Hole making process parameter autonomous optimization method and device, electronic equipment and storage medium

Abstract

The invention provides a hole making process parameter autonomous optimization method and device, electronic equipment and a storage medium. The method comprises the steps that a sample data set is established; obtaining an optimal machine learning regression model according to the autonomous selector; obtaining a Pareto leading edge of the hole making process parameter group and a corresponding quality target value; and obtaining an optimal hole making process parameter group. According to the method, the device, the electronic equipment and the storage medium provided by the embodiment of the invention, high-precision prediction can be provided in a relatively short time, and time and cost required by repeated experiments in a traditional optimization method are reduced; the prediction optimization process can adapt to dynamically changing input data, that is, the model can be adjusted according to new data, the fitness model can be updated at any time, and the data analysis depth and the autonomy level of the whole prediction optimization process are remarkably improved; the requirement for multi-objective optimization is flexibly met, the quality defects in the hole making process are effectively reduced, and therefore the hole making quality and efficiency are remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of aviation engineering, and particularly relates to a method, device, electronic device, and storage medium for autonomous optimization of hole-making process parameters. Background Art

[0002] With the wide application of hole-making technology in the aviation field, the generation of hole-making defects not only reduces the service life of aviation products but also may lead to serious safety hazards. During the hole-making process, the hole-making process parameters have a significant impact on the hole-making quality.

[0003] In the actual application in the aviation field, the optimization mode of hole-making process parameters mostly relies on the empirical judgment of operators. This method not only makes it difficult to find an ideal balance among multiple key objectives such as precision control, production efficiency, material utilization rate, and cost control, but also in actual operation, the adjustment process of process parameters often lacks sufficient autonomy and intelligent support, resulting in low efficiency of the entire optimization process and being difficult to meet the urgent needs of modern manufacturing for high-quality, high-efficiency, and low-cost production. Therefore, exploring new methods for optimizing hole-making process parameters has become the key to improving the level of hole-making technology. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to provide a method, device, electronic device, and storage medium for autonomous optimization of hole-making process parameters.

[0005] In a first aspect, an embodiment of the present invention provides a method for autonomous optimization of hole-making process parameters, the method comprising:

[0006] Establishing a sample data set with the hole-making process parameter group as the feature and the characterization quantification data of the corresponding hole-making quality defect as the label;

[0007] Inputting the sample data set into the best machine learning regression model autonomous selector to obtain the best machine learning regression model;

[0008] Using the best machine learning regression model to autonomously predict and optimize the Pareto front of the hole-making process parameter group and the corresponding quality target values;

[0009] Making an autonomous decision on the Pareto front of the hole-making process parameter group and the corresponding quality target values to obtain the optimal hole-making process parameter group.

[0010] Optionally, the inputting the sample data set into the best machine learning regression model autonomous selector to obtain the best machine learning regression model specifically includes:

[0011] Setting the regression model performance evaluation index of the best machine learning regression model autonomous selector and the hyperparameter optimization space corresponding to multiple regression models therein;

[0012] For each set of hyperparameter combinations in the hyperparameter optimization space, train multiple regression models on the sample data set and evaluate the performance of the regression models according to the performance evaluation metrics, and output the best hyperparameter combination;

[0013] Use the best hyperparameter combination to perform cross-validation on the sample data set, and evaluate the performance of multiple regression models trained with the best hyperparameter combination to determine the best machine learning regression model.

[0014] Optionally, using the best machine learning regression model to autonomously predict and optimize the Pareto front of the hole-making process parameter set and the corresponding quality target value, specifically including:

[0015] Independently initialize multiple small populations, and randomly generate corresponding numbers of individuals for each small population to represent a set of hole-making process parameters, where each small population corresponds to a different exploration area of the solution space;

[0016] Select individuals from each small population according to the results of fast non-dominated sorting and crowding distance calculation to generate a parental small population, and merge the generated multiple parental small populations into a parental population;

[0017] Perform crossover and mutation operations on the parental population to update the parental population, and repeat the above steps until the preset termination condition is met;

[0018] According to the updated parental population, output the Pareto front of the corresponding hole-making process parameter set and the corresponding quality target value.

[0019] Optionally, the step of selecting individuals from each small population according to the results of fast non-dominated sorting and crowding distance calculation to generate a parental small population specifically includes:

[0020] For the individuals in each small population, use the fitness model to evaluate the target values output by the individuals;

[0021] According to the target values, judge the dominance relationship between the individuals in each small population and other individuals in the small population, and perform fast non-dominated sorting for each small population to divide the individuals into different non-dominated ranks;

[0022] For each small population, calculate the crowding distance of each individual in each non-dominated rank;

[0023] Select individuals according to the results of fast non-dominated sorting and crowding distance calculation to generate a parental small population.

[0024] Optionally, the step of performing crossover and mutation operations on the parental population to update the parental population specifically includes:

[0025] Select a certain proportion of individuals in the parental population for crossover to generate new individuals;

[0026] Randomly select individuals in the parental population for mutation according to an adaptive mutation probability;

[0027] Merge the offspring population generated by crossover and mutation with the parental population, and complete the update of the parental population by selecting individuals based on the elite strategy in the merged population.

[0028] Optionally, the adaptive mutation probability is determined according to the distance between individuals.

