genetic algorithm, construct a prediction and optimization model for the deformation uniformity of superalloy disk forgings. Use 40 samples in the dataset as the training set for establishing the SVR model, and 10 samples as the test set. The SVR schematic diagram is as Figure 6 shown. Use the genetic algorithm to construct a GA-SVR prediction model, and use the genetic algorithm to optimize the parameters of SVR, the regularization parameter C, the kernel function parameter γ, and the error tolerance ε. The variation range of each parameter is set to (0, 100), (0, 1000), (0, 10). The setting of parameters in the genetic algorithm directly affects the optimization speed and quality of the algorithm. The parameter settings are determined as follows: the initial population size is 200, the maximum number of genetic generations is 200, the stochastic universal sampling function sus is selected, the two-point crossover probability is 0.8, and the discrete mutation probability is 0.007. Finally, predict the optimal value of the objective function to obtain the best combination of pre-forging geometric parameters.

[0162] The optimal value of the objective function Y = 3.05 predicted by the GA-SVR model, and the corresponding combination of pre-forging geometric parameters is that the lower die H1 is 60.64 mm, the lower die H2 is 24.47 mm, the lower die D1 is 23.88 mm, the upper die H3 is 22.93 mm, the upper die H4 is 11.71 mm, and the upper die D2 is 49.81 mm.

[0163] Draw this pre-forging in CAD software, design the corresponding pre-forging die, and finally import it into the finite element simulation software to simulate and verify this scheme. Compare the optimized simulation results with the simulation results before optimization. The equivalent strain distribution nephogram and grain size distribution nephogram after optimization and before optimization are as Figure 7 shown.

[0164] The above results show that the method proposed by the present invention can effectively perform multi-objective optimization design on the shape of the pre-forgings of superalloy disk parts. The shape of the pre-forgings designed by the method can, under the condition of ensuring complete filling, meet the requirements of uniform deformation, grain refinement and sufficient recrystallization of superalloy disk parts, providing a feasible method for improving the quality of superalloy disk parts.

[0165] The present invention can be applied to the rotary core components formed by forging process in aerospace engines. These core components work in extremely harsh environments for a long time, and relatively strict requirements are imposed on the comprehensive performance of such forgings after forming. The related products mainly include fan disks, compressor disks, turbine disks, journal shafts and rotor support conical walls, etc.

[0166] The geometric parameter combinations of each sample under Latin hypercube sampling and the corresponding objective functions are shown in Table 1. Therefore, the average value of Y in the samples is 3.36, the minimum value of the objective function of Sample 6 is Y = 3.14, and the corresponding geometric parameter combination of the pre-forging is that the lower die H1 is 49.78 mm, the lower die H2 is 20.81 mm, the lower die D1 is 31.20 mm, the upper die H3 is 22.93 mm, the upper die H4 is 13.32 mm, and the upper die D2 is 75.25 mm.

[0167]

[0168]

[0169] Table 1

[0170] As Figure 8 and Figure 9 shown, using the above data set to establish a GA-SVR model for prediction, the mean square error between the predicted value and the actual value of the objective function of the training samples is 0.13161, and the mean square error between the predicted value and the actual value of the objective function of the test samples is 0.1575. The GA-SVR model has a high fitting degree between the predicted pre-forging geometric parameters and the objective function.

[0171] Apply this prediction model to all solutions within the range of pre-forging geometric parameter values, and predict the optimal value of the objective function. The optimal value of the objective function Y = 3.07 predicted by using the GA-SVR model and the corresponding pre-forging geometric parameter combination are shown in Table 2 below.

[0172]

[0173] Table 2

[0174] The comparison of each sub-objective function between the optimal sample of the objective function obtained by Latin hypercube sampling and the optimal predicted geometric parameter combination of the objective function obtained by using the GA-SVR model is shown in Table 3

[0175]

[0176] Table 3

[0177] By comparing the objective function and each sub-objective function before and after optimization through this method, it can be known that the average value of Y in the sampling sample is 3.36, the minimum Y value in the sample is 3.14, and the Y value after the GA-SVR model is optimized is 3.07, showing a relatively large improvement compared with the sample. Moreover, each sub-objective function basically develops in the direction of improving the forging forming quality.

