Process optimization method, device, equipment and system

Abstract

The invention discloses a process optimization method, device, equipment and system, and relates to the field of process optimization. The method comprises the following steps: acquiring input data of a user, wherein the input data comprises process boundary conditions; generating an orthogonal test scheme through a preset orthogonal test design method based on the input data, wherein the orthogonal test scheme comprises a plurality of orthogonal test samples; determining an alternative parameter combination based on the plurality of orthogonal test samples and a prediction model, wherein the prediction model is constructed based on a random forest regression model; and determining the alternative parameter combination as an initial solution of a simulated annealing algorithm, and obtaining an optimized parameter combination through the simulated annealing algorithm, wherein the optimized parameter combination is used for process optimization. According to the method, the optimized parameter combination can be obtained and used for process optimization.

Description

Technical Field

[0001] The present application relates to the technical field of process optimization, and specifically to a process optimization method, device, equipment and system. Background Art

[0002] Wafer cutting is a core link in the field of wafer manufacturing. The precise control of its process parameters will directly affect the quality and cost of the product. Since the optimization of process parameters requires comprehensive consideration of many dimensions, the parameters of multiple dimensions will affect each other, and the parameter space is huge and difficult to verify, it is difficult to find the parameter combination that optimizes the quality. Summary of the invention

[0003] The embodiments of the present application provide a process optimization method, device, equipment and system to obtain an optimization parameter combination and use it for process optimization.

[0004] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions: In a first aspect, a process optimization method is provided, comprising: Obtaining user input data, wherein the input data includes process boundary conditions; Based on the input data, an orthogonal test scheme is generated by a preset orthogonal test design method, wherein the orthogonal test scheme includes a plurality of orthogonal test samples; Determining candidate parameter combinations based on a plurality of the orthogonal test samples and a prediction model, wherein the prediction model is constructed based on a random forest regression model; The candidate parameter combination is determined as an initial solution of a simulated annealing algorithm, and an optimized parameter combination is obtained through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

[0005] In some embodiments, based on the input data, an orthogonal test scheme is generated by a preset orthogonal test design method, including: Determining factors and levels of orthogonal experiments based on the process boundary conditions, wherein the factors are process parameters to be optimized, and the levels are value ranges of the process parameters; Based on the factors and levels of the orthogonal test, a plurality of the orthogonal test samples are selected by using the Latin hypercube sampling method.

[0006] In some embodiments, the prediction model is used to characterize the mapping relationship between process data, process parameters, production process influencing data and total sheet thickness variance; determining an alternative parameter combination based on a plurality of the orthogonal test samples and the prediction model includes: Input each of the orthogonal test samples into the prediction model to obtain a total slice thickness variance prediction value corresponding to each of the orthogonal test samples; Based on the predicted value of the total slice thickness variance corresponding to each of the orthogonal test samples, determine the sum of the total slice thickness variance and the average value of the total slice thickness variance for each factor at each level; Determine the range of each of the factors based on the average value of the total slice thickness variance of each of the factors at each of the levels; According to the order of range from large to small, the alternative levels with the smallest total slice thickness variance under each of the factors are determined in turn, and the first parameter combination formed by the alternative levels of each of the factors is determined as the alternative parameter combination.

[0007] In some embodiments, determining the range of each factor based on the average value of the total slice thickness variance of each factor at each level comprises: Based on the average value of the total slice thickness variance of each factor at each level, determining the maximum value and the minimum value of the average value of the total slice thickness variance in each factor; The difference between the maximum value and the minimum value is determined as the range corresponding to the factor.

[0008] In some embodiments, obtaining an optimized parameter combination through the simulated annealing algorithm includes: A neighborhood structure of a solution is defined, and a candidate parameter combination is determined as one of the candidate solutions, and a total slice thickness variance prediction value of the candidate solution is determined as fitness, wherein the candidate parameter combination is the one with a smaller total slice thickness variance prediction value among the first parameter combination and the second parameter combination, and the second parameter combination is the orthogonal test sample with the smallest total slice thickness variance prediction value among all the orthogonal test samples; In each iteration, the candidate solution with poor fitness is accepted as the new current solution with a preset probability based on the Metropolis criterion; Lowering the temperature according to a set cooling rate so that the iteration result converges to an optimal solution or a suboptimal solution, wherein the preset probability decreases as the temperature decreases; When the iteration reaches a preset end condition, the iteration is ended, and the optimal solution or the suboptimal solution is determined as the optimization parameter combination.

[0009] In some embodiments, the prediction model is trained based on historical process data, historical process parameters, historical production process impact data, and historical inspection data; the step of acquiring the prediction model includes: Constructing the prediction model based on the random forest regression model; Acquire training samples based on the historical process data, the historical process parameters, the historical production process impact data, and the historical inspection data; The prediction model is trained and verified based on the training samples to obtain the trained prediction model.

[0010] In some embodiments, the step of obtaining the training sample includes: Preprocessing the historical process data, the historical process parameters, the historical production process impact data and the historical inspection data to obtain preprocessed data; Select features whose influence on the total sheet thickness variance meets preset conditions from pre-processed historical process data, historical process parameters, and historical production process influence data to obtain an initial feature set; Performing feature extraction and normalization processing on the initial feature set to obtain a training feature set; The training samples are obtained based on the training feature set and the preprocessed historical test data.

[0011] In some embodiments, training and validating the prediction model based on the training samples to obtain the trained prediction model includes: Based on the training samples, the prediction model is trained and verified using a K-fold cross-validation method to obtain the trained prediction model.

[0012] In some embodiments, the process data includes various observed parameters and states of the production equipment during the process, the process parameters include various process setting conditions that affect product quality, the production process influencing data includes man-machine-material-method-environment measurement data, and the inspection data includes the total sheet thickness variance detection value.

[0013] In some embodiments, after obtaining the optimized parameter combination through the simulated annealing algorithm, the method further includes: The optimization parameter combination is displayed on a user interface based on a preset display form, wherein the preset display form includes at least one of a chart and a text.

