JP7395987B2 - Information processing systems, methods, and programs - Google Patents

Description

The present invention relates to an information processing system, method, and program.

2. Description of the Related Art Conventionally, there has been known a technique for predicting the quality of a product before production. Based on the predicted quality, operational quantities related to the production process, such as raw materials and processing conditions, can be determined.

In Patent Document 1, an analysis device predicts whether the state of the production target is good or bad in the process in which a grinder as production equipment produces a crankshaft as the production target (paragraph [0099] of Patent Document 1). . Specifically, the prediction unit predicts whether or not the crankshaft being ground is a good product based on data obtained during the grinding process and various information entered in advance by an operator or the like ( Paragraph [0029] of Patent Document 1). Then, if the operator determines that there are more defective products than usual as a result of checking the analysis results by the analysis device, he performs maintenance on the grinding machine at that point (paragraph [0063] of Patent Document 1).

However, if there is an abnormality in the data used for prediction (in Patent Document 1, data obtained during grinding, various information entered in advance by an operator, etc.), it is difficult to accurately predict quality. I can't. The amount of operation related to the production process is determined based on quality that is not accurately predicted, and as a result, there is a problem in that products of desired quality are not produced.

Accordingly, an embodiment of the present invention aims to improve the accuracy of manipulated variables related to the production process.

In order to solve the above-mentioned problems, a data aggregation unit that aggregates data related to the production of a product, a quality prediction unit that uses machine learning to predict the quality of the product based on the data, and a quality prediction unit that uses machine learning to predict the quality of the product. an operation amount calculation section that calculates the operation amount related to the production process necessary to produce a product of a predetermined quality based on the operation amount; an abnormality detection section that uses machine learning to determine whether or not there is an abnormality in the operation amount; and a process operation instruction unit that outputs a manipulated variable with no error.

According to the present invention, it is possible to improve the accuracy of manipulated variables related to the production process.

1 is an overall configuration diagram including an information processing system according to an embodiment of the present invention. 1 is a hardware configuration diagram of an information processing system according to an embodiment of the present invention. FIG. 1 is a functional block diagram of an information processing system according to an embodiment of the present invention. 3 is a flowchart of information processing according to an embodiment of the present invention. 3 is a flowchart of information processing according to an embodiment of the present invention. 3 is a flowchart of information processing according to an embodiment of the present invention. 1 is a flowchart showing a process for manufacturing a polymerized toner according to an embodiment of the present invention.

Each embodiment will be described below with reference to the accompanying drawings. Note that, in this specification and the drawings, components having substantially the same functional configuration are designated by the same reference numerals and redundant explanation will be omitted.

In this specification, the case of producing toner will be described as an example, but the present invention is applicable not only to toner but also to producing any product (for example, a chemical product manufactured using a chemical reaction). be able to.

<System configuration>
FIG. 1 is an overall configuration diagram including an information processing system 10 according to an embodiment of the present invention. As shown in FIG. 1, the information processing system 10 connects equipment 20a, 20b, 20c, . . . ( Hereinafter, it is communicably connected to the production process equipment 20). Note that the production process may be only one production process, or may include a plurality of production processes. Each will be explained below.

The equipment 20 for each production process includes controllers 21a, 21b, 21c, ... (hereinafter collectively referred to as controllers 21) and production equipment 22a, 22b, 22c, ... (hereinafter collectively referred to as production equipment 22) and sensors 23a, 23b, 23c, . . . (hereinafter collectively referred to as sensors 23).

The controller 21 receives manipulated variables related to the production process from the information processing system 10 . Further, the controller 21 controls the production equipment 22 based on the operation amount received from the information processing system 10.

The production equipment 22 is any equipment for producing products.

Sensor 23 is any sensor. For example, the sensor 23 can detect the state of the production equipment 22 and the state inside the factory or the like where the production equipment 22 is installed.

The information processing system 10 predicts the quality of the product produced in each production process, and calculates the amount of operation related to the production process based on the predicted quality. The information processing system 10 can determine whether there is an abnormality in the calculated manipulated variable. Furthermore, the information processing system 10 can determine whether there is an abnormality in data used to predict quality. The information processing system 10 will be described in detail later with reference to FIG. 3.

