Cutting process optimization methods, devices and equipment

Abstract

This application provides a cutting process optimization method, apparatus, and equipment, relating to the field of multi-wire cutting. The method includes: determining the process boundary conditions for each parameter to be optimized, including wire feeding parameters and wire return parameters; determining an optimized parameter combination based on the process boundary conditions and a prediction model for each parameter to be optimized, wherein the prediction model is constructed based on a support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity; and the optimized parameter combination is used for cutting process optimization. This application can iteratively search for optimized parameter combinations based on the process boundary conditions and prediction model of the parameters to be optimized, thereby achieving autonomous optimization of cutting process parameters before cutting begins, significantly improving the efficiency and quality of cutting process optimization.

Description

Technical Field

[0001] This application relates to the field of multi-wire cutting technology, specifically to a cutting process optimization method, apparatus, and equipment. Background Technology

[0002] As wire diameters in multi-wire EDM become increasingly thinner and the application window for cutting wire narrows, the appropriate combination of steel wire and process becomes particularly important. To continuously improve cutting quality, process parameters need continuous optimization. Process parameter optimization is typically performed after wafer cutting is complete, seeking better process parameters based on cutting process indicators (such as breakage rate, wire bowing value, torque, etc.) and cutting results (such as yield, parameter defect rate, wire mark value, etc.), thus making process optimization somewhat delayed. Summary of the Invention

[0003] This application provides a cutting process optimization method, apparatus, and equipment to improve the problem of lagging process optimization.

[0004] In a first aspect, embodiments of this application provide a cutting process optimization method, including:

[0005] Determine the process boundary conditions for each parameter to be optimized, including wire feeding parameters and wire return parameters;

[0006] Based on the process boundary conditions and prediction model of each parameter to be optimized, an optimal parameter combination is determined. The prediction model is constructed based on a support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The optimal parameter combination is used to optimize the cutting process.

[0007] In some embodiments, determining the combination of optimization parameters based on the process boundary conditions and prediction model for each of the parameters to be optimized includes:

[0008] Based on the process boundary conditions for each of the parameters to be optimized, multiple candidate samples are determined;

[0009] Each candidate sample is input into the prediction model to obtain the predicted value of wire cutting non-uniformity corresponding to each candidate sample;

[0010] Based on the predicted value of wire cutting non-uniformity corresponding to each candidate sample and the process boundary conditions of each parameter to be optimized, the combination of optimization parameters is determined, and the wire cutting non-uniformity corresponding to the combination of optimization parameters satisfies the preset optimization termination condition.

[0011] In some embodiments, multiple candidate samples are determined based on the process boundary conditions of each parameter to be optimized, including:

[0012] Based on the process boundary conditions of each parameter to be optimized, a plurality of candidate samples are determined using the Latin hypercube sampling method.

[0013] In some embodiments, the prediction model is trained based on historical process parameters and the actual values ​​of wire cutting non-uniformity corresponding to the historical process parameters; the steps for obtaining the prediction model include:

[0014] The prediction model is constructed based on the support vector regression model, wherein the support vector regression model uses a radial basis kernel function.

[0015] Training samples are obtained based on the historical process parameters;

[0016] Determine the actual value of the wire cutting non-uniformity corresponding to the training sample;

[0017] Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the prediction model is trained and validated to obtain the trained prediction model.

[0018] In some embodiments, the prediction model is trained and validated based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples to obtain the trained prediction model, including:

[0019] Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the hyperparameters of the prediction model are optimized using a grid search method.

[0020] The prediction model is validated using cross-validation to obtain the trained prediction model.

[0021] In some embodiments, determining the wire cutting non-uniformity corresponding to the training samples includes:

[0022] Based on the training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit steel wire is determined to characterize the wear degree of the unit steel wire through the cumulative cutting volume.

[0023] Based on the cumulative cutting volume of each unit steel wire, the cutting volume difference is determined, and the cutting volume difference is used to characterize the actual value of the steel wire cutting non-uniformity corresponding to the training sample.

[0024] In some embodiments, determining the cutting volume difference based on the cumulative cutting volume of each unit steel wire includes:

[0025] The maximum and minimum values ​​are determined from the cumulative cutting volume of each unit steel wire;

[0026] The cutting volume difference is determined based on the maximum and minimum values.

[0027] In some embodiments, the historical process parameters include the wire feed length and wire return length for each cutting cycle, wherein the wire feed length is determined based on the wire feed speed, and the wire return length is determined based on the wire return speed; based on the training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit wire is determined, including:

[0028] Based on the training samples, the single crystal parameters, and the cutting bow parameters, the total cutting volume for each cutting cycle is determined.

[0029] Based on the wire feed length, wire return length, and total cutting volume of each cutting cycle, the single crystal cutting volume of each unit wire in each cutting cycle is determined.

[0030] The cumulative cutting volume of each unit steel wire is determined based on the single-crystal cutting volume of each unit steel wire in each cutting cycle.

[0031] Secondly, embodiments of this application provide a cutting process optimization apparatus, comprising:

[0032] The boundary condition determination module is used to determine the process boundary conditions for each parameter to be optimized, including the wire feeding parameters and the wire return parameters.

[0033] The optimization parameter determination module is used to determine the combination of optimization parameters based on the process boundary conditions and prediction model of each parameter to be optimized. The prediction model is constructed based on the support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The combination of optimization parameters is used to optimize the cutting process.

[0034] Thirdly, embodiments of this application provide an electronic device, the electronic device comprising:

[0035] At least one processor; and a memory communicatively connected to said at least one processor;

[0036] The memory stores a computer program executable by the at least one processor, which is executed by the at least one processor to enable the at least one processor to perform the cutting process optimization method as described in any of the first aspects, or to enable the processor to configure the cutting process optimization apparatus as described in the second aspect.

[0037] The cutting process optimization method provided in this application can iteratively search for optimal parameter combinations based on the process boundary conditions and prediction models of the parameters to be optimized, thereby enabling autonomous optimization of cutting process parameters before cutting begins, significantly improving the efficiency and quality of cutting process optimization. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the overall process of the cutting process optimization method according to an embodiment of this application;

[0040] Figure 2 This is a schematic diagram of the single crystal structure according to an embodiment of this application;

[0041] Figure 3 This is a schematic diagram illustrating the shape of the wire bow during the cutting process according to an embodiment of this application;

[0042] Figure 4 This is a schematic diagram showing the prediction results of the bow value for each cutting cycle in the embodiments of this application;

[0043] Figure 5 This is a schematic diagram illustrating the change in wire bow during the main blade cycle cutting process according to an embodiment of this application;

[0044] Figure 6 This is a schematic diagram illustrating the change in wire bow during the cutting process in the retraction cycle of this application embodiment;

[0045] Figure 7 This is a schematic diagram illustrating an example of calculating the number of wear cycles per unit steel wire according to an embodiment of this application;

[0046] Figure 8 This is a schematic diagram showing the wear distribution of a unit steel wire after cutting, before the cutting process is optimized, according to an embodiment of this application.