[0014] In a second aspect, a process optimization device is provided, comprising: A data interaction module, used to obtain input data from a user, wherein the input data includes process boundary conditions; An orthogonal design module, used to generate an orthogonal test scheme based on the input data by a preset orthogonal test design method, wherein the orthogonal test scheme includes a plurality of orthogonal test samples; A parameter prediction module, used to determine candidate parameter combinations based on a plurality of the orthogonal test samples and a prediction model, wherein the prediction model is constructed based on a random forest regression model; The parameter optimization module is used to determine the candidate parameter combination as the initial solution of the simulated annealing algorithm, and obtain the optimized parameter combination through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

[0015] In a third aspect, an electronic device is provided, comprising a processor and a memory; a computer program is stored in the memory, and the processor is used to execute the computer program stored in the memory to implement the process optimization method as described in any one of the first aspects.

[0016] In a fourth aspect, a process optimization system is provided, comprising: The equipment layer is used to generate and obtain process data, process parameters, production process influencing data and inspection data; A data layer, the data layer is used to process the data acquired by the device layer; A model layer, wherein the model layer is used to execute the process optimization method as described in any one of the first aspects; The application layer is used to interact with the user to obtain input data and display the optimization parameter combination based on a preset display form.

[0017] A process optimization method of the present application includes obtaining user input data, the input data includes process boundary conditions; based on the input data, generating an orthogonal test scheme through a preset orthogonal test design method, the orthogonal test scheme includes multiple orthogonal test samples; determining alternative parameter combinations based on multiple orthogonal test samples and a prediction model, the prediction model is constructed based on a random forest regression model; determining the alternative parameter combination as the initial solution of the simulated annealing algorithm, and obtaining the optimized parameter combination through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization. The method provided in the present application effectively replaces the traditional practical experiment through the deep integration of orthogonal experiments and prediction models, which can greatly reduce the test time and improve the optimization efficiency of process parameters. At the same time, the use of alternative parameter combinations to provide a high-quality initialization level for the simulated annealing algorithm can significantly improve the convergence speed and solution quality of the algorithm, thereby improving the optimization accuracy of process parameters.

[0018] A process optimization device of the present application can effectively replace traditional practical experiments through the deep integration of orthogonal experiments and prediction models, greatly reduce the test time, and improve the optimization efficiency of process parameters. At the same time, it uses alternative parameter combinations to provide a high-quality initialization level for the simulated annealing algorithm, which can significantly improve the convergence speed and solution quality of the algorithm, thereby improving the optimization accuracy of process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 This is a schematic diagram of the overall structure of the process optimization system of an embodiment of the present application; Figure 2 This is a schematic diagram of the overall process of the process optimization method of an embodiment of the present application; Figure 3 A schematic diagram of the structure of the prediction model of the embodiment of the present application; Figure 4 A schematic diagram of the structure of a process optimization device according to an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device according to an embodiment of the present application; Reference numerals: 10-device layer; 20-data layer; 30-model layer; 40-application layer; 401-data interaction module; 402-orthogonal design module; 403-parameter prediction module; 404-parameter optimization module; 501-memory; 502-processor. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0022] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present application, "multiple" means two or more, and at least one means one, two or more, unless otherwise clearly and specifically defined.

[0023] In the field of chip manufacturing, chip cutting is a core link, and the precise control of its process parameters directly affects the quality and cost of the product. The inventors have conducted more in-depth research and optimization on the cutting process parameters. For example, by adjusting key parameters such as steel wire tension, wire speed, and feed speed, the surface geometry of the chip can be effectively improved, the warpage and line mark depth can be reduced, and the cutting quality can be improved. These optimizations not only improve the yield rate of chips, but also reduce production costs, injecting new vitality into the development of the industry.

[0024] The inventors have been continuously testing and conducting in-depth research on the cutting process parameters, hoping to improve the yield rate of the chip and reduce the production cost by optimizing the cutting process parameters. However, with the continuous expansion of market demand and production scale, the continuous change of cutting methods and the increasing personalization of customer needs, the process parameters of the cutting link have become more complex and diverse. The order of magnitude and difficulty of process parameter optimization have become new challenges faced by process research and development, and the optimization of process parameters under different process boundaries is directly related to the surface quality, geometric morphology, cutting speed and production cost of the chip during the cutting process. The optimization of process parameters in the chip cutting process is a systematic project, which requires comprehensive consideration of multiple factors. Through continuous experiments, analysis and adjustments, the best combination of process parameters is found to achieve the best balance of cutting quality, production efficiency and cost control. However, the optimization of process parameters requires precise balance of multiple parameters that affect each other, and the huge number of parameter spaces is difficult to verify experimentally one by one, which makes it difficult to find the parameter combination that optimizes the quality parameters, and thus makes the efficiency and effect of process parameter optimization unsatisfactory.

[0025] According to research, the optimization methods of process parameters in related technologies mainly include experimental optimization method, numerical simulation method, optimization algorithm and dynamic model method. Among them, the experimental optimization method is based on cutting experiments, and optimizes the experimental process through orthogonal experiments, variance analysis, signal-to-noise ratio analysis and other methods. The more classic methods include factorial analysis, Taguchi method and response surface method. These methods can obtain the best combination of process parameters through experiments, but they require a large number of experiments and long-term accumulation of production practices, with high time and economic costs, and can only find the optimal solution in a given parameter combination, and cannot recommend the optimal solution outside the parameter level setting. The numerical simulation method simulates and optimizes the process by constructing mathematical models and simulation software. This method can predict and evaluate the effects of different parameter combinations without actual production, thereby reducing the number of experiments and costs, but simulation usually requires simplification of complex process processes, and it is difficult to simulate the dynamic changes and external interference of the real environment. With the development of computer technology and artificial intelligence, methods based on optimization algorithms and expert knowledge systems have been more widely used, such as neural networks, genetic algorithms, particle swarm algorithms, simulated annealing and ant colony algorithms, which can determine the optimized process parameters before wafer cutting. However, this method relies on the initial parameter settings and is often prone to local optimality. The optimization method based on the kinetic model establishes a kinetic model of the process and analyzes the dynamic relationship between the parameters to find the optimal parameter combination. This method requires a deep understanding of the physical and chemical mechanisms of the process and the interactions between the parameters.