[0029] Optionally, the autonomous decision-making on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group specifically includes:

[0030] Perform an initial screening of candidate solutions for the optimal hole-making process parameter group on the Pareto front;

[0031] Take the predicted value of the quality target as the evaluation factor of quality, and through fuzzy grey relational analysis, convert it into two groups of grey relational degrees;

[0032] After subjective and objective fusion weighting, linearly weight the two groups of grey relational degrees into a group of comprehensive grey relational degrees, which is used as the comprehensive evaluation index of hole-making quality, and obtain the optimal hole-making process parameter group according to the sorting result of the quality comprehensive evaluation index.

[0033] In a second aspect, an embodiment of the present invention provides a device for autonomous optimization of hole-making process parameters, characterized in that the device includes:

[0034] A data acquisition module, configured to establish a sample data set with the hole-making process parameter group as the feature and the characterization quantification data of the corresponding hole-making quality defect as the label;

[0035] A model determination module, configured to input the sample data set into the best machine learning regression model autonomous selector to obtain the best machine learning regression model;

[0036] A prediction and optimization module, configured to autonomously predict and optimize using the best machine learning regression model to obtain the Pareto front of the hole-making process parameter group and the corresponding quality target value;

[0037] A parameter decision module, configured to perform autonomous decision-making on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group.

[0038] In a third aspect, an embodiment of the present invention provides an electronic device, characterized in that it includes:

[0039] One or more processors;

[0040] A memory for storing one or more programs;

[0041] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method described in the first aspect.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having executable instructions stored thereon, characterized in that when the executable instructions are executed by a processor, the processor is caused to execute the method described in the first aspect.

[0043] The method, device, electronic device, and storage medium for autonomous optimization of hole-making process parameters provided by the embodiments of the present invention aim to combine data-driven, advanced algorithms, and statistical analysis techniques to achieve autonomous prediction, optimization, and decision-making of the optimal process parameter set for multiple quality objectives. This autonomous prediction and optimization method can provide high-precision predictions in a relatively short time, reducing the time and cost required for repeated experiments in traditional optimization methods; the prediction and optimization process can adapt to dynamically changing input data, that is, the model can be adjusted according to new data and the fitness model can be updated at any time, significantly improving the depth of data analysis and the autonomous level of the entire prediction and optimization process; it can flexibly adapt to the needs of multi-objective optimization, effectively reducing quality defects in the hole-making process, thereby significantly improving the hole-making quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below.

[0045] Figure 1 It is a schematic flowchart of the method for autonomous optimization of hole-making process parameters provided by the embodiment of the present invention;

[0046] Figure 2 It is a schematic flowchart of the method for determining the best machine learning regression model provided by the embodiment of the present invention;

[0047] Figure 3 It is a schematic flowchart of the method for determining the Pareto front of the hole-making process parameter set and the corresponding quality target values provided by the embodiment of the present invention;

[0048] Figure 4 It is a schematic flowchart of the method for generating a small parent population provided by the embodiment of the present invention;

[0049] Figure 5 It is a schematic flowchart of the method for updating the parent population provided by the embodiment of the present invention;

[0050] Figure 6Schematic diagram of comparison between Pareto fronts before and after algorithm improvement and their quality target data provided by embodiments of the present invention;

[0051] Figure 7 Schematic flowchart of the decision-making method for the optimal hole-making process parameter set provided by embodiments of the present invention;

[0052] Figure 8 Schematic diagram of the structure of the hole-making process parameter self-optimization device provided by embodiments of the present invention;

[0053] Figure 9 Schematic diagram of the structure of the electronic device provided by embodiments of the present invention. Detailed implementation manners

[0054] Next, the technical solutions in the embodiments of the present application will be described with reference to the accompanying drawings in the embodiments of the present application.

[0055] Similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, terms such as "first" and "second" are only used for differential description and cannot be construed as indicating or implying relative importance.

[0056] With the wide application of hole-making technology in the aviation field, the generation of hole-making defects not only reduces the service life of aviation products but also may lead to serious safety hazards. During the hole-making process, the hole-making process parameters have a significant impact on the hole-making quality.

[0057] In the actual application in the aviation field, the optimization mode of hole-making process parameters mostly relies on the empirical judgment of operators. This method not only makes it difficult to find an ideal balance among multiple key objectives such as precision control, production efficiency, material utilization rate, and cost control, but also in actual operation, the adjustment process of process parameters often lacks sufficient autonomy and intelligent support, resulting in low efficiency of the entire optimization process and being difficult to meet the urgent needs of modern manufacturing for high-quality, high-efficiency, and low-cost production. Therefore, exploring new hole-making process parameter optimization methods has become the key to improving the hole-making process level. Based on this, embodiments of the present invention provide a hole-making process parameter self-optimization method, device, electronic device, and storage medium. The appended Figure 1 shows a schematic flowchart of the hole-making process parameter self-optimization method provided by embodiments of the present invention.

[0058] Step S110, establish a sample data set with the hole-making process parameter set as the feature and the characterization quantification data of the corresponding hole-making quality defect as the label.

[0059] Specifically, in the embodiments of the present invention, a hole-making experiment is conducted to obtain the characterization and quantification data of hole-making quality defects and the corresponding hole-making process parameter data. Taking the process parameters as features and the corresponding defect data as the quality target, a data set is established, and the data is standardized. Subsequently, the data set is randomly divided into a training set and a test set, where 80% of the data is used for training and 20% of the data is used for testing.

