KR102585970B1 - Apparatus and method for optimization of injection process parameter - Google Patents

Abstract

An electronic device is disclosed. This electronic device classifies quality based on process variable data including a plurality of process variables and variable values for each of the plurality of process variables, quality data that distinguishes good or defective products according to the process variable data, and a plurality of process variables. A memory storing a plurality of different classification models learned from process variable data and quality data and a regression model that calculates the value of one process variable among the plurality of process variables using the remaining process variables, and process variables used in the plurality of classification models Variable importance, which is the degree of influence on quality depending on the change in variable value, is calculated for each variable, and based on the calculated variable importance, a plurality of key process variables are selected from a plurality of process variables, and a plurality of key process variables selected from the process variable data are calculated. Using data corresponding to the process variables, a regression model for each major process variable is learned. When a new process variable value is input, a regression model for each major process variable and the new process variable value is used to learn a plurality of major process variables. It includes a processor that calculates the optimal value for each process variable.

Description

Apparatus and method for optimization of injection process parameters}

The present disclosure relates to an apparatus and method for optimizing injection process variables. More specifically, a device for selecting key process variables that affect the quality of injection products and calculating optimal values for the selected key process variables. and methods.

The injection process is one of the plastic molding methods. It is a process of heating thermoplastic resin into a fluid state, injecting it into a mold, and cooling it within the mold to create a molded product according to the shape of the mold.

The injection process is automatically controlled, and many injection factories set the injection conditions for one mold once and do not change them for 2 to 4 weeks or more, resulting in frequent occurrence of defects after a certain point. .

This is due to the aging process equipment, etc., and in order to resolve these defects, the currently set injection conditions must be appropriately modified according to the state of the process equipment.

However, because the process variables for injection conditions are intertwined, it is difficult to determine the main process variables that affect quality, and it is difficult to perform optimization for all process variables.

Therefore, the present disclosure is intended to solve the problems described above, by selecting key process variables that affect the quality of injection molded products based on data obtained from current process equipment and calculating optimal values for the key process variables. The purpose is to provide a device and method for doing so.

The present disclosure is intended to achieve the above object, and the present electronic device distinguishes good or defective products according to process variable data including a plurality of process variables and variable values for each of the plurality of process variables, and the process variable data. Quality is classified based on one quality data and a plurality of process variables, and the process variable data, a plurality of different classification models learned with the quality data, and the value of one process variable among the plurality of process variables are used to use the remaining process variable. Variable importance, which is the degree of influence on quality according to changes in variable values, is calculated for each process variable used in the memory in which the calculating regression model is stored and the plurality of classification models, and the plurality of processes are based on the calculated variable importance. A plurality of key process variables are selected among the variables, a regression model for each of the plurality of key process variables is learned using data corresponding to the selected plurality of key process variables among the process variable data, and new process variable values are calculated. When input, it includes a processor that calculates the optimal value for each of the plurality of main process variables using the new process variable value and a regression model for each of the plurality of main process variables.

In this case, the plurality of different classification models may be models learned by up-sampling the process variable data corresponding to defective products among the quality data and using all data including the up-sampled data.

In this case, the processor may calculate the variable importance based on a change in results resulting from input of random values for each process variable into the plurality of classification models.

In this case, if the classification model is a decision tree model, the processor calculates the variable importance based on at least one of the number and location of nodes using the process variable in the decision tree model, and the classification model is K nearest neighbors. If it is a neighborhood model, the variable importance can be calculated based on data purity when data is classified for each process variable.

In this case, the processor selects a plurality of main classification models based on prediction accuracy among the plurality of different classification models, and selects main process variables commonly included in the selected plurality of main classification models. .

In this case, the plurality of classification models are classification models learned with first data that is part of the process variable data and the quality data, and the prediction accuracy is the second classification model excluding the first data among the process variable data and the quality data. It may be prediction accuracy calculated from data.

Meanwhile, the memory stores a plurality of different regression models, and the processor trains the plurality of different regression models for each major process variable and calculates the plurality of regression models based on error values of the plurality of different regression models. A main regression model can be selected for each main process variable, and the optimal value can be calculated using the selected main regression model.