[0026] Due to the huge experimental verification overhead, easy to fall into local optimality, and the high-dimensional nonlinear coupling characteristics of the parameter system are difficult to simply abstract into a mathematical model during the optimization of relevant process parameters, finding the parameter combination that optimizes quality and improving the efficiency and effect of process parameter optimization are still the main challenges faced by parameter optimization.

[0027] In view of this, the embodiments of the present application provide a process optimization method, device, equipment and system, which replace the traditional practical experiment through the deep integration of orthogonal experiment and prediction model, so as to greatly reduce the test time and improve the optimization efficiency of process parameters. At the same time, the alternative parameter combination screened out by orthogonal experiment and prediction model is used to provide a high-quality initialization level for the simulated annealing algorithm, so as to significantly improve the convergence speed and solution quality of the algorithm, and improve the optimization accuracy of process parameters, thereby solving at least part of the above-mentioned technical problems.

[0028] See also Figure 1 , Figure 1The overall structural diagram of the process optimization system of the embodiment of the present application. The process optimization system includes an equipment layer 10, a data layer 20, a model layer 30 and an application layer 40. The equipment layer 10 is used to obtain process data, process parameters, production process influencing data and inspection data. In other words, the equipment layer 10 is the main production object and data source for generating data. The equipment layer 10 may include production equipment and inspection equipment, such as wire cutting machines and inspection machines, and may also include data generation modules, data perception modules, data acquisition modules and data transmission modules. Among them, process data includes various observed parameters and states of production equipment during the process. For example, process data may include temperature, pressure, flow, speed, position, etc. Process parameters include various process setting conditions that affect product quality. For example, process parameters may include temperature setting values, pressure ranges, time periods, etc. Process parameters are usually set by process personnel and monitored and recorded by equipment sensors or instruments. The data affecting the production process include man-machine-material-method-environment-measurement data. Man-machine-material-method-environment-measurement data refers to 5M1E data, which can cover six aspects that affect the production process, namely man (Man), machine (Machine), material (Material), method (Method), environment (Environment), and measurement (Measurement). "Man" refers to the wire cutting operator, "machine" refers to the wire cutting machine, stick sticking machine and other equipment, "material" refers to the main materials and auxiliary materials such as steel wire groove wheels, "method" refers to the process formula, "environment" corresponds to the production environment including temperature, etc., and "measurement" refers to the influencing factors of the detection link. The inspection data includes the total thickness variance (Total ThicknessVariation, referred to as TTV) detection value. The inspection data is the data obtained when the product is quality inspected, which can be obtained from the inspection machine sensor. The embodiment of this application only focuses on the TTV data.

[0029] The data layer 20 is used to process the data acquired by the equipment layer 10. Specifically, the data layer 20 can select, clean, calculate, convert and aggregate the collected data of various dimensions by cuts, thereby generating a wide data table that associates multiple data sources by process ID (Identity document). The wide table is used as a quasi-input for data analysis and models by placing the fields of process, technology, inspection and 5M1E data in the same table.

[0030] As the core part of the system, the model layer 30 undertakes the main recommendation logic. The model layer 30 is used to execute the process optimization method of the embodiment of the present application.

[0031] The application layer 40 is used as an interface for interacting with the user to obtain input data. The input data may include set process boundary conditions and production process influencing data (i.e., 5M1E data), such as specifications, sheet thickness, cutting machine, etc. By receiving the process boundary conditions and 5M1E important influencing factors input by the user, the model layer 30 performs corresponding process parameter optimization based on the input data. The application layer 40 is also used to display the optimization parameter combination based on a preset display form, which may include at least one of a chart and a text.

[0032] Through the above scheme, the embodiment of the present application provides a complete solution from data collection, processing, model training to intelligent recommendation, which can not only greatly simplify the optimization process of process parameters, but also realize automatic and precise recommendation of process parameters, improve the accuracy and efficiency of recommendations, and help enterprises improve production efficiency and product quality.

[0033] The process optimization method of the embodiment of the present application is introduced below with reference to the accompanying drawings.

[0034] See also Figure 2 , Figure 2 The overall process flow diagram of the process optimization method of the embodiment of the present application is as follows. The process optimization method comprises the following steps: Step 201: Obtain input data from a user, where the input data includes process boundary conditions.

[0035] Specifically, input data refers to data input by the user through the application layer, which may include set process boundary conditions and production process impact data (i.e., 5M1E data). It is understandable that if the user's input data does not include production process impact data, then the default production process impact data and process boundary conditions can be used as input together. If the user's input data includes production process impact data, then the production process impact data can be used to correct the default production process impact data, and the corrected data and process boundary conditions can be used as input together.

[0036] In some embodiments, after executing step 201 and before executing step 202, the method of the embodiment of the present application may further include: Preprocess the input data.

[0037] Exemplarily, the input data may be subjected to preset cleaning, format conversion, normalization and other processing.

[0038] Step 202: Based on the input data, an orthogonal experiment scheme is generated by a preset orthogonal experiment design method, and the orthogonal experiment scheme includes a plurality of orthogonal experiment samples.

[0039] Specifically, the orthogonal experimental design method is a highly systematic experimental method, the core of which is to plan the multi-factor and multi-level experimental process through a carefully designed orthogonal table. This method aims to cover all key factor combinations with the least number of experiments, so as to efficiently explore the influence of each factor on the experimental results and find the optimal experimental conditions. First of all, it is crucial to clarify the purpose of the experiment, because it determines the factors to be investigated and the expected results. Subsequently, according to the purpose of the experiment and the actual situation, the various factors in the experiment and the different levels of each factor are determined. These factors may be process parameters such as temperature, pressure, time, concentration, etc., and the levels are the specific values ​​or conditions that these parameters may take in the experiment. Next, choosing a suitable orthogonal table is a key step in the experimental design. An orthogonal table is a special table with unique properties that ensures that each factor is balanced at each level while ensuring comparability between test points. By selecting an appropriate orthogonal table, the experiment can be systematically arranged to avoid duplication and omission, so as to obtain as much information as possible within a limited number of experiments. After determining the orthogonal table and the experimental plan, the experiment can be started. According to the arrangement of the orthogonal table, the experiments are carried out one by one, and the experimental results are accurately recorded. These results will be the basis for subsequent analysis to reveal the influence of each factor on the test results. Finally, statistical analysis of the test results is the key to drawing conclusions. Statistical methods such as variance analysis are usually used to quantitatively evaluate the test results to determine which factors have a significant impact on the test results, as well as the direction and magnitude of the impact of these factors. By analyzing the results, the optimal combination of factor levels can be found, that is, the factor values ​​that can achieve the optimal state of the test results.