Note that the devices described in the Examples are merely one of a plurality of computing environments for implementing the embodiments disclosed herein. In some embodiments, information handling system 10 includes multiple computing devices, such as a server cluster. The plurality of computing devices are configured to communicate with each other via any type of communication link, including a network, shared memory, etc., to perform the processes disclosed herein.

<Hardware configuration>
FIG. 2 is a hardware configuration diagram of the information processing system 10 according to an embodiment of the present invention.

As shown in FIG. 2, the information processing system 10 is constructed by a computer, and includes a CPU 101, ROM 102, RAM 103, HD 104, HDD (Hard Disk Drive) controller 105, It is equipped with a display 106, an external device connection I/F (Interface) 107, a network I/F 108, a data bus 109, a keyboard 110, a pointing device 111, a DVD-RW (Digital Versatile Disk Rewritable) drive 113, and a media I/F 115. .

Among these, the CPU 101 controls the operation of the information processing system 10 as a whole. The ROM 102 stores programs used to drive the CPU 101, such as IPL. RAM 103 is used as a work area for CPU 101. The HD 104 stores various data such as programs. The HDD controller 105 controls reading and writing of various data to the HD 104 under the control of the CPU 101. The display 106 displays various information such as a cursor, menu, window, characters, or images. The external device connection I/F 107 is an interface for connecting various external devices. The external device in this case is, for example, a USB (Universal Serial Bus) memory, a printer, or the like. Network I/F 108 is an interface for data communication using a communication network. The bus line 109 is an address bus, a data bus, etc. for electrically connecting each component such as the CPU 101 shown in FIG. 2.

Further, the keyboard 110 is a type of input means that includes a plurality of keys for inputting characters, numerical values, various instructions, and the like. The pointing device 111 is a type of input means for selecting and executing various instructions, selecting a processing target, moving a cursor, and the like. The DVD-RW drive 113 controls reading and writing of various data to and from the DVD-RW 112, which is an example of a removable recording medium. Note that it is not limited to DVD-RW, but may be DVD-R or the like. The media I/F 115 controls reading or writing (storage) of data to a recording medium 114 such as a flash memory.

<Functional block>
FIG. 3 is a functional block diagram of the information processing system 10 according to an embodiment of the present invention. As shown in FIG. 3, the information processing system 10 includes a data aggregation unit 11, a quality prediction unit 12, an operation amount calculation unit 13, an abnormality detection unit 14, a process operation instruction unit 15, a data storage unit 16, a calculation result storage unit 17 can be provided. Further, the information processing system 10 can function as a data aggregation unit 11, a quality prediction unit 12, an operation amount calculation unit 13, an abnormality detection unit 14, and a process operation instruction unit 15 by executing programs. Each will be explained below.

The data storage unit 16 stores data related to the production of products (for example, data related to raw materials, data related to processing conditions, etc.) and data related to the quality of the produced products. For example, the data regarding raw materials are inspection results and physical property data of the raw materials. For example, data regarding processing conditions is input in real time from quantitative data and operation performance data for process operation obtained from the process operation instruction unit 15, scheduled data for a process to be started, and various sensors 23 in the production process. Data related to the operation of the production process, such as data, sample data of the production process, quantitative or qualitative data input by operators, prescribed prescription values, and production takt time. For example, data regarding the quality of a produced product is the inspection result of a production process.

The data aggregation unit 11 aggregates data required by the quality prediction unit 12, the operation amount calculation unit 13, and the abnormality detection unit 14 for processing from among the data in the data storage unit 16. Specifically, the data aggregation unit 11 stores the data acquired from the data storage unit 16 in a memory or the like so that the quality prediction unit 12 or the like can refer to the data.

The quality prediction unit 12 predicts the quality of the product at the end of each production process before starting each production process. Specifically, the quality prediction unit 12 uses a learned model generated by machine learning to predict the quality of the product from data related to the production of the product (for example, at least one of data related to raw materials and data related to processing conditions). do. For example, the quality prediction unit 12 can predict each quality of a plurality of items of a product as a numerical value. Note that the quality prediction unit 12 can use a neural network, deep learning, support vector machine, K-means method, decision tree, etc. Furthermore, accuracy can be improved by combining multiple methods through ensemble learning.