[0178] It should be noted that the embodiments of the present invention can be implemented through hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code is provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software, such as firmware.

[0179] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made by those skilled in the art within the technical scope disclosed by the present invention, as long as they are made within the spirit and principle of the present invention, should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the shape of a superalloy disk pre-forging, characterized in that It includes the following steps: Step 1: Simulate the disk-shaped part without the pre-forging process. According to the overall equivalent strain of the disk-shaped part after final forging, use the inverse simulation technology to quickly design the initial geometric shape of the pre-forged part, and design the shape of the pre-forging die according to the shape of the pre-forged part; Step 2: Combine the experimental design method and the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forged part in Step 1 as the key optimization parameters, and determine the value range of the optimization parameters; Step 3: Use the Latin hypercube sampling method to generate approximate random samples within the multi-parameter value range, set the number of samples extracted by the Latin hypercube sampling, and use the finite element simulation method to automatically batch simulate all random samples; Step 4: Establish an objective function describing the forming quality, and extract the data required in the simulation results of all samples in Step 3 as the data set for establishing the prediction model; Step 5: Establish a GA-SVR prediction model between the input variables and the objective function through the data set in Step 4, predict the optimal value of the objective function, and obtain the best combination of geometric parameters of the pre-forged part.

2. The method for optimizing the shape of the superalloy disk pre-forging as described in claim 1, wherein, The simulation of the disk-shaped part without the pre-forging process, according to the overall equivalent strain of the disk-shaped part after final forging, uses the inverse simulation technology to quickly design the initial geometric shape of the pre-forged part, and designs the shape of the pre-forging die according to the shape of the pre-forged part: Directly conduct finite element simulation on the cake blank. Through the forging equivalent strain results obtained by the finite element simulation, reverse track the positions of the small deformation area and the large deformation area in the original blank, and preliminarily design the shape of the pre-forged part. This method can eliminate the small deformation area and the large deformation area, which can significantly improve the deformation uniformity of the forging. The specific design method is as follows: First, divide the final forging into N rectangular regions with equal volume along the radial direction according to the deformation after final forging. If the shape of the final forging is more complex, the number of partitions for final forging is also more. Then, divide the cake blank into N rectangular regions with the same volume according to the partitions of the final forging; Write a script to extract the strain value ∈i of each grid after final forging and calculate the average effective strain value of the forging after final forging Then analyze the deformation of the large deformation area and small deformation area of the final forging, mark the small deformation area and large deformation area smaller than the forging, and use the reverse tracking method to track the positions of the small deformation area and large deformation area in the original blank. Select the points to be tracked after final forging through point tracking in the post-processing process of the finite element simulation software, inversely simulate and track the positions of the selected points in the cake blank, then reduce the material at the corresponding small deformation area of the cake blank, and add the reduced volume to the large deformation area along the radial direction. Finally, smooth the sharp parts of each different rectangular area to obtain the initial pre-forging shape; Average effect change value ∈i is the equivalent strain of the i-th grid, and n is the total number of grids; α and β are parameters for selecting the overall deformation amount, and the value range is 0.1 to 0.

5.

3. The method for optimizing the shape of a superalloy disk pre-forging as described in claim 1, wherein The combination of the experimental design method and the finite element method is used to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forged part in Step 1 as the key optimization parameters, and determine the value range of the optimization parameters; The geometric dimensions that mainly affect the effective strain of the forging during the pre-forging process are selected as the key geometric dimensions. The selection method is as follows: extract all the geometric parameters of the preliminarily designed pre-forging die in Step 1, then set all the parameters as the design variables of the finite element simulation, combine the experimental design method with the finite element method, and conduct a screening experimental design to evaluate the influence degree of the geometric dimensions of the die on the effective strain of the forging. In the post-processing of the finite element simulation, obtain the relationship diagram of the influence degree of all geometric parameters on the effective strain, and use the Spearman correlation coefficient (rs) to evaluate the influence degree of each geometric parameter variable on the effective strain. When the Spearman correlation coefficient is greater than q, select this parameter as the characteristic geometric parameter, where q is the threshold for the geometric parameter to have a greater influence on the effective strain, and the range is 0.1 - 0.

