Multi-target grinding decision method and system based on dynamic clustering and hyper-parameter optimization

Abstract

The invention discloses a multi-target grinding decision-making method and system based on dynamic clustering and hyper-parameter optimization, and relates to the field of grinding process parameter optimization, and the method comprises the steps: obtaining input feature data of a grinding process, and obtaining a standardized feature set; constructing a multi-target prediction model based on the normalized feature set; the multi-target prediction model comprises a regression model and a classification model, and the regression model is trained through a bagging method of dynamic clustering and target priority adjustment; hyper-parameters of the multi-target prediction model are adjusted through cooperation of a coevolution algorithm and Bayesian optimization, and an optimized multi-target prediction model is generated; and inputting a to-be-optimized grinding process parameter range into the optimized multi-target prediction model, predicting multi-target performance indexes, and selecting an optimal process parameter combination through comprehensive scores. Through the cooperation mechanism of dynamic K-Means clustering, target priority adjustment and the co-evolutionary algorithm, the machining precision, stability and efficiency of the ultra-thin piece grinding process are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of optimization of grinding process parameters, and particularly to a multi-objective grinding decision-making method and system based on dynamic clustering and hyperparameter optimization. Background Art

[0002] The statements in this part only provide background technical information related to the present disclosure, and do not necessarily constitute prior art.

[0003] The flexible contact reference surface grinding process processes the ultra-thin sheet reference surface through a flexible gasket and then performs multiple conventional grindings to reduce stress and control the end jump and flatness accuracy. Its effect highly depends on the dynamic adjustment of parameters such as the gasket thickness, the depth of cut, and the number of grinding times.

[0004] In the prior art, the selection of grinding process parameters mostly depends on manual experience or trial-and-error adjustment, with low efficiency and poor consistency. Some studies have introduced algorithms such as multi-objective optimization, Bayesian optimization, or hybrid machine learning, but their data grouping mostly uses static methods such as time windows or cross-validation, lacking a dynamic adjustment mechanism, being difficult to capture non-linear interactions and adapt to the dynamic optimization requirements of multi-objective coupling relationships, and being difficult to meet the complex requirements of multi-objective collaborative optimization such as the end jump difference value, stress difference value, flatness, and positive and negative proportion of end jump in the flexible reference surface grinding process. Summary of the Invention

[0005] To overcome the limitations of the prior art that rely on manual experience and lack of self-adaptability, the present invention provides a multi-objective grinding decision-making method and system based on dynamic clustering and hyperparameter optimization. The method first optimizes data grouping and target priorities through dynamic K-Means clustering, and then efficiently adjusts the hyperparameters of the multi-objective prediction model through a co-evolution algorithm combined with Bayesian optimization, accurately predicting multi-objective performance indicators such as the end jump difference value, stress difference value, flatness, and positive and negative proportion of end jump. The optimal process parameter combination is screened through a comprehensive scoring formula, significantly improving the machining accuracy, stability, and efficiency of the ultra-thin sheet grinding process, and ensuring that the positive and negative proportion of end jump meets the requirements of positive maximum and negative minimum.

[0006] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, the present invention provides a multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization, including: Obtaining input feature data of the grinding process and performing preprocessing to obtain a normalized feature set; Based on the normalized feature set, constructing a multi-objective prediction model; the multi-objective prediction model includes a regression model and a classification model, and training the regression model through a bagging method with dynamic clustering and target priority adjustment; Cooperatively adjust the hyperparameters of the multi-objective prediction model through the co-evolution algorithm and Bayesian optimization to generate an optimized multi-objective prediction model; Input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict the multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

[0007] In a further technical solution, the input feature data includes the thickness of the workpiece before grinding, the target thickness, the thickness of the shim, the number of grinding back-and-forth times, and the depth of cut.

[0008] In a further technical solution, the multi-objective prediction model includes a regression model for the end-jump difference, stress difference, and flatness, and a classification model for the positive and negative proportion of the end-jump.