Meanwhile, the memory further stores information on whether the plurality of process variables are controllable and controllable alternative process variables for uncontrollable process variables, and the processor is configured to determine whether the selected main process variables are uncontrollable process variables. In this case, the regression model can be trained using the process variable data corresponding to alternative process variables for the selected main process variables.

Meanwhile, the electronic device further includes a user interface and a communication device, wherein the processor displays the calculated optimal value for each of the plurality of main process variables on the user interface and commands optimization for the main process variable. When input through the user interface, a command to modify the main process variable value according to the calculated optimal value can be transmitted to the process device through the communication device.

In this case, if the error between the new process variable value and the calculated optimal value is more than a threshold, the processor sends a command to modify the main process variable value according to the calculated optimal value to the process device through the communication device. It can be sent to .

The process variable optimization device according to the present disclosure can select key process variables so that process variables can be adjusted to reduce defects from the time defects frequently occur, and calculate optimal values for the selected key process variables.

In addition, it is possible to minimize the defect rate of injection molded products by providing the user with optimal values for major process variables or actively modifying the currently set process variable values to correspond to the optimal values.

In addition, optimization of process variables can be performed periodically, and even if the mold for the injection process changes, the optimal value can be easily calculated based on data according to the injection process.

1 is a block diagram for explaining the configuration of an electronic device according to an embodiment of the present disclosure.
Figure 2 is a diagram for explaining up-sampling and down-sampling of data according to an embodiment of the present disclosure.
3A and 3B are diagrams for explaining the prediction accuracy of a plurality of different classification models by up-sampling data according to an embodiment of the present disclosure.
4A and 4B are diagrams for explaining the prediction accuracy of a plurality of different classification models by down-sampling data according to an embodiment of the present disclosure.
Figure 5 is a diagram for explaining the importance of variables for each classification model according to an embodiment of the present disclosure.
Figure 6 is a diagram illustrating a screen displayed on a user interface according to an embodiment of the present disclosure.
7 and 8 are flowcharts for explaining a process variable optimization method according to an embodiment of the present disclosure.

The examples described below are illustrative to aid understanding of the present disclosure, and it should be understood that the present disclosure can be implemented with various modifications, different from the examples described herein. However, in describing the present disclosure below, if it is determined that detailed descriptions of related known functions or components may unnecessarily obscure the gist of the present disclosure, the detailed descriptions and specific illustrations will be omitted.

The terms used in this specification and claims are general terms selected in consideration of the function of the present disclosure. However, these terms may vary depending on the intention of technicians working in the field, legal or technical interpretation, and the emergence of new technologies.

Additionally, some terms are arbitrarily selected by the applicant. These terms may be interpreted as defined in this specification, and if there is no specific term definition, they may be interpreted based on the overall content of this specification and common technical knowledge in the relevant technical field.

In addition, since this specification describes the components necessary for explanation of each example of the present disclosure, it is not necessarily limited thereto. Accordingly, some components may be changed or omitted, and other components may be added. It can also be distributed and deployed on different independent devices.

In the description of the present disclosure, the order of each step should be understood as non-limiting unless the preceding step must be performed logically and temporally prior to the subsequent step. In other words, except for the above exceptional cases, even if the process described as a subsequent step is performed before the process described as a preceding step, the nature of disclosure is not affected, and the scope of rights must also be defined regardless of the order of the steps.

In this specification, the term “A or B” is defined not only to selectively indicate either A or B, but also to include both A and B. In addition, in the present disclosure, the term “includes” has the meaning of including additional components other than those listed as included.

And in this disclosure, “value” is defined as a concept that includes not only scalar values but also vectors.

The mathematical operations and calculations of each step of the present disclosure described later may be implemented in computer operations by known coding methods and/or coding designed to be suitable for the present disclosure to perform the corresponding operations or calculations.

Hereinafter, examples of the present disclosure will be described in detail with reference to the accompanying drawings and the contents described in the accompanying drawings, but the present disclosure is not limited or limited by the examples.