3. By this method, filter out the geometric dimensions with less influence on the effective strain to reduce the complexity and diversity of the data, and at the same time reduce the number of samples required for establishing the GA-SVR prediction model between the input variables and the objective function, and improve the prediction accuracy and generalization ability of the final model; and select the value range of the required optimization parameters according to the Spearman correlation coefficient of each characteristic geometric dimension; Spearman correlation coefficient Among them, dj is the range between the variable and the effective strain, j is the number of key geometric dimensions after screening, and k is the number of data points; r s The value ranges from -1 to +1; Positive correlation means that as one variable increases, the other variable also tends to increase; Negative correlation means that as one variable increases, the other variable tends to decrease; r s A value close to 0 indicates no significant monotonic relationship between the two, r s Values close to -1 or +1 indicate a strong correlation between the two; Range of values for the optimized parameter: L j ×(1 + rs) > L , j > L j ×(1 - rs); L j is the original dimension of the j-th critical geometric dimension, L , j is the j-th critical geometric dimension.

4. The method for optimizing the shape of the superalloy disk pre-forging as described in claim 1, wherein The Latin hypercube sampling method is used to generate approximate random samples within the multi-parameter value range, and set the number of samples extracted by the Latin hypercube sampling. Use the finite element simulation method to automatically batch simulate all random samples; Take Latin hypercube sampling to generate approximate random samples from the multi-characteristic geometric parameter distribution determined in Step 2. In this sampling method, divide the value range of each parameter into several probability intervals, randomly select a sample point within each interval, and combine these randomly selected sample points to form the final sample set. It ensures the uniform distribution of the samples in the multi-dimensional space, thus guaranteeing the diversity and representativeness of the samples, and is applicable to complex problems that require a large number of simulations and analyses; the required number of samples depends on multiple factors, including the complexity of the problem, the diversity of the data, the number of features, and the generalization ability of the model. If the complexity of the data set is relatively high, or the data is widely distributed in the input space, more samples are required to ensure that the model can capture all important features; After setting the required number of samples, the simulation samples can be submitted in the corresponding finite element simulation software for automatic batch simulation.

5. The method for optimizing the shape of a superalloy disk pre-forging as described in claim 1, characterized in that, Establish the objective function describing the forming quality, and extract all the required data in the simulation results of all samples in Step 3 as the data set for establishing the prediction model; The input variables required for the above step dataset are the key geometric parameters to be optimized. The objective function of the dataset is mainly the sub-objectives for evaluating the forging forming quality, including but not limited to: the average effective strain y1 describing the deformation degree of the final forging, the standard deviation of the effective strain y2 describing the deformation uniformity, the average grain size y3 describing the tissue uniformity, the standard deviation of the average grain size y4, and the recrystallized volume fraction y5. This involves the problem of multi-objective optimization. Now, each sub-objective is normalized to obtain y , i ∈(0, 1). Then, the normalized sub-objectives are integrated using linear weighting to obtain the objective function M is the number of selected sub-objectives, and the weight w of each sub-objective l The distribution needs to consider the influence of the average effective strain, the standard deviation of the effective strain, the average grain size, the standard deviation of the average grain size, and the recrystallized volume fraction on the final forming quality of the forging. The greater the influence on the final forming quality of the forging, the greater the weight value. The weights of each sub-objective satisfy Since the smaller the standard deviation of the effective strain, the average grain size, and the standard deviation of the average grain size, the better, they are taken as positive numbers. Since the larger the average effective strain and the recrystallized volume fraction, the better, they are taken as negative numbers. The smaller the final objective function Y, the better the forging forming quality. Through the python script, the corresponding data is automatically and batch extracted from the file storing the simulation results through the corresponding keywords. The extracted data includes the geometric size parameters of each sample and the effective strain and grain size of all grids at the end of the simulation. After processing the extracted data according to the required objectives, finally, the obtained data is used as the dataset for establishing the prediction model. This dataset is a multi-input and single-output table file; Average effect change value Standard deviation of equivalent strain Average grain size Standard deviation value of average grain size Recrystallized volume fraction ∈i is the effective strain of the i-th grid, is the average effective strain value, d i is the grain size of the i-th grid, is the average grain size, is the recrystallized volume fraction of the i-th grid, and n is the total number of grids; Min-Max Normalization Among them, y l is each sub-goal, y , l is the normalized data, and this method scales all data into the interval (0, 1).

