Manufacturing system

Abstract

To provide a manufacturing system capable of automatically determining more suitable manufacturing parameters.SOLUTION: A manufacturing system includes: a manufacturing unit; a quality measurement unit; a model generation unit; a first quality prediction unit that predicts quality parameters from manufacturing parameters at a certain point in time using a nonlinear regression model; a value creation unit that creates a plurality of parameter values that fall within a predetermined range from parameter values at a certain point in time for one type of variable manufacturing parameter when the quality parameters predicted by the first quality prediction unit do not satisfy a quality standard; a second quality prediction unit that predicts quality parameters from each parameter value created by the value creation unit and parameter values of the remaining manufacturing parameters at a point in time using a linear regression model; a parameter extraction unit that extracts a parameter value of one type of manufacturing parameter corresponding to a quality parameter that satisfies an extraction criterion from among the quality parameters predicted by the second quality prediction unit; and a parameter setting unit.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a manufacturing system for producing products such as chemical products. [Background technology]

[0002] Conventionally, at manufacturing sites such as factories that manufacture various products, one or more of multiple manufacturing parameters are appropriately changed to manufacture the products so that the product characteristics satisfy the product specifications. For example, when manufacturing an organic EL device, the layer structure is changed so that the product characteristics such as brightness, color temperature, and driving voltage satisfy the product specifications.

[0003] However, many manufacturing parameters are related to multiple product characteristics, and changing a manufacturing parameter to make one product characteristic satisfy the product standard can sometimes cause other product characteristics that originally satisfied the product standard to fluctuate. In other words, when manufacturing a product, each manufacturing parameter interacts with the other to affect the product characteristics, making it extremely difficult to manufacture a product so that multiple product characteristics simultaneously satisfy the product standard.

[0004] Therefore, one means for solving such problems is to construct a prediction model by machine learning, predict the manufacturing results using the prediction model, and determine the manufacturing conditions (manufacturing parameters) so that the manufacturing results have the desired product characteristics (for example, Patent Document 1).The manufacturing system of Patent Document 1 can automatically determine the manufacturing conditions (manufacturing parameters) that are likely to produce the desired results. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2020-166749 Summary of the Invention [Problem to be solved by the invention]

[0006] However, in practical operation of conventional manufacturing systems, there are cases where the desired results are not obtained when a product is manufactured under the manufacturing conditions determined by the manufacturing system. In other words, there is room for improvement in the conventional manufacturing systems in terms of improving the accuracy of determining manufacturing parameters.

[0007] Therefore, an object of the present invention is to provide a manufacturing system that can automatically determine more suitable manufacturing parameters. [Means for solving the problem]

[0008] As a result of intensive research by the present inventors to solve the above problems, it was found that the prediction accuracy can be improved by adopting a nonlinear model (nonlinear regression model) as a prediction model when predicting the product characteristics of products to be manufactured in the future from predetermined manufacturing parameters.Furthermore, it was found that the prediction accuracy can be improved by adopting a linear model (linear regression model) as a prediction model when predicting how product characteristics will change before and after varying some of the manufacturing parameters when varying the manufacturing parameters.

[0009] One aspect of the present invention for solving the above-mentioned problems, which is provided based on this finding, is a manufacturing system including: a manufacturing unit that manufactures products using set manufacturing parameters; a model generation unit that generates a nonlinear regression model and a linear regression model through supervised learning using data sets of past manufacturing parameters and past quality parameters; a first quality prediction unit that predicts quality parameters from the manufacturing parameters at a certain point in time using the nonlinear regression model; a value creation unit that, when the quality parameters predicted by the first quality prediction unit do not satisfy a quality standard, creates a plurality of parameter values ​​for one type of variable manufacturing parameter among the manufacturing parameters, the parameter values ​​falling within a predetermined range from the parameter value at the certain point in time for the one type of variable manufacturing parameter among the manufacturing parameters; a second quality prediction unit that predicts quality parameters from the parameter values ​​created by the value creation unit and the parameter values ​​of the remaining manufacturing parameters at the certain point in time using the linear regression model; a parameter extraction unit that extracts parameter values ​​of the one type of manufacturing parameter that correspond to the quality parameters predicted by the second quality prediction unit and that satisfy an extraction criterion; and a parameter setting unit that sets manufacturing parameters for the manufacturing unit using the parameter values ​​extracted by the parameter extraction unit.