[0009] In a further technical solution, the RUSBoost algorithm is used to train the classification model for the positive and negative proportion of the end-jump.

[0010] In a further technical solution, the training of the regression model by the bagging method with dynamic clustering and target priority adjustment is specifically as follows: Perform initial clustering on the normalized feature set through the clustering algorithm, and dynamically adjust the cluster center according to the multi-objective prediction error; Determine the current priority target based on the target priority, sample data from the cluster with large error contribution based on the current priority target, replace the data in the bagging method to generate multiple new sub-datasets, train decision trees based on the new sub-datasets, and integrate multiple decision trees to obtain the trained regression model; Predict the end-jump difference, stress difference, and flatness based on the trained regression model.

[0011] In a further technical solution, the dynamic adjustment of the cluster center according to the multi-objective prediction error is represented by the formula:

[0012] where, represents the adjustment vector of the th cluster center, represents the adjustment step size, represents the set of data points of the th cluster center, represents the prediction error of the th data point, represents the mean of the th intra-cluster error, represents the feature vector of the th data point, represents the center vector of the th cluster.

[0013] In a further technical solution, the calculation formula for the target priority is represented as:

[0014] Among them, represents the priority of the th target, represents the normalized error of the th target, represents the sum of the normalized errors of all targets, represents the number of targets. ;

[0015] In a second aspect, the present invention provides a multi-objective grinding decision-making system based on dynamic clustering and hyperparameter optimization, including: A data acquisition module, which is configured to: acquire input feature data of a grinding process and perform preprocessing to obtain a normalized feature set; A model construction module, which is configured to: construct a multi-objective prediction model based on the normalized feature set; the multi-objective prediction model includes a regression model and a classification model, and the regression model is trained by a bagging method with dynamic clustering and target priority adjustment; A model optimization module, which is configured to: co-evolve the hyperparameters of the multi-objective prediction model through a co-evolution algorithm and Bayesian optimization to generate an optimized multi-objective prediction model; A parameter optimization module, which is configured to: input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

[0016] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization as described in the first aspect are implemented.

[0017] In a fourth aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in the multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization as described in the first aspect are implemented.

[0018] The above one or more technical solutions have the following beneficial effects: The present invention generates a standardized feature set through data acquisition and preprocessing, providing input for the multi-objective prediction model. By means of dynamic K-Means clustering and target priority adjustment mechanism, combined with multi-objective ensemble learning technology, a multi-objective prediction model is constructed to predict multi-objective performance indicators such as end jump difference, stress difference, flatness, and positive and negative ratios of end jump. And an optimal parameter combination is recommended through comprehensive scoring, overcoming the limitations of traditional methods that rely on manual experience and are difficult to adapt to complex working conditions, and significantly improving the machining accuracy, stability, and efficiency of the flexible contact reference surface grinding process.

[0019] Through the collaborative mechanism of dynamic clustering and target priority adjustment, combined with multi-objective ensemble learning technology, the present invention can accurately predict the end jump difference, stress difference, flatness, and positive and negative ratios of end jump during the ultra-thin sheet grinding process in complex grinding working conditions, significantly improving the accuracy and stability of parameter selection. Dynamic K-Means clustering dynamically adjusts the cluster center through error feedback, and target priority adjustment dynamically optimizes the training process according to multi-objective errors. Combining the global search of the co-evolutionary algorithm and the local refinement of Bayesian optimization, it has higher adaptability and specificity compared with traditional static grouping methods. The present invention can dynamically recommend the gasket thickness, grinding times, and depth of cut according to the incoming material thickness, ensuring that the end jump difference, stress difference, flatness, and positive and negative ratios of end jump for the same ultra-thin sheet reach the optimal distribution (positive for the maximum end jump value and negative for the minimum value), suitable for the rough grinding, finish grinding, and ultra-precision grinding requirements of ultra-thin sheets, and improving the processing efficiency and product qualification rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0021] Figure 1 is a flowchart of the multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to an embodiment of the present invention; Figure 2 is a flowchart of training a regression model by the bagging method through dynamic clustering and target priority adjustment according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0025] Embodiment 1 As Figure 1 shown, this embodiment discloses a multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization. The method includes the following steps: S1: Obtain the input feature data of the grinding process and perform preprocessing to obtain a normalized feature set; In this embodiment, the input feature data includes the thickness of the workpiece before grinding, the target thickness, the thickness of the spacer, the number of grinding round trips, and the depth of cut. The input feature data is preprocessed to generate a normalized feature set.

