Method for identifying key process characteristics of high-temperature alloy turbine disc forge piece and related device

Abstract

The invention relates to the field of high-temperature alloy forging, in particular to a high-temperature alloy turbine disc forge piece key process feature recognition method and a related device. The method comprises the following steps: firstly, constructing a multi-dimensional data set fusing raw material parameters, equipment process parameters and quality detection indexes, and carrying out feature pre-screening in a Filter stage by adopting the combination of a Pearson correlation coefficient and minimum redundancy and maximum correlation to obtain candidate feature subsets; on the basis, a Bayesian optimization strategy taking a random forest as an agent model is introduced, and iterative search is performed on a feature combination space through an expected improved acquisition function to obtain an optimal key process feature set. According to the method provided by the invention, the calculation efficiency is remarkably improved through screening of a hybrid key process feature recognition method, and meanwhile, the product performance consistency is remarkably improved by predicting the key process features and regulating and controlling the process parameters in the production process.

Description

Technical Field

[0001] This application relates to the field of superalloy forging, and particularly to a method for identifying key process characteristics of superalloy turbine disk forgings and related devices. Background Art

[0002] As one of the core load-bearing components of aero-engines, the mechanical properties, microstructure stability and finished product quality of superalloy turbine disks are directly related to the service life and safety of the whole machine. In the actual manufacturing process, the forging forming of turbine disks often involves multi-dimensional and strongly coupled complex process parameters, such as alloy composition ratio, initial tissue state, forging temperature, press pressure, die cooling efficiency, etc. These parameters have non-linear and time-varying effects on quality indicators such as final grain size, defect rate, and tensile strength at different stages. In addition, there are problems such as limited sample quantity, uncontrollable process fluctuations, and multiple collinearity among parameters at the production site. Traditional methods relying on expert experience or single-variable statistics are no longer sufficient to accurately identify key process factors in high-quality precision manufacturing.

[0003] In recent years, with the development of intelligent manufacturing and industrial big data, data-driven process feature recognition methods have gradually become an important means to improve the transparency and controllability of the forging process. However, traditional feature selection methods have insufficient high-dimensional data processing capabilities, and traditional Bayesian optimization initializes randomly, resulting in slow convergence speed and being easily trapped in local optimal traps. A single evaluation criterion cannot comprehensively measure the importance of features, which will lead to the omission of key characteristics. Summary of the Invention

[0004] In order to improve the accuracy and efficiency of key process characteristic recognition, this application provides a method for identifying key process characteristics of superalloy turbine disk forgings.

[0005] In a first aspect, a method for identifying key process characteristics of superalloy turbine disk forgings and related devices provided by this application adopts the following technical solutions: A method for identifying key process characteristics of superalloy turbine disk forgings includes the following steps: S1. Obtain the original data set and perform preprocessing to construct a feature set including raw material parameters, equipment process parameters and quality indicators; S2. Based on the Pearson correlation coefficient and the minimum redundancy maximum correlation double-index evaluation method, perform feature pre-screening in the Filter stage on the feature set to obtain a candidate feature subset; S3. Using the candidate feature subset as the initial input, construct an objective function based on a random forest model, and use an improved Bayesian optimization algorithm to iteratively optimize the objective function. Select the feature combination with the best performance in the Wrapper stage and output the key quality characteristic set.

[0006] By adopting the above technical solutions, by obtaining raw material parameters, equipment process parameters and final quality indicators, a multi-dimensional feature set is constructed, realizing the preparation of basic data for quality modeling; then, the Pearson correlation coefficient and the minimum redundancy maximum correlation double evaluation index are introduced for preliminary feature screening, which respectively measure the linear correlation and non-linear redundancy between features and target variables in principle, and can retain variables with high information gain while minimizing the feature dimension as much as possible; then, the objective function based on random forest and the Bayesian optimization search mechanism are adopted to achieve efficient global search in the feature combination space, further removing redundant features and improving the prediction accuracy.

[0007] Through the two-stage feature selection process, the input dimension required during model calculation is significantly reduced, thereby reducing the training time and resource consumption; at the same time, by quantitatively identifying key process variables, it provides an operable parameter basis for process window control and quality consistency management in the forging production process, effectively avoiding the uncertainty interference from the source of quality defects. The identification mechanism based on feature contribution degree proposed in this solution is feasible and applicable to forging scenarios with complex coupling relationships in engineering practice.