[0040] In some embodiments, step 202 may be implemented in the following manner: Step 1: Determine the factors and levels of the orthogonal experiment based on the process boundary conditions, where the factors are the process parameters to be optimized and the levels are the value ranges of the process parameters.

[0041] In some examples, there are 9 process parameters to be optimized, that is, there are 9 factors, expressed as: B={B1,B2,B3,B4,B5,B6,B7,B8,B9}, each factor takes 10 levels, that is, each parameter factor is divided into 10 different value ranges within its possible range, and the 10 levels of each factor can be expressed as: A={A1,A2,A3,A4,A5,A6,A7,A8,A9,A10}.

[0042] Step 2: Based on the factors and levels of the orthogonal experiment, the Latin hypercube sampling method is used to select multiple orthogonal experiment samples.

[0043] Specifically, Latin hypercube sampling is a method for approximate random sampling from a multivariate parameter distribution. It belongs to the stratified sampling technique and is often used in computer experiments or Monte Carlo integration.

[0044] Through the above scheme, the initial sample points can be selected uniformly and randomly in the entire parameter space, and the entire range of each parameter can be ensured to be fully covered, thereby improving the effect of the orthogonal experiment.

[0045] Step 203: Determine candidate parameter combinations based on multiple orthogonal test samples and a prediction model, where the prediction model is constructed based on a random forest regression model.

[0046] Among them, the prediction model is trained based on historical process data, historical process parameters, historical production process influencing data and historical inspection data. The prediction model is used to characterize the mapping relationship between process data, process parameters and production process influencing data and the total sheet thickness variance.

[0047] Before introducing the specific implementation of step 203, the steps of obtaining the prediction model in the embodiment of the present application are first introduced.

[0048] In some embodiments, the step of obtaining the prediction model includes: Step 1: Build a prediction model based on the random forest regression model.

[0049] See also Figure 3 , Figure 3Schematic diagram of the structure of the prediction model of the embodiment of the present application. The prediction model of the embodiment of the present application is constructed based on a random forest regression model, which includes multiple decision trees. Random forest regression is an ensemble learning algorithm that improves the prediction performance of a single decision tree by combining the prediction results of multiple decision trees (base learners). This ensemble method mainly relies on two key concepts: integration and randomness. In ensemble learning, multiple base learners (decision trees in the embodiment of the present application) are trained, and their prediction results are combined in some way to form the final prediction. For random forest regression, the final prediction result is the average of the prediction results of all decision trees. This ensemble method helps to reduce the risk of overfitting, because different decision trees may make mistakes on different sample subsets and feature subsets, and these mistakes may offset each other. The "randomness" in the random forest is mainly reflected in two aspects: the randomness of the sample and the randomness of the feature, where the sample randomness means that each decision tree randomly extracts a sample subset from the original data set as a training sample during training. This process is achieved through Bootstrap sampling, that is, each sampling is to extract the same number of samples as the original data set with replacement from the original data set. Because of the sampling with replacement, the same sample may appear multiple times in the training of the same tree, or it may not appear at all. This randomness of the samples helps to increase the differences between different decision trees. The randomness of the features means that when each node of the decision tree is split, instead of using all the features to find the optimal split point, a subset of features is randomly selected, and then the optimal split point is found within this subset. This randomness of the features further increases the differences between decision trees and helps to improve the generalization ability of the entire random forest. When using the random forest regression model for prediction, each decision tree will give its own prediction result (i.e., regression value) based on the input feature vector. Then, these prediction results are averaged to form the final prediction result. Since the decision trees in the random forest are trained independently and have randomness, their prediction results may be different. However, by averaging these prediction results, we can get a more stable and accurate prediction value.

[0050] Step 2: Obtain training samples based on historical process data, historical process parameters, historical production process impact data, and historical inspection data.

[0051] In some examples, the step of obtaining a training sample may include: The first step is to preprocess the historical process data, historical process parameters, historical production process impact data and historical inspection data to obtain preprocessed data.

[0052] Specifically, historical process data, historical process parameters, historical production process impact data, and historical inspection data may originate from the equipment layer 10 .

[0053] Exemplarily, preprocessing may include data cleaning, data integration, data formatting and data reduction. Data cleaning is to remove or correct errors, outliers and missing values ​​in data, mainly including eliminating abnormal data, selecting appropriate samples, filling missing values, etc., where error values ​​and outliers can be identified according to the judgment rules given by business experience, and missing values ​​can be deleted at a ratio of 5%. Data integration is to integrate data from different data sources to form a complete data set, mainly including steps such as data merging, data conversion and data integration, where data merging is to reorganize data from multiple sources according to process ID, and data integration is to centrally process and store data from different characteristic sources. Data formatting is to unify the granularity of multi-source heterogeneous data, ensure that data from different sources are consistent in terms of fineness, and also maintain uniformity in statistical caliber, ensure that the standards, methods or rules used in the process of data collection, calculation or analysis are consistent, and convert character and time data types into numerical formats. Data reduction is to aggregate data and select samples, exclude extremely individual samples that are not representative, and maintain the representativeness and accuracy of data.

[0054] In the second step, the features whose influence on the total sheet thickness variance meets the preset conditions are selected from the pre-processed historical process data, historical process parameters, and historical production process influence data to obtain the initial feature set.