[0026] Based on the processing data of the ultra-thin sheet grinding process, 500 pieces of input feature data are obtained through a data acquisition device as a training data set. Each piece of data includes the thickness before grinding, the target thickness, the thickness of the spacer, the number of grinding round trips, and the depth of cut. Taking one piece of data as an example, the thickness before grinding is 3.0 mm, the target thickness is 2.88 mm, the thickness of the spacer is 0.2 mm, the number of grinding round trips is 1 time, and the depth of cut is 0.01 mm. The grinding thickness difference is calculated based on the data, and the grinding thickness difference is the difference between the thickness before grinding and the target thickness.

[0027] All the input feature data is standardized to eliminate the differences in magnitude and unit and ensure the balanced contribution of each feature to the subsequent model. The standardization formula is used:

[0028] where represents the standardized feature value, represents the original feature value, represents the original feature mean, represents the original feature standard deviation.

[0029] S2: Based on the normalized feature set, construct a multi-objective prediction model; the multi-objective prediction model includes a regression model and a classification model, and the regression model is trained by the bagging method of dynamic clustering and target priority adjustment; In this embodiment, a multi-objective prediction model is constructed using an ensemble learning algorithm based on a normalized feature set. Dynamic clustering and target-prioritized bagging are used to train regression models for end-jump difference, stress difference, and flatness. A RUSBoost algorithm is used to train a classification model for the positive and negative end-jump ratio. Specifically, the multi-objective prediction model includes three regression models (for predicting end-jump difference, stress difference, and flatness) and one classification model (for predicting the positive and negative end-jump ratio). The regression models are trained using the bagging method, and the classification model is trained using the RUSBoost algorithm.

[0030] Based on the normalized feature set, the dynamic clustering and target-priority adjustment bagging method is used to train the regression model to predict the end jump difference, stress difference and flatness. Specifically: (1) Initially cluster the normalized feature set using a clustering algorithm, and dynamically adjust the cluster center based on the multi-target prediction error; use the dynamic K-Means clustering method: set the number of clusters , perform initial clustering on the normalized feature set using the K-Means clustering algorithm to generate clusters; and dynamically adjust the cluster centers according to the multi-target prediction errors.

[0031] The cluster center is dynamically adjusted according to the multi-objective prediction error, and the adjustment formula is:

[0032] in, Indicates the The adjustment vector of the cluster center, Indicates the adjustment step size, Indicates the A set of data points at the center of a cluster, Indicates the The prediction error for each data point, Indicates the The mean of the intra-cluster errors, Indicates the The number of data points in a cluster, Indicates the The feature vector of the data point, Indicates the The center vector of the cluster, Indicates the cluster number.

[0033] Furthermore, the regression target error: For the end runout difference, stress difference, and flatness, the regression model uses the bagging method to predict their values. The error is calculated by the mean square error (MSE), which is expressed as:

[0034] in, Indicates the The true value of a data point on the th target, indicating the th data point on the th target, , corresponding to the end jump difference, stress difference, and flatness respectively.

[0035] Classification target error: For the positive and negative proportions of end jump, the model uses RUSBoost to predict the probability of the positive class, and the error is calculated through the logistic loss:

[0036] where represents the true label (0 or 1) of the th data point on the positive and negative proportion target of end jump, represents the predicted probability of the positive class.