[0008] Optionally, the S2 includes: S21. Calculate the Pearson correlation coefficient between the quality feature set and the quality index feature set, sort them in descending order according to the absolute value of the Pearson correlation coefficient, and select the top several quality features as the first candidate subset of data to be processed; S22. Calculate the minimum redundancy maximum correlation between features and prediction targets, and select the top several features according to the minimum redundancy maximum correlation as the second candidate subset of data to be processed; S23. Take the union of the first candidate subset of data to be processed and the second candidate subset of data to be processed as the candidate feature subset.

[0009] By adopting the above technical solutions, the Pearson correlation coefficient is used to measure the linear relationship between each quality feature and the target index, and at the same time, the information redundancy between features is evaluated through the minimum redundancy maximum correlation (MRMR), constructing a candidate feature subset from two dimensions. This combination method can retain key information variables while removing features with high redundancy and strong mutual dependence. Its technical effect is reflected in improving the feature screening efficiency, reducing the interference introduced by redundant features, improving the prediction accuracy and feature interpretability of subsequent modeling, and further enhancing the ability to identify key influencing factors in the forging process.

[0010] Optionally, the S21 includes: S211. Determine the quality prediction target of the turbine disk forging and construct a feature set FeatureSet including raw material parameters and process-equipment parameters, and collect m groups of sample data composed of n groups of quality features to form a quality characteristic data set And expressed as Among them, x i is the sample data corresponding to each quality prediction target, x i ∈R 2 is an n-dimensional vector; y i is the quality parameter of turbine disk forging; S212. Set X = {x1, x2, ...} as the quality feature set, and Y = {y1, y2, ...} as the quality indicator feature set; S213. Calculate quality characteristics x i and quality index y j Linear correlation Where n is the number of samples; x m,i is the i-th eigenvalue of sample m; is the average value of the i-th feature; y m,j is the j-th eigenvalue of sample m; is the average value of the jth feature; r(i,j) represents the correlation between the i-th quality characteristic feature and the j-th quality indicator feature; S214. Sort in descending order by PCC absolute value, and retain the first several features of PCC absolute value as the first data candidate subset

[0011] By adopting the above technical solution, the linear correlation between each quality characteristic and each quality indicator is statistically analyzed to quantify the degree of single-variable influence on the target performance. Compared with the multivariate modeling process, PCC is simple to calculate and converges quickly, and can quickly eliminate redundant features that are almost irrelevant to the target in the unsupervised stage. Its effect is reflected in: without the need to build a complex model, it can efficiently reduce the feature dimension and significantly reduce the computational cost required for subsequent feature combination optimization (such as MRMR or Bayesian optimization). At the same time, it retains the key parameters that directly contribute to the change of the target indicator, making it easier to directly associate quality causes in the industrial process.

[0012] Optionally, the S22 includes: S221. For a given dataset D, select a feature subset S from it; S222. Calculate the feature correlation I(x) of the feature subset S based on the optimization objective function i ,y) and feature redundancy I(x i ,x j ), where the optimization objective function is P(x i ,y) is the joint probability; P(x i ) and P(y) are marginal probabilities; I(x i ,y) is the quality characteristic x iThe mutual information with the quality index y reflects the importance of the feature to the target variable; I(x i ,x j ) is the quality characteristic x i and quality characteristics x j Mutual information between features, measuring the redundancy between features; S223. Calculate the minimum redundancy and maximum correlation of feature subset S S224. Gradually select features that meet the maximum relevance and minimum redundancy conditions in a greedy manner until the preset number of features is reached or the stopping condition is met, thereby obtaining the optimal feature subset solved from the full feature set FeatureSet So that: Where P(·) is the performance evaluation function of the prediction model; S225. By calculating each quality characteristic x i The MRMR values between the product quality index y are sorted in descending order by the MRMR absolute value, and several features with the highest MRMR score are retained as the second data candidate subset.

[0013] By adopting the above technical solution, mutual information is used to measure the correlation between features and the target, while also evaluating information redundancy between features. An optimization objective function is constructed, and features are greedily selected based on their strong correlation with the target variable and low mutual redundancy. Compared with traditional correlation analysis, mutual information can capture nonlinear dependencies and avoid missing features with weak linear but strong nonlinear correlations. Its technical effect is to screen out feature combinations that have strong independence and significant impact on quality indicators, providing a balanced information distribution input foundation for subsequent modeling, thereby reducing modeling bias and improving the explanatory power of feature selection.