[0010] According to this aspect, when manufacturing a product, it is possible to obtain highly accurate predicted values ​​of quality parameters and to predict with high accuracy future quality degradation of the manufactured product. Furthermore, when the future quality parameters do not satisfy the quality standards and quality degradation is predicted, it is possible to predict with high accuracy the parameter values ​​of manufacturing parameters that are inferred to cause the future quality parameters to satisfy the quality standards, and change the parameter values ​​of the manufacturing parameters to the predicted values. In other words, according to the manufacturing system of this aspect, it is possible to prevent quality degradation by automatically determining more preferable parameter values ​​and manufacturing products using those parameter values.

[0011] In a preferred aspect, the nonlinear regression model is generated by at least one algorithm selected from the group consisting of a decision tree, a support vector machine, and a neural network.

[0012] In a preferred aspect, the linear regression model is generated by at least one algorithm selected from the group consisting of partial least squares regression, logistic regression, and Lasso regression. [Effects of the Invention]

[0013] According to the present invention, it is possible to provide a manufacturing system that can automatically determine more suitable manufacturing parameters. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a block diagram illustrating a manufacturing system according to an embodiment of the present invention. [Figure 2] 2 is a cross-sectional view schematically showing an organic EL device manufactured by the manufacturing system of FIG. 1. FIG. [Figure 3] 4 is a flowchart showing the procedure of a quality adjustment operation executed by the manufacturing system of FIG. 1. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described in detail.

[0016] The manufacturing system 1 of this embodiment manufactures an organic EL device 100, which is a product, and as shown in FIG. 1, mainly includes a manufacturing unit 2 that automatically performs each manufacturing process of the organic EL device 100, and a control unit 3 that controls the manufacturing processes in the manufacturing unit 2. 2, the organic EL device 100 includes an organic EL element 111 laminated on a transparent insulating substrate 110, and the organic EL element 111 is sealed with the transparent insulating substrate 110 and a sealing layer 112. The organic EL element 111 includes a transparent electrode layer 113, an organic functional layer 114, and a back electrode layer 115 superimposed thereon, and the organic functional layer 114 includes, in this order from the transparent insulating substrate 110 side, a blue-green light-emitting unit 116, an intermediate layer 117, and a red-green light-emitting unit 118. The blue-green light-emitting unit 116 has, in this order from the transparent electrode layer 113 side, a hole injection layer 119 , a hole transport layer 120 , a blue light-emitting layer 121 , and an electron transport layer 122 . Intermediate layer 117 has electron injection layer 123 and connecting layer 124 in this order from the blue-green light-emitting unit 116 side. The red / green light-emitting unit 118 has, from the intermediate layer 117 side, a hole injection layer 125 , a hole transport layer 126 , a red / green light-emitting layer 127 , a hole blocking layer 128 , an electron transport layer 129 , and an electron injection layer 130 .

[0017] The manufacturing section 2 has a substrate introduction device 10, a pretreatment device 11, a first film-forming device 12, a second film-forming device 13, a third film-forming device 14, a sealing film-forming device 15, a product removal device 16, a transfer device 17, and a temporary storage section 18.

