Apparatus and method for deriving optimal control guide in manufacturing process

Abstract

According to the present disclosure, an apparatus and a method for deriving an optimal control guide in a manufacturing process are disclosed. The apparatus may comprise at least one processor for executing computer-executable commands. The processor can define a condition variable, a control variable and a target variable from process data, determine stratification groups by performing data stratification using the condition variable, extract a main control variable by means of machine learning model training using the control variable for each stratification group, perform segmentation of the main control variable for each stratification group, determine an optimal control condition by evaluating subsets combined on the basis of the segmentation for each stratification group, and derive an optimal control guide by connecting each determined optimal control condition to the respective stratification groups.

Description

Device and method for deriving optimal control guide in manufacturing process

[0001] The present disclosure relates to optimization of a manufacturing process, and more particularly, to a device and method for deriving an optimal control guide in a manufacturing process.

[0002] In manufacturing, improving quality and reducing costs are crucial goals, and achieving these goals requires optimizing manufacturing processes. However, traditionally, decisions to optimize manufacturing processes often relied on the experience and intuition of process managers. This conventional approach is inefficient and, in particular, makes it difficult to quickly respond to changing market conditions and manufacturing conditions.

[0003] Recently, data-driven approaches have been developed and used to optimize manufacturing processes. However, when applied to actual manufacturing data, they often fail to achieve expected performance. Therefore, there is a need to develop solutions that can collect and analyze diverse data from manufacturing sites to derive reliable control guidance.

[0004] The present disclosure, which aims to solve these problems, aims to provide a device and method for deriving an optimal control guide in a manufacturing process.

[0005] According to one embodiment of the present disclosure, a device for deriving an optimal control guide in a manufacturing process may be provided. The device may include a memory configured to store computer-executable instructions; and at least one processor connected to the memory and configured to execute the computer-executable instructions. According to the computer-executable instructions, the processor may be configured to define condition variables, control variables, and target variables from process data, perform data stratification using the condition variables to determine stratification groups, extract key control variables through machine learning model training using the control variables for each stratification group, perform binning of the key control variables for each stratification group, evaluate combined subsets based on the binning for each stratification group to determine optimal control conditions, and associate each determined optimal control condition with each stratification group, thereby deriving an optimal control guide.

[0006] Additionally, the processor may be configured to perform target variable encoding to determine a target variable threshold and classify a class based on the determined target variable threshold.

[0007] In addition, the processor may be configured to calculate an information gain for each condition variable based on an entropy calculated from the target variable encoded data, determine a stratification rule based on a condition variable with the largest information gain and a condition variable threshold, and generate dichotomized lower nodes according to the stratification rule.

[0008] In addition, the processor may be configured to calculate the information gain, determine the stratification rule, and repeat the process of generating the dichotomized subnodes until the F1 score of the stratification rule determined for each subnode becomes less than a preset value, and determine leaf nodes among the generated subnodes as stratification groups.

[0009] In addition, the processor may be configured to determine the importance of each control variable and the correlation coefficient with the target variable through learning a logistic regression model using each control variable for each stratified group, sort the control variables in descending order according to scores determined based on the importance and the correlation coefficient, determine the change rate of each sorted control variable based on the score of the upper-sorted control variable and the score of the control variable, and extract control variables whose change rate is less than a preset rate as the main control variables.

[0010] In addition, the processor may be configured to determine, for each stratified group, data subsets corresponding to combinations of the segmented main control variables, calculate an evaluation score for each subset based on the number of data in each subset and the model result probability average determined through the logistic regression model for each subset, and determine a combination of the segmented main control variables corresponding to the subset with the highest evaluation score as an optimal control condition for the corresponding stratified group.

[0011] Additionally, the optimal control guide may include visualized information of a range including upper and lower limits of the main control variables according to the determined optimal control conditions.

[0012] According to one embodiment of the present disclosure, a method for deriving an optimal control guide in a manufacturing process executable by a computing device may be provided. The method may include the steps of: defining condition variables, control variables, and target variables from process data; performing data stratification using the condition variables to determine stratification groups; extracting key control variables through machine learning model training using the control variables for each stratification group; performing binarization of the key control variables for each stratification group; determining optimal control conditions by evaluating combined subsets based on the binarization for each stratification group; and deriving an optimal control guide by connecting each determined optimal control condition to each stratification group.