The operation amount calculation unit 13 calculates the operation amount regarding the production process based on the quality (that is, the quality predicted by the quality prediction unit 12). Specifically, the operation amount calculation unit 13 calculates the operation amount related to the production process necessary to produce a product of a predetermined quality based on the predicted quality (that is, the quality predicted by the quality prediction unit 12). do. Note that the manipulated variable refers to the optimal control condition value for the production process. For example, the amount of manipulation is the amount of manipulation of at least one of raw materials (amount, proportion, etc.) and processing conditions in order to ensure that the numerical values of each quality of multiple items of the product are within the standard range. . Hereinafter, the case where the manipulated variable is calculated by <machine learning> and the case where the manipulated variable is calculated by <rule base> will be explained separately.

<Machine learning>
The operation amount calculation unit 13 can calculate the operation amount related to the production process from the quality using a neural network, deep learning, support vector machine, K-means method, decision tree, etc. Furthermore, accuracy can be improved by combining multiple methods through ensemble learning.

<Rule base>
The manipulated variable calculation unit 13 can calculate the manipulated variable related to the production process from the quality based on the correspondence between the quality and the manipulated variable related to the production process.

At this time, for example, if there is an error in the data aggregated by the data aggregation unit 11 due to an operator's input error or sensor failure, quality prediction will be performed based on the incorrect data, so the operation amount will be incorrect. The quality is not as expected due to uncalculated and inappropriate control. Furthermore, product quality changes depending on many factors such as physical property values of raw materials, valve conditions in the process, and sensor values. The more processes there are, the more factors affect quality, and the more complex the relationship between quality and control variables becomes. If data without a learning record is input, the predicted quality results will deviate significantly, making it impossible to correctly calculate the amount of operation in the production process. Furthermore, if the calculation result by the operation amount calculation unit 13 is a local solution, the operation amount will deviate from the target and the quality will not be as predicted. Therefore, in one embodiment of the present invention, an abnormality in the manipulated variable is detected.

Note that checking whether there is a track record, checking errors or deficiencies, checking fluctuation trends, etc. is referred to as abnormality detection. In addition, regarding data or calculation results, a state in which there is no track record, a state in which errors or omissions are observed, a state in which rapid fluctuations are observed, etc. is expressed as abnormal.

<Detection of abnormality in manipulated variable>
The abnormality detection unit 14 detects an abnormality in the amount of operation. Specifically, the abnormality detection unit 14 uses a learned model generated by machine learning to detect from the manipulated variable whether there is an abnormality in the manipulated variable. For example, the abnormality detection unit 14 detects the presence or absence of an abnormality by checking the presence or absence of a track record of the manipulated variable, errors, defects, and fluctuation trends.

<Detection of anomalies in data used to predict quality>
Further, the anomaly detection unit 14 can also detect the presence or absence of an anomaly in the data used to predict quality (that is, the data aggregated by the data aggregation unit 11). Further, the anomaly detection unit 14 can also detect the presence or absence of an anomaly in data for generating a trained model for predicting quality (that is, data aggregated by the data aggregation unit 11). Specifically, the anomaly detection unit 14 uses a learned model generated by machine learning to detect the presence or absence of an anomaly in the data aggregated by the data aggregation unit 11. For example, the anomaly detection unit 14 detects the presence or absence of an anomaly by checking the track record, errors, defects, and fluctuation trends of the data aggregated by the data aggregation unit 11.

The calculation result storage unit 17 stores manipulated variables related to the production process. Specifically, the calculation result storage section 17 stores the operation amount calculated by the operation amount calculation section 13.

The process operation instruction unit 15 transmits the operation amount determined to be normal to the controller 21. Note that the process operation instruction unit 15 may be configured to select whether or not to reflect the operation amount determined to be abnormal in the production process. For example, the process operation instruction unit 15 selects whether to reflect, not reflect, or recalculate the operation amount determined to be abnormal in the production process in response to an instruction from an operator's terminal or the like. be able to. The process operation instruction unit 15 transmits the manipulated variable to the controller 21 in the case of acceptance (that is, when it is to be reflected), and does not reflect the manipulated variable in the production process when it is not adopted (that is, when it is not to be reflected). When the process is finished and recalculation is to be performed, quality prediction and operation amount calculation are performed again.

