Predictive model generation method, control method, predictive model generation apparatus, control apparatus, predictive model generation program, and control program

Abstract

To obtain a prediction model generation method capable of including integrated values of measurement values in teaching data for machine learning, thereby improving prediction accuracy.SOLUTION: A processor acquires teaching data including a measured value of at least one parameter when processing an object to be processed by a processing device, an integrated value of the measured value, a surface state of the object to be processed, and a processing state of the object to be processed, and generates a prediction model using the measured value, the integrated value and the surface state as explanatory variables and the processing state as an objective variable by machine learning using the acquired teaching data.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present disclosure relates to a prediction model generation method, a control method, a prediction model generation device, a control device, a prediction model generation program, and a control program. [Background Art]

[0002] Patent Document 1 discloses a film thickness estimation system for estimating the film thickness of an electrodeposition paint for each part of a vehicle body, the system including a learning unit that performs learning by associating data relating to compositional condition factors of the electrodeposition paint, data relating to the film thickness, and data relating to adjustment factors of electrodeposition coating equipment with each other. [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2022-107869 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] However, Patent Document 1 does not discuss at all the inclusion of not only measured parameter values but also integrated values of the measured values in teacher data for machine learning.

[0005] An object of the present disclosure is to obtain a prediction model generation method, a prediction model generation device, a prediction model generation program, a control method, a control device, and a control program, wherein the prediction model generation method, prediction model generation device, and prediction model generation allow integrated values of measured values to be included in teacher data for machine learning, which makes it possible to improve prediction accuracy, and the control method, control device, and control program allow parameters to be controlled with high accuracy using a prediction model having high prediction accuracy. [Means for Solving the Problem]

[0006] A predictive model generation method according to one aspect of the present disclosure involves a processor acquiring training data including measured values ​​of at least one parameter when processing an object by a processing device, the integrated value of the measured values, the surface state of the object, and the processing state of the object, and generating a predictive model using machine learning with the acquired training data, with the measured values, the integrated value, and the surface state as explanatory variables and the processing state as the objective variable.

[0007] A control method according to another aspect of the present disclosure involves a processor predicting the processing state of an object using a machine learning-trained predictive model, where the predictive variable is the processing state of the object, and the predictive variable is the processing state of the object, with the measured value of at least one parameter used when the processing device processes the object, the integrated value of the measured values, and the surface state of the object. The processor predicts the processing state of the object and controls the parameter based on the predicted processing state. [Effects of the Invention]

[0008] According to this disclosure, the cumulative value of measured values ​​can be included in training data for machine learning, thereby improving prediction accuracy.

[0009] Furthermore, according to this disclosure, parameters can be controlled with high precision using a predictive model with high prediction accuracy, thereby stabilizing the processing state of the object being processed. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows the configuration of the processing system according to the embodiment of this disclosure. [Figure 2] This diagram shows a simplified configuration of a predictive model generation device. [Figure 3] This diagram shows a simplified configuration of the control device. [Figure 4] This flowchart shows the processes executed by the information processing unit of the predictive model generation device. [Figure 5] This flowchart shows the processes executed by the information processing unit of the control unit. [Figure 6] This graph shows an example of setting control parameters by the control unit. [Modes for carrying out the invention]

[0011] (Knowledge that forms the basis of this disclosure) In the process of forming a chemical conversion coating on an electrolytic capacitor element, for example, the electrolytic capacitor element is placed in an aqueous phosphoric acid solution, and a voltage is applied between the anode and cathode connected to the electrolytic capacitor element to form a chemical conversion coating on the surface of the electrolytic capacitor element. The values ​​of parameters such as the applied voltage are determined by prior verification, and after the start of the process, the process is carried out under these determined fixed conditions.

[0012] However, even when the process is carried out under consistent conditions, variations may occur in the quality (especially the film thickness) of the chemical conversion coating formed on the surface of the electrolytic capacitor element. In other words, the processing state of the object being processed may not be stable.