[0047] Figure 9 This is a schematic diagram showing the wear distribution of a unit steel wire after cutting, following the optimization of the cutting process, according to an embodiment of this application.

[0048] Figure 10 This is a schematic diagram of the cutting process optimization device according to an embodiment of this application;

[0049] Figure 11 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0050] Explanation of reference numerals in the attached figures:

[0051] 901 - Boundary condition determination module; 902 - Optimization parameter determination module; 1001 - Processor; 1002 - Memory. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0054] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.

[0055] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0056] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0057] As wire diameters in multi-wire EDM become increasingly thinner and the application window for cutting wire narrows, the appropriate combination of steel wire and process becomes particularly important. To continuously improve cutting quality, process parameters need continuous optimization. Process parameter optimization is typically performed after wafer cutting is complete, seeking better process parameters based on cutting process indicators (such as breakage rate, wire bowing value, torque, etc.) and cutting results (such as yield, parameter defect rate, wire mark value, etc.), thus making process optimization somewhat delayed.

[0058] In view of this, embodiments of this application provide a cutting process optimization method, apparatus, and equipment. By replacing the traditional comparison of post-cutting result indicators with digital simulation prediction before cutting, and by iteratively searching and optimizing parameter combinations through big data, the cutting process parameters can be autonomously optimized before cutting, thereby improving the efficiency and quality of cutting process optimization and solving at least some of the above-mentioned technical problems.

[0059] Please see Figure 1 , Figure 1 This is a schematic diagram of the overall flow of the cutting process optimization method according to an embodiment of this application. The cutting process optimization method specifically includes the following steps:

[0060] Step 101: Determine the process boundary conditions for each parameter to be optimized. The parameters to be optimized include wire feeding parameters and wire return parameters.

[0061] Specifically, the process boundary conditions for each parameter to be optimized can be set according to actual cutting requirements. The parameters to be optimized may include wire feeding parameters and wire return parameters, and may also include a process parameter vector. The wire feeding parameters may include wire feeding speed, the wire return parameters may include wire return speed, and the process parameter vector may include at least one other process parameter such as tension, wire speed, or position.

[0062] For example, the process boundary conditions may include parameter range constraints and process constraints. For instance, the wire feeding speed X1 satisfies: X1_min≤X1≤X1_max, the wire return speed X2 satisfies: X2_min≤X2≤X2_max, and the ratio of the wire feeding speed X1 to the wire return speed X2 satisfies: 0.8≤X1 / X2≤1.2.

[0063] Step 102: Based on the process boundary conditions and prediction model of each parameter to be optimized, determine the combination of optimization parameters. The prediction model is constructed based on the support vector regression model. The prediction model is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The combination of optimization parameters is used to optimize the cutting process.

[0064] Specifically, the Support Vector Regression (SVR) model is a regression analysis method based on Support Vector Machines (SVM). It is used to solve regression problems by finding an optimal hyperplane to minimize the error between predicted and actual values, while keeping the model's complexity as low as possible.

[0065] The following section will first introduce the construction of the prediction model.

[0066] In some examples, the prediction model was trained based on historical process parameters and the actual values ​​of wire cutting non-uniformity corresponding to those historical process parameters.

[0067] For example, historical process parameters refer to the process parameters used in historical single-crystal cutting processes. These parameters may include the wire feed length and wire return length for each cutting cycle. The wire feed length is determined based on the wire feed speed, and the wire return length is determined based on the wire return speed. Historical process parameters may also include the target distance between the winding wheel and the contact point between the wire and the crystal rod, wire mesh tension, table speed, average wire speed, slot pitch, and kerf width. The table speed, average wire speed, wire feed length, wire return length, wire mesh tension, slot pitch, and kerf width can all be extracted from the wire cutting machine.

[0068] In some embodiments, the step of obtaining the prediction model includes:

[0069] Step 1: Construct a prediction model based on the support vector regression model, which uses the radial basis kernel function.

[0070] Specifically, the Radial Basis Function (RBF) is represented by the following formula:

[0071] ;

[0072] Here, γ is the parameter of the kernel function, which controls the range of similarity and determines the sensitivity of the similarity between sample points. K(x, x') is the output of the kernel function, representing the similarity between input sample points x and x'. Through the kernel function, the SVR model can map the original data to a high-dimensional space, thereby performing linear regression in the high-dimensional space. The two vectors x and x' represent two input sample points, each of which is usually a vector composed of multiple features. For example, if the input data is two-dimensional, then x and x' might be vectors like (x_1, x_2) and (x'_1, x'_2). It is the square of the Euclidean distance between sample points x and x', and the formula is:

[0073] ;

[0074] Where n represents the feature dimension of the sample point (for example, if two-dimensional data has x_1 and x_2, then n=2). This distance metric reflects the similarity between two sample points; the closer the distance, the higher the similarity and the smaller the value; the farther the distance, the lower the similarity and the larger the value.

[0075] The parameters of the SVR model can be configured, such as penalty factor (C): 1.0, kernel function parameter (γ): 'auto', precision parameter (ε): 0.1, tolerance parameter (tolerance): 0.001, and maximum number of iterations: 1000.

[0076] Step 2: Obtain training samples based on historical process parameters.

[0077] Specifically, training samples can be constructed based on the wire feed speed, wire return speed, and other process parameters for each cutting cycle. Understandably, each training sample represents a combination of different values ​​for multiple historical process parameters.

[0078] In some examples, after obtaining the training samples, all input features can be standardized using MinMax, as shown in the following formula:

[0079] ;

[0080] Among them, X norm X represents the standardized input features, and X represents the input features from the training samples. min X is the smallest input feature in the training samples. max The largest input feature in the training samples.

[0081] Step 3: Determine the actual value of the wire cutting non-uniformity corresponding to the training sample.

[0082] In some examples, step three is implemented as follows:

[0083] The first step is to determine the cumulative cutting volume of each unit steel wire based on the training samples, single crystal parameters, and cutting bow parameters, so as to characterize the wear degree of the unit steel wire through the cumulative cutting volume.

[0084] For example, the single crystal parameters include single crystal width and single crystal length, the training samples include slot pitch and kerf width, and the cutting cycle includes main blade cycle and cut-off cycle.