Note that the process operation instruction unit 15 may be configured to display the operation amount determined to be normal on a display means such as an operator's terminal. In this case, the operator views the operation amount displayed on a display means such as the operator's terminal and inputs an instruction to the controller 21. The process operation instruction section 15 is said to output both the transmission to the controller 21 and the display on a display means such as an operator's terminal.

<Processing method>
The method executed by the information processing system 10 will be described below with reference to FIGS. 4 to 6.

FIG. 4 is a flowchart of information processing according to an embodiment of the present invention. In the information processing shown in FIG. 4, it is possible to automatically reflect the operation amount with no abnormality in the production process.

In step 101 (S101), the data aggregation unit 11 aggregates data necessary for processing by the quality prediction unit 12, operation amount calculation unit 13, and abnormality detection unit 14 from among the data in the data storage unit 16. do.

In step 102 (S102), the quality prediction unit 12 uses the learned model generated by machine learning to determine the quality of the product based on data related to production of the product (for example, at least one of data related to raw materials and data related to processing conditions). Predict.

In step 103 (S103), the manipulated variable calculation unit 13 calculates the manipulated variable related to the production process based on the quality predicted in S102.

In step 104 (S104), the abnormality detection unit 14 executes processing for detecting an abnormality in the manipulated variable calculated in S103.

In step 105 (S105), the abnormality detection unit 14 determines whether or not there is an abnormality in the manipulated variable calculated in S103, based on the process in S104. If there is no abnormality, the process proceeds to step 106, and if there is an abnormality, the operation amount is not reflected in the production process and the process is terminated.

In step 106 (S106), the process operation instruction unit 15 transmits to the controller 21 the operation amount determined to be normal in S105. Thereafter, the manipulated variables with no abnormalities are reflected in the production process.

FIG. 5 is a flowchart of information processing according to an embodiment of the present invention. In the information processing shown in FIG. 5, it is possible to determine whether or not there is an abnormality in the data used to predict quality, and to automatically reflect the operation amount with no abnormality in the production process.

In step 201 (S201), the data aggregation unit 11 aggregates data necessary for processing by the quality prediction unit 12, operation amount calculation unit 13, and abnormality detection unit 14 from among the data in the data storage unit 16. do.

In step 202 (S202), the abnormality detection unit 14 executes processing for detecting an abnormality in the data aggregated in S201.

In step 203 (S203), the abnormality detection unit 14 determines whether or not there is an abnormality in the data aggregated in S201, based on the process in S202. If there is no abnormality, the process proceeds to step 204, and if there is an abnormality, the process proceeds to step 208.

In step 208 (S208), the abnormality detection unit 14 displays the data determined to be abnormal in S203 on a display means such as an operator's terminal.

In step 209 (S209), the abnormality detection unit 14 takes action on the abnormal data displayed in S208. For example, the anomaly detection unit 14 can delete or modify abnormal data in response to an instruction from an operator's terminal or the like. Thereafter, the process advances to step 204.

In step 204 (S204), the quality prediction unit 12 uses the trained model generated by machine learning to determine whether the data determined to be normal in S203 or the data deleted or modified in S209 (of the Predict the quality of a product from data related to production (for example, data related to raw materials and/or data related to processing conditions).

In step 205 (S205), the manipulated variable calculation unit 13 calculates the manipulated variable related to the production process based on the quality predicted in S204.

In step 206 (S206), the abnormality detection unit 14 executes a process for detecting an abnormality in the manipulated variable calculated in S205, and determines whether or not there is an abnormality in the manipulated variable calculated in S205. If there is no abnormality, the process proceeds to step 207, and if there is an abnormality, the operation amount is not reflected in the production process and the process is terminated.

In step 207 (S207), the process operation instruction unit 15 transmits to the controller 21 the operation amount determined to be normal in S206. Thereafter, the manipulated variables with no abnormalities are reflected in the production process.

In this manner, in the information processing shown in FIG. 5, the accuracy of quality prediction and operation amount calculation is improved, and abnormalities in the operation amount calculation results can be reduced.

FIG. 6 is a flowchart of information processing according to an embodiment of the present invention. In the information processing shown in FIG. 6, it is possible to determine whether or not there is an abnormality in the data used to predict quality, and to select whether or not to reflect the operation amount with no abnormality in the production process.