[0055] Exemplarily, the features that have the greatest impact on the prediction target are selected from the preprocessed raw data to form the initial feature set. Specifically, first, the factors that have a significant impact on the label in the business can be sorted out through the prior knowledge of business logic; at the same time, the collected 5M1E data, process data and process data are subjected to correlation analysis, variance analysis, chi-square test, etc. with the label to analyze the degree of influence of factors in each dimension on the label; then, based on the significant factors sorted out based on business experience and the significant factors sorted out in the literature in this field, the results of the correlation analysis are screened to reduce the complexity of the model. According to business logic, statistical analysis and literature sorting, a total of 6 process boundary factors are screened out, including specifications, wire diameters, etc., 9 process parameter factors, including wire speed, feed speed, etc., 30 5M1E factors and process factors, including the use time of the groove wheel and the steel wire manufacturer, etc.

[0056] The third step is to extract and normalize the initial feature set to obtain the training feature set.

[0057] Exemplarily, feature extraction means creating new features by transforming or combining original features, including feature binning, numerical transformation, feature fusion, etc. Normalization processing means performing maximum-minimum standardized transformation on features to eliminate the dimensional differences between different features. For the specific value of each feature, the value can be subtracted from the minimum value of the feature, and then divided by the maximum value of the feature minus the minimum value of the feature to obtain the quotient of the normalization processing result. For details, please refer to the following formula: ; Among them, X normalization is the normalized result, X i is the i-th eigenvalue, X max is the maximum value of the feature, X min is the minimum value of the feature.

[0058] The fourth step is to obtain training samples based on the training feature set and the preprocessed historical test data.

[0059] Exemplarily, the training feature set may include 6 process boundary features D, 9 process parameter factors B and 10 other features C, such as 6 process boundary features D={D1, D2, D3, D4, D5, D6}, 9 process parameter factors B={B1, B2, B3, B4, B5, B6, B7, B8, B9}, and 10 other features are generated according to 30 5M1E and process factors and recorded as C={C1, C2, C3, C4, C5, C6, C7, C8, C9, C10}. The pre-processed historical inspection data includes TTV inspection results, which are represented by Y labels.

[0060] Step three: train and verify the prediction model based on the training samples to obtain the trained prediction model.

[0061] In some examples, the prediction model may be trained and validated based on training samples using a K-fold cross-validation method to obtain a trained prediction model.

[0062] Specifically, the K-fold cross-validation method is to divide the training samples into K subsets, use K-1 subsets to train the model each time, and use the remaining subset for validation. Repeat this process K times, calculate the performance index of each validation, and take the average as the final validation result.

[0063] When using K-1 subsets to train the model, the data can be further divided into a training set and a test set. The training set data is used to train the model, and the test set data is used to test the model. When testing and evaluating the model, the root mean square error (RMSE) and R² score can be used as the prediction model evaluation indicators to measure the difference between the model prediction value and the actual value. For details, please refer to the following formula: ; ; Where RMSE is the root mean square error, i is the index or iteration variable, N is the total number of samples in the data set; R 2 is the coefficient of determination, n is the nth sample in the data set, y i is the observed value of the i-th sample label in the data set, is the predicted value of the i-th sample label in the data set, is the mean of all sample label observations in the dataset.

[0064] After using the validation set to verify the model, you can also compare the performance indicators on the training set and the validation set to check for overfitting. You can alleviate the overfitting problem by adjusting the model complexity, adding regularization terms, collecting more data, etc.

[0065] Based on the validation results, model parameters can be adjusted using methods such as grid search, random search, or Bayesian optimization to find the best parameter combination and further improve model performance.

[0066] Through the above scheme, the embodiment of the present application introduces an advanced prediction model, which can successfully replace traditional practical experiments and significantly reduce experimental costs and time consumption. The prediction model can also quickly generate reliable process parameter recommendations based on historical data and algorithm optimization, effectively avoiding waste of resources and unnecessary experimental risks.

[0067] After obtaining the prediction model in the embodiment of the present application, the specific implementation method of step 203 is introduced below.

[0068] In some embodiments, step 203 may be implemented in the following manner: Step 1: Input each orthogonal test sample into the prediction model to obtain the total slice thickness variance prediction value corresponding to each orthogonal test sample.

[0069] Specifically, the prediction model is used to predict the TTV of each orthogonal test sample in the orthogonal test scheme, and the obtained TTV prediction value is used as the evaluation basis.

[0070] Step 2: Based on the predicted value of the total slice thickness variance corresponding to each orthogonal test sample, determine the sum of the total slice thickness variance and the average value of the total slice thickness variance of each factor at each level.

[0071] Specifically, the total thickness variance of each factor at each level is expressed by the following formula: ; The mean value of the total thickness variance for each factor at each level is expressed by the following formula: ; Among them, TTV(B t ,A j ) is factor B t At level A j The corresponding total thickness variance prediction value, KSUM(t,j) is factor B t At level A j The total thickness variance under the condition, KAVG(t,j) is the factor B t At level A j The average value of the total slice thickness variance under .

[0072] It can be understood that each orthogonal test sample includes multiple factors and the levels corresponding to each factor. The same level of the same factor may exist in multiple orthogonal test samples. For example, in orthogonal test samples 1, 2, and 4, the sum of the TTV prediction values ​​corresponding to orthogonal test samples 1, 2, and 4, respectively, can be used as the sum of the total slice thickness variance of the factor at this level, and the average of the TTV prediction values ​​corresponding to orthogonal test samples 1, 2, and 4, respectively, can be used as the average of the total slice thickness variance of the factor at this level.

[0073] Step 3: Determine the range of each factor based on the mean of the total slice thickness variance at each level of each factor.

[0074] For example, the maximum value Max(t) and the minimum value Min(t) of the total slice thickness variance average value in each factor can be determined based on the total slice thickness variance average value at each level of each factor. The difference between the maximum value Max(t) and the minimum value Min(t) is determined as the range R of the corresponding factor.

[0075] Step 4: Determine the alternative levels with the smallest total slice thickness variance under each factor in descending order of range, and determine the first parameter combination formed by the alternative levels of each factor as the alternative parameter combination.