[0060] Step S120: Input the sample data set into the best machine learning regression model autonomous selector to obtain the best machine learning regression model.

[0061] Specifically, as shown in the appendix Figure 2 The method for determining the best machine learning regression model provided in step S120 is specifically implemented through steps S121 to S123.

[0062] Step S121: Set the regression model performance evaluation index of the best machine learning regression model autonomous selector and the hyperparameter optimization space corresponding to multiple regression models therein.

[0063] Step S122: For each group of hyperparameter combinations in the hyperparameter optimization space, train multiple regression models on the sample data set and evaluate the performance of the regression models according to the performance evaluation index, and output the best hyperparameter combination.

[0064] Step S123: Use the best hyperparameter combination to perform cross-validation on the sample data set, evaluate the performance of multiple regression models trained with the best hyperparameter combination, and determine the best machine learning regression model.

[0065] Specifically, in the embodiments of the present invention, a best machine learning regression model autonomous selector needs to be established. This selector can autonomously train multiple machine learning regression models and evaluate their performance. Through the comparison of the performance evaluation results, the model with the best performance is autonomously selected. Multiple machine learning regression models include models such as least squares linear regression, decision tree regression, random forest regression, support vector machine regression, neural network regression, and K-nearest neighbor regression.

[0066] In this step, first, the performance evaluation index of the regression model needs to be set, which can be evaluation indexes such as root mean square error, coefficient of determination, average determination error, and relative error. For the convenience of explanation, the root mean square error and the coefficient of determination are used as examples of the evaluation indexes for explanation here.

[0067] The root mean square error (RMSE) represents the standard deviation of the prediction error. The smaller the RMSE value, the better the prediction performance of the model. The formula is

[0068]

[0069] Among them, n is the number of samples, and y i is the actual observed value, and

[0070] The coefficient of determination R 2 is a statistic used to evaluate the ability of a regression model to explain the variability of data.

[0071] R 2 The value range of R is (0, 1). The closer the R 2 value is to 1, the better the model fits the data. The formula is

[0072]

[0073] Among them, n is the number of samples, and y i is the actual observed value, is the model predicted value, and

[0074] Next, the embodiments of the present invention need to define the hyperparameter optimization space of the regression model.

[0075] The model needs to pre-define hyperparameters during training, and these hyperparameters have an important impact on the performance of the model. Through grid search, different combinations of hyperparameters can be systematically tried to find the combination of hyperparameters that makes the model perform best. After finding the best combination of hyperparameters, the prediction ability of the model will be significantly improved, thereby obtaining higher accuracy and better generalization ability.

[0076] The goal of least squares linear regression is to accurately define the model parameters by minimizing the loss function, that is, the mean square error between the predicted value and the true value. This process does not depend on hyperparameters such as the learning rate and regularization that need to be adjusted, and directly performs calculations. Therefore, only the definition of the hyperparameter optimization space of the remaining five regression models is described.

[0077] Set the hyperparameter optimization space of the decision tree regression model as shown in the following table.

[0078]

[0079] Set the hyperparameter optimization space of the random forest regression model as shown in the following table.

[0080]

[0081] Set the hyperparameter optimization space of the support vector machine regression model as shown in the following table.

[0082]

[0083] Set the hyperparameter optimization space for the neural network regression model as shown in the following table.

[0084]

[0085] Set the hyperparameter optimization space for the K-nearest neighbor regression model as shown in the following table.

[0086]

[0087] After determining the performance evaluation metrics and hyperparameter optimization space of the regression model, the embodiments of the present invention need to search for the best hyperparameter combination. Specifically, for each group of hyperparameter combinations in the hyperparameter optimization space, train the regression model on the training set and evaluate its performance, and output the hyperparameter combination that makes the model performance the best as the best hyperparameter combination.

[0088] The following shows the best hyperparameter combinations of five regression models. Since two quality objectives are specifically used as examples in this example, the best hyperparameter combinations for the two quality objectives are listed in the table respectively.

[0089]

[0090] After determining the best hyperparameter combinations for the two quality objectives of various machine learning regression models, perform 10-fold cross-validation on the training set using the best hyperparameter combinations, and output the average root mean square error value and coefficient of determination value of the 10-fold cross-validation to evaluate the performance of the regression model trained with the best hyperparameter combinations. Exemplarily, after verification, train the final regression model using the complete training set data. Complete the target prediction on the test set and output the corresponding performance evaluation results as shown in the following table.

[0091]

[0092] Finally, the embodiments of the present invention need to compare the performance evaluation results of six regression models as shown in the following table to determine the best machine learning regression model.

[0093]

[0094] For the two quality objectives in this example, the random forest regression model has the best prediction performance.

[0095] The algorithm provided in this step is an ensemble learning-based algorithm, which performs regression analysis by combining multiple base learners (decision trees in this example). It improves the prediction accuracy and stability of the model and reduces the risk of overfitting by making predictions on different decision tree models and then averaging these prediction results.

[0096] The training of each decision tree adopts the bagging method and random feature selection, that is, multiple samples are randomly drawn from the training set (with replacement) to create a different training set for each decision tree, increasing the diversity of the model; randomly select a subset of features for splitting internal nodes instead of using all features, reducing the correlation between trees and improving the generalization ability of the overall model; select the best splitting feature, split at the node, and continue to build until the preset stopping condition is reached.