[0014] Optionally, the S23 includes the following steps: S231. Calculate each quality characteristic x separately i The PCC and MRMR values between the product quality index y, and sort the features in descending order according to the absolute value of PCC and the absolute value of MRMR, retaining the features of the first α and first β ratios, respectively, and recording them as sets and collection in, S232. For the independently selected set and collection Take the union to form a pre-screened feature subset

[0015] By adopting the above technical solutions, the linear correlation analysis based on the Pearson correlation coefficient (PCC) and the minimum redundancy maximum relevance (MRMR) method based on mutual information are used in combination to perform double-path screening on features. After separately retaining the feature subsets with the top rankings in PCC and MRMR and taking the union, it is possible to take into account the features capturing strong linear relationships and complex non-linear effects, and make up for the bias in the selection results of a single evaluation index. The effect of this method is to improve the coverage and effectiveness of candidate features, ensure that the pre-screened feature set finally used for modeling contains various types of key factors, enhance the initial quality of the search in the subsequent optimization process, and contribute to improving the representativeness and stability of the finally identified key process parameters.

[0016] Optionally, S3 includes the following steps: S31. Select the top k features with the highest scores from as the initial feature subset S32. Based on the initial feature subset construct a surrogate model, where the surrogate model is an objective function prediction model based on a random forest model, and the objective function is the mean squared error on the validation set; S33. Adopt a Bayesian optimization strategy and use the expected improvement acquisition function to select the next set of candidate feature subsets from the feature space; S34. Evaluate the objective function for the new feature subset and add its result to the historical evaluation record; update the surrogate model based on historical data and correct the prediction of the objective function using the maximum likelihood estimation method; S35. Repeat S33 and S34 until the termination condition is met. The termination condition includes reaching the set maximum number of iterations or the improvement amplitude of the objective function being less than the preset threshold in several consecutive iterations; S36. Output the feature subset with the optimal performance and its objective function value during the historical evaluation as the final key quality characteristic set.

[0017] By adopting the above technical solutions, a surrogate model of the objective function is constructed using a random forest, and the expected improvement (EI) acquisition function is used to balance exploration and exploitation in the feature space; the surrogate model is continuously updated through historical evaluation data to improve the search accuracy. This method avoids the problems of low efficiency or being easily trapped in local optima in exhaustive or greedy searches. Its effect is to effectively identify the feature combination that contributes the most to the model performance within a limited number of iterations, significantly reduce the interference of redundant inputs on the prediction accuracy, and provide the most representative set of process parameters for quality control.

[0018] In a second aspect, a computer device provided by the present application adopts the following technical solutions: A computer device, which includes: One or more processors; Memory; One or more applications, where the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to: Execute the above-mentioned method for identifying key process characteristics of superalloy turbine disk forgings.

[0019] In a third aspect, a computer-readable storage medium provided by the present application adopts the following technical solution: A computer-readable storage medium stores a computer program that can be loaded and executed by a processor to perform the above-mentioned method.

[0020] The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to achieve: The method for identifying key process characteristics of superalloy turbine disk forgings as described above.

[0021] The present application has the following beneficial effects: 1. In the Filter stage, the feature dimension is reduced by 50 - 70%, and in the Wrapper stage, the number of optimization iterations is reduced by 40%; 2. By combining Filter and Wrapper, the accuracy and stability of feature selection are improved; 3. The calculation efficiency is significantly improved through screening by the hybrid key process characteristic identification method; 4. By predicting key process characteristics and regulating process parameters during the production process, the product performance consistency is significantly improved. Description of the Drawings

[0022] Figure 1 Schematic diagram of the application environment of the method for identifying key process characteristics of superalloy turbine disk forgings in an embodiment of the present invention.

[0023] Figure 2 Flowchart of the method for identifying key process characteristics of superalloy turbine disk forgings in an embodiment of the present invention.

[0024] Figure 3 Schematic diagram of a computer device in an embodiment of the present invention. Detailed Description of the Embodiments

[0025] The following further details the present application in conjunction with the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0026] In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the inventive concept. As part of this specification, some of the figures in the accompanying drawings of the present disclosure represent structures and devices in block diagram form to avoid obscuring the disclosed principles. For clarity, not all features of actual specific implementations need to be described. In addition, the language used in the present disclosure has been mainly selected for readability and guidance purposes and may not have been selected to delimit or define the subject matter of the invention, and thus recourse is had to the required claims to determine such inventive subject matter. References in the present disclosure to "a specific implementation" or "specific implementations" mean that the particular features, structures, or characteristics described in connection with that specific implementation are included in at least one specific implementation, and multiple references to "a specific implementation" or "specific implementations" should not be construed as necessarily all referring to the same specific implementation.