[0037] Total prediction error: , where represents the weight of the th target. In this embodiment, the weight of the end jump difference is set to 0.5, the stress difference is set to 1, the flatness is set to 1, and the positive and negative of end jump is set to 0.5. It can be flexibly set according to specific situations and is not specifically limited.

[0038] After adjustment, the updated cluster center is ← , and re-cluster to make the clustering result dynamically adapt to the multi-target error distribution. Use the clusters generated by dynamic clustering to sample data according to the priority target.

[0039] (2) Determine the current priority target based on the target priority, sample data from the clusters with large error contributions based on the current priority target to replace the data in the bagging method to generate multiple new sub-datasets, train decision trees based on the sub-datasets, and integrate multiple decision trees to obtain a trained regression model; Predict the end jump difference, stress difference, flatness, and positive and negative proportions of end jump through the validation set, calculate the mean square error and classification error as the multi-target prediction error; calculate the priority of each target according to the prediction error of each target, and the formula is expressed as:

[0040] where represents the priority of the th target, represents the standardized error of the th target, represents the sum of the standardized errors of all targets, represents the number of targets (in this embodiment is 4, corresponding to four objectives: jump difference value, stress difference value, flatness, and positive / negative ratio of end jump).

[0041] Example: Assume the standardized errors are respectively (end jump value error), (stress difference value), (flatness), (positive / negative ratio), then the total error is , and the priority of the positive / negative ratio of end jump is . With the highest priority, the positive / negative ratio of end jump is used as the current priority objective.

[0042] The bagging method is used to train the regression model. Specifically: Set the number of decision trees and the maximum number of splits . Multiple sub-datasets are generated by sampling with replacement from the normalized feature set. For each sub-dataset, a decision tree is independently trained, and the predicted values of the end jump difference, stress difference, and flatness are calculated. The bagging method predicts each objective value through the integration of multiple decision trees (taking the average of the predicted values of all decision trees). The formula is expressed as:

[0043] where, represents the predicted value, represents the predicted value of the th decision tree, represents the decision tree serial number; the units of the end jump difference, stress difference, and flatness are mm, MPa, and mm respectively. The bagging method (Bagging) generates the prediction result of the final regression model by taking the average of the predicted values of multiple decision trees.

[0044] The one with the highest priority is the current priority objective. Based on the current priority objective, data is sampled from the clusters with a larger error contribution, replacing part of the data in the "bag" in the bagging method to generate multiple new sub-datasets. Each new sub-dataset retrains a decision tree, and multiple decision trees are integrated to obtain a trained regression model. Specifically, using the clusters generated by dynamic clustering, calculate the error contribution of each cluster to the positive / negative ratio of end jump, select the two clusters with the largest error contribution (such as cluster 1 and cluster 3), randomly extract 10 data points from each cluster, replace part of the data in the "bag" in the bagging method, generate multiple new sub-datasets, and each new sub-dataset retrains a decision tree. Multiple decision trees are integrated to obtain a trained regression model. Repeat dynamic clustering and target priority adjustment until the multi-objective error converges (error change < 0.001) or reaches the preset number of iterations (such as 10 times).

[0045] Furthermore, the error contribution is defined as: the error contribution of each cluster to the current priority objective (determined by the objective priority The error contribution (determined) is the mean of the prediction errors of the data points within the cluster calculated , where is the th cluster, and is the prediction error of the

[0046] Specifically: Determine the priority target: According to the target priority formula, select the one with the highest priority as the current priority target; Calculate the within-cluster error: For each cluster , only calculate its error contribution on the priority target , that is , where is the th data point's prediction error on the target ; Sort and select: For all clusters , sort them in descending order according to , and select the clusters with the largest error contributions (such as the first two clusters).

[0047] Furthermore, replace part of the data in the "bag" of the bagging method, and its replacement strategy is as follows: Determine the replacement ratio: Each "bag" (sub-dataset) in the bagging method is generated by sampling with replacement from the normalized feature set, and usually contains samples of the size of the original dataset (about 500). In this embodiment, 10% of the data is replaced each time, that is, 50 data are replaced in each bag.