[0097] In random forest regression, assuming there are M trees and the input feature matrix X, the prediction of each tree for the input data X is f m (X) (where m = 1, 2,... M), then, the final prediction result of the random forest can be expressed as

[0098] Step S130, use the best machine learning regression model to autonomously predict and optimize the Pareto front of the hole-making process parameter group and the corresponding quality target value.

[0099] In the embodiment of the present invention, the non-dominated sorting genetic algorithm NSGA-II (Non-dominated Sorting Genetic Algorithm-II) with an elite strategy is improved, and the best machine learning regression model is used to predict the target values of the individuals (each individual is a process parameter group) generated during the optimization process. Through a data-driven method, autonomously predict and optimize the Pareto front (the candidate solution set of the optimal process parameter group) of the process parameter group and the corresponding quality target value.

[0100] In order to improve the performance of the algorithm, two improvements are proposed:

[0101] The first improvement is to propose a multi-local population exploration strategy for the problem of local concentration of population individuals that may occur in the initialization stage of NSGA-II. This strategy generates the required initial population in the form of independently initializing multiple small populations to reduce the probability of individuals concentrating in the same area. Subsequently, fitness evaluation and fast non-dominated sorting are performed on each small population to select individuals to generate the parent small population. Finally, these parent small populations are merged to generate the parent population, providing a basis for subsequent crossover and mutation operations. This strategy can effectively avoid the premature convergence of the initial population on a local solution, thus improving the exploration ability of the population and increasing the possibility of finding the global optimal solution.

[0102] The second improvement is to propose an adaptive mutation probability adjustment strategy for the problem of insufficient population diversity that may occur in the mutation stage of NSGA-II. This strategy dynamically measures the diversity of the population during the iterative process, evaluates the distribution of the population in the objective space in real time, and adaptively adjusts the mutation probability according to the population diversity. This strategy can promote the generation of a more diverse population, thus expanding the exploration space of the population and increasing the possibility of finding the global optimal solution.

[0103] Specifically, as shown in the appendix Figure 3 The Pareto front of the set of hole-making process parameters provided in step S130 and the corresponding quality target value determination method are specifically implemented through steps S131 to S134.

[0104] Step S131, independently initialize multiple small populations, and randomly generate for each small population a corresponding number of individuals used to represent a set of hole-making process parameters, where each small population corresponds to a different exploration area of the solution space.

[0105] Exemplarily, adopt a multi-local population exploration strategy. First, independently initialize 5 small populations (each containing 10 individuals). Each small population can be randomly generated in different areas of the solution space, thereby reducing the possibility that all individuals are concentrated in a certain specific area, and store the individuals of these 5 small populations in a data structure containing multiple lists or arrays.

[0106] Step S132, select individuals from each small population according to the results of fast non-dominated sorting and crowding distance calculation to generate a parental small population, and merge the generated multiple parental small populations into a parental population.

[0107] Specifically, as shown in the appendix Figure 4 The method for generating the parental small population provided in step S132 is specifically implemented through steps S132-1 to S132-4.

[0108] Step S132-1, for the individuals in each small population, use the fitness model to evaluate the target value output by the individual.

[0109] For the individuals in each small population, the fitness model can be used to evaluate their performance on each target. Exemplarily, in this example, the fitness model is a trained optimal random forest regression model, which predicts the target value of the individual and stores the target value of each individual for subsequent fast non-dominated sorting and selection.

[0110] Step S132-2, according to the target value, judge the dominance relationship between the individuals in each small population and other individuals in the small population, and perform fast non-dominated sorting for each small population, dividing the individuals into different non-dominated levels.

[0111] This step requires establishing the dominance relationship among individuals in the small population. For each individual x in the small population i , determine its dominance relationship with other individuals in the small population. Individual x i dominates x j (i≠j), that is, if x i is superior to x j in all objectives, or is superior to x j in some objectives and not inferior to x j in other objectives

[0112] After determining the dominance relationship, fast non-dominated sorting can be performed for each small population to divide the individuals into different non-dominated levels, and the individuals in each level are non-dominant to other individuals

[0113] Step S132-3: For each small population, calculate the crowding distance of each individual in each non-dominated level

[0114] For each small population, calculate the crowding distance of each individual in each non-dominated level to evaluate the diversity of the population. The crowding degree represents the density between individuals, and a higher crowding degree means that the position of the individual in this non-dominated level is relatively sparse

[0115] Step S132-4: Select individuals according to the results of fast non-dominated sorting and crowding distance calculation to generate the parent small population

[0116] This step adopts the tournament selection strategy. For each small population, select individuals according to the results of fast non-dominated sorting and crowding distance calculation to generate the parent small population. When performing tournament selection, give priority to selecting individuals with a lower non-dominated level, and within the same level, give priority to selecting individuals with a higher crowding degree

[0117] Exemplarily, then the 5 generated parent small populations can be merged into 1 parent population

[0118] Step S133: Perform crossover and mutation operations on the parent population to update the parent population, and repeat the above steps until the preset termination condition is met

[0119] Specifically, as shown in the appendix Figure 5 , the method for updating the parent population provided in step S133 is specifically implemented through steps S133-1 to S133-2

[0120] Step S133-1: Select a certain proportion of individuals in the parent population for crossover to generate new individuals

[0121] Step S133-2: Randomly select individuals in the parent population for mutation according to the adaptive mutation probability

[0122] Step S133-3: Combine the offspring population generated after crossover and mutation with the parent population, and complete the update of the parent population by selecting individuals based on the elitist strategy in the combined population.