[0018] The substrate introduction device 10 is a device that introduces a transparent conductive substrate, which is the substrate of the organic EL device 100, into the production department 2 from the outside. The pretreatment device 11 is a cleaning device that cleans the transparent conductive substrate prior to the formation of each layer. The first to third film forming apparatuses 12 to 14 are film forming apparatuses, specifically vacuum deposition apparatuses. The first film forming apparatus 12 is an apparatus that forms each layer of the blue-green light emitting unit 116 and the red-green light emitting unit 118. The second film forming apparatus 13 is an apparatus that forms the intermediate layer 117. The third film forming apparatus 14 is an apparatus that forms the back electrode layer 115.

[0019] The sealing film forming apparatus 15 is a film forming apparatus that forms the sealing layer 112, and specifically, is a plasma CVD apparatus. The product removal device 16 is a device that removes the completed organic EL device 100 from the manufacturing department 2 to the outside. The transfer device 17 is a device that transfers a transparent conductive substrate, a work-in-progress substrate that is a transparent conductive substrate on which some layers have been formed, or an organic EL device 100, and transfers these between the above-mentioned devices (substrate introduction device 10 to product removal device 16). The temporary placement section 18 is a section (area) where the substrate is temporarily placed during transfer when the substrate is transferred between the devices.

[0020] The control unit 3 is a computer having, as its hardware configuration, a control device that controls each device in the manufacturing unit 2, a central processing unit consisting of an arithmetic unit that performs calculations on data, a storage device that stores data, an input device that inputs data from the outside, an output device that outputs data to the outside, and a display device that displays various information. The control unit 3 is connected to a management device that manages each of the above-mentioned devices in the manufacturing unit 2 via a network so that signals (information) can be sent and received. That is, the control unit 3 controls each device in the manufacturing unit 2. In addition, in the manufacturing system 1 of this embodiment, the control unit 3 can output manufacturing parameters to be used by each device in the manufacturing unit 2 (details will be described later), and the manufacturing unit 2 can manufacture a product (organic EL device 100) using the manufacturing parameters output by the control unit 3. The "manufacturing parameters" referred to here are parameters used in the manufacturing process of the organic EL device 100. That is, the "manufacturing parameters" are various parameters such as the flow rate and temperature of material gases, the temperature and / or pressure of film formation chambers belonging to one or more film formation devices, the temperature and / or pressure of the substrate, the film thickness, and the film formation speed.

[0021] As shown in FIG. 1, the control unit 3 has a quality measurement unit 30, a model generation unit 31, a first quality prediction unit 32, a value creation unit 33, a second quality prediction unit 34, a parameter extraction unit 35, and a data accumulation unit 36.

[0022] The quality measurement unit 30 is a section that measures the quality parameters of the product, and more specifically, a section that measures and acquires the quality parameters of the organic EL device 100 that is actually produced. The "quality parameters" referred to here are parameters that serve as evaluation indexes for the quality of the product, that is, the organic EL device 100. In other words, the "quality parameters" are various parameters such as luminance when lit, color coordinates (X, Y), and drive voltage.

[0023] The model generation unit 31 is a component that generates a first prediction model, which is a nonlinear regression model (nonlinear model), and a second prediction model, which is a linear regression model (linear model). The model generation unit 31 functions as a learning unit capable of performing supervised learning in accordance with a machine learning algorithm. The model generation unit 31 constructs a first prediction model using a set of data on manufacturing parameters used when a product was previously manufactured by the manufacturing system 1 and quality parameters of a product actually manufactured using those manufacturing parameters. In this embodiment, there are multiple manufacturing parameters, or in other words, a group consisting of the respective values ​​of the multiple manufacturing parameters. In this embodiment, there are also multiple quality parameters, or a group consisting of the respective values ​​of the multiple quality parameters. That is, the model generation unit 31 uses the above data sets as training data and performs supervised learning in accordance with a machine learning algorithm to learn features from the above data sets (data sets) given in large quantities, and inductively acquires the relationship between input and result, thereby generating a first prediction model. The first prediction model is generated by at least one algorithm selected from the group consisting of a decision tree, a support vector machine, and a neural network.