Hereinafter, the present disclosure will be described in more detail with reference to the attached drawings.

1 is a block diagram for explaining the configuration of an electronic device according to an embodiment of the present disclosure.

Referring to FIG. 1 , an electronic device 100 according to an embodiment of the present disclosure includes a memory 110, a processor 120, a user interface 130, and a communication device 140. The electronic device 100 may be a device such as a PC, laptop, or server, as well as a device included in the process equipment itself.

At least one instruction regarding the electronic device 100 may be stored in the memory 110 . Specifically, the memory 110 may store various programs (or software) for operating the electronic device 100 according to various embodiments of the present disclosure. This memory 110 may be implemented in various forms such as RAM, ROM, flash memory, HDD, external memory, memory card, etc., and is not limited to any one.

In addition, the memory 110 may store process variable data including a plurality of process variables and variable values for each of the plurality of process variables obtained from the injection processing device, and quality data that distinguishes good or defective products according to the process variable data. .

Process variable data obtained from the injection processing device can be obtained using an Internet of Things sensor (IoT sensor) included in the processing device, but is not limited to this and can also be obtained from a Manufacturing Execution System (MES).

The plurality of process variables may include temperature-related, pressure-related, time-related, rate-related and/or volume-related data. According to one example, filling time is the filling time when the contents are injected from the injection machine into the mold, max back pressure is the maximum pressure to prevent the screw from being pushed out while the resin is being metered, and the screw is pushed out while the resin is being metered. are the average back pressure, which is the average pressure to prevent the phenomenon; the plasticizing position, which is the position of the screw that has finished weighing; and the clamp, which is the time it takes for the injection machine to close the open mold after the product is produced and to hold the fixed and moving axes without leaving any gaps. It may include, but is not limited to, closing time, barrel temperature, which is the temperature at which the resin maintains constant melting during metering and injection.

Quality data is data that distinguishes whether the injection product is a good product or a defective product according to the variable values of the plurality of variable data, and can be stored in correspondence with the process variable data.

The memory 110 may store a plurality of different classification models that have been learned to classify quality based on a plurality of process variables.

A classification model is a type of supervised learning model and refers to a model that divides the boundaries of variables by identifying the correlation between variables for the result values of a series of data. According to one example, a decision tree classifier, a random forest classifier, a K-nearest neighbor classifier, a light gradient boosting machine ), etc., but is not limited thereto.

According to one example, a classification model with different algorithms refers to a classification model learned with the same process variable data and quality data, and can receive process variable values as input and output quality data for good or defective products, and each The accuracy of each classification model may be different.

The memory 110 may store a regression model that calculates the value of one process variable among a plurality of process variables using the remaining process variables.

A regression model refers to a model that learns the relationship between one dependent variable and a plurality of independent variables and predicts the dependent variable value for the input independent variable value. According to one example, a linear regression model, nonlinear regression model, decision tree regression model, random forest regression model, k-nearest neighbors It includes, but is not limited to, a regression model (k-nearest neighbor regression model).

A regression model according to an example may be learned using process variable data corresponding to data corresponding to non-defective products among quality data. Since training is performed using multiple process variable values included in the process variable data, the learned regression model can calculate the value of one process variable using the remaining process variables.

The memory 110 may store one type of regression model, but is not limited thereto and may store a plurality of different regression models.

Additionally, the memory 110 may store information on whether a plurality of process variables are controllable and information on controllable alternative process variables for uncontrollable process variables. Controllable process variables refer to process conditions that can be set directly for the process equipment. In one example, injection pressure, heating temperature, etc. are included, but are not limited thereto. Uncontrollable process variables include injection speed, material temperature, etc., and refer to process conditions whose variable values cannot be directly controlled. The memory 110 may store information on controllable alternative process variables according to correlations between variables for uncontrollable process variables. According to one example, the injection speed is related to the injection pressure and the temperature of the material, and the temperature of the material is related to the heating temperature, so the injection pressure and heating temperature can be stored as alternative process variables for the injection speed condition.