[0085] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a single crystal according to an embodiment of this application. Specifically, the single crystal parameters refer to the dimensional parameters of the single crystal. For example, taking a cuboid-shaped single crystal as an example, the single crystal parameters may include the single crystal width W, the single crystal height H, and the single crystal length L. The single crystal width W is the dimension along the first direction X of the single crystal, the single crystal length L is the dimension along the second direction Y of the single crystal, and the single crystal height H is the dimension along the third direction Z of the single crystal. The first direction X, the second direction Y, and the third direction Z intersect each other. For example, the single crystal width W can typically be 210 mm or 182 mm.

[0086] For example, the cumulative cutting volume of each unit of steel wire is determined as follows:

[0087] First, based on the training samples, single crystal parameters, and cutting bow parameters, the total cutting volume for each cutting cycle is determined.

[0088] Specifically, during the cutting process, the single crystal moves along the third direction Z, while the steel wire moves back and forth along the first direction X to cut the single crystal in the XZ plane. The cutting cycle refers to the time corresponding to one feed and one return of the steel wire. For example, if the feed length of the steel wire is 1000 meters (m) and the return length is 500 meters, then the process of the steel wire being fed out for 1000 meters and returning for 500 meters constitutes one cutting cycle. Exemplarily, the cutting cycle includes the main cutting cycle and the retraction cutting cycle. The main cutting cycle represents the cutting cycle during which the steel wire has not yet cut to the top surface T of the single crystal, and the retraction cutting cycle represents the cutting cycle during which the steel wire has cut to the top surface T of the single crystal.

[0089] In some examples, the cutting bow parameters include the predicted bow variation for each cutting cycle.

[0090] Specifically, the predicted change in the bow wire refers to the difference between the bow wire value at the end of each cutting cycle and the bow wire value at the beginning of each cutting cycle.

[0091] Please see Figure 3 , Figure 3This is a schematic diagram illustrating the shape of the wire bow during the cutting process according to an embodiment of this application. In this embodiment, the shape of the wire bow during the cutting process is simulated as an arc. The wire bow value can be characterized by the wire bow height b, which refers to the height along the third direction Z between the winding wheel P and the contact point C between the steel wire and the crystal rod.

[0092] In some examples, the predicted change in bow length for each cutting cycle can be determined using the following steps:

[0093] The training samples are input into the pantograph prediction model to obtain the output of the pantograph prediction model. The pantograph prediction model is built based on the XGBoost model and trained based on historical process parameters and actual pantograph values.

[0094] The output results are denormalized to obtain the predicted change in the bow line for each cutting cycle.

[0095] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating the predicted bow value for each cutting cycle in this embodiment. After obtaining the predicted bow change through the bow prediction model, the bow value for each cutting cycle can be obtained by combining it with the initial bow prediction value. Combined with the grooved wheel position and cutting depth, the following can be obtained: Figure 4 The diagram shown is shown in the image.

[0096] The following describes how the pantograph prediction model is obtained according to an embodiment of this application.

[0097] For example, obtaining the bow prediction model includes the following steps:

[0098] (1) Obtain the training dataset.

[0099] First, historical process parameters are collected for each product specification, each cutting wire diameter, and under different process conditions (e.g., table speed, average line speed, wire feed length, wire return length, bow value, wire mesh tension, and wire mesh location). Table speed, average line speed, wire feed length, wire return length, and wire mesh tension can all be extracted from the wire cutting machine. Actual bow value and wire mesh location can be collected using a bow detection device linked to the wire cutting machine. The bow detection device collects one set of data per cutting cycle, and the wire mesh location is collected by the bow detection device with sequential numbering along the wire mesh direction, and the location density is adjustable.

[0100] Then, the collected data undergoes preprocessing. This includes data cleaning and feature engineering. Data cleaning may include deleting samples with missing values ​​exceeding a preset proportion (e.g., 20%), filling missing values ​​of numerical features with the median, filling missing values ​​of categorical features with the mode, identifying outliers based on the 3σ criterion, and replacing outliers using the upper and lower quartiles method. Feature engineering includes defining the input feature data, data transposition and reconstruction, and feature processing. The input feature data may include numerical and categorical features. Numerical features include, for example, tension (rightZL), average table speed (agvFeedSpdde), front-cut bow (xianGongData), average line speed (agvLineSpeed), and millimeter line consumption (lineQuantity). Categorical features include, for example, left and right sides (zuoYouXuanZe), cutting status (qiegeMode), and point location variables (xianGongData_1-90).

[0101] Data transposition and reconstruction includes the following steps:

[0102] First, the feature columns are defined. Based on the time-series data of the cutting process, the following feature columns are constructed: a cutting mode feature column, containing cutting status data for at least some cutting points; and a bow data feature column, containing actual bow values ​​(i.e., bow measurement values) for at least some cutting points. Simultaneously, basic feature columns are defined, including tension value (rightZL), average table speed (agvFeedSpdde), average line speed (agvLineSpeed), line quantity per millimeter (lineQuantity), and left / right sides (zuoYouXuanZe).

[0103] Secondly, perform the following operations on each cut record:

[0104] a. Extract the cutting position markers, for example, the range is 0 to 90, or the range is 11 to 80, which can be determined according to actual needs.

[0105] b. Transpose the feature column data of each group to convert them into the following identifiers: cutting mode (qiegeMode) and bow data (xianGongData).

[0106] c. Add a location index column (location) to record the corresponding cutting location identifier.

[0107] d. Copy the basic feature data to each cutting position record.

[0108] e. Merge all transposed data records.

[0109] Finally, null value cleanup is performed on the reconstructed data, deleting data records containing any null values ​​to ensure data integrity.

[0110] The above method can convert the original wide table format data into a long table format that is easier for modeling and analysis, with each record containing complete feature information at a specific cutting position. The converted data structure is more suitable for subsequent machine learning modeling processes and can make full use of the position-related feature information during the cutting process.

[0111] Feature processing includes the following steps:

[0112] The numerical features are Robust standardized using the following formula:

[0113] x_robust=(x-median) / IQR;

[0114] Where x_robust represents the feature after Robust standardization, median is the feature median, and IQR is the interquartile range (Q3-Q1). This method is less sensitive to outliers compared to traditional standardization methods.

[0115] Categorical features are processed using the built-in categorical feature processing methods of the XGBoost model, enabled via the `enable_categorical` parameter. In categorical features, left and right sides are set to category type with values ​​{'0', '1'}; cutting states are set to category type with values ​​{'cutting in', 'main cutting', 'cutting out'}; and point variables are set to category type with values ​​{'0-90'}.

[0116] The above approach is more efficient than one-hot encoding in obtaining training datasets, can automatically process category features, and learns the optimal feature segmentation method during model training.

[0117] (2) Construct the XGBoost regression model.