[0024] Furthermore, the model generation unit 31 constructs a second prediction model using a set of data on manufacturing parameters used when a product was previously manufactured by the manufacturing system 1 and quality parameters of a product actually manufactured using those manufacturing parameters. That is, the model generation unit 31 generates the second prediction model by using the set of data as training data and performing supervised learning in accordance with a machine learning algorithm using a large amount of the set of data provided. The second prediction model is generated by an algorithm different from that of the first prediction model, and is generated by at least one algorithm selected from the group consisting of partial least squares regression, logistic regression, and Lasso regression.

[0025] The first quality prediction unit 32 is a part that uses the above-mentioned first prediction model to predict the quality parameters of a product when manufactured under specified manufacturing parameters (a group consisting of the values ​​of multiple types of manufacturing parameters, i.e., manufacturing conditions).

[0026] The value creating unit 33 is a component that creates a plurality of virtual manufacturing parameters based on the manufacturing parameters used for prediction in the first quality predicting unit 32. Specifically, the value creation unit 33 creates a plurality of virtual manufacturing parameters by changing the value of one of the original manufacturing parameters. For example, assume that the original manufacturing parameters are the values ​​of four types of manufacturing parameters (flow rate of material gas: X1, temperature of substrate: Y1, thickness of film: Z1, deposition rate: A1). In this case, for example, the value creation unit 33 creates manufacturing parameters by changing only the value of the predetermined type, the thickness of film, i.e., manufacturing parameters such as (flow rate of material gas: X1, temperature of substrate: Y1, thickness of film: Z2, deposition rate: A1). That is, multiple virtual manufacturing parameters (groups of values ​​of virtual manufacturing parameters) are created, each differing only in the value of the film thickness, such as (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z2, film deposition rate: A1), (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z3, film deposition rate: A1), (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z4, film deposition rate: A1). At this time, the value of one type of manufacturing parameter is changed so that it falls within a predetermined range. For example, in the above example, it is changed so that it falls within the range of α1 or more and α2 or less (α1 and α2 are fixed values). Alternatively, it may be changed so that it falls within the range of Z1 plus or minus α3 (α3 is a fixed value) based on the original value (Z1 in the above example). The fixed value may be set in advance for each type of manufacturing parameter. In this case, multiple manufacturing parameters (a group of multiple manufacturing parameters) may be created by changing the amount of change (variation range) to a predetermined amount. For example, in the above example, if Z1 is 75, Z2, Z3, Z4, etc. will become 75.1, 75.2, 75.3, etc.

[0027] The second quality prediction unit 34 is a component that predicts the quality parameters of a completed product when the product is manufactured using the virtual manufacturing parameter values, using the values ​​(virtual manufacturing parameter values) created by the value creation unit 33 and the second prediction model described above. That is, the second quality prediction unit 34 predicts the virtual quality parameters that are predicted to become the quality parameters of the completed product when the product is manufactured using the virtual manufacturing parameter values.

[0028] The parameter extraction unit 35 is a component that evaluates the quality parameters (a group of virtual quality parameters) predicted by the second quality prediction unit 34, and identifies (extracts) virtual manufacturing parameter values ​​that are estimated to be the most desirable quality parameters. The parameter extraction unit 35 evaluates the quality parameters based on actual values ​​of quality parameters measured in the past. That is, the evaluation criterion for the quality parameters may be whether or not all of multiple types of quality parameters exceed a reference value (threshold value) set based on past actual measurement values. The evaluation criterion for the quality parameters may also be how close each quality parameter is to an ideal value set based on past actual measurement values, or whether or not each quality parameter is within a range set based on past actual measurement values. Furthermore, the quality parameters may be evaluated based on multiple evaluation criteria.

[0029] The data storage unit 36 ​​is a part that stores (stores) various data such as past set values ​​and actual measured values ​​of each manufacturing parameter, each quality parameter (product characteristic) of the organic EL device 100 manufactured in the past, each predictive model generated by the model generation unit 31, and each predictive model.