[0123] In the embodiment of the present invention, it is necessary to perform crossover and mutation processing on the generated parent population. Among them, the crossover operation refers to using the single-point crossover method, that is, first randomly selecting a crossover point from two parent individuals. The new first offspring individual is copied from the starting part of the first parent individual to the crossover point, and then from the crossover point of the second parent individual to the end part. The new second offspring individual is the opposite, first copied from the starting part of the second parent individual to the crossover point, and then from the crossover point of the first parent individual to the end part.

[0124] The mutation operation refers to performing a mutation operation on the individuals in the parent population, that is, randomly selecting individuals for mutation operation according to the set mutation probability to increase the diversity of the population. The mutation operation in this step adopts an adaptive adjustment of the mutation probability strategy. It is necessary to initialize the mutation probability P m = 0.01, and then use the Euclidean distance to calculate the distance between each pair of individuals in the population. For a pair of individuals x i and x j in the population, the distance d ij is

[0125] d ij = ||x i - x j ||

[0126] Measure the diversity of the population by calculating the average distance between all individuals. For N individuals in the population, the average distance D avg is

[0127]

[0128] For the average distance D avg , set the diversity threshold T d = 0.1. Adjust the mutation probability P m according to the diversity. If D avg is lower than T d , increase the mutation probability, and set the mutation probability increase step size to ΔP m = 0.01, and the maximum mutation probability is

[0129] Perform fast non-dominated sorting again in the new population and calculate the new crowding distance.

[0130] This step is obtained by improving the NSGA-II multi-objective optimization algorithm. This algorithm has a better encounter ratio than the earlier version of the NSGA algorithm. NSGA-II mainly overcomes the disadvantages of high computational complexity, lack of elitism, and the need for users to define sharing parameters. Therefore, when selecting elite individuals in this step, individuals with a lower non-dominated rank in the combined population are preferentially selected; within the same non-dominated rank, individuals with a larger crowding distance are preferentially selected. This selection mechanism not only ensures the retention of elite solutions but also improves the diversity of individual selection. During the iterative execution of steps S131 to S133, the termination condition is to reach the maximum number of iterations g = 50 or stop when the fitness change of the initialized population is less than the threshold preset by the user, thus completing the autonomous prediction optimization.

[0131] Step S134: According to the updated parent population, output the Pareto front of the corresponding hole-making process parameter group and the corresponding quality target value.

[0132] After completing the autonomous prediction optimization, the Pareto front can be obtained, and the corresponding quality target prediction results are output. The following table shows the Pareto front output by the improved algorithm and the corresponding target space data.

[0133]

[0134] Hypervolume is an important performance index in multi-objective optimization. It measures the size of the area covered by the solution set in the target space. The larger the hypervolume value, the wider the area covered by the solution set in the target space, and the higher the diversity and quality of the solution set. After calculation, the hypervolume values of the algorithms before and after improvement are 802.78 and 2548.07 respectively. The hypervolume value after improvement increases, indicating that a better balance is achieved among the objectives, and the diversity and quality of the Pareto front are significantly improved. As shown in the appendix Figure 6 is a schematic diagram comparing the Pareto front and its quality target data before and after the algorithm improvement.

[0135] Step S140: Make an autonomous decision on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group.

[0136] By performing fuzzy grey relational analysis on the predicted values of multiple quality targets corresponding to the Pareto front, they are transformed into multiple grey relational degrees. After weighted by the subjective and objective fusion, the multiple grey relational degrees are linearly weighted into a comprehensive grey relational degree, which is used as a comprehensive evaluation index of quality. Finally, according to the comprehensive evaluation ranking result, the optimal process parameter group is autonomously determined.

[0137] Specifically, as shown in the appendix Figure 7 the decision-making method of the optimal hole-making process parameter group provided in step S140 is specifically implemented through steps S141 to S143.

[0138] Step S141, conduct the initial screening of the candidate solutions of the optimal hole-making process parameter groups for the Pareto front.

[0139] Take the average values of the experimental data of quality target 1 and quality target 2 as the initial screening conditions, that is, the process parameter groups with the predicted value of quality target 1 greater than 41.96 and the predicted value of quality target 2 greater than 13.92 are no longer candidates. The screened candidate solutions and the corresponding quality target prediction results are shown in the following table.

[0140]

[0141] Step S142, take the predicted values of the quality targets as the quality evaluation factors, and through fuzzy grey relational analysis, convert them into two groups of grey relational degrees.

[0142] Fuzzy grey relational analysis solves the complex problems faced in making decisions in fuzzy and information-incomplete systems. Although fuzzy processing can already convert two evaluation factors into dimensionless membership degrees that have no influence on the weight effect, the spaces to which these two membership degrees belong are different, and there is no comparability between them. Grey relational analysis aims to convert the membership degree data of two evaluation factors into the same grey relational degree space, so that there is a certain comparability between them.