[0048] Randomly select the data to be replaced: In each bag, randomly select 50 data as the objects to be replaced to ensure an unbiased replacement process.

[0049] Sample new data: Randomly draw 25 data (a total of 50) from each of the clusters with larger error contributions (such as cluster 1 and cluster 3), and replace the 50 selected data in the bag.

[0050] (3) Predict the end jump difference, stress difference, and flatness based on the trained regression model. Train an independent regression model for each target, namely the end jump difference, stress difference, and flatness. Each regression model contains multiple decision trees (100 decision trees in each model in this embodiment). Three regression models are trained for the three targets respectively, and each model contains multiple decision trees and integrates to obtain the prediction result.

[0051] Use the RUSBoost algorithm to train the classification model of the positive and negative ratios of the end jump, specifically: To achieve the accurate classification of the positive and negative ratios of end - jump, the present invention adopts the RUSBoost algorithm, which optimizes the logical loss function and improves the classification performance by iteratively updating the sample weights. This algorithm combines random undersampling and a weighted mechanism to effectively address the problem of data imbalance and ensure the prediction accuracy for the minority class.

[0052] The specific training process of the classification model for the positive and negative ratios of end - jump is as follows: (1)Initialization Initialize the learning rate and the number of training rounds M, using decision trees as weak learners. For the training data set , which comes from the data collection and pre - processing in S1 and contains grinding experiment records. Each record includes the pre - grinding thickness, target thickness, shim thickness, number of grinding back - and - forths, feed per pass, and thickness difference after standardization as feature vectors , as well as the positive and negative labels obtained through end - jump value measurement . Initialize the sample weights:

[0053] where represents the initial weight of the -th sample, represents the total number of samples, a positive integer representing the number of grinding experiments.

[0054] Initialize the cumulative output: .

[0055] (2)Iterative training The RUSBoost algorithm iteratively trains the weak learner through the following steps: 1)Random undersampling to generate a balanced sub - data set In each iteration, based on the number of positive - class samples (end - jump positive, ) and the number of negative - class samples (end - jump negative, ), randomly draw samples from the negative - class samples and combine them with all positive - class samples to form a balanced sub - data set with a size of 2 to alleviate the problem of the majority class dominance.

[0056] 2)Train the weak learner On the balanced sub - data set , use the current samples to train the weak learner . To match , map to , so map to .

[0057] 3) Calculate the weighted error rate

[0058] in, represents the weighted error rate, represents the sample weight, is the decision tree prediction.

[0059] Weighted error rate: For samples with incorrect predictions, the weight will be added to It is used to measure the performance of weak learners.

[0060] 4) Calculate the weight of weak learners According to the weighted error rate, calculate the weak learner The weight is expressed as:

[0061] Among them, if , the weak learner has poor performance and can be adjusted Or stop the iteration.

[0062] 5) Update sample weights

[0063] And normalize the updated sample weights:

[0064] After normalization Assign to , and update the cumulative output: .

[0065] 6) Optimize the logistic loss function:

[0066] in, Indicates the true label, 0 or 1, 1 indicates a positive end jump, 0 indicates a negative end jump; Represents the probability of the positive class, with a value of (0,1), Represents the feature vector and normalizes the grinding parameters. After multiple iterations, the logistic loss decreases and the classification performance is improved.

[0067] (3) Output the final classification model After iterating M times and minimizing the logistic loss function, the final positive class probability and predicted label are output:

[0068]

[0069] Among them, represents the positive class probability, and its value range is (0, 1).