[0076] Specifically, according to the order of range from large to small, the importance of each factor can be ranked, for example, R(2)>R(5)>R(6)>R(1)>R(7)>R(9)>R(8)>R(3)>R(4). According to the results of the range analysis, according to the importance of the factors from large to small, the alternative level with the smallest total thickness variance under each factor is selected in turn. Assume that under factor B(2), the alternative level corresponding to the minimum value of KSUM(2, j) is 3; under factor B(5), the alternative level corresponding to the minimum value of KSUM(5, j) is 2; under factor B(6), the alternative level corresponding to the minimum value of KSUM(6, j) is 5; under factor B(1), the alternative level corresponding to the minimum value of KSUM(1, j) is The alternative level is 7; under factor B(7), the alternative level corresponding to the minimum value of KSUM(7, j) is 9; under factor B(9), the alternative level corresponding to the minimum value of KSUM(9, j) is 10; under factor B(8), the alternative level corresponding to the minimum value of KSUM(8, j) is 1; under factor B(3), the alternative level corresponding to the minimum value of KSUM(3, j) is 2; under factor B(4), the alternative level corresponding to the minimum value of KSUM(4, j) is 4. Then the first parameter combination is {'B(1)': 7, 'B(2)': 3, 'B(3)': 2, 'B(4)': 4, 'B(5)': 2, 'B(6)': 5, 'B(7)': 9, 'B(8)': 1, 'B9': 10}.

[0077] Through the above scheme, through the deep integration of orthogonal experiments and prediction models, the parameter search is replaced by model prediction, which can save computing time to a certain extent, and can also effectively reduce the high cost of comprehensive experiments, and achieve efficient and economical process parameter optimization.

[0078] Step 204: determine the candidate parameter combination as the initial solution of the simulated annealing algorithm, and obtain the optimized parameter combination through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

[0079] Specifically, the simulated annealing algorithm is a global optimization algorithm inspired by the physical annealing process. It simulates the process of slowly cooling a metal after heating to find the lowest energy state. The algorithm starts with an initial solution, which can be randomly selected or obtained based on some heuristic method. As the algorithm proceeds, it continuously generates new candidate solutions through random perturbations, and accepts new solutions with a certain probability according to the Metropolis criterion based on the difference in the objective function value between the current temperature and the old and new solutions (i.e., the "energy difference"). When the new solution is better than the current solution, the algorithm accepts it unconditionally; when the new solution is worse, the algorithm accepts it with a probability inversely proportional to the temperature, which enables the algorithm to jump out of the local optimal solution during the search process and approach the global optimal solution. In the simulated annealing process, temperature is a key parameter that determines the probability of the algorithm accepting a worse solution. Initially, the temperature is set high, allowing the algorithm to accept a wider range of solutions so that it can explore extensively in the solution space. As the algorithm iterates, the temperature gradually decreases, the probability of accepting a poor solution also decreases, and the algorithm gradually tends to find a more refined solution space area. This temperature control mechanism enables the algorithm to quickly locate the approximate optimal solution area in the early stage of the search, and to fine-tune the quality of the solution in the later stage, so as to find the global optimal solution or a satisfactory solution close to the global optimal solution. The simulated annealing algorithm continuously iterates and cools down until the termination condition is met (such as reaching the preset number of iterations, the temperature drops to a low enough level, or the quality of the solution is no longer significantly improved). Due to its advantages such as global search capability, flexibility and universality, the simulated annealing algorithm has been widely used in engineering design, machine learning, signal processing, economics and other fields, and has become an important means to solve complex optimization problems.

[0080] In some embodiments, step 204 is implemented by: Step 1, define the neighborhood structure of the solution, and determine the candidate parameter combination as one of the candidate solutions, and determine the total thickness variance prediction value of the candidate solution as the fitness. The candidate parameter combination is the one with the smaller total thickness variance prediction value between the first parameter combination and the second parameter combination, and the second parameter combination is the orthogonal test sample with the smallest total thickness variance prediction value among all orthogonal test samples.

[0081] Specifically, before the iteration begins, the initial solution is set, as well as the algorithm's initial temperature, cooling rate, number of iterations, acceptance probability function and other parameters. The domain structure that defines the solution can set the upper and lower limits of the initial solution based on business experience, and combine candidate parameters as part of the candidate solution. The total thickness variance prediction value of the candidate solution is used as its fitness and compared with the fitness of the current solution.

[0082] Step 2: In each iteration, based on the Metropolis criterion, the candidate solution with poor fitness is accepted as the new current solution with a preset probability.

[0083] In this way, the global search capability of the algorithm can be increased.

[0084] Step three, lowering the temperature according to the set cooling rate so that the iterative result converges to the optimal solution or the suboptimal solution, wherein the preset probability decreases as the temperature decreases.

[0085] Specifically, the temperature is gradually reduced according to the set cooling rate. As the temperature decreases, the probability of accepting a worse solution also decreases, so that the algorithm gradually converges to the optimal solution or suboptimal solution.

[0086] Step 4: When the iteration reaches a preset end condition, the iteration is terminated, and the optimal solution or suboptimal solution is determined as the optimization parameter combination.

[0087] Exemplarily, the preset termination condition includes that the number of iterations reaches a maximum number of iterations or the change of the solution is less than a predetermined threshold.

[0088] Through the above scheme, in the application of the simulated annealing algorithm, the embodiment of the present application innovatively combines the orthogonal experiment with the range analysis, and uses the range analysis method of the orthogonal experiment to accurately locate the optimal parameter combination, providing high-quality initialization conditions for the simulated annealing algorithm, ensuring the stability and reliability of the algorithm initialization parameters, and significantly improving the algorithm convergence speed and solution quality. This not only helps the algorithm to quickly converge to the global optimal solution, but also significantly reduces the risk of falling into the local optimum, thereby improving the accuracy and practicality of process parameter recommendations.

[0089] In some embodiments, after executing step 204, the method of the embodiment of the present application further includes: The optimization parameter combination is displayed on the user interface based on a preset display form, where the preset display form includes at least one of a chart and a text.