In step 301 (S301), the data aggregation unit 11 aggregates data necessary for processing by the quality prediction unit 12, operation amount calculation unit 13, and abnormality detection unit 14 from among the data in the data storage unit 16. do.

In step 302 (S302), the abnormality detection unit 14 executes processing for detecting an abnormality in the data aggregated in S301.

In step 303 (S303), the abnormality detection unit 14 determines whether or not there is an abnormality in the data aggregated in S301, based on the process in S302. If there is no abnormality, the process proceeds to step 304, and if there is an abnormality, the process proceeds to step 309.

In step 309 (S309), the abnormality detection unit 14 displays the data determined to have an abnormality in S303 on a display means such as an operator's terminal.

In step 310 (S310), the abnormality detection unit 14 takes action on the abnormal data displayed in S309. For example, the anomaly detection unit 14 can delete or modify abnormal data in response to an instruction from an operator's terminal or the like. Thereafter, the process advances to step 304.

In step 304 (S304), the quality prediction unit 12 selects the data determined to be normal in S303 or the data deleted or modified in S310 (data related to the production of products (for example, data related to raw materials and data related to processing conditions). Predict the quality of an object from at least one of the data)).

In step 305 (S305), the manipulated variable calculation unit 13 calculates the manipulated variable related to the production process based on the quality predicted in S304.

In step 306 (S306), the abnormality detection unit 14 executes processing for detecting an abnormality in the manipulated variable calculated in S305.

In step 307 (S307), the abnormality detection unit 14 determines whether or not there is an abnormality in the manipulated variable calculated in S305, based on the process in S306. If there is no abnormality, the process advances to step 308, and if there is an abnormality, the process advances to step 311.

In step 308 (S308), the process operation instruction unit 15 transmits to the controller 21 the operation amount determined to be normal in S307. Thereafter, the manipulated variables with no abnormalities are reflected in the production process.

In step 311 (S311), the process operation instruction unit 15 selects whether or not the manipulated variable determined to be abnormal in S307 is to be reflected in the production process. For example, the process operation instruction unit 15 may, in response to an instruction from an operator's terminal or the like, reflect the operation amount determined to be abnormal in S307 in the production process, not reflect it, or recalculate it. You can choose. If it is accepted (that is, if it is to be reflected), the process advances to S308; if it is not adopted (that is, if it is not to be reflected), the process is terminated without reflecting the manipulated variable in the production process, and if it is to be recalculated, the process proceeds to step S308. Proceed to 312.

In step 312 (S312), the process operation instruction unit 15 adds constraints to the calculation conditions and returns to S304 to perform recalculation. For example, if an error occurs during calculation of the manipulated variable, or if the calculation result becomes a local solution and is judged as NG, it is possible to obtain the correct manipulated variable by adding constraints to the calculation conditions and recalculating. You will be able to do this.

In this way, in the information processing shown in FIG. 6, even if the abnormality detection result is NG, detailed quality control can be performed by separately providing a judgment as to whether or not to adopt the abnormality detection result.

Note that in the information processing of FIG. 6, S302 and S303 may be omitted.

<Manufacture of polymerized toner>
FIG. 7 is a flowchart showing a process for manufacturing a polymerized toner according to an embodiment of the present invention. Although the production system of the present invention is applicable to any process, an example in the oil phase creation process will be described. Note that the following processing is not limited to the production of toner, but can be applied to the production of any product.

[Example 1]
Toner particles are produced by dissolving or dispersing a toner raw material composition containing at least two types of resins with different molecular weights, colorants, and release agents in an organic solvent, and then dissolving or dispersing the dissolved or dispersed material (hereinafter referred to as an oil phase). is produced by continuous emulsification in an aqueous medium (hereinafter referred to as aqueous phase) in which a solid resin fine particle dispersant is present.

In the manufacturing method of polymerized toner, it is known that process variations such as the content of impurities in raw materials, the polymerization reaction in the previous process, the control of particle shape, the manufacturing conditions in the emulsification process, and the dispersion process in the previous process affect the quality. It is being Conventionally, manufacturing has been carried out by evaluating quality after the final process, establishing a process control range, and controlling the quality through feedback control based on the manufacturing conditions of the emulsification process to ensure that the quality does not fall outside the control range. Since feedback control causes control delays, there were issues with quality stabilization, etc. Therefore, by introducing the information processing system 10 according to the present invention into the oil phase preparation process and predicting the quality and calculating the optimal control conditions before the start of the process, the recipe in the oil phase process can be automatically adjusted and the quality can be stabilized. conduct.