[0027] Unless explicitly defined otherwise, the terms "a", "an", and "the" are not intended to refer to a singular entity but include the general category for which a particular example can be used for illustration. Thus, the use of the term "a" or "an" can mean any number of at least one, including "one", "one or more", "at least one", and "one or more than one". The term "or" means any one of the alternatives and any combination of the alternatives, including all the alternatives, unless the alternatives are explicitly indicated as mutually exclusive. The phrase "at least one of" when combined with a list of items refers to a single item in the list or any combination of the items in the list. The phrase does not require all of the listed items, unless explicitly so defined.

[0028] An embodiment of the present application discloses a method for identifying key process characteristics of a superalloy turbine disk forging. The method for identifying key process characteristics of a superalloy turbine disk forging provided by an embodiment of the present invention can be applied in an application environment such as Figure 1 In the application environment, the method for identifying key process characteristics of a superalloy turbine disk forging is applied in a system for identifying key process characteristics of a superalloy turbine disk forging. The system for identifying key process characteristics of a superalloy turbine disk forging includes a client and a server. Among them, the client communicates with the server through a network. The client is also called the user side and refers to a program that provides local services corresponding to the server. Further, the client is a computer-side program, an APP program of a smart device, or a third-party small program embedded in another APP. The client can be installed on, but not limited to, various computer devices such as personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0029] Refer to Figure 2, the key process feature identification method for the superalloy turbine disk forging includes the following steps S1 - S3.

[0030] S1. Obtain the original data set and perform preprocessing to construct a feature set including raw material parameters, equipment process parameters, and quality indicators.

[0031] This step first determines three types of variables related to the forging process of the superalloy turbine disk: raw material parameters, equipment process parameters, and quality indicators. Among them, raw material parameters include alloy composition (such as the percentage of Ni, Cr, Al), grain size grade, initial microstructure, etc.; process parameters cover forging temperature, holding time, pressure grade, equipment number, die cooling rate, etc.; and quality indicators select quantifiable final inspection indicators such as grain size uniformity, defect rate, hardness distribution, tensile strength, etc.

[0032] In the actual production environment, data is collected from the enterprise's MES system (Manufacturing Execution System) and quality inspection system. Assume that m batches of forging samples are collected, and each batch records n process and raw material parameters, as well as the corresponding target quality indicators, to construct a data set For example: The i - th sample data is (x i , y i ), where x i = [T i , P i , t i , C i respectively represent the forging temperature (in degrees Celsius), applied pressure (in MPa), holding time (in s), and Ni content (%) of this batch; y i is its final grain size grade.

[0033] After data collection, data preprocessing is carried out, including missing value filling (such as mean filling or interpolation method), outlier removal (such as the 3σ principle or IQR method), and dimension unification (such as standardization or normalization) to eliminate the scale differences between different physical quantities. For example, the holding time range is [60, 180] seconds, while the Ni content is only in [50, 55]%, and the standard score is needed to unify the scale. In addition, for nominal variables (such as equipment number), one - hot encoding can be used to facilitate entry into the subsequent model.

[0034] After the above processing, the final feature set FeatureSet = {x1, x2, …, x n}, which contains the structured input variables of all samples and serves as the input basis for feature selection in the Filter and Wrapper stages. The execution effect of S1 is to restore the key data patterns in the production process to the greatest extent, and to eliminate noise, anomalies, and redundant variables, providing a high-quality data basis for subsequent analysis and avoiding the invalidation or bias of the model due to the quality problems of the original data.

[0035] S2. Perform pre-screening of features in the Filter stage on the feature set based on the dual-index evaluation method of Pearson correlation coefficient and minimum redundancy maximum relevance, and obtain a candidate feature subset.

[0036] For the high-dimensional turbine disk forging product quality data set, the PM-IBO key quality characteristic identification method proposed in this application performs preliminary feature screening on the full feature set in the Filter stage and provides prior information for the subsequent Wrapper feature selection stage.

[0037] This step adopts a dual-index fusion strategy, using the Pearson correlation coefficient (Pearson Correlation Coefficient, PCC) based on linear correlation measurement and the minimum redundancy maximum relevance (Minimum Redundancy Maximum Relevance, MRMR) based on non-linear measurement as the measurement criteria for feature importance in the Filter stage.

[0038] Specifically, in one embodiment, S2 includes the following sub-steps S21 - S23.

[0039] S21. Calculate the Pearson correlation coefficient between the quality feature set and the quality index feature set, sort them in descending order according to the absolute value of the Pearson correlation coefficient, and select the top several quality features as the first data candidate subset.

[0040] Further, in one embodiment, S21 includes sub-steps S211 - S214.