[0030] Next, the operation of manufacturing a product (organic EL device 100) by the manufacturing system 1 of this embodiment will be described in detail with reference to FIG. In the manufacturing system 1 of this embodiment, when manufacturing a product, the first quality prediction unit 32 uses the manufacturing parameters actually used in manufacturing to execute a prediction operation to predict the quality parameters of the product to be manufactured (STEP 1). Then, a determination operation is performed to determine whether the predicted quality parameters satisfy the standards (quality standards), and if the standards are not satisfied (Yes in STEP 2), the manufacturing system 1 changes the manufacturing parameters and manufactures the product in the current or subsequent manufacturing operations.

[0031] That is, after the production of a product is completed, it usually takes a certain period of time (e.g., two weeks) until the respective product characteristics (quality parameters) are measured. In particular, in the case of long-run production, the period until the product characteristics are measured becomes even longer. In this case, there is a risk that an operator will not notice if some factor has caused the product characteristics to deteriorate before the product characteristics are measured. In particular, in the case of long-run production, it may be necessary to adjust the production conditions (production parameters) without stopping the process, which may result in the deterioration of the product characteristics. In such a case, there is a risk that many products with undesirable product characteristics will be produced until the measurement of the product characteristics is completed and the operator confirms the deterioration of the product characteristics. In contrast, in the present invention, by performing a predictive operation by the first quality prediction unit 32 at the same time as or before or after the start of production, it is possible to predict future deterioration of product characteristics and respond to the deterioration of product characteristics as quickly as possible.

[0032] In detail, if the predicted quality parameters do not meet the standards (Yes in STEP 2), the following operations are performed to change the manufacturing parameters: a value creation operation to create virtual manufacturing parameters (STEP 3), a prediction operation to predict quality parameters using the virtual manufacturing parameters (STEP 4), a parameter extraction operation to identify virtual manufacturing parameters that will result in desirable quality (STEP 5), and an operation to change the manufacturing parameters to manufacturing parameters that will result in desirable quality (STEP 6).

[0033] The determination operation (STEP 2) of whether the predicted quality parameters satisfy the criteria may be performed based on whether one or more of the values ​​of multiple types of quality parameters are equal to or greater than a reference value set for each type. The determination criterion for the determination operation may be how close one or more quality parameters are to an ideal value set based on past actual measurement values, or whether they are within a range set based on past actual measurement values. Furthermore, the determination operation may be performed based on multiple criteria. Furthermore, the discrimination operation may determine that the discrimination criterion is not met when the value of one or more quality parameters is tending to deteriorate. Note that the term "the quality parameter value is deteriorating" as used herein includes, for example, when a predetermined number of discrimination operations are performed, the quality parameter value of the manufactured product changes in a direction that does not satisfy the reference value (approaches a value that does not satisfy the reference value), such as when the quality parameter value predicted in the discrimination operation performed in the second manufacturing operation is lower than the quality parameter value predicted in the discrimination operation performed in the first manufacturing operation, and when the quality parameter value predicted in the discrimination operation performed in the third manufacturing operation is even lower than the quality parameter value predicted in the discrimination operation performed in the second manufacturing operation. Similarly, it includes a change in a direction that is away from the ideal value or a change in a direction that is outside the range.

[0034] The value creation operation (STEP 3) is an operation performed by the value creation unit 33, and as described above, is an operation for creating a plurality of virtual manufacturing parameters by changing the value of one type of manufacturing parameter.