The processor 120 controls the overall operation of the electronic device 100. Specifically, the processor 120 may generally control the operation of the electronic device 100 by executing at least one instruction stored in the memory 110. The processor 120 may be composed of a single device such as a central processing unit (CPU) or an application-specific integrated circuit (ASIC), or may be composed of a plurality of devices such as a CPU or a graphics processing unit (GPU).

The user interface 130 provides various UI screens, and the user can input various selection commands by directly touching the UI screen or manipulating buttons provided on the user interface 130. The UI screen displayed on the user interface 130 may display process variable values currently set in the process device, or the user may input separate process variable values, but is not limited thereto.

The communication device 140 is formed to connect the electronic device 100 with an external device or process device (not shown), and is connected to the external device or process device through a local area network (LAN) and the Internet network. In addition to the form, it can also be connected through a USB (Universal Serial Bus) port or wireless communication (eg, WiFi 802.11a/b/g/n, NFC, Bluetooth) port. This communication device 140 may also be referred to as a transceiver.

The communication device 140 may receive data from the outside or transmit commands to control process equipment.

Figure 2 is a diagram for explaining up-sampling and down-sampling of data according to an embodiment of the present disclosure.

As for quality data obtained from processing equipment, data on good products may be greater than data on defective products. According to one example, the data for good products may be 99% and the data for defective products may be 1%, and this is because the initial process variables are set to generate good injection products.

If quality data is severely imbalanced, the number of data samples can be adjusted to ensure objectivity in model performance. Upsampling is a method of reducing data imbalance by increasing the number of data on the side with less data belonging to the corresponding category, and downsampling is a method of reducing data imbalance by reducing the number of data on the side with more data belonging to the corresponding category.

Referring to FIG. 2, the number of data is shown in the case of up-sampling the defective product data (210) and the case of down-sampling the good product data (220) from the initial data (200) with severe data imbalance.

3A and 3B are diagrams for explaining the prediction accuracy of a plurality of different classification models by up-sampling data according to an embodiment of the present disclosure.

Referring to FIG. 3A, the table 310 shows 13 different classification models with different algorithms, such as a random forest classification model, a light gradient boosting classification model, and an extra trees classifier. This is the result of data learning using a model, and the number of models is not limited to this.

As shown in table 310 of FIG. 3A, even if a plurality of different classification models are trained with the same upsampled process variable data and quality data, the prediction accuracy of each classification model may differ depending on the characteristics of the data.

The accuracy for each classification model can be determined by averaging the accuracy values for multiple learning sessions, as shown in table 320 of FIG. 3B.

According to an example, the table 320 of FIG. 3B shows the accuracy judgment results for the random forest classification model 330 that showed the highest accuracy. According to an example, the average value of prediction accuracy (340) is obtained by learning a total of 10 times. can be calculated.

A multiple-time learning method according to an example is to learn a classification model with first data that is part of the process variable data and quality data, and input second data excluding the first data among the process variable data into the learned classification model to obtain a classification model. Prediction accuracy can be determined by comparing the results with the second data among the quality data.

Table 320 of FIG. 3B according to one example shows prediction accuracy judgment performed a total of 10 times. This divides the process variable data and quality data into 10 equal parts, trains a classification model using the remaining data sets excluding one of the divided data sets, and uses the excluded data set to determine the accuracy of the classification model. In this way, each 10-part data set was repeated 10 times.

Referring to FIG. 3A, it can be seen that when a plurality of different classification models are learned after up-sampling the data, the accuracy of the classification models ranges from 0.8532 to 0.9957.

4A and 4B are diagrams for explaining the prediction accuracy of a plurality of different classification models by down-sampling data according to an embodiment of the present disclosure.

Referring to FIG. 4A, the table 410 is the result of learning using 13 different classification models as in FIG. 3A, and the table 420 in FIG. 4B is the extra tree classification model with the highest accuracy. This shows the accuracy judgment results for . Regarding the accuracy determination process, since it is similar to the content described in FIGS. 3A and 3B, redundant description will be omitted.

Referring to Figure 4a, it can be seen that when the data is down-sampled and then a plurality of different classification models are learned, the accuracy of the classification models is 0.7467 to 0.9178.