First, using the information processing system 10 shown in FIGS. 1 to 3, the data aggregation unit 11 uses the previous equipment settings, raw material information, raw material prescription, production results, and product inspection results stored in the data storage unit 16. , tank internal pressure, liquid temperature, various sensor data installed in equipment such as weight scales (hereinafter referred to as past performance data), and raw material physical properties of the oil phase creation process to be carried out from now on held by the process operation instruction unit 15. Collect information such as information, raw material prescriptions, equipment setting values, etc. (hereinafter referred to as planned data). Although the process operation instruction section 15 and the data storage section 16 can be constructed using means such as cloud computing and a network system, in this embodiment, they were constructed in an on-premises environment.

In FIG. 1, each production process A, B, C, . . . becomes each manufacturing process shown in FIG. 7. Furthermore, information processing is assumed to be performed according to the operational flow shown in FIG.

The data storage unit 16 stores past performance data not only for the oil phase creation process but also for each process in FIG. Further, although the controller 21 of the production process equipment 20 in FIG. 1 may be a PLC, DCS, microcomputer, etc., a PLC was used in this embodiment. The data acquired from the sensor 23 is stored in the data storage section 16 via the controller 21, but may be stored directly in the data storage section 16 without going through the controller 21.

Next, the quality prediction unit 12 performs predictive calculation using the collected data. If there is a strong correlation between the final quality and the collected data and there is no interaction with other factors, univariate analysis would be possible, but a multivariate analysis method was adopted for prediction during the oil phase creation process. In general, various methods can be used for quality prediction, including linear regression and nonlinear regression, which are classified as machine learning that can handle more complex and nonlinear processes. We adopted predictive calculation using random forest as the learning method. In addition, all of the collected data mentioned above can be used as data for building a random forest model, but for time series data, it is necessary to convert it to feature values, reduce dimensionality by principal component analysis, etc., and analyze prediction errors and predictions from machine learning. It is also effective to improve prediction accuracy by adopting a method of selecting variables based on information such as correlation coefficients. In this example, a method was adopted in which variables were selected using indicators based on feature value conversion and prediction errors, a model was created from past performance data using random forest, and product quality prediction calculations were performed from planned data.

Based on the quality prediction results, optimal control conditions are calculated for the scheduled data. The operation amount calculation unit 13 calculates control conditions based on the calculation results of the quality prediction unit 12 so that the quality is as desired. The operation amount calculation unit 13 calculates physical phenomena that occur in the production process, the relationship between the process conditions and final quality confirmed from past performance data, and the process conditions and final quality confirmed from the model constructed by the quality prediction unit 12. Optimal control conditions were calculated using a rule-based algorithm based on quality relationships. Although it is possible to calculate optimal conditions using machine learning, it is preferable to select them according to the characteristics of the application target, taking into consideration interactions between variables, presence or absence of constraint conditions, etc. When calculating the optimal control conditions, the items to be controlled and the amount of change are determined so that all final quality items satisfy the target range.

Algorithm models for quality prediction and operation amount calculation can be either fixed or updateable, but in this example, seasonal fluctuations, changes due to aging of processing machines, etc. are incorporated into the model to create a flexible system that can respond to changes. Therefore, we adopted a sequential update type. In the case of sequential updates, calculation results that were not expected at the time of development may be calculated depending on the model built when new data is imported, and the values used for prediction may be within the range of machine learning, regardless of fixed updates. There is a possibility that incorrect predictions and control calculation results may be output, such as when a local solution occurs when the value is outside the range or when the variables are complex for the physical property value to be predicted. In such cases, production is performed based on incorrect calculation results, making it impossible to create an oil phase with stable quality.

Therefore, the problem of the model update type was solved by detecting an abnormality in the abnormality detection unit 14 under optimal control conditions. In addition, by combining a model-updating prediction system and anomaly detection, we can solve the other problem of machine learning methods, which is that the estimation accuracy of areas not found in the training data is significantly reduced. By performing detection, it is possible to detect when the estimation accuracy is decreasing and switch to manual control, thereby preventing operations under incorrect conditions and using quality prediction to produce products with stable quality. can be produced.