[0013] To solve these problems, the inventors have found that by generating a predictive model using machine learning with the processing state of the object to be processed as the target variable, the processing state can be stabilized by dynamically controlling the parameters based on the processing state predicted using the predictive model, and furthermore, the prediction accuracy of the predictive model can be improved by including the integrated value of the measured values ​​of the parameters in the training data for machine learning, leading to the present disclosure.

[0014] Next, we will describe each aspect of this disclosure.

[0015] In the prediction model generation method according to the first aspect of the present disclosure, a processor acquires teacher data including measured values of at least one parameter when a processing apparatus executes processing on a processing object, an integrated value of the measured values, a surface state of the processing object, and a processing state of the processing object, and generates a prediction model that uses the measured values, the integrated value, and the surface state as explanatory variables and the processing state as an objective variable through machine learning using the acquired teacher data.

[0016] According to the first aspect, the teacher data for machine learning of the prediction model includes the measured value of at least one parameter, the integrated value of the measured values, the surface state of the processing object, and the processing state of the processing object. Therefore, not only the measured values of parameters but also the integrated values of the measured values can be included in the teacher data for machine learning, thereby making it possible to improve the prediction accuracy of the prediction model.

[0017] In the prediction model generation method according to the second aspect of the present disclosure, in the first aspect, it is preferable that the processing includes a film formation process on the surface of the processing object, and the processing state includes the film thickness of the film.

[0018] According to the second aspect, it is possible to achieve stabilization of the film thickness in the film formation process on the surface of the processing object.

[0019] In the prediction model generation method according to the third aspect of the present disclosure, in the first or second aspect, it is preferable that the parameter includes at least one of voltage, current, temperature, concentration, flow rate, and a production amount of reaction product, and the surface state includes at least one of surface color, gloss, and electrical conductivity.

[0020] According to the third aspect, it is possible to improve the prediction accuracy of the prediction model through machine learning in which appropriate parameters are included in the teacher data.

[0021] A control method according to a fourth aspect of the present disclosure involves a processor predicting the processing state of an object using a machine learning-based predictive model, in which the processor uses a measured value of at least one parameter when the processing device processes the object, the integrated value of the measured values, and the surface state of the object as explanatory variables, and the processing state of the object as the objective variable, and controls the parameter based on the predicted processing state.

[0022] According to the fourth aspect, the explanatory variables of the prediction model include measured values ​​of at least one parameter, the integrated value of the measured values, and the surface state of the object being processed. Therefore, parameters can be controlled with high precision using a prediction model that includes the integrated value of the measured values, and as a result, the processing state of the object being processed can be stabilized.

[0023] A predictive model generation apparatus according to a fifth aspect of the present disclosure includes: an acquisition unit that acquires training data including measured values ​​of at least one parameter when processing an object to be processed by a processing apparatus, an integrated value of the measured values, the surface state of the object to be processed, and the processing state of the object to be processed; and a generation unit that generates a predictive model using machine learning with the training data acquired by the acquisition unit, with the measured values, the integrated value, and the surface state as explanatory variables and the processing state as the objective variable.

[0024] According to the fifth aspect, the training data for machine learning a predictive model includes measured values ​​of at least one parameter, the integrated value of the measured values, the surface state of the object being processed, and the processed state of the object being processed. Therefore, the training data for machine learning can include not only measured values ​​of parameters but also the integrated value of the measured values, thereby improving the prediction accuracy of the predictive model.

[0025] A control device according to a sixth aspect of this disclosure includes: a storage unit that stores a machine learning-trained predictive model in which the measured values ​​of at least one parameter used when processing an object by a processing device, the integrated value of the measured values, and the surface state of the object are used as explanatory variables, and the processing state of the object is used as the objective variable; a prediction unit that predicts the processing state of the object using the predictive model read from the storage unit; and a control unit that controls the parameters based on the processing state predicted by the prediction unit.

[0026] According to the sixth aspect, the explanatory variables of the prediction model include measured values ​​of at least one parameter, the integrated value of the measured values, and the surface state of the object being processed. Therefore, parameters can be controlled with high precision using a prediction model that includes the integrated value of the measured values, and as a result, the processing state of the object being processed can be stabilized.

