Learning management program, learning management device, and learning system

Abstract

A learning management program according to the present invention causes a computer to function as: a first preprocessing unit (101) for associating first production site information with productivity information, the first production site information being acquired by executing preprocessing on production site information collected as training data, thereby creating first training data used for generating a trained model for diagnosing productivity information from production site information; a second preprocessing unit (102) for associating second production site information with productivity information, the second production site information being acquired by executing preprocessing on production site information collected as inference data to be subjected to diagnosis by using the trained model, thereby creating second training data; a comparison unit (103) for determining whether or not there is a difference between a first tendency of the first production site information and a second tendency of the second production site information; and an instruction unit (104) for instructing that a trained model be regenerated by using the first training data and the second training data when it is determined that there is a difference between the first tendency and the second tendency.

Description

Learning management program, learning management device, and learning system

[0001] The present disclosure relates to a learning management program, a learning management device, and a learning system.

[0002] There is known a technology that collects data indicating the operating status of equipment at a production site from sensors installed at the production site, and uses the collected data to generate a trained model that predicts equipment abnormalities, diagnoses the quality of products, etc. For example, Patent Literature 1 discloses a technology that uses process signals acquired from sensors installed in a plant to generate a trained model that predicts future trends in the process signals, and detects signs of abnormalities using the output of the trained model.

[0003] Japanese Patent Application Laid-Open No. 2021-47523

[0004] However, technologies such as those described in Patent Document 1 do not take into account changes over time that occur in production sites when generating trained models. Generally, in production sites, various changes over time occur due to changes in equipment operation methods, deterioration of equipment components, changes in the environment, and so on. For this reason, the trends of the data used when generating the trained model before the changes over time may differ from the trends of the data after the changes over time. Therefore, applying the data after the changes over time to the trained model may result in an inaccurate output. For example, when data after the changes over time occur is applied to a trained model that detects signs of abnormality, the trained model may not detect the signs of abnormality even if the data after the changes over time indicate a sign of abnormality. Alternatively, the trained model may detect a sign of abnormality even if the data indicates normality. To prevent such situations, the trained model must be retrained when changes over time occur.

[0005] The present disclosure has been made in consideration of the above circumstances, and aims to provide a learning management program, a learning management device, and a learning system that are capable of performing re-learning in a timely manner based on the occurrence of changes over time in the production site.

[0006] In order to achieve the above object, a learning management program according to the present disclosure includes a computer including: a first preprocessing means for acquiring first production site information by performing preprocessing on production site information indicating the operating status of equipment at a production site, collected in order to create learning data, and associating the first production site information with productivity information indicating productivity at the production site of the first production site information, thereby creating first learning data to be used in generating a trained model for inferring productivity information indicating productivity at the production site of the production site information from the production site information; a second preprocessing means for acquiring second production site information by performing the preprocessing on production site information collected in order to obtain inference data for making inference using the trained model, and associating the second production site information with productivity information indicating productivity at the production site of the second production site information, thereby creating second learning data; a comparison means for comparing a first trend that is a trend of the first production site information with a second trend that is a trend of the second production site information, and determining whether or not there is a difference between the first trend and the second trend; and an instruction means for instructing regeneration of the trained model using the first training data and the second training data when the comparison means determines that there is a difference between the first trend and the second trend.

[0007] According to the present disclosure, it is possible to provide a learning management program, a learning management device, and a learning system that are capable of performing re-learning in a timely manner based on the occurrence of changes over time in the production site.

[0008] FIG. 1 is a diagram showing the functional configuration of a learning system according to an embodiment; FIG. 2 is a block diagram showing an example of the hardware configuration of an information processing device according to an embodiment; FIG. 3 is a schematic diagram for explaining the correspondence between time-series data of extrusion speed, flag data, and time-series data of quality information according to an embodiment; FIG. 4 is a diagram showing an example of a neural network according to an embodiment; FIG. 5 is a diagram showing an example of the relationship between the average value of time-series data of extrusion speed, the average value of time-series data of resin viscosity, and quality information according to an embodiment;

[0009] 1, a learning system 1 according to an embodiment includes a learning management device 100, a collection device 200, a learning device 300, and an inference device 400. The learning management device 100 is communicatively connected to the collection device 200, the learning device 300, and the inference device 400 via a network cable (not shown) or wirelessly. The learning device 300 is communicatively connected to the inference device 400 via a network cable (not shown) or wirelessly.

[0010] The learning management device 100 is a device that performs preprocessing on the production site information collected by the collection device 200, and uses the preprocessed production site information to create learning data that the learning device 300 uses to generate a trained model.

[0011] Production site information is information that indicates the operating status of equipment installed at the production site. Equipment at the production site is, for example, factory automation (FA) equipment connected to a programmable logic controller (PLC). Production site information includes information acquired from equipment installed at the production site and information that can be acquired from external devices of the equipment, such as cameras and sensors installed at the production site. Production site information is, for example, information that indicates the status of processing, transportation, etc. by equipment, and information about materials used in production.