With the configuration of the information processing system 10, areas in the learning data of machine learning are automatically controlled, and if the estimation accuracy decreases in an unlearned area, it is detected by abnormality detection and the control is switched to human control. By incorporating human control results into the learning data by updating the model, we have created a system configuration that can expand the learning range and expand the range that can be handled by the automatic control system.

Various methods can be used to calculate the anomaly detection model, including simple upper and lower limit settings, QC methods such as X-bar control charts, classification methods using machine learning such as Hotelling, one-class SVM, and the neighborhood method. In this example, there are many target items, and the task of setting and maintaining thresholds for each item becomes complicated. We adopted two methods: determining the degree of deviation between the predicted value and the actual result using the average) model, and determining the likelihood of the past actual density distribution using kernel density estimation. Control limits can be set arbitrarily depending on the number of items to be managed and control change frequency, but in this example, in the ARIMA model, if the deviation between the predicted value and the actual value exceeds 3 based on the root mean square error standard, or the kernel density A case where the estimated interval probability was not included in the top 99.7% of the cumulative probability density was considered abnormal. In addition, in this example, since direct interactions between items were not included, calculations and judgments were made for each item, but if necessary, principal component analysis, which is a multivariate analysis method, or an anomaly detection method using an autoencoder may be applied. By using this, it is also possible to detect abnormalities taking into account the relationships between variables.

As a result of abnormality detection, if the optimal control conditions for the manipulated variable calculated by the manipulated variable calculation unit 13 are within the normal range, the optimal control conditions are reflected in the oil phase creation process to bring the final quality of the toner particle diameter into the target quality range. Can be adjusted. If the optimal control conditions are within the normal range, the control conditions are stored in the calculation result storage unit 17 and reflected in the oil phase creation process via the process operation instruction unit 15.

Although the process operation instruction section 15 and the calculation result storage section 17 can be constructed using means such as cloud computing and a network system, in this embodiment, they were constructed in an on-premises environment. Further, in this embodiment, the optimal control conditions are reflected through the calculation result storage section 17, but it is also possible to directly reflect them from the manipulated variable calculation section 13 or the abnormality detection section 14 to the process operation instruction section 15. be.

If an abnormality is detected in the manipulated variable optimum control conditions calculated by the manipulated variable calculation unit 13, the abnormality is not reflected in the process.

[Example 2]
In Example 2, similarly to Example 1, the information processing system 10 shown in FIGS. 1 to 3 is used to automatically adjust the prescription in the oil phase process and stabilize the quality.

In the second embodiment, unlike the first embodiment, information processing is performed according to the operation flow shown in FIG.

First, using the information processing system 10 shown in FIGS. 1 to 3, the data aggregation unit 11 uses the previous equipment settings, raw material information, raw material prescription, production results, and product inspection results stored in the data storage unit 16. etc. (hereinafter referred to as past performance data), and information held by the process operation instruction unit 15 such as raw material property information, raw material prescription, equipment setting values, etc. (hereinafter referred to as planned data) for the oil phase creation process to be implemented from now on. .

Next, the abnormality detection unit 14 performs abnormality detection on the collected past performance data and scheduled data. By performing abnormality detection before quality prediction, it is possible to detect data input errors and failures in sensors and production equipment in advance. Furthermore, by determining in advance whether the data is within the learning range, it is possible to determine whether accurate prediction is possible. If an abnormality is determined, the operator is notified of the details, making it possible to correct input errors and respond to failures, enabling production with stable quality.

Next, quality prediction is performed using past performance data and scheduled data that have been confirmed to be free of abnormalities. Then, the operation amount calculation unit 13 calculates optimal control conditions based on the quality prediction results. The abnormality detection unit 14 performs abnormality detection again with respect to the calculated optimal conditions. Even if there is no abnormality in the input data, local solutions may sometimes occur due to the complex variables associated with the physical property values that you want to predict. Therefore, abnormality detection can prevent operations under incorrect conditions and ensure stable quality. production becomes possible.

[Example 3]
In Example 3, similarly to Example 2, the information processing system 10 shown in FIGS. 1 to 3 is used to automatically adjust the prescription in the oil phase process and stabilize the quality.