[0013] According to one embodiment of the present disclosure, a computer program stored on a computer-readable medium including computer-executable instructions for executing a method for deriving an optimal control guide in a manufacturing process may be provided.

[0014] According to the present disclosure, a solution can be provided that maximizes the efficiency of the manufacturing process and improves quality by analyzing various process data from a manufacturing site in real time and providing optimal control guidance based on the analysis.

[0015] In addition, according to the present disclosure, a solution can be implemented that can effectively separate controllable and uncontrollable variables from process data, apply a stratified method, and derive an optimal control group by combining ensemble tree-based machine learning and optimal binning.

[0016] FIG. 1 is an exemplary diagram illustrating a computing device for deriving an optimal control guide in a manufacturing process according to one embodiment of the present disclosure.

[0017] FIG. 2 is an exemplary block diagram illustrating components of the computing device of FIG. 1.

[0018] Figure 3 is an exemplary diagram showing condition variables, control variables, and target variables of process data.

[0019] FIG. 4 is an exemplary diagram illustrating target variable encoding according to one embodiment of the present disclosure.

[0020] FIG. 5 is an exemplary diagram illustrating refining a data range by identifying the start and end of a process by extracting change points according to one embodiment of the present disclosure.

[0021] FIG. 6 is an exemplary diagram illustrating data stratification using condition variables according to one embodiment of the present disclosure.

[0022] FIG. 7 is an exemplary diagram showing the segmentation of key control variables according to one embodiment of the present disclosure.

[0023] FIG. 8 is an exemplary diagram showing determining optimal control conditions for each layered group according to one embodiment of the present disclosure.

[0024] FIG. 9 is an exemplary diagram showing a visualization of the labeling ratio of original data and data to which optimal control has been applied according to one embodiment of the present disclosure.

[0025] FIG. 10 is an exemplary diagram showing target average scores of original data and data to which optimal control is applied according to one embodiment of the present disclosure.

[0026] FIGS. 11a and 11b are exemplary diagrams showing the distribution range of major control variables by optimal control conditions according to one embodiment of the present disclosure.

[0027] FIG. 12 is an exemplary flowchart illustrating a method for deriving an optimal control guide in a manufacturing process according to one embodiment of the present disclosure.

[0028] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. First, when assigning reference numerals to components in each drawing, it should be noted that, where possible, identical components are assigned the same reference numerals even if they appear in different drawings. Furthermore, when describing the present invention, if a detailed description of a related, well-known structure or function is judged to obscure the gist of the present invention, such detailed description will be omitted.

[0029] Various aspects of the present invention are described below. It is to be understood that the inventions presented herein may be embodied in a wide variety of forms, and that any specific structure, function, or both presented herein are merely exemplary. Based on the inventions presented herein, one of ordinary skill in the art will appreciate that one aspect presented herein may be implemented independently of any other aspects, and that two or more of these aspects may be combined in various ways. For example, a device may be implemented or a method may be implemented using any number of the aspects described herein. Furthermore, such a device may be implemented or such a method may be implemented using structures, functions, or structures and functions other than or in addition to one or more of the aspects described herein.

[0030] FIG. 1 is an exemplary diagram illustrating a computing device for deriving an optimal control guide in a manufacturing process according to one embodiment of the present disclosure.

[0031] Process data generated at a manufacturing site may be sensed by sensors attached to various equipment in the manufacturing process or may be various data output from such equipment. To derive an optimal control guide according to the present disclosure, such process data may be defined as condition variables, control variables, and target variables. Condition variables may be defined as variables that cannot be controlled under the current situation (e.g., raw material mass, etc.). Control variables may be defined as variables that can be controlled under the current situation (e.g., pressure control, temperature control, etc.). Target variables may be defined as target variables (e.g., performance target values, etc.) for achieving a desired level of quality.