[0027] A predictive model generation program according to a seventh aspect of this disclosure is a program that causes a processor mounted on a predictive model generation device to function as an acquisition means for acquiring training data including measured values ​​of at least one parameter when processing an object to be processed by a processing device, the integrated value of the measured values, the surface state of the object to be processed, and the processing state of the object to be processed; and a generation means for generating a predictive model using machine learning with the training data acquired by the acquisition means, with the measured values, the integrated value, and the surface state as explanatory variables and the processing state as the objective variable.

[0028] According to the seventh aspect, the training data for machine learning a predictive model includes measured values ​​of at least one parameter, the integrated value of the measured values, the surface state of the object being processed, and the processed state of the object being processed. Therefore, the training data for machine learning can include not only measured values ​​of parameters but also the integrated value of the measured values, thereby improving the prediction accuracy of the predictive model.

[0029] The control program according to the eighth aspect of this disclosure is a program for causing a processor mounted on a control device to function as follows: the control device has a storage unit that stores a machine learning-trained predictive model in which the measured value of at least one parameter when the processing device performs processing on the object to be processed, the integrated value of the measured value, and the surface state of the object to be processed are used as explanatory variables, and the processing state of the object to be processed is used as the objective variable; the control device has a storage unit that stores a machine learning-trained predictive model in which the measured value of at least one parameter when the processing device performs processing on the object to be processed is used as an explanatory variable, and the processing state of the object to be processed is used as an objective variable; and the control program is a program for causing a processor mounted on a control device to function as follows: predictive means that predicts the processing state of the object to be processed using the predictive model read from the storage unit; and control means that controls the parameters based on the processing state predicted by the predictive means.

[0030] According to the eighth aspect, the explanatory variables of the prediction model include a measured value of at least one parameter, the integrated value of the measured values, and the surface state of the object being processed. Therefore, parameters can be controlled with high precision using a prediction model that includes the integrated value of the measured values, and as a result, the processing state of the object being processed can be stabilized.

[0031] This disclosure can also be implemented as a program that causes a computer to execute each characteristic configuration included in such a method or apparatus, or as a system that operates using such a program. It goes without saying that such a computer program can be distributed via a computer-readable, non-temporary recording medium such as a CD-ROM, or via a communication network such as the Internet.

[0032] (Embodiments of the present disclosure) Embodiments of this disclosure will be described in detail below with reference to the drawings. Elements denoted by the same reference numeral in different drawings refer to the same or corresponding elements. Furthermore, the components, their arrangement, connection configurations, and operating sequences shown in the following embodiments are examples and are not intended to limit this disclosure. This disclosure is limited only by the claims. Therefore, among the components in the following embodiments, those not described in the independent claims representing the highest-level concepts of this disclosure are described as constituting a more preferable configuration, even though they are not necessarily required to achieve the object of this disclosure.

[0033] Figure 1 is a diagram showing the configuration of a processing system according to an embodiment of the present disclosure. The processing system comprises a predictive model generation device 11, a control device 12, and a processing device 20. The predictive model generation device 11 and the control device 12 may be configured as dedicated terminals, as general-purpose personal computers, or as server devices such as edge servers or cloud servers. The predictive model generation device 11 and the control device 12 may be configured individually as separate devices, or they may be configured commonly as a single integrated device.

[0034] The processing apparatus 20 is a device that performs processing on an object to be processed. In this embodiment, the object to be processed is an electrolytic capacitor element 100 such as a tantalum capacitor, and the processing is the formation of a chemical conversion film 101 on the surface of the electrolytic capacitor element 100. The present disclosure can also be applied to the formation of a conductive polymer film, which is performed after the formation of the chemical conversion film 101. The processing apparatus 20 includes a processing tank 21, a cathode 23, an anode 24, wires 25, a power supply circuit 26, a regulator 27, sensors 31, 32, and a camera 33. Each of the sensors 31, 32 may be composed of multiple sensors, and the camera 33 may be composed of multiple cameras.

[0035] The processing tank 21 is filled with a phosphoric acid aqueous solution 22, and the electrolytic capacitor element 100 is placed in the phosphoric acid aqueous solution 22. The electrolytic capacitor element 100 is connected to an anode 24 such as an aluminum carrier bar via a wire 25 such as tantalum.