In the third embodiment, unlike the second embodiment, information processing is performed according to the operation flow shown in FIG. If the abnormality detection result for the optimal control conditions is abnormal, the operator determines whether or not to apply the optimal control conditions. If necessary, by adding constraints to the calculation conditions and performing quality prediction and operation amount calculation again, more optimal control conditions can be obtained and production with stable quality becomes possible.

As a result of utilizing this information processing system 10 in the toner manufacturing process, it was confirmed that the process capability index (C_pk) improved and the man-hours for toner quality managers were reduced by 15%.

As described above, in one embodiment of the present invention, the quality of the product to be produced is predicted and the optimal control conditions for the production process are calculated from at least one of data related to raw materials and data related to processing conditions, and the calculation results are A production system that detects abnormalities can be provided.

Each function of the embodiments described above can be realized by one or more processing circuits. Here, the term "processing circuit" as used herein refers to a processor programmed to execute each function by software, such as a processor implemented by an electronic circuit, or a processor designed to execute each function explained above. This includes devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules.

Note that the present invention is not limited to the configurations shown here, such as combinations of other elements with the configurations listed in the above embodiments. These points can be modified without departing from the spirit of the present invention, and can be appropriately determined depending on the application thereof.

10 Information processing system 20 Production process equipment 21 Controller 22 Production equipment 23 Sensor 30 Network 11 Data aggregation unit 12 Quality prediction unit 13 Operation amount calculation unit 14 Abnormality detection unit 15 Process operation instruction unit 16 Data storage unit 17 Calculation result storage unit

JP 2017-151962 Publication

Claims (9)

In the production of a product that includes multiple production processes, a data aggregation unit that aggregates data regarding the production of the product in each production process ;
a quality prediction unit that uses machine learning to predict the quality of the object in each of the production processes based on the data;
an operation amount calculation unit that calculates an operation amount related to each of the production steps necessary to produce a product of a predetermined quality based on the predicted quality;
an abnormality detection unit that uses machine learning to determine whether or not there is an abnormality in the manipulated variable;
An information processing system comprising: a process operation instruction unit that outputs a manipulated variable that is free of abnormalities; The data related to the production of the product includes at least one of data related to raw materials and data related to processing conditions,
The information processing system according to claim 1, wherein the manipulated variable includes a manipulated variable of at least one of raw materials and processing conditions. The information processing system according to claim 1 or 2, wherein the abnormality detection unit uses machine learning to determine whether there is an abnormality in the data aggregated by the data aggregation unit. The process operation instruction unit is
Receives an instruction as to whether or not to output the manipulated variable with an abnormality,
The information processing system according to any one of claims 1 to 3, wherein the information processing system outputs the abnormal operation amount when the output instruction is received. The process operation instruction unit receives an instruction to recalculate a manipulated variable with an abnormality,
The information processing system according to any one of claims 1 to 3, wherein the operation amount calculation unit recalculates the operation amount. Displaying whether or not there is an abnormality in the manipulated variable;
The information processing system according to any one of claims 1 to 5, wherein the information processing system displays a screen having: a display for instructing whether or not to output the manipulated variable having the abnormality. The information processing system according to any one of claims 1 to 6, wherein the object is a toner. A method performed by an information processing system, the method comprising:
In the production of a product that includes multiple production processes, a step of aggregating data regarding the production of the product in each production process ;
predicting the quality of the object in each production process based on the data using machine learning;
calculating the amount of operation necessary for each production process to produce a product of a predetermined quality based on the predicted quality;
a step of determining whether or not there is an abnormality in the manipulated variable using machine learning;
A method including the steps of outputting a manipulated variable with no abnormality. computer
In the production of a product that includes multiple production processes, a data aggregation unit that aggregates data regarding the production of the product in each production process ;
a quality prediction unit that uses machine learning to predict the quality of the object in each production process based on the data;
an operation amount calculation unit that calculates an operation amount related to each of the production steps necessary to produce a product of a predetermined quality based on the predicted quality;
an abnormality detection unit that uses machine learning to determine whether or not there is an abnormality in the manipulated variable;
A program that functions as a process operation instruction unit that outputs manipulated variables that are free of abnormalities.

Images