[0035] The quality parameter prediction operation (STEP 4) is an operation in which the second quality prediction unit 34 uses the above-mentioned second prediction model to predict the quality parameters of each product when the product is manufactured using each of the multiple manufacturing parameters (group of manufacturing parameters) created by the value creation operation (created by the value creation unit 33). In detail, this prediction operation predicts the quality parameters of the product manufactured using each of the manufacturing parameters, for example, when the manufacturing parameters are (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z2, deposition speed: A1), the quality parameters of the product manufactured are (brightness: α1, color coordinate (X): β2, color coordinate (Y): γ1, drive voltage: Δ1); when the manufacturing parameters are (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z3, deposition speed: A1), the quality parameters of the product manufactured are (brightness: α2, color coordinate (X): β1, color coordinate (Y): γ3, drive voltage: Δ2); and so on.

[0036] The parameter extraction operation (STEP 5) is an operation performed by the parameter extraction unit 35, and includes a discrimination operation for discriminating the most preferable quality parameter from among the multiple quality parameters predicted by the second quality prediction unit 34 in the operation of STEP 4). Furthermore, the parameter extraction operation includes an operation for identifying (extracting) a manufacturing parameter (a virtual manufacturing parameter created by the value creation unit 33) for manufacturing a product that will become the quality parameter, based on the quality parameter identified by the discrimination operation. In other words, the parameter extraction operation is an operation for executing the discrimination operation and the manufacturing parameter identification operation in this order. The determination operation for determining the most preferable quality parameter is an operation for determining the most preferable quality parameter based on the above-described quality parameter evaluation method. Furthermore, the operation of identifying the manufacturing parameters is, for example, in the above example, if the most desirable quality parameters are as above (brightness: α2, color coordinate (X): β1, color coordinate (Y): γ3, driving voltage: Δ2), the manufacturing parameters (virtual manufacturing parameters) set to manufacture a product with these quality parameters will be the manufacturing parameters used to predict those quality parameters (flow rate of material gas: X1, temperature of substrate: Y1, film thickness: Z3, film formation speed: A1).In this way, the operation of identifying the virtual manufacturing parameters used to predict those quality parameters based on the quality parameters identified in the discrimination operation.

[0037] Then, when the virtual manufacturing parameters predicted to be the most desirable quality parameters are identified, the manufacturing parameters used when manufacturing the product in the manufacturing unit 2 are changed to the identified virtual manufacturing parameters (STEP 6). This prevents the problem of manufacturing products with undesirable product characteristics (quality parameters), or reduces the number of manufactured products with undesirable product characteristics. [Explanation of symbols]

[0038] 1. Manufacturing System 2 Manufacturing Department 30 Quality measurement department 31 Model Generation Unit 32 1st Quality Prediction Department 33 Value Creation Unit 34 2nd Quality Prediction Department 35 Parameter Extraction Unit

Claims

1. a manufacturing department that manufactures products using the set manufacturing parameters; a model generation unit that generates a nonlinear regression model and a linear regression model by supervised learning using data sets of past manufacturing parameters and past quality parameters; a first quality prediction unit that predicts quality parameters from manufacturing parameters at a point in time using the nonlinear regression model; a value creation unit that, when the quality parameter predicted by the first quality prediction unit does not satisfy the quality standard, creates a plurality of parameter values ​​for one variable manufacturing parameter among the manufacturing parameters, the parameter values ​​being within a predetermined range from the parameter value at the one time point; a second quality prediction unit that predicts quality parameters from the parameter values ​​generated by the value generation unit and the parameter values ​​of the remaining manufacturing parameters at the time point using the linear regression model; a parameter extraction unit that extracts a parameter value of the one type of manufacturing parameter corresponding to a quality parameter that satisfies an extraction criterion from among the quality parameters predicted by the second quality prediction unit; a parameter setting unit that sets manufacturing parameters for the manufacturing unit using the parameter values ​​extracted by the parameter extraction unit;

2. The manufacturing system of claim 1 , wherein the nonlinear regression model is generated by at least one algorithm selected from the group consisting of a decision tree, a support vector machine, and a neural network.

3. The manufacturing system according to claim 1 or 2, wherein the linear regression model is generated by at least one algorithm selected from the group consisting of partial least squares regression, logistic regression, and Lasso regression.

Images