[0090] It can be understood that the method of the embodiment of the present application constructs an accurate prediction model of chip cutting process parameters and product quality through big data and AI algorithm technology, and can perform comprehensive quality evaluation on different process parameter combinations, replacing traditional actual verification, and on this basis, combined with orthogonal experiments, search globally in the parameter space to provide reasonable initialization parameters for the simulated annealing optimization algorithm, and finally generate the final precise recommendation results through the simulated annealing algorithm, so as to realize intelligent optimization recommendation of process parameters, aiming to ensure the flexibility, refinement and systematicness of the optimization results to a certain extent, thereby ensuring the global or approximate global optimality of the search results to the greatest extent, significantly improving the optimization quality and efficiency, and reducing production costs.

[0091] Accordingly, the embodiment of the present application also provides a process optimization device, which can be arranged in the model layer 30. Figure 4 , Figure 4 The process optimization device provided in the embodiment of the present application includes a data interaction module 401 , an orthogonal design module 402 , a parameter prediction module 403 and a parameter optimization module 404 .

[0092] The data interaction module 401 is used to obtain the user's input data, and the input data includes process boundary conditions; An orthogonal design module 402 is used to generate an orthogonal test scheme based on input data by a preset orthogonal test design method, wherein the orthogonal test scheme includes a plurality of orthogonal test samples; A parameter prediction module 403 is used to determine candidate parameter combinations based on multiple orthogonal test samples and a prediction model, where the prediction model is constructed based on a random forest regression model; The parameter optimization module 404 is used to determine the candidate parameter combination as the initial solution of the simulated annealing algorithm, and obtain the optimized parameter combination through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

[0093] In some embodiments, the orthogonal design module 402 is specifically used to: Determine the factors and levels of the orthogonal experiment based on the process boundary conditions, where the factors are the process parameters to be optimized and the levels are the value ranges of the process parameters; Based on the factors and levels of the orthogonal experiment, Latin hypercube sampling method is used to select multiple orthogonal experiment samples.

[0094] In some embodiments, the prediction model is used to characterize the mapping relationship between process data, process parameters, production process influencing data and total slice thickness variance; the parameter prediction module 403 is specifically used to: Each orthogonal test sample is input into the prediction model to obtain the total slice thickness variance prediction value corresponding to each orthogonal test sample; Based on the predicted value of the total slice thickness variance corresponding to each orthogonal test sample, determine the sum of the total slice thickness variance and the average value of the total slice thickness variance of each factor at each level; Based on the mean of the total slice thickness variance for each factor at each level, the range of each factor was determined; According to the order of range from large to small, the alternative levels with the smallest total slice thickness variance under each factor are determined in turn, and the first parameter combination formed by the alternative levels of each factor is determined as the alternative parameter combination.

[0095] In some embodiments, the parameter prediction module 403 is specifically used to: Based on the average value of the total slice thickness variance of each factor at each level, determine the maximum and minimum values ​​of the average value of the total slice thickness variance in each factor; The difference between the maximum and minimum values ​​is determined as the range of the corresponding factor.

[0096] In some embodiments, the parameter optimization module 404 is specifically used to: A neighborhood structure of a solution is defined, and a candidate parameter combination is determined as one of the candidate solutions, and a total slice thickness variance prediction value of the candidate solution is determined as fitness, wherein the candidate parameter combination is the one with the smaller total slice thickness variance prediction value among the first parameter combination and the second parameter combination, and the second parameter combination is the orthogonal test sample with the smallest total slice thickness variance prediction value among all orthogonal test samples; In each iteration, the candidate solution with poor fitness is accepted as the new current solution with a preset probability based on the Metropolis criterion; The temperature is reduced according to a set cooling rate so that the iteration result converges to an optimal solution or a suboptimal solution, wherein the preset probability decreases as the temperature decreases; When the iteration reaches the preset end condition, the iteration ends and the optimal solution or suboptimal solution is determined as the optimization parameter combination.

[0097] In some embodiments, the prediction model is trained based on historical process data, historical process parameters, historical production process impact data, and historical inspection data; the device also includes a model acquisition module, which is used to acquire the prediction model through the following steps: Build a prediction model based on the random forest regression model; Obtain training samples based on historical process data, historical process parameters, historical production process impact data, and historical inspection data; The prediction model is trained and verified based on the training samples to obtain the trained prediction model.

[0098] In some embodiments, the model acquisition module is specifically used to: Preprocess historical process data, historical process parameters, historical production process impact data and historical inspection data to obtain preprocessed data; Select features whose influence on the total sheet thickness variance meets preset conditions from pre-processed historical process data, historical process parameters, and historical production process influence data to obtain an initial feature set; Perform feature extraction and normalization on the initial feature set to obtain a training feature set; Based on the training feature set and the preprocessed historical test data, training samples are obtained.

[0099] In some embodiments, the model acquisition module is specifically used to: Based on the training samples, the K-fold cross-validation method is used to train and verify the prediction model to obtain the trained prediction model.

[0100] In some embodiments, process data includes various observed parameters and states of production equipment during the process, process parameters include various process setting conditions that affect product quality, production process influencing data include human-machine-material-method-environment measurement data, and inspection data include total sheet thickness variance detection value.

[0101] In some embodiments, the device further comprises a display module: The display module is used to display the optimization parameter combination on the user interface based on a preset display form, and the preset display form includes at least one of a chart and a text.

[0102] It can be understood that the process optimization device of the embodiment of the present application, through the deep integration of orthogonal experiments and prediction models, can effectively replace traditional practical experiments, greatly reduce the test time, and improve the optimization efficiency of process parameters. At the same time, the use of alternative parameter combinations to provide a high-quality initialization level for the simulated annealing algorithm can significantly improve the convergence speed and solution quality of the algorithm, thereby improving the optimization accuracy of process parameters.

[0103] Accordingly, see Figure 5 , Figure 5 Schematic diagram of the structure of an electronic device of an embodiment of the present application. An electronic device provided in an embodiment of the present application includes a memory 501 and a processor 502. The memory 501 is used to store a computer program. The processor 502 is used to execute the computer program stored in the memory 501. When the computer program stored in the memory 501 is executed, the processor 502 executes the process optimization method of the aforementioned embodiment of the present application.

[0104] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to implement the process optimization method as described in the aforementioned embodiment of the present application when executed.