6. The method for optimizing the shape of the superalloy disk pre-forging as described in claim 1, wherein Establish the GA-SVR prediction model between the input variables and the objective function through the data set in Step 4, predict the optimal value of the objective function, and obtain the best combination of pre-forging die geometric parameters; Call the data set established in step 4, combine the support vector regression machine with the genetic algorithm, and construct a prediction and optimization model for the deformation uniformity of superalloy disk forgings; Establish a sample set for the model: (a1, b1), (a2, b2)...(a p , b p )...(a t , b t ), a p , b p ∈R The objective function of SVR can be expressed as: Subject to the following conditions: Kernel function of SVR: where a p is the input variable value of the support vector regression machine, b p is the corresponding output variable value, φ(a p ) is the kernel function that maps the data to a high-dimensional space, ω is the weight vector, μ is the bias term, ξ p and are slack variables used to handle points outside the insensitive margin; C is the regularization parameter that controls the penalty intensity for margin violations; the kernel function parameters γ and the error tolerance ε; Search for the optimal solution of the problem through the selection, crossover, and mutation operations in the genetic algorithm iteration process; when optimizing the SVR model, use the genetic algorithm to optimize the parameters of SVR, including the regularization parameter C, the kernel function parameter γ, and the error tolerance ε. The change range of each parameter is set to (0, 100), (0, 1000), (0, 10). The optimization process is divided into the following steps Coding: Encode the parameters of SVR, the regularization parameter C, the kernel function parameter γ, and the error tolerance ε into chromosomes; Initializing the population: Randomly generate a set of solutions as the initial population; Fitness calculation: Use the cross-validation method to calculate the fitness of each individual, usually the prediction accuracy; Selection, crossover, mutation: The chromosomes in the population evolve towards the optimal values of C, γ, and ε through selection, crossover, and mutation, and generate a new population; Termination condition: Repeat the above process until the termination condition is met, reaching the maximum number of iterations or the fitness reaches a certain threshold; After using the genetic algorithm to iteratively optimize the parameters of the support vector regression model to obtain the optimal model, predict the optimal value of the objective function to obtain the best combination of pre-forging geometry parameters.

7. A superalloy disk preform shape optimization system for implementing the superalloy disk preform shape optimization method according to any one of claims 1-6, characterized in that, The superalloy disk pre-forging shape optimization system includes: A design module for simulating the disk parts without pre-forging process, quickly designing the initial geometry of the pre-forging according to the overall effective strain of the disk parts after final forging, and designing the pre-forging die shape according to the pre-forging shape; A screening module for combining the experimental design method with the finite element method to conduct screening experimental design to determine the characteristic geometric parameters of the pre-forging as the key optimization parameters, and determine the value range of the optimization parameters; A simulation and simulation module for the number of samples extracted by hypercube sampling, and automatically batch simulating all random samples using the finite element simulation method; An extraction module for establishing an objective function describing the forming quality, and extracting the required data from all sample simulation results as the data set for establishing the prediction model; A prediction module for establishing a GA-SVR prediction model between the input variables and the objective function through the data set, predicting the optimal value of the objective function, and obtaining the best combination of pre-forging geometry parameters.

8. A computer device, characterized in that, The computer device includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the superalloy disk pre-forging shape optimization method according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, which when executed by a processor causes the processor to execute the steps of the superalloy disk pre-forging shape optimization method according to any one of claims 1-6.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the superalloy disk pre-forging shape optimization system according to claim 7.

Images