Comparing the contents of FIG. 3A with the contents of FIG. 4A, the accuracy of the classification model appears to be high when the classification model is learned by up-sampling defective product data. Accordingly, a plurality of different classification models in the memory 110 may up-sample process variable data corresponding to defective products among quality data and store the learned model using all data including the up-sampled data.

Figure 5 is a diagram for explaining the importance of variables for each classification model according to an embodiment of the present disclosure.

The processor 120 may calculate variable importance for each process variable used in each classification model for a plurality of different learned classification models stored in the memory 110. Variable importance refers to the degree to which changes in the variable value of a process variable affect the results, and variable importance may be judged differently depending on the type of algorithm on which the classification model is based.

According to one example, variable importance may be determined based on result changes resulting from input of random values for each process variable into a plurality of classification models. When quality data is maintained and the value of one process variable among the plurality of process variables included in the corresponding process variable data is replaced with a random value and input into the learned classification model, the accuracy of the classification model is compared to the accuracy at the time of learning. This may indicate low accuracy. The above-described process is performed for each of a plurality of process variables, and the importance of the variable can be calculated based on the numerical value with reduced accuracy. In other words, if the accuracy is significantly lowered when the variable value of a process variable is randomly replaced, it can be seen that the process variable has a significant influence on the calculation results.

According to one example, if the classification model is a decision tree classification model, variable importance may be calculated based on at least one of the number and location of nodes that use process variables in the decision tree model. Alternatively, if the classification model is a K-nearest neighbor classification model, variable importance may be calculated based on data purity when data is classified for each process variable, but is not limited to this.

In FIG. 5 according to an example, the first graph 510 shows variable importance values for each process variable for the random forest classification model, the second graph 520 shows variable importance for the extra tree classification model, and the third graph 530 shows variable importance values for each process variable for the random forest classification model. This shows variable importance for the light gradient boosting classification model.

The processor 120 may select a plurality of key process variables from a plurality of process variables based on the calculated variable importance. According to one example, 10 process variables with high variable importance may be selected from each of a plurality of different learned classification models, and process variables commonly included in each classification model may be selected as main process variables. The number of selected process variables with high variable importance is not limited to this and the user can select them separately.

Additionally, the processor 120 may select a plurality of main classification models based on prediction accuracy among a plurality of different classification models and select main process variables commonly included in the plurality of selected main classification models.

According to one example, in Figure 5, three models that showed high prediction accuracy in Figure 3 are selected, and the importance of variables for each classification model is shown. According to an example, the variables with high variable importance in common in each classification model are average back pressure, plasticizing position, max back pressure, filling time, clamp close time, and barrel temperature 5, and each process variable can be selected as a major process variable. there is.

The processor 120 uses the controllability and alternative process variable information for the process variable stored in the memory 110 to select a replacement process variable for the selected main process variable if the selected main process variable is an uncontrollable variable. It can also be selected as a key process variable.

The processor 120 may learn a regression model for each of the plurality of major process variables using data corresponding to a plurality of major process variables selected from among the process variable data stored in the memory 110. According to one example, when the selected main process variable is an uncontrollable process variable, a regression model may be trained using process variable data corresponding to an alternative process variable for the selected main process variable.

According to one example, the regression model for each major variable learned may calculate the value of one process variable among a plurality of process variables using the remaining process variables.

When a new process variable value is input, the processor 120 may calculate the optimal value for each of the plurality of main process variables using the new process variable value and a regression model learned for each of the plurality of main process variables. The calculated optimal value can be calculated as the optimal value for each major variable included in the new process variable.

According to one example, the processor 120 may learn each of a plurality of different regression models stored in the memory 110 for each major process variable. In this case, the processor 120 may learn the error values of the plurality of different regression models. Based on this, you can also select the main regression model. The error value can be determined by inputting the process variable value into the model and comparing the calculated result value with the actual process variable value.

The main regression model can be selected for each main process variable, and the error value can be determined by learning the regression model multiple times with different data and judging the average of the error values as the error value of the regression model, as described in Figure 3. You can.