[0041] S211. Determine the quality prediction target of the turbine disk forging and construct a feature set FeatureSet including raw material parameters and process-equipment parameters, and collect m groups of sample data composed of n groups of quality features to form a quality characteristic data set And it is expressed as where, x i is the sample data corresponding to each quality prediction target, x i ∈R 2 is an n-dimensional vector; y i is the quality parameter of the turbine disk forging; S212. Set X = {x1, x2,...} as the quality feature set and Y = {y1, y2,...} as the quality index feature set; S213. Calculate the quality characteristic x i and the linear correlation with the quality index y j where n is the number of samples; x m,i is the i-th eigenvalue of sample m; is the average value of the i-th characteristic; y m,j is the j-th eigenvalue of sample m; is the average value of the j-th characteristic; r(i, j) represents the correlation between the i-th quality characteristic feature and the j-th quality index feature; S214. Sort in descending order by the absolute value of PCC, and retain a number of features with the largest absolute values of PCC as the first data candidate subset

[0042] The Pearson correlation coefficient is used to measure the linear correlation between two variables, and its value range is [-1, 1]. The calculation of PCC is based on the ratio of covariance to standard deviation, which is defined as the mean of the dot product of two variables in the standardized space, thus reflecting the linear change relationship between variables. If the coefficient is 1, it means a perfect positive correlation; if it is -1, it is a perfect negative correlation; and 0 means there is no linear relationship between the two variables.

[0043] The formula derivation of PCC is as follows: Let X = {x1, x2,...} be the set of quality characteristics, and Y = {y1, y2,...} be the set of quality index characteristics. Then the correlation between the i-th quality characteristic x i in X and the j-th quality index y j in Y is defined as the ratio of the covariance Cov(x i , y j ) to the standard deviation , that is: The covariance Cov(x i , y j ) is: The standard deviations are respectively: Substitute the expressions of covariance and standard deviation into the original formula, and after simplification, the calculation formula of the linear correlation PCC between the quality characteristic x i and the quality index y j is obtained: In the formula: n is the number of samples; x m,i is the i-th eigenvalue of sample m; ​is the average value of the i-th feature; y m,j is the j-th eigenvalue of sample m; is the average value of the jth feature; r(i,j) represents the correlation between the i-th quality characteristic feature and the j-th quality indicator feature.

[0044] By calculating the PCC value between each quality characteristic x and product quality index y, sorting them in descending order by PCC absolute value, and retaining the first k1 features of PCC absolute value, recorded as set

[0045] S22. Calculate the minimum redundancy and maximum correlation between the features and the predicted target, and select the top several features based on the minimum redundancy and maximum correlation as the second data candidate subset.

[0046] MRMR achieves feature selection by balancing the maximum relevance of features to the target variable with the minimum redundancy between features. This step aims to extract a subset of features that are discriminative and non-redundant from high-dimensional data, thereby improving model performance.

[0047] Furthermore, in one embodiment, S22 includes the following sub-steps S221 - S225 .

[0048] S221. For a given dataset D, select a feature subset S from it; S222. Calculate the feature correlation I(x) of the feature subset S based on the optimization objective function i ,y) and feature redundancy I(x i ,x j ), where the optimization objective function is P(x i ,y) is the joint probability; P(x i ) and P(y) are marginal probabilities; I(x i ,y) is the quality characteristic x i The mutual information with the quality index y reflects the importance of the feature to the target variable; I(x i ,x j ) is the quality characteristic x i and quality characteristics x j Mutual information between features, measuring the redundancy between features; S223. Calculate the minimum redundancy and maximum correlation of feature subset S S224. Gradually select features that meet the maximum relevance and minimum redundancy conditions in a greedy manner until the preset number of features is reached or the stopping condition is met, thereby obtaining the optimal feature subset solved from the full feature set FeatureSet So that: Among them, P(·) is a prediction model performance evaluation function; S225. By calculating the MRMR value between each quality characteristic x i and the product quality index y, arranging them in descending order according to the absolute value of MRMR, and retaining the first several features with the MRMR score, which is denoted as the second data candidate subset

[0049] The theoretical basis of MRMR includes two aspects: correlation and redundancy. Correlation refers to the degree of dependence between features and the target variable, which is usually measured by mutual information (MI); redundancy refers to the similarity or information overlap degree between features, which can also be quantified by mutual information. Its optimization objective function can be represented by the above formula as shown.

[0050] In the formula: P(x i , y) is the joint probability; P(x i ) and P(y) are marginal probabilities.