[0032] Process data collected in the manufacturing process can be input into a computing device (100), and the computer device (100) can be configured to derive and provide an optimal control guide for the manufacturing process based on such process data, as will be described later.

[0033] FIG. 2 is an exemplary block diagram illustrating components of the computing device of FIG. 1.

[0034] As illustrated in FIG. 2, the computing device (100) may include a processor (110), a storage medium (120), a memory (130), and a network interface (140), which may be connected to each other via a system bus (150).

[0035] An operating system (OS) (122) and a computer program (124) may be installed in the storage medium (120). The storage medium (120) may be a data storage device such as a hard disk or a solid state drive (SSD) that can store computer programs and related data. The operating system (122) may be operating system software such as Windows, IOS, Linux, etc. for operating the computing device (100). The computer program (124) may include not only functional modules for deriving an optimal control guide in a manufacturing process according to the present disclosure, but also computer-executable instructions therefor. In addition, the computer program (124) may be loaded into the memory (130) so that it may be executed by the processor (110). When the computer-executable instructions of the computer program (124) are executed by the processor (110), they may cause the processor (110) to perform a method for deriving an optimal control guide in a manufacturing process according to the present disclosure. The processor (110) may be configured to provide computing and control capabilities to support the execution of the entire computing device (100). The processor (110) may be a data processing device such as a Central Processing Unit (CPU), a Microprocessor Unit (MPU), an Application Processor (AP), etc., and may be configured with one processor or multiple processors. When configured with multiple processors, the processors (110) may operate as parallel processing processors. The network interface (140) may provide an interface that can be connected to an external device (e.g., a database storing process data, an external entity that receives optimal control guidance, a display device, another wired or wireless communication device that can be connected via a network, etc.) to communicate data.

[0036] Figure 3 is an exemplary diagram showing condition variables, control variables, and target variables of process data.

[0037] As illustrated in FIG. 3, process data may include a condition variable (310), a control variable (320), and a target variable (330), and each variable may have multiple data values ​​measured in a time series.

[0038] The computing device (100) can select condition variables (310), control variables (320), and target variables (330) for applying optimal guide derivation according to the present disclosure, and can perform target variable encoding as illustrated in FIG. 4.

[0039] FIG. 4 is an exemplary diagram illustrating target variable encoding according to one embodiment of the present disclosure.

[0040] The computing device (100) may be configured to determine a target variable threshold and perform target variable encoding to classify classes based on the determined target variable threshold. To this end, the computing device (100) may determine the most distinct threshold by analyzing the distribution of the target variable, and may apply an analysis model capable of automatically searching for threshold values, such as a Gaussian mixture model (GMM).

[0041] As illustrated in FIG. 4, the computing device (100) can determine a target variable threshold (410) from the distribution of the target variable, and can encode the target variable based on this threshold (410). For example, the computing device (100) can obtain target variable encoded data (target_binary value) (420) classified as class 0 (good product) if the threshold (410) is greater than or equal to the threshold, and as class 1 (defective product) if the threshold (410) is less than the threshold.

[0042] Next, for data purification, the computing device (100) can optionally perform duplicate data processing to remove duplicate data, and can be configured to detect and remove process start and end points that cause learning degradation. These process start and end points may be points where inflection points (Change Points) occur, and the computing device (100) can detect these points by applying an inflection point analysis model, such as the Prophet model. As illustrated in FIG. 5, when the process start point (510) and end point (520) are detected, the computing device (100) can purify the data by removing the range outside the corresponding points (510, 520).

[0043] Next, the computing device (100) can be configured to perform data stratification using condition variables to determine stratification groups.

[0044] FIG. 6 is an exemplary diagram illustrating data stratification using conditional variables according to one embodiment of the present disclosure. As illustrated in FIG. 6 , the computing device (100) can determine data stratification groups for subsequent learning and analysis through tree-structured stratification.

[0045] The computing device (100) may be configured to calculate an information gain for each condition variable based on an entropy calculated from target variable encoded data, determine a stratification rule based on a condition variable with the greatest information gain and a condition variable threshold, and generate dichotomized lower nodes according to the determined stratification rule. In addition, the computing device (100) may be configured to repeat the processes of calculating the above-described information gain, determining the stratification rule, and generating dichotomized lower nodes until the F1 score of the stratification rule determined for each generated lower node becomes less than a preset value, and determine leaf nodes among the generated lower nodes as stratification groups.