[0118] Specifically, the parameters for configuring the XGBoost regression model include: learning rate (learning_rate) set to 0.1, maximum tree depth (max_depth) set to 6, minimum number of leaf node samples (min_child_weight) set to 3, subsample ratio (subsample) set to 0.8, feature sampling ratio (colsample_bytree) set to 0.8, number of iterations (n_estimators) set to 100, and regularization parameters including L1 regularization coefficient (alpha) = 0.01 and L2 regularization coefficient (lambda) = 1.

[0119] (3) Use the training dataset to train the XGBoost regression model to obtain the line bow prediction model.

[0120] Specifically, the training dataset can be divided into a training set and a validation set in a 7:3 ratio. The XGBoost regression model can then be trained and validated using the training and validation sets respectively. For example, a 5-fold cross-validation method can be used to evaluate model performance, and Bayesian optimization can be used to optimize hyperparameters, with the optimization objective being to minimize MAPE (Mean Absolute Percentage Error). Finally, a well-trained line bow prediction model can be obtained.

[0121] (4) Perform denormalization on the output of the pantograph prediction model to obtain the pantograph prediction change for each cutting cycle.

[0122] Specifically, the output results can be destandardized using the following formula:

[0123] y_actual = y_pred * IQR + median;

[0124] Where y_actual is the predicted change in bow length for each cutting cycle, y_pred is the output result of the model prediction for each cutting cycle, IQR is the interquartile range of the target variable, and median is the median of the target variable.

[0125] (5) Evaluate the predicted changes in the bow line for each cutting cycle.

[0126] Specifically, calculable reliability metrics for prediction results include MAPE (Mean Absolute Percentage Error) and R... 2 The system calculates the following parameters: Coefficient of Determination (CDE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Finally, it outputs the predicted bow data, reliability assessment metrics for the prediction results, and the prediction confidence interval to generate a prediction report.

[0127] Through the above-described scheme, the embodiments of this application can achieve accurate simulation of the cutting process, providing reliable data support for the identification of design defects. The accuracy of the prediction results is verified by actual data; the MAPE predicted by the bow data can be controlled within 15%.

[0128] It is understandable that before inputting the training samples into the bow prediction model for prediction, the same preprocessing process as the historical process parameters can be performed on the training samples. For details, please refer to the relevant content mentioned above, which will not be repeated here.

[0129] Please refer to the following: Figure 3 and Figure 5 , Figure 5 This is a schematic diagram illustrating the change in line bow during the main cutter cycle cutting process according to an embodiment of this application. In some embodiments, the total cutting volume of the main cutter cycle is determined through the following steps:

[0130] Based on the training samples, the single crystal width W, and the predicted change Δb of the main cutting cycle, the total cutting area S of the main cutting cycle is determined.

[0131] The total cutting volume V of the main cutter cycle is determined based on the total cutting area S, single crystal length L, groove spacing d1, and kerf width d2 of the main cutter cycle.

[0132] For example, a method for determining the total cutting area S of the main tool cycle includes:

[0133] Based on the table speed and the duration of the main cutting cycle, the feed distance s for each main cutting cycle is determined.

[0134] Based on the first arc prediction value b1 at the start of each master cutter cycle, the second arc prediction value b2 at the end of each master cutter cycle, the target distance c, the single crystal width W, and the feed distance s, the total cutting area S of each master cutter cycle is determined. The first arc prediction value b1 and the second arc prediction value b2 are determined based on the arc prediction change Δb of the master cutter cycle and the initial arc prediction value b0.

[0135] The feed distance *s* for each main cutter cycle can be calculated by multiplying the table speed by the duration of the main cutter cycle. The target distance *c* is the distance along the first direction *X* between the winding wheel P and the contact point C between the steel wire and the crystal rod. Typically, the initial bow prediction value *b0* is 0. It can be understood that the first bow prediction value *b1* at the beginning of each main cutter cycle is the same as the second bow prediction value *b2* at the end of the previous main cutter cycle. The first bow prediction value *b1* at the beginning of the first main cutter cycle is the initial bow prediction value *b0*, which is 0.

[0136] Specifically, the total cutting area S for each master cutter cycle is determined through the following steps:

[0137] Based on the first predicted value b1 of the linear bow, the target distance c, and the single crystal width W at the beginning of each master cutter cycle, the first angle θ1 and the first radius R1 are determined.

[0138] Based on the second line bow prediction value b2, target distance c, and single crystal width W at the end of each master cutter cycle, the second angle θ2 and the second radius R2 are determined.

[0139] Based on the first angle θ1, the first radius R1, the second angle θ2, the second radius R2, and the feed distance s, the total cutting area S for each master tool cycle is determined.

[0140] The formulas for determining the first angle θ1, the first radius R1, the second angle θ2, and the second radius R2 are as follows:

[0141] θ1=arctan(b1 / c), θ2=arctan(b2 / c);

[0142] R1=W / 2 / sinθ1, R2=W / 2 / sinθ2.

[0143] The formula for determining the total cutting area S for each master cutting cycle is as follows:

[0144] S = W × s + S1 - S2.

[0145] S1 is the first area enclosed by the first arc and the single crystal at the beginning of the main cutting cycle, and S2 is the second area enclosed by the second arc and the single crystal at the end of the main cutting cycle. The specific formula is as follows:

[0146] S1=πR1 2 ×θ1 / 2π×2-W×R1×cosθ1 / 2;

[0147] S² = πR² 2 ×θ2 / 2π×2-W×R2×cosθ2 / 2.

[0148] Understandable Figure 3 In the diagram, 'a' represents the height from the contact point C between the steel wire and the crystal rod to the lowest point A of the wire bow, 'h' represents the actual feed amount at the lowest point A of the wire bow, and 'k' represents the feed distance of the winding wheel P relative to the bottom surface of the single crystal. Therefore, 'h' = kab = k - (R - Rcosθ) - b. At the beginning of the main cutting cycle, 'b' represents b1 and 'R' represents R1; at the end of the main cutting cycle, 'b' represents b2 and 'R' represents R2.

[0149] Specifically, the formula for the total cutting volume V of the main tool cycle is as follows:

[0150] V = S × (n-1) × d2.

[0151] Where n is the number of slices into which the single crystal is cut, the specific calculation formula is as follows:

[0152] n=L / d1.

[0153] Please refer to the following: Figure 3 and Figure 6 , Figure 6This is a schematic diagram illustrating the change in wire bow during the cutting process in the retraction cycle according to an embodiment of this application. In other embodiments, the total cutting volume of the retraction cycle is determined through the following steps:

[0154] Based on the training samples and the change Δb in the bow curve prediction during the cut-out cycle, the total cutting area S during the cut-out cycle is determined.