[0051] In the formula: S is the selected feature subset; I(x i , y) is the mutual information between the quality characteristic x i and the quality index y, reflecting the importance of the feature to the target variable; I(x i , x j ) is the mutual information between the quality characteristic x i and the quality characteristic x j , measuring the redundancy between features.

[0052] The first term in the formula represents the maximum relevance (Max-Relevance), which evaluates the discriminative ability of features by summing the mutual information between features and the target variable; the second term represents the minimum redundancy (Min-Redundancy), which controls the redundancy degree of information by reducing the mutual information between features. Finally, the MRMR algorithm gradually selects features that meet the conditions of maximum relevance and minimum redundancy in a greedy manner until the preset number of features is reached or the stop condition is satisfied. The MRMR method is especially suitable for feature selection of high-dimensional datasets because of its good theoretical interpretability and low computational complexity.

[0053] By calculating the MRMR value between each quality characteristic x i and the product quality index y, arranging them in descending order according to the absolute value of MRMR, and retaining the first k2 features with the MRMR score, which is denoted as the set

[0054] S23. Take the union of the first candidate data subset and the second candidate data subset as the candidate feature subset.

[0055] Further, in one embodiment, S23 includes the following sub-steps S231 - S232.

[0056] S231. Calculate the PCC and MRMR values between each quality characteristic x i and the product quality index y respectively, and sort the features in descending order according to the absolute values of PCC and MRMR. Retain the features in the top α and top β proportions respectively, and denote them as sets and set respectively. Among them,

[0057] S232. Take the union of the independently selected sets and set to form a pre-screened feature subset

[0058] Due to the limitation that the traditional Filter method uses a single evaluation criterion, which is not comprehensive in evaluating features in feature selection and lacks important feature information. This embodiment proposes an evaluation strategy for the Filter stage that fuses dual evaluation indicators, using the linear correlation coefficient PCC and the information entropy index MRMR coefficient as evaluation indicators for feature importance. Calculate the PCC value and MRMR value between each quality characteristic x i and the product quality index y and sort them in descending order. Select a certain number of feature subsets in proportion and take their union to obtain a feature subset after initial screening. For example, the feature screening process based on PCC - MRMR is divided into the following two steps.

[0059] Step1: Independent screening. Calculate the PCC and MRMR values between each quality characteristic x i and the product quality index y respectively, and sort the features in descending order according to the absolute values of PCC and MRMR. Retain the features in the top α and top β proportions respectively, and denote them as sets and set Step2: Result merging. Take the union of the independently selected sets and set to perform result merging to form a pre-screened feature subset. Improve feature diversity through the dual evaluation criteria of PCC and MRMR, and control information redundancy at the same time.

[0060] The feature subset initially screened in the Filter stage will provide prior information of the feature subset with initial weights for the subsequent Wrapper feature selection stage.

[0061] S3. Using the candidate feature subset as the initial input, construct an objective function based on the random forest model, and use the improved Bayesian optimization algorithm to iteratively optimize the objective function. Select the feature combination with the best performance in the Wrapper stage and output the key quality characteristic set.

[0062] The Wrapper feature selection method based on the learner feedback mechanism can usually effectively screen out the feature subset that makes the learner perform better by traversing and searching the global feature space. The Wrapper method mainly uses intelligent optimization algorithms to achieve iterative optimization of the feature subset. Among them, algorithms such as the Bayesian optimization framework based on the probability model, the genetic evolution algorithm inspired by bionics, the particle swarm algorithm and the ant colony algorithm based on intelligent driving are representative. Because of their good global optimization ability, they are widely used in feature selection. Bayesian optimization has significant advantages in feature selection. It effectively models the performance of the feature subset through the surrogate model, balances exploration and exploitation using the acquisition function, significantly reduces the number of objective function evaluations, and reduces the computational cost. Its global optimization ability and strong adaptability make it particularly suitable for scenarios with high-dimensional feature spaces and limited computing resources. In this embodiment, the Bayesian optimization algorithm is improved based on the characteristics of CTQ identification. The Improved Bayesian Optimization (IBO) is used in the Wrapper stage, combined with the prior information in the Filter stage. Specifically, in one embodiment, S3 is implemented through the following sub-steps S31 - S36.

[0063] S31. Select the top k features with the highest score rankings from as the initial feature subset

[0064] S32. Based on the initial feature subset construct a surrogate model, where the surrogate model is an objective function prediction model based on the random forest model, and the objective function is the mean squared error on the validation set.

[0065] S33. Adopt the Bayesian optimization strategy and use the expected improvement acquisition function to select the next group of candidate feature subsets from the feature space.