The processor 120 may calculate the optimal value for the new input process variable value using a main regression model selected for each key variable.

Figure 6 is a diagram illustrating a screen displayed on a user interface according to an embodiment of the present disclosure.

When a new process variable value is input (610), the processor 120 may display the calculated optimal value (620) for each of a plurality of major process variables on the user interface (130). It is not limited to this, and the processor 120 may display the error 630 between the new process variable value and the calculated optimal value on the user interface 130.

When an optimization command for the main process variable is input through the user interface 130, the processor 120 transmits a command to modify the main process variable value according to the calculated optimal value to the process device through the communication device 140. You can.

Optimization commands may be entered by the user, or may be set to be entered at a certain cycle.

When a correction command is input, the process device modifies the currently set process variable value according to the calculated optimal variable value, thereby minimizing the defect rate of the injection molded product.

If the error between the input new process variable value and the calculated optimal value is greater than or equal to a threshold, the processor 120 may transmit a command to modify the main process variable value according to the calculated optimal value to the process device through a communication device. The error threshold can be set by the user, but is not limited thereto.

7 and 8 are flowcharts for explaining a process variable optimization method according to an embodiment of the present disclosure.

According to FIG. 7, the process variable optimization method for an electronic device includes process variable data including a plurality of process variables and variable values for each of the plurality of process variables, quality data distinguishing good or defective products according to the process variable data, and a plurality of process variable data including variable values for each of the plurality of process variables. Classifies quality based on process variables and stores a plurality of different classification models learned with process variable data and quality data and a regression model that calculates the value of one process variable among the plurality of process variables using the remaining process variables. Step (S710), variable importance, which is the degree of influence on quality according to changes in variable values, is calculated for each process variable used in a plurality of classification models, and a plurality of main processes among the plurality of process variables are calculated based on the calculated variable importance. A step of selecting variables (S720), a step of learning a regression model for a plurality of major process variables using data corresponding to a plurality of major process variables selected among the process variable data (S730), and input of new process variable values If so, it may include calculating the optimal value for each of the plurality of main process variables using the new process variable values and a regression model for each of the plurality of main process variables (S740).

Contents related to process variable data and quality data, contents related to classification models and regression models, and contents related to major process variables and optimal values have been described in the above section, so duplicate explanations will be omitted.

According to FIG. 8, the process variable optimization method for an electronic device selects a plurality of main classification models based on prediction accuracy among a plurality of different classification models, and determines variable importance for each process variable used in the selected plurality of main classification models. Calculating and selecting key process variables commonly included in a plurality of main classification models selected among a plurality of process variables based on the calculated variable importance (S820) and the error between the new process variable value and the calculated optimal value If is greater than or equal to the threshold, a step (S850) of transmitting a command to modify the main process variable value according to the calculated optimal value to the process device may be further included. Contents related to selecting major classification models and modification orders have been described in the above section, so redundant explanations will be omitted.

Meanwhile, the process variable optimization method according to the various embodiments described above may be implemented in the form of program code for performing each step, and may be stored and distributed on a recording medium. In this case, a device equipped with a recording medium can perform the above-described processing operations.

These recording media may be various types of computer-readable media such as ROM, RAM, memory chips, memory cards, external hard drives, hard drives, CDs, DVDs, magnetic disks, or magnetic tapes.

Although the present disclosure has been described above with reference to the accompanying drawings, the scope of rights of the present disclosure is determined by the scope of the patent claims described later and should not be construed as being limited to the above-described embodiments and/or drawings. In addition, it should be clearly understood that improvements, changes and modifications of the disclosure described in the patent claims, which are obvious to those skilled in the art, are also included in the scope of rights of the present disclosure.