[0155] The total cutting volume V of the cut-out cycle is determined based on the total cutting area S, single crystal length L, and groove spacing d1 of the cut-out cycle.

[0156] For example, a method for determining the total cutting area S of the cutter retraction cycle includes:

[0157] Based on the first bow prediction value b1 at the start of each cut-and-close cycle, the second bow prediction value b2 at the end of each cut-and-close cycle, the target distance c, and the single crystal width W, the total cutting area S of each cut-and-close cycle is determined. The first bow prediction value b1 and the second bow prediction value b2 are determined based on the bow prediction change Δb and the initial bow prediction value b0 of the cut-and-close cycle.

[0158] Specifically, the total cutting area S for each cutter cycle is determined through the following steps:

[0159] Based on the first predicted value b1 of the linear bow, the target distance c, and the single crystal width W at the beginning of each cut-off cycle, the first angle θ1 and the first radius R1 are determined.

[0160] Based on the second bow prediction value b2, target distance c, and single crystal width W at the end of each cut-off cycle, the second angle θ2 and the second radius R2 are determined.

[0161] Based on the first angle θ1 and the first radius R1, the second angle θ2 and the second radius R2, the total cutting area S for each cutter retraction cycle is determined.

[0162] The formulas for determining the first angle θ1, the first radius R1, the second angle θ2, and the second radius R2 are as follows:

[0163] θ1=arctan(b1 / c), θ2=arctan(b2 / c);

[0164] R1=W / 2 / sinθ1, R2=W / 2 / sinθ2.

[0165] The formula for determining the total cutting area S for each cutter cycle is as follows:

[0166] S = S1 - S2.

[0167] S1 is the first area enclosed by the first arc and the single crystal at the beginning of the termination cycle, and S2 is the second area enclosed by the second arc and the single crystal at the end of the termination cycle. The specific formula is as follows:

[0168] S1=πR1 2 ×θ1 / 2π×2-R1 2 ×sinθ1×cosθ1;

[0169] S² = πR² 2 ×θ² / 2π×2-R² 2 ×sinθ2×cosθ2.

[0170] Specifically, the formula for the total cutting volume V during the tool recovery cycle is as follows:

[0171] V = S × (n-1);

[0172] Where n is the number of slices into which the single crystal is cut, the specific calculation formula is as follows:

[0173] n=L / d1.

[0174] Using the above scheme, the total cutting volume of different cutting cycles is calculated using single crystal parameters and training samples according to different methods, thus providing a data basis for subsequent calculation of the cutting volume of a unit steel wire.

[0175] Next, based on the wire feed length, wire return length, and total cutting volume of each cutting cycle, the single crystal cutting volume of each unit wire in each cutting cycle is determined.

[0176] Specifically, a unit steel wire refers to the steel wire corresponding to a unit length. The unit length can be set according to actual needs. For example, if the unit length is 1m, then for 1000m of steel wire, there are 1000 unit steel wires. When calculating the degree of steel wire wear in each cycle, the movement of the steel wire on the wire mesh is considered as the wear caused by cutting single crystals.

[0177] For example, the step of determining the single-crystal cutting volume of each unit steel wire in each cutting cycle includes:

[0178] Based on the wire feed length, wire return length, and wire length located inside the single crystal in each cutting cycle, the number of wear cycles per unit wire in each cutting cycle is determined.

[0179] The total cutting volume of each cutting cycle is allocated according to the number of wear cycles per unit steel wire in the corresponding cutting cycle, thereby determining the single crystal cutting volume per unit steel wire in each cutting cycle.

[0180] Please see Figure 7 , Figure 7 This is a schematic diagram illustrating an example of calculating the number of wear cycles per unit steel wire according to an embodiment of this application. In some examples, determining the number of wear cycles per unit steel wire in each cutting cycle can be achieved through the following steps:

[0181] In each cutting cycle, based on the order of cutting into the single crystal, the position values ​​of each unit steel wire corresponding to the wire feeding length, each unit steel wire located inside the single crystal, and each unit steel wire corresponding to the wire return length are determined, so as to characterize the number of wear times of the unit steel wire in the corresponding cutting cycle through the position values.

[0182] For example, the unit length of the steel wire is set to 1m, and a value of 1 is assigned for every 1000m the unit steel wire moves on the wire. Then, in one cutting cycle, the wire feed length is 1000m, and the wire return length is 500m. This means that in one cutting cycle, 1000m of steel wire will newly enter the single crystal, and 1000m of steel wire will exit the single crystal, with the entire steel wire within the single crystal moving forward 1000m. For better understanding, the 1000m of steel wire about to enter the single crystal is divided into 1000 segments by unit length, from closest to furthest from the single crystal, and named sequentially as segment 1, segment 2, segment 3... segment 999, segment 1000. During the 1000m wire delivery process, the first segment of steel wire advances 1000m and is assigned a value of 1; the second segment advances 999m and is assigned a value of 0.999; the third segment advances 998m and is assigned a value of 0.998; ... the 999th segment advances 1m and is assigned a value of 0.001; the 1000th segment advances 0m and is assigned a value of 0. Following this assignment logic, 1000m of steel wire will leave the single crystal. The assignment results for this 1000m of steel wire are 1, 0.999, 0.998 ... 0.001, 0. Excluding the 1000m of steel wire that leaves, the remaining unit length of steel wire within the single crystal moves 1000m, so all of them are assigned a value of 1. The number of 1s is the length of the steel wire within the single crystal minus 1000. This assignment method can be understood as all the steel wires on the spool being within the single crystal. The steel wires entering during the wire feeding cycle enter the single crystal sequentially, and the steel wires leaving the single crystal leave the single crystal sequentially.

[0183] Based on this assignment method, within the return wire cutting cycle, the values ​​are assigned as 0.5, 0.449, 0.448…0.001, 0, 0.5, 0.5…0.5, 0.5, 0.449, 0.448…0.001, 0. The number of times the steel wire moves within the single crystal with an assignment of 0.5 is equal to the length of the steel wire within the single crystal minus 500. The total number of times the steel wire wears within this cycle is defined as the sum of these assignments.

[0184] In some examples, the single-crystal cutting volume per unit steel wire in each cutting cycle can be determined by the following steps:

[0185] The weight of each unit steel wire in the cutting cycle is determined based on the point assignment of each unit steel wire and the sum of the point assignments of all unit steel wires corresponding to the cutting cycle.