[0066] S34. Evaluate the objective function of the new feature subset and add its result to the historical evaluation record; update the surrogate model based on the historical data and correct the prediction of the objective function using the maximum likelihood estimation method.

[0067] S35. Repeat the execution of S33 and S34 until the termination condition is met. The termination condition includes reaching the set maximum number of iterations or the improvement amplitude of the objective function being less than the preset threshold in several consecutive iterations.

[0068] S36. Output the feature subset with the optimal performance and its objective function value during the historical evaluation process as the final set of critical quality characteristics.

[0069] For example, the main process of the algorithm is as follows.

[0070] First, initialize the feature subset. The initial feature subset of traditional Bayesian optimization is usually randomly generated, resulting in low search efficiency in the initial stage. This paper uses the feature scoring results of the Filter stage to select the top k features with the highest scores from as the initial feature subset to provide prior information for optimization.

[0071] Definition: Suppose the feature scores calculated in the Filter stage are S = {s1, s2,... s n}, and the feature ranking results are R = {f (1) , f (2) ,... f (n)}, where f (1) is the feature with the highest score. The initial feature subset selects the top k features, where k = αn and α ∈ (0, 1) is the proportionality coefficient. The initial feature subset is as follows: This improvement significantly reduces the search space in the initial stage and improves the optimization efficiency.

[0072] Then, construct the objective function and the surrogate model. Objective function: The random forest model is an ensemble learning algorithm that mainly realizes the prediction of the target variable through the ensemble of multiple decision trees. For the feature subset the random forest trains multiple decision trees T1, T2,..., T m , and combines voting or averaging methods to output the predicted value. Its objective function value can be defined as the mean squared error of the validation set In the formula: y i is the actual value, and is the model predicted value.

[0073] Surrogate model: The Gaussian Process (GP) is used to construct the objective function because it can provide the prediction mean and variance Support the calculation of acquisition functions. The hyperparameters θ of GP include kernel function parameters, which are updated by maximum likelihood estimation: In the formula: is the historical evaluation record.

[0074] Then optimize the acquisition function. The core of Bayesian optimization is to globally model the objective function using a surrogate model and select the next candidate feature subset by optimizing the acquisition function. In this paper, the Expected Improvement (EI) is adopted as the acquisition function, which depends on the predicted mean and variance of GP: In the formula: is the objective function value of the current optimal solution.

[0075] EI can be analytically expressed as: In the formula: Φ(·) and φ(·) are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0076] By optimizing EI(X), the exploration of unknown regions and the exploitation of the current optimal solution can be balanced, and the search efficiency can be improved.

[0077] The algorithm needs to be iteratively optimized until the stopping condition is reached. First, evaluate the new feature subset, select the candidate subset through the acquisition function Calculate its objective function value Then update the surrogate model. Add to Re-fit the GP model until the stopping condition is reached: ① Reach the maximum number of iterations T max ; ② Reach the objective function improvement threshold, the improvement amount of the validation set MSE for k consecutive iterations (e.g., k = 5) (e.g., ∈ = 0.001); ③ The acquisition function converges, and the standard deviation σ of the EI value EI < δ (e.g., δ = 0.01).

[0078] Finally, output the optimal feature subset. The algorithm finally outputs the feature subset with the best performance in the historical evaluation and its objective function value That is: This feature subset significantly reduces the data dimension while retaining the key information with the strongest predictive ability for the target variable.

[0079] It should be understood that the sequence numbers of the steps in the above embodiments do not indicate the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0080] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as Figure 3 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for data related to the key process feature recognition method of the superalloy turbine disk forging. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a key process feature recognition method for superalloy turbine disk forgings.

[0081] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the key process feature recognition method of the superalloy turbine disk forging in the above embodiment.

[0082] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, it implements the key process feature recognition method of the superalloy turbine disk forging in the above embodiment.

[0083] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. This computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments of the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0084] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for identifying key process characteristics of a superalloy turbine disk forging, characterized in that It includes the following steps: S1. Obtain the original data set and perform preprocessing to construct a feature set including raw material parameters, equipment process parameters, and quality indicators; S2. Based on the Pearson correlation coefficient and the minimum redundancy maximum relevance double-index evaluation method, perform feature pre-screening in the Filter stage for the feature set to obtain a candidate feature subset; S3. Using the candidate feature subset as the initial input, construct an objective function based on the random forest model, and use an improved Bayesian optimization algorithm to iteratively optimize the objective function. Select the feature combination with the optimal performance in the Wrapper stage and output the key quality characteristic set.