[0036] The power supply circuit 26 and the sensor 31 are connected between the cathode 23 and the anode 24. The power supply circuit 26 can adjust the output voltage and current according to the control signal S21 input from the control device 12. The sensor 31 measures the voltage and current, which are parameters of the power supply circuit 26, and inputs the measured value data D11, which represents the measured values, to the prediction model generation device 11 and the control device 12.

[0037] The regulator 27 and sensor 32 are located inside the processing tank 21. The regulator 27 can adjust the temperature, concentration, and flow rate of the phosphoric acid aqueous solution 22 by a control signal S22 input from the control device 12. The sensor 32 measures the temperature, concentration, flow rate, and amount of reaction products (e.g., hydrogen) produced, which are parameters of the phosphoric acid aqueous solution 22, and the conductivity, which is the surface state of the electrolytic capacitor element 100, and inputs the measured value data D12, which represents these measured values, to the prediction model generation device 11 and the control device 12.

[0038] Camera 33 photographs the electrolytic capacitor elements 100 placed inside the processing tank 21, and inputs the image data D13 acquired by the photography to the prediction model generation device 11 and the control device 12.

[0039] Figure 2 is a simplified diagram showing the configuration of the prediction model generation device 11. The prediction model generation device 11 comprises an information processing unit 41, a storage unit 42, a communication unit 43, and an input unit 44.

[0040] The information processing unit 41 is configured using a processor (information processing device) such as a CPU or GPU. The storage unit 42 is configured using an HDD, SSD, or semiconductor memory. The communication unit 43 is configured using a communication module compatible with any communication standard such as Bluetooth (registered trademark) or Wi-Fi. The input unit 44 is configured using any input device such as a touch panel, mouse, or keyboard.

[0041] The memory unit 42 stores and stores measurement data 61, integrated value data 62, surface condition data 63, and processing condition data 64.

[0042] The measured data 61 corresponds to the measured data D11 of the voltage and current of the power supply circuit 26. The measured data 61 also corresponds to the measured data D12 of the temperature, concentration, flow rate, and amount of reaction product produced of the phosphoric acid aqueous solution 22.

[0043] The cumulative value data 62 shows the cumulative value from the start of processing to the present for each of the multiple parameters included in the measured value data 61.

[0044] The surface condition data 63 is data indicating the color, gloss, and conductivity of the surface of the electrolytic capacitor element 100, which is the object to be processed. The color and gloss of the surface of the electrolytic capacitor element 100 can be obtained by analyzing the image data D13, and the conductivity of the surface of the electrolytic capacitor element 100 can be extracted from the measured value data D12.

[0045] The processing status data 64 is data indicating the thickness of the chemical conversion coating 101 formed on the surface of the electrolytic capacitor element 100. The thickness of the chemical conversion coating 101 can be measured by non-destructive measurement using X-rays, or it can be estimated from the color of the surface of the electrolytic capacitor element 100 by analyzing the image data D13.

[0046] The information processing unit 41 includes an acquisition unit 51, a generation unit 52, and an output unit 53, which are functions realized by the processor executing a program read from a computer-readable non-volatile recording medium such as ROM. In other words, the above program is a program that causes the information processing unit 41, which is an information processing device mounted on the prediction model generation device 11, to function as an acquisition unit 51 (acquisition means), a generation unit 52 (generation means), and an output unit 53 (output means). Details of the processing performed by each processing unit will be described later.

[0047] Figure 3 is a simplified diagram showing the configuration of the control device 12. The control device 12 comprises an information processing unit 71, a storage unit 72, a communication unit 73, and an input unit 74.

[0048] The information processing unit 71 is configured using a processor (information processing device) such as a CPU or GPU. The storage unit 72 is configured using an HDD, SSD, or semiconductor memory. The communication unit 73 is configured using a communication module compatible with any communication standard such as Bluetooth (registered trademark) or Wi-Fi. The input unit 74 is configured using any input device such as a touch panel, mouse, or keyboard.

[0049] The memory unit 72 stores the prediction model 91 input from the prediction model generation device 11. The memory unit 72 also stores measurement data 61, integrated value data 62, and surface condition data 63, similar to those stored in the memory unit 42.