[0046] In this regard, the computing device (100) can calculate entropy by calculating the ratio of classes in data (i.e., target variable encoded data), and the calculation formula for this is as follows.

[0047]

[0048] Here, S is the data set, and p i is the proportion of class i.

[0049] In addition, the computing device (100) can calculate the information gain of each condition variable based on this entropy, and the calculation formula for this is as follows.

[0050]

[0051] Here, A is the independent variable to be partitioned (i.e., the condition variable), and S v is a subset of data corresponding to the value v of variable A.

[0052] For convenience of explanation, assume that the target variable has data S divided into classes “0” and “1”, and the example data is as shown in Table 1.

[0053]

[0054] In this case, in relation to mathematical expression 1, among the 5 data of the target variable, class "0" is 3 and class "1" is 2. and Therefore, entropy can be calculated as follows.

[0055]

[0056] Next, the condition variable (A) illustrated in Table 1 is a numeric variable. Data can be dichotomized by considering all possible split points, and all split points can be explored to select the split point with the highest information gain. Furthermore, even if the condition variable is a categorical variable (e.g., data is one of large, medium, or small), the data can be dichotomized by defining a split criterion, such as (large, remaining), and the information gain can be calculated in the same manner.

[0057] In the example in Table 1, the possible split points in the condition variable (A) can be 15, 20, 25, and 30, and the data is split into S with 'variable value > 20' as the split criterion (i.e. threshold). small and S large It is assumed that it is divided into . In this case, S according to mathematical formula 1 small and S large The entropy for and the information gain for the condition variable (A) according to mathematical expression 2 can be calculated as follows.

[0058]

[0059] The computing device (100) can calculate the information gain for each condition variable in the same manner as described above, and determine a stratification rule by selecting the condition variable and condition variable threshold (i.e., the corresponding dichotomization condition) with the greatest information gain (at this time, stratification is not performed if the information gain is 0). The computing device (100) can generate dichotomized subnodes according to the determined stratification rule.

[0060] Next, the computing device (100) can determine a stratification rule by repeating the aforementioned processes for each generated sub-node and can generate dichotomized sub-nodes, and can evaluate the determined stratification rule for the corresponding sub-node (e.g., calculate an F1 score) and terminate the stratification of the corresponding sub-node when the evaluated F1 score is less than a preset value (e.g., 0.7). As a result, when all stratification tasks are completed, the computing device (100) can determine leaf nodes (620, 640, 650) in the tree structure as illustrated in FIG. 6 as stratification groups for subsequent learning and analysis.

[0061] In the example of Fig. 6, the first and second stage stratification operations were performed. In the first stage stratification operation, the entire dataset (610) was divided into lower nodes (620, 630), and the stratification rule was defined by the condition variable ('condition variable 1' in the example of Fig. 6) and the condition variable threshold (621). As a result of the first stage stratification operation, the leaf node (620) (N11) was selected as the first stratification group, and the second stage stratification operation was performed on the other lower nodes (630). In the second stage stratification operation, the node (630) was divided into lower nodes (640, 650), and the stratification rule (641) was determined, and all lower nodes were selected as the second and third stratification groups as leaf nodes (640, 650) (N12_11, N12_12).

[0062] Once the stratified groups are selected, the computing device (100) can be configured to extract key control variables for each stratified group through machine learning model learning using control variables. Here, the machine learning model may be a logistic regression model, but is not limited thereto, and other learning models capable of learning the relationship between the control variables and the target variables may be applied.

[0063] The computing device (100) may optionally perform an additional data cleaning process to determine and remove outliers or missing values ​​present in the data (i.e., exclude them from the learning target). Next, the computing device (100) may be configured to determine, for each stratified group, the importance of each control variable and the correlation coefficient with the target variable through learning a logistic regression model using each control variable, sort the control variables in descending order according to the scores determined based on the importance and the correlation coefficient, determine the rate of change of the control variable based on the score of the upper-sorted control variable and the score of the control variable for each sorted control variable, and extract control variables for which the determined rate of change is less than a preset rate as main control variables.