[0070] S3: Co-evolve the hyperparameters of the multi-objective prediction model in cooperation with Bayesian optimization to generate an optimized multi-objective prediction model; In this embodiment, the hyperparameters of the regression model are optimized, and the number of decision trees is set in the range of [100, 500], and the maximum number of splits is set in the range of [1, 20]; the hyperparameters of the classification model are optimized, and the number of decision trees is set in the range of [100, 300], the learning rate is set in the range of [0.01, 0.5], and the maximum number of splits is set in the range of [1, 10]; the optimization process is realized by the cooperation of co-evolution algorithm and Bayesian hyperparameter optimization, specifically as follows: The co-evolution algorithm optimizes the hyperparameters of the regression and classification models through two populations, and each population contains 30 hyperparameter combinations. Population 1 is for the hyperparameters of the regression model ( , ), with the goal of minimizing the mean square error; Population 2 is for the hyperparameters of the classification model ( , , ), with the goal of minimizing the log loss. The fitness function formula is:

[0071] Among them, represents the fitness value of the th hyperparameter combination, represents the hyperparameter combination, represents the mean square error, represents the log loss. The dynamic K-Means clustering adjusts the fitness weights of the target priorities, and preferentially optimizes the targets with larger errors (such as the positive and negative ratios of end jump). The evolutionary operations include tournament selection (select 3 combinations and retain the one with the highest fitness), single-point crossover (probability 0.8), random mutation (probability 0.1), and retaining the original combination (probability 0.1). The random tournament selection randomly selects 3 hyperparameter combinations from the population, and screens out the one with the optimal performance as the parent combination by comparing the fitness values, balancing the selection efficiency and population diversity. The population update formula is:

[0072] Among them, represents the New hyperparameter combinations for each generation, indicating the generation combination, indicating the single-point crossover operation, indicating the random mutation operation. Every 5 generations, 3 high-fitness combinations are shared among populations to form the Pareto front, adapting to multi-objective trade-offs.

[0073] Bayesian hyperparameter optimization is based on the hyperparameter combinations generated by the co-evolution algorithm (the top 5 high-fitness combinations in each generation). A surrogate model is constructed through a tree-structured Parzen estimator to predict the hyperparameter performance. The expected improvement function is used, and the formula is:

[0074] where represents the expected improvement value, represents the hyperparameter combination, represents the mathematical expectation, represents the hyperparameter under the cross-validation loss, represents the loss of the current optimal hyperparameter combination. The loss is calculated through 5-fold cross-validation, and high-potential combinations are selected by maximizing the expected improvement value. The refined combinations (the top 3) from Bayesian hyperparameter optimization are fed back to the co-evolution population to replace the low-fitness combinations.

[0075] The optimization process terminates after 50 iterations or when the increase in the expected improvement value is less than 0.001. Initially, 10 groups of hyperparameter combinations are randomly evaluated, and then through the collaborative iteration of the co-evolution algorithm and Bayesian hyperparameter optimization, the optimal hyperparameter combination is determined, generating a high-precision prediction model to improve the prediction performance of the end-jump difference, stress difference, flatness, and the positive and negative ratios of the end-jump, providing accurate input for screening the optimal process parameters in the comprehensive scoring formula.

[0076] S4: Input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict the multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

[0077] In this embodiment, according to the processing requirements, the range of grinding process parameters to be optimized is input, the multi-objective performance indicators are predicted, and the optimal parameter combination is selected through comprehensive scoring to ensure that the positive and negative ratios of the end-jump meet the requirements of the maximum positive and minimum negative values.

[0078] Obtain the incoming material thickness according to the processing requirements and the target thickness , set the range of parameters to be optimized, including the range of shim thickness (e.g., [0.2, 1.2] mm, step size 0.1 mm), the range of grinding back-and-forth times (e.g., [2, 6] times when the depth of cut is 0.005 mm, [1, 3] times when the depth of cut is [0.01, 0.015] mm), and the range of depth of cut (e.g., [0.005, 0.015] mm, step size 0.005 mm), and generate a set of parameter combinations.