[0012] The learning management device 100 also preprocesses the production site information collected by the collection device 200 in real time, and uses the preprocessed production site information to obtain inference data that will be used by the inference device 400 to perform inference. The learning management device 100 also compares past and current trends in the production site information collected by the collection device 200, and instructs the learning device 300 to re-learn.

[0013] The collection device 200 is a device that collects time-series data of production site information that indicates the operating status of equipment at the production site. For example, the collection device 200 is communicably connected to the equipment at the production site and external devices such as sensors and cameras.

[0014] The learning device 300 is a device that uses the learning data created by the learning management device 100 to generate a trained model that infers productivity information from production site information.

[0015] Productivity information is information indicating productivity at a production site. Productivity at a production site refers, for example, to the operating efficiency of equipment installed at the production site and the quality accuracy of products produced at the production site. Productivity information is provided by a user of the learning system 1 or acquired by the collection device 200 from equipment installed at the production site or an external device. Productivity information is, for example, information indicating the presence or absence of abnormalities in equipment installed at the production site, or information indicating the quality of products produced by the equipment. Therefore, the trained model generated by the learning device 300 is a model that infers productivity information from production site information, and can also be said to be a model that diagnoses the productivity of equipment based on production site information. Note that users of the learning system 1 include users of the learning management device 100, the collection device 200, the learning device 300, and the inference device 400.

[0016] The inference device 400 is a device that infers or diagnoses productivity information of a production site by applying the inference data acquired by the learning management device 100 to the trained model generated by the learning device 300.

[0017] FIG. 2 shows an example of the hardware configuration of an information processing device 10 in which the learning management device 100, the collection device 200, the learning device 300, and the inference device 400 are realized.

[0018] The information processing device 10 has a processor 11 that executes various processes, a main memory unit 12 used as a work area for the processor 11, an auxiliary memory unit 13 that stores various data used in the processes of the processor 11, a communication unit 14 for communicating with external devices, an input unit 15 that acquires input information, and an output unit 16 that presents various information. The main memory unit 12, the auxiliary memory unit 13, the communication unit 14, the input unit 15, and the output unit 16 are all connected to the processor 11 via a bus 17.

[0019] The processor 11 includes a CPU (Central Processing Unit). The processor 11 executes programs stored in the auxiliary storage unit 13 to realize various functions of the information processing device 10.

[0020] The main memory unit 12 includes a RAM (Random Access Memory). Programs are loaded into the main memory unit 12 from the auxiliary memory unit 13. The main memory unit 12 is used as a working area for the processor 11.

[0021] The auxiliary storage unit 13 includes a non-volatile memory such as an EEPROM (Electrically Erasable Programmable Read-Only Memory). In addition to programs, the auxiliary storage unit 13 stores various data used in the processing of the processor 11. In accordance with instructions from the processor 11, the auxiliary storage unit 13 supplies the processor 11 with data used by the processor 11 and stores the data supplied from the processor 11.

[0022] The communication unit 14 includes a network interface circuit for communicating with an external device. The communication unit 14 receives a signal from the external device and outputs data indicated by the signal to the processor 11. The communication unit 14 also transmits a signal indicating the data output from the processor 11 to the external device.

[0023] The input unit 15 includes input devices such as input keys, a pointing device, etc. The input unit 15 acquires information input by the user of the information processing device 10 and notifies the processor 11 of the acquired information.

[0024] The output unit 16 includes output devices such as an LCD (Liquid Crystal Display) and a speaker. The output unit 16 may be configured as a touch screen integrally formed with a pointing device constituting the input unit 15. The output unit 16 presents various information to the user in accordance with instructions from the processor 11.

[0025] Next, the functions of each device included in the learning system 1 of FIG. 1 will be described.

[0026] The learning management device 100 functionally comprises a first preprocessing unit 101 that performs preprocessing on production site information to create learning data, a second preprocessing unit 102 that performs preprocessing on production site information to obtain inference data, a comparison unit 103 that compares the production site information preprocessed by the first preprocessing unit 101 with the production site information preprocessed by the second preprocessing unit 102, an instruction unit 104 that instructs the learning device 300 to regenerate the trained model, and a first notification unit 105 that notifies the result of the comparison by the comparison unit 103.

[0027] Functionally, the collection device 200 comprises a first collection unit 201 that collects production site information to create learning data, and a second collection unit 202 that collects production site information to obtain inference data.

[0028] Functionally, the learning device 300 comprises a learning data storage unit 301 that stores learning data created by the learning management device 100, a model generation unit 302 that generates a trained model using the learning data, and a trained model storage unit 303 that stores the trained model.

[0029] Functionally, the inference device 400 comprises an inference data storage unit 401 that stores inference data acquired by the learning management device 100, an inference unit 402 that performs inference using the inference data and the trained model, and a second notification unit 403 that notifies the inference result.

[0030] Below, the functions of the learning system 1 are explained using the example of learning the relationship between the production site information of the injection molding machine and the productivity information of the injection molding machine, and inferring the productivity information of the injection molding machine from the production site information of the injection molding machine.

[0031] First, the functions of the learning system 1 during learning will be described.