[0064] For convenience of explanation, the control variables, variable importance, correlation coefficient, score and change rate for the respective stratified groups can be illustrated as in Table 2.

[0065]

[0066] In the example of Table 2, the computing device (100) can calculate a score by multiplying the variable importance of each control variable (X2, X3, ....) and the correlation coefficient with the target variable, and can sort the control variables in descending order (i.e., X7, X15, X2, ...) according to the score as illustrated in Table 2. Next, the computing device (100) can calculate (change rate) = [top score - current score] / (top score) for each control variable according to the sorted order (here, the change rate of the top variable can be set to 0). The computing device (100) can determine control variables of which the calculated change rate is less than a preset rate (e.g., 0.5) as the main control variables. In the example of Table 2, the control variables X7, X15, and X2 can be determined as the main control variables.

[0067] Once the primary control variables are determined, the computing device (100) can be configured to perform binarization of the primary control variables for each stratified group. Here, optimal binning can be applied to the binarization, through which the optimal boundary values ​​can be determined that maximize the correlation between each control variable interval and the feature difference between intervals.

[0068] FIG. 7 is an exemplary diagram illustrating the segmentation of a key control variable according to one embodiment of the present disclosure. As illustrated in FIG. 7, the key control variable (control variable X2 in the example of FIG. 7) can be segmented into multiple segments, and each segment can be used to determine subsequent optimal control conditions.

[0069] The computing device (100) may be configured to evaluate subsets combined based on the segmentation for each stratified group to determine optimal control conditions. To this end, the computing device (100) may be configured to determine data subsets corresponding to combinations of segmented key control variables for each stratified group, calculate an evaluation score for each subset based on the number of data in each subset and the model result probability average determined through a logistic regression model trained for each subset, and determine the combination of segmented key control variables corresponding to the subset with the highest evaluation score as the optimal control condition for the corresponding stratified group.

[0070] For convenience of explanation, the combined subsets, number of data, model result probability average, and evaluation score can be illustrated as in Table 3.

[0071]

[0072] In the example of Table 3, the meaning of X2_1 may mean the first interval condition of the main control variable X2. In addition, the meaning of the combination subsets X2_1, X7_1, and X15_1 may mean data subsets that satisfy the first interval condition of the main control variable X2, the first interval condition of the main control variable X7, and the first interval condition of the main control variable X15.

[0073] Here, the model output probability average for each combination subset can be the average of the 0-label classification probabilities obtained through the trained logistic regression model. By applying the average of the 0-label classification probabilities of this regression model instead of the number of 0-labels of the binary-classified target variable, we can avoid the potential for distorted results depending on the size and distribution of the subset. Furthermore, by calculating the probability of being classified as 0-labeled by reflecting the characteristics of each data point based on the relationships between the patterns and variables learned by this model, we can derive more reliable results.

[0074] In the example of Table 3, the computing device (100) can calculate (evaluation score) = (number of data) / (model result probability average) for each subset according to the combination of the segmented main control variables, and can determine the combination of X2_3, X7_1, and X15_1 with the highest evaluation score of 10612 points as the optimal control condition for the corresponding stratified group.

[0075] FIG. 8 is an exemplary diagram showing determining optimal control conditions for each layered group according to one embodiment of the present disclosure.

[0076] As illustrated in FIG. 8, the computing device (100) can determine optimal control conditions (820, 840, 860) for each layered group (620, 640, 650), and can be configured to derive an optimal control guide by connecting each optimal control condition to each layered group.

[0077] Such optimal control guides can be configured to include visualized data such as range information including upper and lower limits of key control variables according to determined optimal control conditions, labeling ratio information of data to which optimal control is applied compared to original data, and target average score information of data to which optimal control is applied compared to original data.

[0078] Figure 9 is an exemplary diagram illustrating a visualization of the labeling ratios of original data and data subjected to optimal control according to one embodiment of the present disclosure. As illustrated in Figure 9, the 0-label distribution in the original data was 52%, while the 0-label distribution in the data subjected to optimal control according to the present disclosure was maximized to 98.12%.