[0105] The above is a detailed introduction to a process optimization method, device, equipment and system provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the technical solution and its core idea of ​​the present application. Ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A process optimization method, characterized in that: include: Obtaining user input data, wherein the input data includes process boundary conditions; Based on the input data, an orthogonal test scheme is generated by a preset orthogonal test design method, wherein the orthogonal test scheme includes a plurality of orthogonal test samples; Determining candidate parameter combinations based on a plurality of the orthogonal test samples and a prediction model, wherein the prediction model is constructed based on a random forest regression model; The candidate parameter combination is determined as an initial solution of a simulated annealing algorithm, and an optimized parameter combination is obtained through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

2. The process optimization method according to claim 1, characterized in that: Based on the input data, an orthogonal test scheme is generated by a preset orthogonal test design method, including: Determining factors and levels of orthogonal experiments based on the process boundary conditions, wherein the factors are process parameters to be optimized, and the levels are value ranges of the process parameters; Based on the factors and levels of the orthogonal test, a plurality of the orthogonal test samples are selected by using the Latin hypercube sampling method.

3. The process optimization method according to claim 1, characterized in that: The prediction model is used to characterize the mapping relationship between process data, process parameters, production process influencing data and total sheet thickness variance; Determining an alternative parameter combination based on a plurality of the orthogonal test samples and the prediction model comprises: Input each of the orthogonal test samples into the prediction model to obtain a total slice thickness variance prediction value corresponding to each of the orthogonal test samples; Based on the predicted value of the total slice thickness variance corresponding to each of the orthogonal test samples, determine the sum of the total slice thickness variance and the average value of the total slice thickness variance for each factor at each level; Determine the range of each of the factors based on the average value of the total slice thickness variance of each of the factors at each of the levels; According to the order of range from large to small, the alternative levels with the smallest total slice thickness variance under each of the factors are determined in turn, and the first parameter combination formed by the alternative levels of each of the factors is determined as the alternative parameter combination.

4. The process optimization method according to claim 3, characterized in that: Determining the range of each factor based on the average value of the total slice thickness variance of each factor at each level includes: Based on the average value of the total slice thickness variance of each factor at each level, determining the maximum value and the minimum value of the average value of the total slice thickness variance in each factor; The difference between the maximum value and the minimum value is determined as the range corresponding to the factor.

5. The process optimization method according to claim 3, characterized in that: Obtaining an optimization parameter combination through the simulated annealing algorithm includes: A neighborhood structure of a solution is defined, and a candidate parameter combination is determined as one of the candidate solutions, and a total slice thickness variance prediction value of the candidate solution is determined as fitness, wherein the candidate parameter combination is the one with a smaller total slice thickness variance prediction value among the first parameter combination and the second parameter combination, and the second parameter combination is the orthogonal test sample with the smallest total slice thickness variance prediction value among all the orthogonal test samples; In each iteration, the candidate solution with poor fitness is accepted as the new current solution with a preset probability based on the Metropolis criterion; Lowering the temperature according to a set cooling rate so that the iteration result converges to an optimal solution or a suboptimal solution, wherein the preset probability decreases as the temperature decreases; When the iteration reaches a preset end condition, the iteration is ended, and the optimal solution or the suboptimal solution is determined as the optimization parameter combination.

6. The process optimization method according to claim 1, characterized in that: The prediction model is trained based on historical process data, historical process parameters, historical production process impact data and historical inspection data; the steps of obtaining the prediction model include: Constructing the prediction model based on the random forest regression model; Acquire training samples based on the historical process data, the historical process parameters, the historical production process impact data, and the historical inspection data; The prediction model is trained and verified based on the training samples to obtain the trained prediction model.

7. The process optimization method according to claim 6, characterized in that: The step of obtaining the training sample comprises: Preprocessing the historical process data, the historical process parameters, the historical production process impact data and the historical inspection data to obtain preprocessed data; Select features whose influence on the total sheet thickness variance meets preset conditions from pre-processed historical process data, historical process parameters, and historical production process influence data to obtain an initial feature set; Performing feature extraction and normalization processing on the initial feature set to obtain a training feature set; The training samples are obtained based on the training feature set and the preprocessed historical test data.

8. The process optimization method according to claim 6, characterized in that: Training and verifying the prediction model based on the training samples to obtain the trained prediction model includes: Based on the training samples, the prediction model is trained and verified using a K-fold cross-validation method to obtain the trained prediction model.

9. The process optimization method according to claim 6, characterized in that: The process data includes various observed parameters and states of the production equipment during the process, the process parameters include various process setting conditions that affect product quality, the production process influencing data includes man-machine-material-method-environment measurement data, and the inspection data includes the total sheet thickness variance detection value.

10. The process optimization method according to claim 1, characterized in that: After obtaining the optimization parameter combination through the simulated annealing algorithm, the method further includes: The optimization parameter combination is displayed on a user interface based on a preset display form, wherein the preset display form includes at least one of a chart and a text.

11. A process optimization device, characterized in that: include: A data interaction module, used to obtain input data from a user, wherein the input data includes process boundary conditions; An orthogonal design module, used to generate an orthogonal test scheme based on the input data by a preset orthogonal test design method, wherein the orthogonal test scheme includes a plurality of orthogonal test samples; A parameter prediction module, used to determine candidate parameter combinations based on a plurality of the orthogonal test samples and a prediction model, wherein the prediction model is constructed based on a random forest regression model; The parameter optimization module is used to determine the candidate parameter combination as the initial solution of the simulated annealing algorithm, and obtain the optimized parameter combination through the simulated annealing algorithm, and the optimized parameter combination is used for process optimization.

12. An electronic device, characterized in that: The electronic device includes a processor and a memory; a computer program is stored in the memory, and the processor is used to execute the computer program stored in the memory to implement the process optimization method according to any one of claims 1 to 10.

13. A process optimization system, characterized in that: include: The equipment layer is used to generate and obtain process data, process parameters, production process influencing data and inspection data; A data layer, the data layer is used to process the data acquired by the device layer; A model layer, wherein the model layer is used to perform the process optimization method according to any one of claims 1 to 10; The application layer is used to interact with the user to obtain input data and display the optimization parameter combination based on a preset display form.

Images