100: electronic device 110: memory
120: processor 130: user interface
140: communication device

Claims (13)

In electronic devices,
Process variable data including a plurality of process variables and variable values for each of the plurality of process variables, quality data that distinguishes good or defective products according to the process variable data, and classifying quality based on the plurality of process variables and the process a memory storing variable data, a plurality of different classification models learned from the quality data, and a regression model that calculates the value of one process variable among the plurality of process variables using the remaining process variables;
processor; and
Includes a communication device;
The processor,
Selecting a plurality of main classification models based on prediction accuracy among the plurality of different classification models, and selecting main process variables commonly included in the selected plurality of main classification models,
Using data corresponding to the plurality of key process variables selected among the process variable data, a regression model for each of the plurality of main process variables is learned,
An electronic device that, when a new process variable value is input, calculates an optimal value for each of the plurality of main process variables using the new process variable value and a regression model for each of the plurality of main process variables,
The main process variables are,
Calculate variable importance, which is the degree of influence on quality according to changes in variable values, for each process variable used in the plurality of classification models, and select among the plurality of process variables based on the calculated variable importance,
The variable importance is,
By replacing the value of one of the plurality of process variables with a random value in the plurality of classification models and calculating it based on the decrease in accuracy value that appears when inputting it into the learned classification model,
If the error between the new process variable value and the calculated optimal value is greater than or equal to a threshold, an electronic device transmits a command to modify the main process variable value according to the calculated optimal value to a process device through the communication device. According to paragraph 1,
The plurality of different classification models are,
An electronic device, which is a model learned by up-sampling the process variable data corresponding to defective products among the quality data and using all data including the up-sampled data. delete According to paragraph 1,
The processor,
If the classification model is a decision tree model, the variable importance is calculated based on at least one of the number and location of nodes using process variables in the decision tree model,
If the classification model is a K nearest neighbor model, the variable importance is calculated based on data purity when data is classified for each process variable. delete According to paragraph 1,
The plurality of classification models are classification models learned with first data that is part of the process variable data and the quality data,
The prediction accuracy is,
An electronic device, wherein the prediction accuracy is calculated using second data excluding the first data among the process variable data and the quality data. According to paragraph 1,
The memory is,
Stores multiple different regression models,
The processor,
Learning the plurality of different regression models for each major process variable,
Selecting a main regression model for each main process variable based on the error values of the plurality of different regression models,
An electronic device that calculates the optimal value using the selected main regression model. According to paragraph 1,
The memory is,
Further storing information on whether the plurality of process variables are controllable and controllable alternative process variables for uncontrollable process variables,
The processor,
When the selected main process variable is an uncontrollable process variable, an electronic device that trains the regression model using the process variable data corresponding to an alternative process variable for the selected main process variable. According to paragraph 1,
user interface; and a communication device;
The processor,
Displaying the calculated optimal value for each of the plurality of main process variables on the user interface,
When an optimization command for the main process variable is input through the user interface, an electronic device transmits a command to modify the main process variable value according to the calculated optimal value to a process device through the communication device. delete In a method for optimizing process variables of an electronic device,
Process variable data including a plurality of process variables and variable values for each of the plurality of process variables, quality data that distinguishes good or defective products according to the process variable data, and classifying quality based on the plurality of process variables and the process storing a plurality of different classification models learned from variable data and the quality data, and a regression model that calculates the value of one process variable among the plurality of process variables using the remaining process variables;
Selecting a plurality of main classification models based on prediction accuracy among the plurality of different classification models and selecting main process variables commonly included in the selected plurality of main classification models;
Using data corresponding to the plurality of selected main process variables among the process variable data, learning a regression model for each of the plurality of main process variables; and
When a new process variable value is input, calculating an optimal value for each of the plurality of main process variables using the new process variable value and a regression model for each of the plurality of main process variables;
If the error between the new process variable value and the calculated optimal value is more than a threshold, transmitting a command to modify the main process variable value according to the calculated optimal value to the process device through a communication device; ,
The main process variables are,
Variable importance, which is the degree of influence on quality according to changes in variable values, is calculated for each process variable used in the plurality of classification models, and a plurality of key process variables among the plurality of process variables are calculated based on the calculated variable importance. select,
The variable importance is,
A method for optimizing process variables, wherein the value of one of the plurality of process variables is replaced with a random value in the plurality of classification models, and the value is calculated based on the decrease in accuracy that appears when inputting the learned classification model.
delete delete

Images