[0050] The information processing unit 71 includes an acquisition unit 81, a prediction unit 82, a control unit 83, and an output unit 84, which are functions realized by the processor executing a program read from a computer-readable non-volatile recording medium such as ROM. In other words, the above program is a program that causes the information processing unit 71, which is an information processing device mounted on the control device 12, to function as an acquisition unit 81 (acquisition means), a prediction unit 82 (prediction means), a control unit 83 (control means), and an output unit 84 (output means). Details of the processing performed by each processing unit will be described later.

[0051] Figure 4 is a flowchart showing the process executed by the information processing unit 41 of the predictive model generation device 11.

[0052] First, in step SP11, the acquisition unit 51 acquires training data, including measurement data 61, integrated value data 62, surface condition data 63, and processing condition data 64, by reading them from the storage unit 42.

[0053] Next, in step SP12, the generation unit 52 generates a prediction model 91 using machine learning with the training data acquired by the acquisition unit 51. Any machine learning algorithm, such as a neural network, can be applied. The prediction model 91 is a prediction model that uses the measured value data 61, the integrated value data 62, and the surface state data 63 as explanatory variables, and the film thickness of the chemical conversion coating 101, which is the processing state of the object to be processed, as the objective variable.

[0054] Next, in step SP13, the output unit 53 outputs the prediction model 91 generated by the generation unit 52. The prediction model 91 output by the output unit 53 is input to the control device 12 via the communication unit 43.

[0055] Figure 5 is a flowchart showing the process executed by the information processing unit 71 of the control device 12.

[0056] First, in step SP21, the acquisition unit 81 acquires the measured value data 61, the integrated value data 62, and the surface condition data 63 by reading them from the storage unit 72.

[0057] Next, in step SP22, the prediction unit 82 inputs the measured value data 61, integrated value data 62, and surface condition data 63 acquired by the acquisition unit 81 into the prediction model 91 read from the storage unit 72, thereby predicting the film thickness of the chemical conversion coating 101, which is the processing state of the object to be processed.

[0058] Next, in step SP23, the control unit 83 sets the control parameters for the power supply circuit 26 and the regulator 27 based on the thickness of the chemical conversion coating 101 predicted by the prediction unit 82. The control parameters for the power supply circuit 26 include voltage and current, and the control parameters for the regulator 27 include temperature, concentration, and flow rate of the phosphoric acid aqueous solution 22, and at least one of these control parameters is set by the control unit 83.

[0059] Figure 6 is a graph showing an example of setting control parameters by the control unit 83. The horizontal axis represents the elapsed time since the start of processing, and the vertical axis represents the film thickness of the chemical conversion coating 101. The dashed line characteristic K0 shows the time-series change of the target value of the film thickness, and the solid line characteristic K1 shows the time-series change of the predicted value of the film thickness by the prediction unit 82.

[0060] At time T1, the predicted value is higher than the target value. In this case, the control unit 83 changes the control parameter to be set from its previous setting. For example, it sets the output voltage value of the power supply circuit 26 to a value lower than the output voltage value set at the previous sampling timing. This brings characteristic K1 closer to characteristic K0.

[0061] At time T2, the predicted value is lower than the target value. In this case, the control unit 83 changes the control parameter to be set from its previous setting. For example, it sets the output voltage value of the power supply circuit 26 to a value higher than the output voltage value set at the previous sampling timing. This brings characteristic K1 closer to characteristic K0.

[0062] Referring to Figure 5, in step SP24, the output unit 84 outputs the control parameters set by the control unit 83. The control parameters output by the output unit 84 are input to the power supply circuit 26 and the regulator 27 via the communication unit 73.

[0063] According to the prediction model generation device 11 of this embodiment, the training data for machine learning the prediction model 91 includes measured value data 61, integrated value data 62, surface state data 63, and processing state data 64. Therefore, the training data for machine learning can include not only measured values ​​of parameters but also integrated values ​​of measured values, thereby improving the prediction accuracy of the prediction model 91.