[0079] Figure 10 is an exemplary diagram illustrating the target average scores of original data and data subjected to optimal control according to one embodiment of the present disclosure. As illustrated in Figure 10 , the target score (i.e., productivity index score) of the original data was 1.4845, while the target score of the data subjected to optimal control according to the present disclosure improved to 1.5752.

[0080] FIGS. 11A and 11B are exemplary diagrams showing distribution ranges of major control variables under optimal control conditions according to one embodiment of the present disclosure. FIG. 11A is an example visualizing a distribution range (1110) under optimal control conditions of a major control variable (control variable X7 in the example of FIG. 11A), and FIG. 11B is an example visualizing a distribution range (1120) under optimal control conditions of a major control variable (control variable X15 in the example of FIG. 11B).

[0081] In this regard, decision tree-based modeling, which is generally used to derive conventional rules, has limitations in that it can only present one of the upper and lower limits of the control factor because it processes the data by dividing it based on specific values ​​of variables at each branch point. On the other hand, the optimal control guide derived according to the present disclosure can derive an optimal control group by combining a machine learning model and optimal binning as described above, and thus can provide an optimal control range that includes both the upper and lower limits of the main control variables, as illustrated in FIGS. 11a and 11b.

[0082] FIG. 12 is an exemplary flowchart illustrating a method for deriving an optimal control guide in a manufacturing process executable by a computing device (100) according to one embodiment of the present disclosure.

[0083] As illustrated in FIG. 12, the computing device (100) can define condition variables, control variables, and target variables from process data (1210). The computing device (100) can perform target variable encoding (1220) and perform data stratification using the condition variables to determine stratification groups (1230). The computing device (100) can perform machine learning model training (e.g., logistic regression model training) using the control variables for each stratification group (1240), thereby extracting key control variables (1250). The computing device (100) can perform binarization of the key control variables for each stratification group (1260). The computing device (100) can extract combined subsets based on the binarization for each stratification group (1270), and can evaluate these subsets to determine the control value combination corresponding to the subset with the highest evaluation score as the optimal control condition (1280). The computing device (100) can derive an optimal control guide for the manufacturing process by connecting each determined optimal control condition to each layered group (1290).

[0084] It should be understood that any specific order or hierarchy of steps in any of the presented processes is merely an example of exemplary approaches. It should be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present invention based on design priorities. The appended method claims present elements of various steps in an exemplary order, but are not intended to be limited to the specific order or hierarchy presented.

[0085] The terms "component," "unit (or part)," "module," "system," and the like, as used herein, may refer to computer-related entities, hardware, firmware, software, a combination of software and hardware, or an execution of software. For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, a thread of execution, a program, and / or a computer. For example, both an application running on a computing device and the computing device may be a component. One or more components may reside within a processor and / or thread of execution, and a component may be localized within a computer or distributed between two or more computers. Furthermore, these components may execute from various computer-readable media having various data structures stored therein.

[0086] The description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments disclosed herein, but is to be construed in the broadest scope consistent with the principles and novel features disclosed herein.

[0087] (Explanation of symbols)

[0088] 100: Computing Device

[0089] 110: Processor

[0090] 120: Storage media

[0091] 122: Operating System

[0092] 124: Computer program

[0093] 130: Memory

[0094] 140: Network interface

[0095] 150: System bus

Claims

1. A device for deriving optimal control guide in a manufacturing process. A memory configured to store computer-executable instructions; and At least one processor connected to the memory and configured to execute the computer-executable instructions, In accordance with the above computer-executable instructions, the processor: Define condition variables, control variables, and target variables from process data, Perform data stratification using the above condition variables to determine the stratification group, For each stratified group, the main control variables are extracted through machine learning model training using the above control variables, For each stratified group, the above major control variables are binned, For each stratified group, the combined subsets are evaluated based on the interval to determine the optimal control conditions, It is configured to derive an optimal control guide by connecting each determined optimal control condition to each layered group. device.

2. In paragraph 1, The above processor, A device configured to perform target variable encoding, which determines a target variable threshold and classifies a class based on the determined target variable threshold.