2. The method for identifying key process characteristics of a superalloy turbine disk forging according to claim 1, wherein The S2 includes: S21. Calculate the Pearson correlation coefficient between the quality feature set and the quality indicator feature set, sort them in descending order according to the absolute value of the Pearson correlation coefficient, and select the top several quality features as the first data candidate subset; S22. Calculate the minimum redundancy maximum relevance between the features and the prediction target, and select the top several features according to the minimum redundancy maximum relevance as the second data candidate subset; S23. Take the union of the first data candidate subset and the second data candidate subset as the candidate feature subset.

3. The method for identifying key process characteristics of a superalloy turbine disk forging according to claim 2, wherein The S21 includes: S211. Determine the quality prediction objective of the turbine disk forging and construct a feature set FeatureSet including raw material parameters and process-equipment parameters, and collect m groups of sample data composed of n groups of quality features to form a quality characteristic data set and represent it as where, x i is the sample data corresponding to each quality prediction objective, and x i ∈R 2 is an n-dimensional vector; y i is the quality parameter of the turbine disk forging; S212. Set X = {x1, x2,...} as the quality feature set and Y = {y1, y2,...} as the quality indicator feature set; S213. Calculate quality characteristics x i and quality index y j Linear correlation Where n is the number of samples; x m,i is the i-th eigenvalue of sample m; is the average value of the i-th feature; y m,j is the j-th eigenvalue of sample m; is the average value of the jth feature; r(i,j) represents the correlation between the i-th quality characteristic feature and the j-th quality indicator feature; S214. Sort in descending order by the absolute value of PCC, and retain a number of features before the absolute value of PCC as the first data candidate subset 4. The method for identifying key process characteristics of a superalloy turbine disk forging according to claim 3, wherein The S22 includes: S221. For a given data set D, select a feature subset S from it; S222. Calculate the feature correlation I(x i , y) and feature redundancy I(x i , x j ) of the feature subset S based on the optimization objective function, where the optimization objective function is P(x i , y) is the joint probability; P(x i ) and P(y) are marginal probabilities; I(x i , y) is the mutual information between the quality feature x i and the quality indicator y, reflecting the importance of the feature to the target variable; I(x i , x j ) is the mutual information between the quality feature x i and the quality feature x j , measuring the redundancy between features; S223. Calculate the minimum redundancy and maximum relevance of the feature subset S S224. Gradually select features that meet the maximum relevance and minimum redundancy conditions in a greedy manner until the preset number of features is reached or the stopping condition is met, thereby obtaining the optimal feature subset solved from the full feature set FeatureSet So that: Among them, P(·) is the prediction model performance evaluation function; S225. By calculating the MRMR value between each quality characteristic x i and the product quality index y, arranging them in descending order according to the absolute value of MRMR, and retaining several features with the top MRMR scores, which is denoted as the second data candidate subset 5. The method for identifying key process characteristics of a superalloy turbine disk forging according to claim 4, characterized in that, The S23 includes the following steps: S231. Calculate each quality characteristic x i and the product quality index y i The PCC and MRMR values between them are calculated, and the features are sorted in descending order according to the absolute values of PCC and MRMR. The features with the top α and top β proportions are retained respectively, and are denoted as sets and set Among them, S232. Take the union of the independently selected set and the set to form a pre-screened feature subset 6. The method for identifying key process characteristics of a superalloy turbine disk forging according to claim 5, wherein The S3 includes the following steps: S31. Select the top k features with the highest scores from as the initial feature subset S32. Based on the initial feature subset Construct a surrogate model, where the surrogate model is an objective function prediction model based on a random forest model, and the objective function is the mean squared error on the validation set; S33. Adopt the Bayesian optimization strategy, use the expected improvement acquisition function to select the next group of candidate feature subsets from the feature space; S34. Evaluate the objective function for the new feature subset and add its result to the historical evaluation record; update the surrogate model based on the historical data, and use the maximum likelihood estimation method to correct the prediction of the objective function; S35. Repeat S33 and S34 until the termination condition is met. The termination condition includes reaching the set maximum number of iterations or the improvement amplitude of the objective function being less than the preset threshold in several consecutive iterations; S36. Output the feature subset with the optimal performance and its objective function value in the historical evaluation process as the final key quality characteristic set.

7. A computer device, characterized in that, It includes: One or more processors; A memory; One or more applications, where the one or more applications are stored in the memory and are configured to be executed by the one or more processors. The one or more programs are configured to: execute the method for identifying key process characteristics of a superalloy turbine disk forging according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, a code set, or an instruction set. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement: the method for identifying key process characteristics of a superalloy turbine disk forging according to any one of claims 1 to 6.

Images