[0143] Specifically, it is necessary to first establish an evaluation factor matrix D

[0144]

[0145] Perform fuzzy processing on the evaluation factor matrix D to establish a fuzzy membership degree matrix R. Calculate the membership degrees for the evaluation factor values in the evaluation factor matrix D through the membership degree function.

[0146]

[0147] Since the two evaluation factors have the characteristic of expecting small values, that is, the smaller the evaluation factor value, the better the hole quality, the following membership degree function formula is used for calculation.

[0148]

[0149] Among them, d ij is the j-th evaluation factor value of the i-th individual (i = 1, 2, 3; j = 1, 2); max d j , min d j are the maximum and minimum values of the j-th evaluation factor among all individuals.

[0150] The smaller the evaluation factor value is, the closer the calculated membership degree is to 1; the larger the evaluation factor value is, the closer the calculated membership degree is to 0. The membership degree data can be compared with the evaluation factor data in terms of the numerical size within the matrix column, and the same conclusion about the size comparison between individual quality objectives can be obtained. That is, after solving the dimension problem through fuzzy processing, there is no loss of data information.

[0151] Perform grey relational analysis. Take the maximum membership degree of each group of evaluation factors as the reference value to establish a reference sequence, and take all the membership degrees of each group of evaluation factors as the comparison values to establish a comparison sequence. The deviation value between the reference sequence and the comparison sequence is obtained according to the following formula

[0152] Δ ij =|c ij -r ij |

[0153] where, Δ ij is the deviation value calculated from the j-th reference value and comparison value of the i-th individual; c ij is the j-th reference value of the i-th individual.

[0154] Use the membership degree data to establish a reference sequence and a comparison sequence. That is, take the maximum membership degree of each group of evaluation factors as the reference value to establish a reference sequence, which is to select the minimum value of each quality objective as the reference; take all the membership degrees of each group of evaluation factors as the comparison values to establish a comparison sequence, which is to select all the values of each quality objective and calculate the deviation from the minimum value.

[0155] Calculate the grey relational degree, that is, quantify the degree of association between the reference sequence and the comparison sequence

[0156]

[0157] where, ξ ij is the j-th grey relational degree of the i-th individual; Δ minj , Δ maxj are the minimum and maximum values of the j-th deviation sequence; μ is the resolution coefficient between 0 and 1, generally taking 0.5.

[0158] Quantify the degree of association between the reference sequence and the comparison sequence, that is, quantify the degree of association between the situation with the smallest quality objective and all individual situations. The larger the grey relational degree of an individual is, the closer the individual situation is to the situation with the smallest quality objective, that is, the smaller the quality objective result of this individual. Therefore, the smaller the evaluation factor value is, the larger the transformed grey relational degree is; the larger the evaluation factor value is, the smaller the transformed grey relational degree is. The transformed grey relational degree values are shown in the following table.

[0159]

[0160] By analyzing the grey relational degree data, conclusions on the comparison of the magnitudes among individual quality objectives consistent with the evaluation factor data can be obtained and analyzed, and they belong to the grey relational degree space, excluding the influence of the inherent spatial characteristics of their respective membership spaces. The entire process of transforming the evaluation factor values into fuzzy membership degrees through fuzzy processing and then into grey relational degrees through grey relational analysis does not change the original information of the data, but makes the data of the two quality objectives comparable, and also makes the subsequent linear weighted sum calculation scientific and effective.

[0161] Step S143: After the subjective and objective fusion for weight assignment, linearly weight the two groups of grey relational degrees into a group of comprehensive grey relational degrees as the comprehensive evaluation index for the hole-making quality, and obtain the optimal hole-making process parameter group according to the sorting result of the quality comprehensive evaluation index.

[0162] In this step, the methods for setting weights are generally divided into two types: subjective methods and objective methods. Subjective methods rely on experience and judgment, while objective methods rely on data analysis and statistical models. In this example, the method for setting weights by fusing subjective and objective weights is adopted, aiming to make full use of the two data information sources of subjective and objective, and avoid the limitations of a single method. Through the fusion of subjective and objective weights, the setting of weights is more comprehensive and representative, reducing the risk of decision-making deviation.

[0163] When subjectively assigning weights, considering the influence of quality objectives on material properties and the overall quality of holes, let the subjective weight w 1-subjective of the grey relational degree of quality objective 1 be 0.6, and the subjective weight w 2-subjective of the grey relational degree of quality objective 2 be 0.4.

[0164] When objectively assigning weights, the entropy weight method is adopted. This method is an objective weight setting method based on information theory. It measures the importance of each index by calculating the information entropy of each index, and then assigns weights to the evaluation indexes in the decision-making. Entropy is an important concept in physics and information theory, used to measure the uncertainty or chaos degree of a system. In the entropy weight method, entropy is used to quantify the uncertainty of index information. If the value of an index changes greatly, it indicates that the index provides more discrimination information, so its entropy value is lower and its importance is higher; on the contrary, if the value of an index changes little, it indicates that the index provides less discrimination information, so its entropy value is higher and its importance is lower. The entropy weight method is completely based on data, and can objectively assign weights to each evaluation index by quantifying the information entropy of each index.

[0165] Standardize the grey relational degree data. The standardization method adopted is Min-Max standardization, which can scale the data to the range of 0 to 1.