[0079] Predict for the set of parameter combinations, calculate the predicted values of the end jump difference, stress difference, flatness, and the positive and negative ratio of end jump, where the positive and negative ratio of end jump is expressed as the positive class probability; perform standardization processing on the predicted values. Specifically: calculate the minimum value and the maximum value , and use the formula:

[0080] where, represents the standardized predicted value, represents the original predicted value, represents the minimum value of the predicted value, represents the maximum value of the predicted value; Set the weight vector (e.g., ), calculate the comprehensive score value, and the formula is expressed as:

[0081] where, represents the comprehensive score value, , [[ID=3), represent the weights of the corresponding objectives, and the weight vector can be set according to the processing requirements or determined through experiments; represents the standardized end jump difference, represents the standardized stress difference, represents the standardized flatness, represents the standardized positive and negative ratio of end jump. Determine the optimal grinding process parameters by minimizing the comprehensive score value .

[0082] For a certain set of input parameters (such as shim thickness, grinding times, depth of cut), the predicted value of the final prediction model is fixed, but different parameter combinations will result in different predicted values. Minimizing the comprehensive score value is achieved by traversing the set of parameter combinations. The range of grinding process parameters to be optimized will generate multiple parameter combinations (such as 165 groups), input each group of parameters into the model, predict the multi-objective values and calculate the comprehensive score value, and select the parameter combination with the minimum comprehensive score value.

[0083] ​Example: The input incoming material thickness is 3.0 mm, the target thickness is 2.8 mm, and the set parameter range is: shim thickness 0.2 - 1.2 mm (step size 0.1 mm), grinding times 1 - 6 times, depth of cut 0.005 - 0.015 mm (step size 0.005 mm), generating 165 groups of parameter combinations. After normalizing the predicted values, set the weight vector (1, 1, 0.5, 0.5) and calculate the comprehensive score. The optimal parameters are: shim thickness 1 mm, grinding times 2 times, depth of cut 0.015 mm, and the comprehensive score is 0.115, achieving the optimal group that best balances the stress difference, flatness, end run difference, and positive and negative values of end run.

[0084] In order to solve the technical problems in the prior art that the optimization of grinding process parameters lacks a dynamic adjustment mechanism and it is difficult to achieve multi-objective collaborative optimization, the training data set in the present invention is derived from grinding experiment records, including standardized process parameters and end run positive and negative labels, generating a standardized feature set through data acquisition and preprocessing, providing input for the multi-objective prediction model; the present invention constructs a multi-objective prediction model through dynamic K-Means clustering and target priority adjustment mechanism, combined with multi-objective ensemble learning technology, predicts multi-objective performance indicators such as end run difference, stress difference, flatness, and end run positive and negative ratio, and recommends the optimal parameter combination through comprehensive scoring, overcoming the limitations of traditional methods that rely on manual experience and are difficult to adapt to complex working conditions, and significantly improving the machining accuracy, stability, and efficiency of the flexible contact reference surface grinding process.

[0085] Dynamic K-Means clustering adaptively adjusts data grouping through error feedback, target priority adjustment dynamically optimizes the prediction focus according to multi-objective errors, the co-evolution algorithm and Bayesian hyperparameter optimization cooperate to generate a high-precision prediction model, form a Pareto front, optimize multi-objective performance, overcome the limitations of traditional methods, dynamically recommend process parameters, ensure the optimal distribution of multiple objectives, and are applicable to rough grinding, finish grinding, and super-finish grinding, improving the product qualification rate and machining efficiency.

[0086] Example Two This example discloses a multi-objective grinding decision-making system based on dynamic clustering and hyperparameter optimization, including: A data acquisition module, which is configured to: acquire input feature data of the grinding process and perform preprocessing to obtain a standardized feature set; A model construction module, which is configured to: construct a multi-objective prediction model based on the standardized feature set; the multi-objective prediction model includes a regression model and a classification model, and the regression model is trained by the bagging method of dynamic clustering and target priority adjustment; A model optimization module, which is configured to: cooperatively adjust the hyperparameters of the multi-objective prediction model through a co-evolution algorithm and Bayesian optimization to generate an optimized multi-objective prediction model; A parameter optimization module, which is configured to: input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

[0087] Embodiment III The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method in Embodiment I are implemented.