[0064] Furthermore, according to the prediction model generation apparatus 11 of this embodiment, it is possible to stabilize the film thickness in the process of forming a chemical conversion film 101 on the surface of the electrolytic capacitor element 100, which is the object to be processed.

[0065] Furthermore, according to the prediction model generation device 11 of this embodiment, it is possible to improve the prediction accuracy of the prediction model 91 by machine learning that includes appropriate parameters in the training data.

[0066] According to the control device 12 of this embodiment, the explanatory variables of the prediction model 91 include measured value data 61, integrated value data 62, and surface state data 63. Therefore, the control parameters can be controlled with high precision using the prediction model 91 which includes integrated value data 62, and as a result, the processing state of the electrolytic capacitor element 100, which is the object to be processed (such as the thickness of the chemical conversion coating 101), can be stabilized. [Industrial applicability]

[0067] This disclosure is particularly useful for application to processing systems such as chemical treatment systems. [Explanation of Symbols]

[0068] 11. Predictive Model Generator 12 Control device 20 Processing Units 41 Information Processing Section 51 Acquisition Department 52 Generation part 61 Measurement Data 62. Cumulative Value Data 63 Surface condition data 64 Processing status data 71 Information Processing Department 72 Memory section 81 Acquisition Department 82 Prediction Section 83 Control Unit 91 Predictive Models 100 electrolytic capacitor elements 101 Chemical conversion coating

Claims

1. The processor, Training data is obtained that includes measured values ​​of at least one parameter when processing an object using a processing device, the integrated value of the measured values, the surface state of the object, and the processing state of the object. A method for generating a predictive model, which generates a predictive model using machine learning with the acquired training data, with the measured value, the integrated value, and the surface state as explanatory variables and the processing state as the target variable.

2. The process includes a process of forming a film on the surface of the object to be processed. The method for generating a predictive model according to claim 1, wherein the processing state includes the film thickness of the film.

3. The aforementioned parameters include at least one of voltage, current, temperature, concentration, flow rate, and amount of reaction product produced. The method for generating a predictive model according to claim 1, wherein the surface condition includes at least one of the surface color, gloss, and conductivity.

4. The processor, Using a machine learning-trained predictive model, with the measured value of at least one parameter used when processing the object by the processing apparatus, the integrated value of the measured values, and the surface state of the object as explanatory variables, and the processing state of the object as the objective variable, the processing state of the object is predicted. A control method for controlling the parameters based on the predicted processing state.

5. An acquisition unit that acquires training data including measured values ​​of at least one parameter when processing an object to be processed by a processing device, the integrated value of the measured values, the surface state of the object to be processed, and the processing state of the object to be processed. A generation unit generates a predictive model using machine learning with the training data acquired by the acquisition unit, with the measured value, the integrated value, and the surface state as explanatory variables and the processing state as the target variable. A predictive model generation device equipped with the following features.

6. A storage unit that stores a machine learning-based predictive model in which the measured values ​​of at least one parameter used when processing an object by a processing device, the integrated value of the measured values, and the surface state of the object are used as explanatory variables, and the processing state of the object is used as the objective variable, A prediction unit predicts the processing state of the object to be processed using the prediction model read from the storage unit, A control unit that controls the parameters based on the processing state predicted by the prediction unit, A control device equipped with the following features.

7. The processor installed in the predictive model generation device is Acquisition means for acquiring training data including measured values ​​of at least one parameter when processing an object to be processed by a processing device, the integrated value of the measured values, the surface state of the object to be processed, and the processing state of the object to be processed. A generation means generates a predictive model using machine learning with the training data acquired by the acquisition means, with the measured value, the integrated value, and the surface state as explanatory variables and the processing state as the target variable. A predictive model generation program designed to function as such.

8. The processor installed in the control unit The control device has a storage unit that stores a machine learning-based predictive model in which the measured values ​​of at least one parameter used when the processing device processes the object to be processed, the integrated value of the measured values, and the surface state of the object to be processed are used as explanatory variables, and the processing state of the object to be processed is used as the objective variable. A prediction means that predicts the processing state of the object to be processed using the prediction model read from the storage unit, A control means that controls the parameters based on the processing state predicted by the prediction means, A control program to make it function as such.

Images