[0186] For example, taking a cutting cycle of 1000m of steel wire fed and 500m of wire returned as an example, the sum of the point assignments for all unit steel wires in this cycle is 1+0.999+0.998+……+0.001+0+1×(total length of the monocrystalline inner wire mesh - 1000)+1+0.999+0.998+……+0.001+0 (this is the total assignment result of the steel wire during wire feeding and cutting)+0.5+0.449+0.448+……+0.001+0+0.5×(total length of the monocrystalline inner wire mesh - 500)+0.5+0.449+0.448+……+0.001+0. The quotient of the point assignment of each unit steel wire to the sum of the point assignments of all unit steel wires corresponding to this cutting cycle is determined as the weight of each unit steel wire in the cutting cycle.

[0187] Based on the weight of each unit steel wire in the cutting cycle and the total cutting volume of the cutting cycle, the single crystal cutting volume of each unit steel wire in the cutting cycle is determined.

[0188] For example, the single-crystal cutting volume of each unit steel wire in the cutting cycle is determined by multiplying the weight of each unit steel wire in the cutting cycle with the total cutting volume of the cutting cycle.

[0189] Finally, the cumulative cutting volume of each unit steel wire is determined based on the single-crystal cutting volume of each unit steel wire in each cutting cycle.

[0190] For example, the step of determining the cumulative cutting volume for each unit of steel wire includes:

[0191] The cumulative cutting volume of a unit steel wire of the same length is determined by summing the single-crystal cutting volumes in each cutting cycle.

[0192] Specifically, each unit of steel wire has a corresponding length mark. For example, if a spool is 300,000 meters long, then each meter of steel wire has a corresponding length mark, namely the 1st meter, the 2nd meter, ..., the 300,000th meter.

[0193] It is understandable that a unit steel wire of the same length will participate in multiple cutting cycles, and its wear count will differ in each cycle. For example, in one cutting cycle, the 1000th unit steel wire has a wear count of 1, while in the next cutting cycle, the 500th unit steel wire has a wear count of 0.5. Therefore, by summing the single-crystal cutting volumes of the unit steel wire of the same length in each cutting cycle, the cumulative cutting volume of the unit steel wire can be determined.

[0194] To more clearly describe the single-crystal cutting volume per unit steel wire in each cutting cycle and the cumulative cutting volume per unit steel wire, please refer to [link to relevant documentation]. Figure 8 , Figure 8This diagram illustrates the wear distribution of a unit steel wire after cutting, before the optimization of the cutting process, according to an embodiment of this application. The horizontal axis represents the length of the steel wire (meters), and the vertical axis represents the volume (in cubic millimeters). Curves a1 to a5 represent the single-crystal cutting volume (i.e., the number of wear cycles per cut) of each unit steel wire in the five cutting cycles before the cutting process optimization, respectively. Curve e1 represents the cumulative cutting volume (i.e., the total number of wear cycles) of each unit steel wire in the five cutting cycles before the cutting process optimization.

[0195] It is understood that the method in this application embodiment can quantify the wear degree of a unit steel wire by determining the cumulative cutting volume of a unit steel wire, thereby enabling a more scientific assessment of the impact of the steel wire on the process and seeking process optimization solutions.

[0196] The second step is to determine the cutting volume difference based on the cumulative cutting volume of each unit steel wire. The cutting volume difference is used to characterize the actual value of the steel wire cutting non-uniformity corresponding to the training sample.

[0197] For example, the difference in cutting volume can be determined in the following way:

[0198] The maximum value V is determined from the cumulative cutting volume of each unit steel wire. max and minimum value V min .

[0199] Based on the maximum value V max and minimum value V min Determine the difference in cutting volume f(X).

[0200] Specifically, the difference in cutting volume f(X) can be determined using the following formula:

[0201] f(X) = (V max -V min ) / V max ×100%.

[0202] In other words, based on the current training samples, the cumulative cutting volume V of each unit steel wire is obtained, and the maximum value V is determined from it. max and minimum value V min Then, the cutting volume difference f(X) is determined using the above formula to characterize the actual value of the wire cutting non-uniformity corresponding to the training sample.

[0203] In other examples, the actual value of wire cutting non-uniformity corresponding to the training samples can also be determined in other ways, and this application embodiment does not specifically limit this.

[0204] Step 4: Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, train and validate the prediction model to obtain the trained prediction model.

[0205] In some examples, step four is implemented as follows:

[0206] Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the hyperparameters of the prediction model are optimized using the grid search method.

[0207] The prediction model is validated using cross-validation to obtain the trained prediction model.

[0208] Specifically, grid search is a systematic hyperparameter tuning method primarily used to find the optimal combination of hyperparameters for machine learning models. It iterates through and verifies all possible parameter combinations within a predefined range of parameter values ​​in an exhaustive manner, ultimately selecting the parameter configuration with the best performance.

[0209] The model's workflow includes: first, the user defines the range of hyperparameters to be optimized; then, a grid search iterates through all possible parameter combinations, training the model one by one; finally, cross-validation is used to evaluate the performance of each parameter group, and the best-performing parameter combination is selected based on the evaluation results. The model's input features are training samples, and its output features are the predicted values ​​of wire cutting non-uniformity corresponding to the training samples. Based on any given hyperparameter combination, the difference between the predicted and actual values ​​of wire cutting non-uniformity is used to evaluate the current hyperparameter combination, ultimately finding the optimal hyperparameter combination and obtaining the trained prediction model.

[0210] The prediction model in this application is based on a support vector regression model. Its core idea is to introduce a "tolerance margin" to fit the data points while minimizing model complexity. The goal is to find a function that is as smooth as possible, ensuring that all training data points fall within a defined error tolerance range (ε-margin). Only data points exceeding the tolerance margin will affect the model's final result; these data points are called support vectors. The model is constructed by minimizing model complexity (the norm of the weight vectors) and prediction error (the deviation beyond the tolerance range). The model's workflow includes defining the error tolerance range (ε), constructing the objective function, and prediction. Defining the error tolerance range (ε) allows the error between the predicted and true values ​​to remain unpenalized within this range. Constructing the objective function involves optimizing constraints to find the hyperplane that best fits the data. Prediction is performed using the support vectors and the optimized model.

[0211] In this context, support vectors are key sample points in the training data that directly affect the model's predictions. The high-dimensional data points of the support vectors are the mapping results of the support vectors (i.e., key sample points) retained during the training phase in a high-dimensional space. During prediction, new sample points are mapped to the same high-dimensional space, and their similarity with the high-dimensional representations of these support vectors is calculated (through a kernel function). Finally, a weighted sum is obtained to obtain the predicted value.

[0212] In some embodiments, after the prediction model is constructed, step 102 can be implemented through the following steps:

[0213] Step 1: Based on the process boundary conditions of each parameter to be optimized, determine multiple candidate samples.

[0214] In some examples, multiple candidate samples can be determined using the Latin hypercube sampling method based on the process boundary conditions of each parameter to be optimized.