[0166]

[0167] Among them, ξ ij is the j-th grey correlation degree of the i-th individual; max ξ j , min ξ j are the maximum and minimum values of the j-th grey correlation degree among all individuals; δ ij is the value of the j-th quality evaluation index of the i-th individual after standardization.

[0168] Calculate the proportion p ij occupied by δ ij

[0169]

[0170] Among them, m is the number of individuals for comprehensive evaluation of hole quality, and m = 3.

[0171] Calculate the entropy value e j

[0172]

[0173] Among them, if p ij = 0, then define p ij ·ln p ij = 0.

[0174] Calculate the objective weight w j-objective :

[0175]

[0176] Among them, n is the number of quality evaluation indicators, that is, n = 2.

[0177] To sum up, when objectively assigning weights, the objective weight w 1-objective of the grey correlation degree of quality target 1 is calculated to be 0.51, and the objective weight w 2-objective of the grey correlation degree of quality target 2 is 0.49.

[0178] Fuse the subjective weight and the objective weight into a comprehensive weight

[0179]

[0180] Calculate the comprehensive weight w 1 of the grey correlation degree of quality target 1 to be 0.61, and the comprehensive weight w 2 of the grey correlation degree of quality target 2 to be 0.39. According to the comprehensive weight, linearly weight the two grey correlation degrees to obtain the comprehensive evaluation index S i

[0181] S i = w 1·ξ i1 +w 2 ·ξ i2

[0182] The comprehensive evaluation index has the characteristic of maximizing, that is, the larger the index value, the better the hole quality. Since the entropy weight method used for objective weighting is completely data-based, as the quality target data for carrying out the quality comprehensive evaluation changes, the objective weight will be recalculated automatically, and the comprehensive weight will be updated accordingly. The final quality comprehensive evaluation index and sorting results are shown in the following table.

[0183]

[0184] According to the quality comprehensive sorting results, individual 1 is automatically determined as the optimal process parameter group, as shown in the following table.

[0185]

[0186] The automatic optimization method for hole-making process parameters provided by the embodiment of the present invention aims to combine data-driven, advanced algorithms and statistical analysis techniques to achieve automatic prediction, optimization and decision-making of the optimal process parameter group for multiple quality targets. This automatic prediction and optimization method can provide high-precision prediction in a relatively short time, reducing the time and cost required for repeated experiments in traditional optimization methods; the prediction and optimization process can adapt to dynamically changing input data, that is, the model can be adjusted according to new data and update the fitness model at any time, significantly improving the depth of data analysis and the automatic level of the entire prediction and optimization process; it can flexibly adapt to the needs of multi-objective optimization, effectively reducing quality defects in the hole-making process, thereby significantly improving the hole-making quality and efficiency.

[0187] Based on any of the above embodiments, the Figure 8 structural schematic diagram of the automatic optimization device for hole-making process parameters provided by the embodiment of the present invention is shown, and the specific content is as follows:

[0188] The data acquisition module 810 is used to establish a sample data set with the hole-making process parameter group as the feature and the characterization and quantification data of the corresponding hole-making quality defect as the label;

[0189] The model determination module 820 is used to input the sample data set into the best machine learning regression model automatic selector to obtain the best machine learning regression model;

[0190] The prediction and optimization module 830 is used to automatically predict and optimize the Pareto front of the hole-making process parameter group and the corresponding quality target value by using the best machine learning regression model;

[0191] The parameter decision module 840 is used to automatically make a decision on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group.

[0192] The hole-making process parameter self-optimization device provided by the embodiment of the present invention aims to combine data-driven, advanced algorithms and statistical analysis techniques to achieve autonomous prediction, optimization and decision-making of the optimal process parameter set for multiple quality objectives. This autonomous prediction and optimization method can provide high-precision prediction in a relatively short time, reducing the time and cost required for repeated experiments in traditional optimization methods; the prediction and optimization process can adapt to dynamically changing input data, that is, the model can be adjusted according to new data and update the fitness model at any time, significantly improving the depth of data analysis and the autonomous level of the entire prediction and optimization process; it can flexibly adapt to the requirements of multi-objective optimization, effectively reducing quality defects in the hole-making process, thereby significantly improving the hole-making quality and efficiency.

[0193] Based on any of the above embodiments, the attached Figure 9 shows a schematic physical structure diagram of an electronic device provided by the embodiment of the present invention. The electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940. Among them, the processor 910, the communication interface 920, and the memory 930 complete mutual communication through the communication bus 940. The processor 910 can call the logical instructions in the memory 930 to execute the following methods:

[0194] Establish a sample data set with the hole-making process parameter set as the feature and the characterization quantization data of the corresponding hole-making quality defect as the label;

[0195] Input the sample data set into the best machine learning regression model automatic selector to obtain the best machine learning regression model;

[0196] Use the best machine learning regression model to autonomously predict and optimize to obtain the Pareto front of the hole-making process parameter set and the corresponding quality target values;

[0197] Autonomously make a decision on the Pareto front of the hole-making process parameter set and the corresponding quality target values to obtain the optimal hole-making process parameter set.