[0088] Embodiment IV The purpose of this embodiment is to provide a computer-readable storage medium. A computer-readable storage medium has a computer program stored thereon. When the program is executed by a processor, the steps of the method in Embodiment I are executed.

[0089] The steps involved in the devices in the above Embodiments III and IV correspond to those in Method Embodiment I. For specific implementation manners, reference may be made to the relevant description part of Embodiment I. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0090] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device. Thus, they can be stored in a storage device and executed by a computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to be implemented. The present invention is not limited to any specific combination of hardware and software.

[0091] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0092] Although the specific implementation manners of the present invention are described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization, characterized in that Including: Obtain the input feature data of the grinding process and perform preprocessing to obtain a normalized feature set; Based on the normalized feature set, construct a multi-objective prediction model; The multi-objective prediction model includes a regression model and a classification model, and the regression model is trained by the bagging method of dynamic clustering and target priority adjustment; Co-evolution algorithm and Bayesian optimization are used to jointly adjust the hyperparameters of the multi-objective prediction model to generate an optimized multi-objective prediction model; Input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict the multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

2. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 1, characterized in that The input feature data includes the thickness of the workpiece before grinding, the target thickness, the thickness of the gasket, the number of grinding back-and-forth times, and the depth of cut.

3. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 1, wherein The multi-objective prediction model includes a regression model of the end jump difference, stress difference, and flatness, and a classification model of the positive and negative ratio of the end jump.

4. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 3, characterized in that, The RUSBoost algorithm is used to train the classification model of the positive and negative ratio of the end jump.

5. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 1, wherein, The specific process of training the regression model by the bagging method of dynamic clustering and target priority adjustment is as follows: Perform initial clustering on the normalized feature set through the clustering algorithm, and dynamically adjust the cluster center according to the multi-objective prediction error; Determine the current priority target based on the target priority, sample data from the clusters with large error contributions based on the current priority target, replace the data in the bagging method to generate multiple new sub-datasets, train decision trees based on the new sub-datasets, and integrate multiple decision trees to obtain a trained regression model; Predict the end jump difference, stress difference, and flatness based on the trained regression model.

6. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 5, characterized in that, Dynamically adjust the cluster center according to the multi-objective prediction error, and the formula is expressed as: Among them, represents the adjustment vector of the th cluster center, represents the adjustment step size, represents the set of data points of the th cluster center, represents the prediction error of the th data point, represents the mean of the within-cluster errors of the th cluster, represents the feature vector of the th data point, represents the center vector of the th cluster.

7. The multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization according to claim 5, characterized in that, The formula for calculating the target priority is expressed as: Among them, represents the priority of the th target, represents the normalized error of the th target, represents the sum of the normalized errors of all targets, represents the number of targets.

8. A multi-objective grinding decision-making system based on dynamic clustering and hyperparameter optimization, characterized in that, Including: A data acquisition module, which is configured to: obtain the input feature data of the grinding process and perform preprocessing to obtain a normalized feature set; A model construction module, which is configured to: based on the normalized feature set, construct a multi-objective prediction model; the multi-objective prediction model includes a regression model and a classification model, and the regression model is trained by the bagging method of dynamic clustering and target priority adjustment; A model optimization module, which is configured to: co-evolution algorithm and Bayesian optimization are used to jointly adjust the hyperparameters of the multi-objective prediction model to generate an optimized multi-objective prediction model; A parameter optimization module, which is configured to: input the range of grinding process parameters to be optimized into the optimized multi-objective prediction model, predict the multi-objective performance indicators, and select the optimal process parameter combination through comprehensive scoring.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the multi-objective grinding decision-making method based on dynamic clustering and hyperparameter optimization as described in any one of claims 1-7.

Images