[0215] Specifically, the Latin hypercube sampling method can be used to generate candidate samples within the parameter constraints corresponding to the process boundary conditions of each parameter to be optimized. Latin hypercube sampling is a method of approximately random sampling from a multivariate parameter distribution; it belongs to stratified sampling techniques and is commonly used in computer experiments or Monte Carlo integration. This ensures that the candidate samples can uniformly cover the search space.

[0216] Step 2: Input each candidate sample into the prediction model to obtain the predicted value of wire cutting non-uniformity for each candidate sample.

[0217] Specifically, before inputting candidate samples into the prediction model, all input features can be standardized using MinMax to unify feature parameters of different dimensions.

[0218] Step 3: Based on the predicted value of wire cutting non-uniformity corresponding to each candidate sample and the process boundary conditions of each parameter to be optimized, determine the combination of optimized parameters. The wire cutting non-uniformity corresponding to the combination of optimized parameters satisfies the preset optimization termination condition.

[0219] Specifically, after predicting the percentage difference in cutting volume for each candidate sample, the search direction can be adjusted based on the prediction results. Optimization can be stopped when the prediction results meet the preset termination condition, and the optimal parameter combination can be obtained. During the search process, the predicted value of wire cutting non-uniformity can be determined as the objective function, and the search is performed with the goal of minimizing the objective function. This determines the candidate sample corresponding to the minimization of the objective function, and this candidate sample is then used as the optimal parameter combination. The formula for minimizing the objective function is as follows:

[0220] .

[0221] The objective function aims to reduce uneven cutting volume differences during the cutting process, for example, by reducing the percentage difference from 20% to below 5%.

[0222] In other words, the preset optimization termination condition can be set to reduce the wire cutting non-uniformity to below a preset threshold, which can be set to 5%. In addition, the preset optimization termination condition can also be set to other termination conditions, such as stopping when a preset number of iterations is met. This application embodiment does not specifically limit this.

[0223] Please see Figure 9 , Figure 9 This diagram illustrates the wear distribution of a unit steel wire after cutting, following the optimization of the cutting process in this embodiment of the application. The horizontal axis represents the length of the steel wire (meters), and the vertical axis represents the volume (in cubic millimeters). Curves a6 to a10 represent the single-crystal cutting volume (i.e., the number of wear cycles per cut) of each unit steel wire in the five cutting cycles after the cutting process optimization, respectively. Curve e2 represents the cumulative cutting volume (i.e., the total number of wear cycles) of each unit steel wire in the five cutting cycles after the cutting process optimization. Compared to... Figure 8 The wear distribution shown before optimization demonstrates a significant improvement in cutting volume variation.

[0224] In some embodiments, the method of this application can also output and display the optimized parameter combination and the predicted value of wire cutting non-uniformity corresponding to the optimized parameter combination. Alternatively, an optimization report can be generated, which includes a comparison of parameter combinations before and after optimization, the improvement of wire cutting non-uniformity, and suggestions for adjusting process parameters. The specific output format can be set according to actual needs, and this application embodiment does not specifically limit it.

[0225] It is understood that the cutting process optimization method of this application can achieve autonomous optimization of the cutting process before cutting begins, significantly reducing the difference in cutting volume and improving the stability, quality, and efficiency of the production process. Experimental results show that this method can reduce the percentage of cutting volume difference from the initial 20% to below 5%, and the optimized process parameters meet actual production needs, exhibiting strong repeatability and stability, and providing effective technical support for the continuous improvement of the cutting process.

[0226] Accordingly, this application also provides a cutting process optimization device. Please refer to... Figure 10 , Figure 10 This is a schematic diagram of the cutting process optimization device according to an embodiment of this application. The cutting process optimization device provided in this embodiment includes a boundary condition determination module 901 and an optimization parameter determination module 902.

[0227] The boundary condition determination module 901 is used to determine the process boundary conditions for each parameter to be optimized, including the wire feeding parameters and the wire return parameters.

[0228] The optimization parameter determination module 902 is used to determine the combination of optimization parameters based on the process boundary conditions and prediction model of each parameter to be optimized. The prediction model is constructed based on the support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The combination of optimization parameters is used to optimize the cutting process.

[0229] In some embodiments, the optimization parameter determination module 902 is specifically used for:

[0230] Based on the process boundary conditions for each parameter to be optimized, multiple candidate samples are determined.

[0231] Each candidate sample is input into the prediction model to obtain the predicted value of wire cutting non-uniformity for each candidate sample.

[0232] Based on the predicted value of wire cutting non-uniformity corresponding to each candidate sample and the process boundary conditions of each parameter to be optimized, the combination of optimized parameters is determined, and the wire cutting non-uniformity corresponding to the combination of optimized parameters satisfies the preset optimization termination condition.

[0233] In some embodiments, the optimization parameter determination module 902 is specifically used for:

[0234] Based on the process boundary conditions for each parameter to be optimized, the Latin hypercube sampling method is used to determine multiple candidate samples.

[0235] In some embodiments, the prediction model is trained based on historical process parameters and the actual values ​​of wire cutting non-uniformity corresponding to those historical process parameters. The device also includes a model acquisition module, which acquires the prediction model through the following steps:

[0236] A prediction model is constructed based on the support vector regression model, which employs a radial basis kernel function.

[0237] Training samples are obtained based on historical process parameters.

[0238] Determine the actual value of the wire cutting non-uniformity corresponding to the training samples.

[0239] Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the prediction model is trained and validated to obtain the trained prediction model.

[0240] In some embodiments, the model acquisition module is specifically used for:

[0241] Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the hyperparameters of the prediction model are optimized using the grid search method.

[0242] The prediction model is validated using cross-validation to obtain the trained prediction model.

[0243] In some embodiments, the model acquisition module is specifically used for:

[0244] Based on training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit steel wire is determined to characterize the wear degree of the unit steel wire through the cumulative cutting volume.

[0245] Based on the cumulative cutting volume of each unit steel wire, the cutting volume difference is determined. The cutting volume difference is used to characterize the actual value of the steel wire cutting non-uniformity corresponding to the training sample.

[0246] In some embodiments, the model acquisition module is specifically used for:

[0247] The maximum and minimum values ​​are determined from the cumulative cutting volume of each unit of steel wire.

[0248] The difference in cutting volume is determined based on the maximum and minimum values.

[0249] In some embodiments, historical process parameters include the wire feed length and wire return length for each cutting cycle. The wire feed length is determined based on the wire feed speed, and the wire return length is determined based on the wire return speed. The model acquisition module is specifically used for:

[0250] Based on the training samples, single crystal parameters, and cutting bow parameters, the total cutting volume for each cutting cycle is determined.