3. In paragraph 2, The above processor, Calculate the information gain for each condition variable based on the entropy calculated from the target variable encoded data, Determine the stratification rule based on the condition variable and condition variable threshold with the greatest information gain, It is configured to create dichotomized sub-nodes according to the above layering rule. device.

4. In paragraph 3, The above processor, For each sub-node, the process of calculating the information gain, determining the stratification rule, and generating the dichotomized sub-nodes is repeated until the F1 score of the stratification rule determined for the sub-node is less than a preset value, It is configured to determine leaf nodes among the generated subnodes as stratified groups. device.

5. In paragraph 1, The above processor, for each layered group By learning a logistic regression model using each control variable, the importance of each control variable and the correlation coefficient with the target variable are determined. Sort the control variables in descending order according to the scores determined based on the above importance and correlation coefficient, For each sorted control variable, the rate of change of the control variable is determined based on the score of the upper sorted control variable and the score of the control variable. It is configured to extract control variables whose change rate is less than a preset rate as the main control variables. device.

6. In paragraph 5, The above processor, for each layered group Determine data subsets corresponding to combinations of the binned key control variables, Calculate the evaluation score of each subset based on the number of data in each subset and the average of the model result probability determined through the logistic regression model for each subset, It is configured to determine the combination of the segmented main control variables corresponding to the subset with the highest evaluation score as the optimal control condition for the corresponding stratified group. device.

7. In paragraph 1, The above optimal control guide includes visualized information of a range including the upper and lower limits of the main control variables according to the determined optimal control conditions. device.

8. A method for deriving an optimal control guide in a manufacturing process executable by a computing device, A step of defining condition variables, control variables and target variables from process data; A step of determining a stratification group by performing data stratification using the above condition variables; A step of extracting key control variables through machine learning model training using the above control variables for each stratified group; A step of performing binarization of the above main control variables for each stratified group; A step of determining optimal control conditions by evaluating the combined subsets based on the segmentation for each stratified group; and A step of deriving an optimal control guide by connecting each determined optimal control condition to each stratified group, method.

9. In paragraph 8, Before the step of determining the above stratification group, A method further comprising the step of determining a target variable threshold and performing target variable encoding to classify classes based on the determined target variable threshold.

10. In paragraph 9, The step of determining the above stratification group is: A step of calculating information gain for each condition variable based on entropy calculated from the target variable encoded data; A step of determining a stratification rule based on a condition variable and a condition variable threshold having the greatest information gain; and A method comprising the step of generating dichotomized subnodes according to the above stratification rule.

11. In paragraph 10, The step of determining the above stratification group is: For each sub-node, a step of calculating the information gain, a step of determining the stratification rule, and a step of generating the dichotomized sub-nodes are repeated until the F1 score of the stratification rule determined for the sub-node becomes less than a preset value; and A method further comprising a step of determining leaf nodes among the generated subnodes as stratified groups.

12. In paragraph 8, The steps for extracting the above key control variables are: A step of determining the importance of each control variable and the correlation coefficient with the target variable by learning a logistic regression model using each control variable; A step of sorting the control variables in descending order according to scores determined based on the above importance and the above correlation coefficient; A step of determining the rate of change of the control variable based on the score of the upper-sorted control variable and the score of the control variable for each sorted control variable; and A method comprising a step of extracting control variables having a change rate less than a preset rate as the main control variables.

13. In paragraph 12, The step of determining the above optimal control conditions is: A step of determining data subsets corresponding to combinations of the interval-determined key control variables; A step of calculating an evaluation score for each subset based on the number of data in each subset and the model result probability average determined through the logistic regression model for each subset; and A method comprising the step of determining a combination of segmented key control variables corresponding to the subset with the highest evaluation score as an optimal control condition for the corresponding stratified group.

14. In paragraph 8, A method wherein the above optimal control guide includes visualized information of a range including upper and lower limits of the main control variables according to the determined optimal control conditions.

15. A computer program stored on a computer-readable medium, comprising computer-executable instructions for executing a method according to any one of claims 8 to 14.

Images