[0198] In addition, when the logical instructions in the above-mentioned memory 930 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0199] On the other hand, the embodiments of the present invention also provide a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the methods provided in the above embodiments, for example, including:

[0200] Establishing a sample data set with the hole-making process parameter group as the feature and the characterization quantization data of the corresponding hole-making quality defect as the label;

[0201] Inputting the sample data set into the optimal machine learning regression model autonomous selector to obtain the optimal machine learning regression model;

[0202] Using the optimal machine learning regression model to autonomously predict and optimize to obtain the Pareto front of the hole-making process parameter group and the corresponding quality target value;

[0203] Making an autonomous decision on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group.

[0204] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0205] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0206] Finally, it should be noted that 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.

Claims

1. A method for autonomous optimization of hole making process parameters, characterized in that: The method comprises: A sample data set is established with the hole-making process parameter group as a feature and the corresponding characterization quantitative data of hole-making quality defects as a label; Inputting the sample data set into the best machine learning regression model autonomous selector to obtain the best machine learning regression model; Using the optimal machine learning regression model to autonomously predict and optimize to obtain the Pareto frontier of the hole-making process parameter group and the corresponding quality target value; An autonomous decision is made on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain an optimal hole-making process parameter group.

2. The method for autonomous optimization of hole making process parameters according to claim 1, characterized in that: The step of inputting the sample data set into the optimal machine learning regression model autonomous selector to obtain the optimal machine learning regression model specifically includes: Setting the regression model performance evaluation index of the optimal machine learning regression model autonomous selector and the hyperparameter optimization space corresponding to the multiple regression models; For each set of hyperparameter combinations in the hyperparameter optimization space, multiple regression models are trained on the sample data set and regression model performance is evaluated according to performance evaluation indicators, and the optimal hyperparameter combination is output; The optimal hyperparameter combination is used to perform cross validation on the sample data set, and the performance of multiple regression models trained with the optimal hyperparameter combination is evaluated to determine the optimal machine learning regression model.

3. The method for autonomous optimization of hole making process parameters according to claim 1, characterized in that: The autonomous prediction and optimization using the optimal machine learning regression model to obtain the Pareto frontier of the hole-making process parameter group and the corresponding quality target value specifically includes: Initialize multiple small populations independently, and randomly generate a corresponding number of individuals for each small population to characterize a set of hole-making process parameters, wherein each small population corresponds to a different exploration area of ​​the solution space; According to the results of fast non-dominated sorting and crowding distance calculation, individuals are selected from each small population to generate a parent small population, and the generated multiple parent small populations are merged into a parent population; Performing crossover and mutation operations on the parent population to update the parent population, and repeating the above steps until a preset termination condition is met; According to the updated parent population, the Pareto frontier of the corresponding hole-making process parameter group and the corresponding quality target value are output.

4. The method for autonomous optimization of hole making process parameters according to claim 3, characterized in that: The step of selecting individuals from each small population to generate a parent small population according to the fast non-dominated sorting and crowding distance calculation results specifically includes: For each individual in the small population, the fitness model is used to evaluate the target value output by the individual; According to the target value, the dominance relationship between the individuals in each small population and other individuals in the small population is determined, and a fast non-dominated sort is performed on each small population to divide the individuals into different non-dominated levels; For each small population, the crowding distance of each individual in each non-dominated class is calculated; Individuals are selected to generate a parent subpopulation based on the results of fast non-dominated sorting and crowding distance calculation.

5. The method for autonomous optimization of hole making process parameters according to claim 3, characterized in that: The performing crossover and mutation operations on the parent population to update the parent population specifically includes: Select a certain proportion of individuals in the parent population for crossover to produce new individuals; According to the adaptive mutation probability, individuals in the parent population are randomly selected for mutation; The offspring population generated after crossover and mutation is merged with the parent population, and the parent population is updated after individuals are selected in the merged population based on the elite strategy.

6. The method for autonomous optimization of hole making process parameters according to claim 5, characterized in that: The adaptive mutation probability is determined according to the distance between individuals.

7. The method for autonomous optimization of hole making process parameters according to claim 1, characterized in that: The autonomous decision-making of the Pareto front and the corresponding quality target value of the hole-making process parameter group to obtain the optimal hole-making process parameter group specifically includes: Perform the initial screening of candidate solutions for the optimal hole-making process parameter group on the Pareto front; The predicted value of the quality target is used as the quality evaluation factor and converted into two groups of grey relational degrees through fuzzy grey relational analysis. After subjective and objective fusion weighting, the two groups of grey correlation degrees are linearly weighted into a group of comprehensive grey correlation degrees as the comprehensive evaluation index of hole making quality. The optimal hole making process parameter group is obtained according to the ranking results of the comprehensive quality evaluation index.

8. An autonomous optimization device for hole making process parameters, characterized in that: The device comprises: A data acquisition module is used to establish a sample data set with a hole-making process parameter group as a feature and with corresponding quantitative data representing hole-making quality defects as a label; A model determination module, used for inputting the sample data set into the optimal machine learning regression model autonomous selector to obtain the optimal machine learning regression model; A prediction and optimization module, used to use the optimal machine learning regression model to autonomously predict and optimize to obtain the Pareto frontier of the hole-making process parameter group and the corresponding quality target value; The parameter decision module is used to make an autonomous decision on the Pareto front of the hole-making process parameter group and the corresponding quality target value to obtain the optimal hole-making process parameter group.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors execute the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, characterized in that: When the executable instructions are executed by a processor, the processor executes the method according to any one of claims 1 to 7.

Images