[0251] Based on the wire feed length, wire return length, and total cutting volume of each cutting cycle, the single crystal cutting volume of each unit wire in each cutting cycle is determined.

[0252] The cumulative cutting volume of each unit steel wire is determined based on the single crystal cutting volume of each unit steel wire in each cutting cycle.

[0253] It is understood that the apparatus in this application embodiment can iteratively search for optimal parameter combinations based on the process boundary conditions and prediction model of the parameters to be optimized, thereby enabling autonomous optimization of cutting process parameters before cutting begins, significantly improving the efficiency and quality of cutting process optimization.

[0254] Accordingly, please refer to Figure 11 , Figure 11This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. An electronic device provided in this application includes at least one processor 1001 and a memory 1002 communicatively connected to the at least one processor 1001. The memory 1002 stores a computer program executable by the at least one processor 1001. The computer program is executed by the at least one processor 1001 to enable the at least one processor 1001 to execute the cutting process optimization method of this application embodiment, or to enable the processor 1001 to configure the cutting process optimization device of this application embodiment.

[0255] Accordingly, embodiments of this application also provide a computer-readable storage medium storing computer instructions, which are used to cause a processor to implement the cutting process optimization method of the embodiments of this application when executed, or to enable a processor to configure the cutting process optimization apparatus of the embodiments of this application.

[0256] It should be noted that the electronic device provided in this application embodiment and the cutting process optimization method in the above embodiments belong to the same concept. Any method provided in the cutting process optimization method embodiment can be run on the electronic device. For details of its implementation process, please refer to the cutting process optimization method embodiment, which will not be repeated here. The embodiments, implementation methods and related technical features of this application can be combined and substituted with each other without conflict.

[0257] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0258] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Although this application has disclosed preferred embodiments as above, it is not intended to limit this application. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the technical solution of this application. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for optimizing a cutting process, characterized in that, include: Determine the process boundary conditions for each parameter to be optimized, including wire feeding parameters and wire return parameters; Based on the process boundary conditions and prediction model for each parameter to be optimized, a combination of optimization parameters is determined. The prediction model is constructed based on a support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The combination of optimization parameters is used to optimize the cutting process. The prediction model is trained based on historical process parameters and the actual values ​​of wire cutting non-uniformity corresponding to the historical process parameters. Determining the actual values ​​of wire cutting non-uniformity corresponding to the training samples includes: Based on the training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit steel wire is determined to characterize the wear degree of the unit steel wire through the cumulative cutting volume; the training samples are obtained based on the historical process parameters, and the cutting bow parameters include the predicted change in bow for each cutting cycle; Based on the cumulative cutting volume of each unit steel wire, the cutting volume difference is determined, and the cutting volume difference is used to characterize the actual value of the steel wire cutting non-uniformity corresponding to the training sample.

2. The cutting process optimization method according to claim 1, characterized in that, Based on the process boundary conditions and prediction model for each of the parameters to be optimized, the combination of optimized parameters is determined, including: Based on the process boundary conditions for each of the parameters to be optimized, multiple candidate samples are determined; Each candidate sample is input into the prediction model to obtain the predicted value of wire cutting non-uniformity corresponding to each candidate sample; Based on the predicted value of wire cutting non-uniformity corresponding to each candidate sample and the process boundary conditions of each parameter to be optimized, the combination of optimization parameters is determined, and the wire cutting non-uniformity corresponding to the combination of optimization parameters satisfies the preset optimization termination condition.

3. The cutting process optimization method according to claim 2, characterized in that, Based on the process boundary conditions for each of the parameters to be optimized, multiple candidate samples are determined, including: Based on the process boundary conditions of each parameter to be optimized, a plurality of candidate samples are determined using the Latin hypercube sampling method.

4. The cutting process optimization method according to claim 1, characterized in that, The steps for obtaining the prediction model include: The prediction model is constructed based on the support vector regression model, wherein the support vector regression model uses a radial basis kernel function. Training samples are obtained based on the historical process parameters; Determine the actual value of the wire cutting non-uniformity corresponding to the training sample; Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the prediction model is trained and validated to obtain the trained prediction model.

5. The cutting process optimization method according to claim 4, characterized in that, Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the prediction model is trained and validated to obtain the trained prediction model, including: Based on the training samples and the actual values ​​of wire cutting non-uniformity corresponding to the training samples, the hyperparameters of the prediction model are optimized using a grid search method. The prediction model is validated using cross-validation to obtain the trained prediction model.

6. The cutting process optimization method according to claim 1, characterized in that, Based on the cumulative cutting volume of each unit steel wire, the difference in cutting volume is determined, including: The maximum and minimum values ​​are determined from the cumulative cutting volume of each unit steel wire; The cutting volume difference is determined based on the maximum and minimum values.

7. The cutting process optimization method according to claim 1, characterized in that, The historical process parameters include the wire feeding length and wire return length for each cutting cycle. The wire feeding length is determined based on the wire feeding speed, and the wire return length is determined based on the wire return speed. Based on the training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit steel wire is determined, including: Based on the training samples, the single crystal parameters, and the cutting bow parameters, the total cutting volume for each cutting cycle is determined. Based on the wire feed length, wire return length, and total cutting volume of each cutting cycle, the single crystal cutting volume of each unit wire in each cutting cycle is determined. The cumulative cutting volume of each unit steel wire is determined based on the single-crystal cutting volume of each unit steel wire in each cutting cycle.

8. A cutting process optimization device, characterized in that, include: The boundary condition determination module is used to determine the process boundary conditions for each parameter to be optimized, including the wire feeding parameters and the wire return parameters. The optimization parameter determination module is used to determine the combination of optimization parameters based on the process boundary conditions and prediction model of each parameter to be optimized. The prediction model is constructed based on the support vector regression model and is used to characterize the mapping relationship between process parameters and wire cutting non-uniformity. The combination of optimization parameters is used to optimize the cutting process. The prediction model is trained based on historical process parameters and the actual values ​​of wire cutting non-uniformity corresponding to the historical process parameters. The device further includes a model acquisition module, which is used for: Based on training samples, single crystal parameters, and cutting bow parameters, the cumulative cutting volume of each unit steel wire is determined to characterize the wear degree of the unit steel wire through the cumulative cutting volume. The training samples are obtained based on the historical process parameters, and the cutting bow parameters include the predicted change in bow for each cutting cycle. Based on the cumulative cutting volume of each unit steel wire, the cutting volume difference is determined, and the cutting volume difference is used to characterize the actual value of the steel wire cutting non-uniformity corresponding to the training sample.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the cutting process optimization method as described in any one of claims 1-7, or to enable the processor to configure the cutting process optimization device as described in claim 8.

Images