[0032] The first collection unit 201 of the collection device 200 collects production site information indicating the operating status of equipment at the production site in order to create learning data. The first collection unit 201 is realized by the processor 11 and the communication unit 14. The first collection unit 201 is an example of a first collection means.

[0033] For example, the first collection unit 201 collects time-series data of production site information from an injection molding machine to create first learning data 3011, which will be described later. The production site information collected from the injection molding machine includes, for example, the extrusion speed, set torque, resin viscosity, resin temperature, and flag data. The flag data is data indicating periods during which an operation by a device is being performed and periods during which no operation is being performed. The flag data is, for example, time-series data indicating periods during which the injection molding machine is performing an extrusion operation and periods during which no extrusion operation is being performed. In the following, it is assumed that the first collection unit 201 collects time-series data of the extrusion speed, time-series data of the resin viscosity, and flag data from the injection molding machine.

[0034] The first collection unit 201 also collects productivity information indicating productivity at the production site from devices at the production site.

[0035] Productivity information collected from equipment at the production site is, for example, quality information collected from injection-molded products indicating whether the product is good or defective. The first collection unit 201 collects time-series data of quality information from the injection molding machine. The quality information is determined to be good if the weight of the injection-molded product exceeds a predetermined value, and whether the product is good or defective is automatically determined based on the weight measured by a sensor installed in the injection molding machine. Therefore, the quality information is collected for each operation, i.e., for each extrusion operation of the injection molding machine.

[0036] FIG. 3 is a schematic diagram illustrating the correspondence between the time-series data of extrusion speed, flag data, and time-series data of quality information collected by the first collection unit 201. The broken line waveform shown by the solid line in FIG. 3 represents the time-series data of extrusion speed, and the rectangular waveform shown by the dotted line in FIG. 3 represents the flag data. The ON period of the flag data is the period during which the injection molding machine is performing the extrusion operation, and the OFF period of the flag data is the period during which the injection molding machine is not performing the extrusion operation. Furthermore, quality information is collected in association with the time-series data of extrusion speed during the ON period of the flag data. Therefore, FIG. 3 shows that quality information of "defective product," "good product," "good product," and "defective product" has been collected for the time-series data of extrusion speed in the rectangle 501, rectangle 502, rectangle 503, and rectangle 504, respectively.

[0037] The first pre-processing unit 101 of the learning management device 100 acquires first production site information by performing pre-processing on the production site information collected by the first collection unit 201 of the collection device 200. Then, the first pre-processing unit 101 creates first learning data 3011 by associating the first production site information with productivity information indicating productivity at the production site of the first production site information. The first pre-processing unit 101 is realized by the processor 11 and the communication unit 14. The first pre-processing unit 101 is an example of a first pre-processing means.

[0038] The preprocessing includes extracting the production site information for each operation performed by the equipment. The first preprocessing unit 101 acquires first production site information based on the extracted production site information for each operation and uses the first production site information as training data. By performing this preprocessing, it is possible to exclude production site information for a state in which no operation is being performed from the training data used for the trained model. The preprocessing is, for example, extracting the production site information for each operation based on a change in the value of flag data indicating periods in which an operation is being performed by the equipment and periods in which no operation is being performed.

[0039] For example, the first pre-processing unit 101 extracts the time-series data of the extrusion speed for each operation, i.e., for each period during which the injection molding machine is performing the extrusion operation, by setting the start point of the time when the flag data switches from OFF to ON in the time-series data of the extrusion speed and the end point of the time when the flag data switches from ON to OFF. Taking the schematic diagram of Figure 3 as an example, the first pre-processing unit 101 extracts the time-series data of the extrusion speed for the rectangular portions 501, 502, 503, and 504. Similarly, the first pre-processing unit 101 extracts the time-series data of the resin viscosity for each operation, i.e., for each period during which the injection molding machine is performing the extrusion operation, by setting the start point of the time when the flag data switches from OFF to ON and the end point of the time when the flag data switches from ON to OFF in the time-series data of the resin viscosity.

[0040] The extraction process included in the preprocessing may be a process of extracting production site information for each operation based on a change in the value of the production site information relative to a predetermined value, a change in the file in which the production site information is stored, or a predetermined interval. The predetermined value and the predetermined interval are set arbitrarily by the user of the learning system 1.

[0041] For example, the first pre-processing unit 101 extracts time-series data of the extrusion speed for each operation, starting from the point when the extrusion speed value becomes equal to or greater than a predetermined value and ending from the point when the extrusion speed value falls below the predetermined value. For example, if the time-series data of one extrusion operation of an injection molding machine is configured to be stored in a single file, the first pre-processing unit 101 extracts the time-series data of the extrusion operation stored in the single file as time-series data of the extrusion speed for each operation. For example, the first pre-processing unit 101 extracts time-series data of the extrusion speed for a predetermined second interval (e.g., 1 second) after a predetermined first interval (e.g., 0.5 seconds) from the start of recording the time-series data of the extrusion speed, and then extracts time-series data of the extrusion speed for the second interval after another first interval. This process is repeated to extract time-series data of the extrusion speed for each operation.

[0042] Next, the first pre-processing unit 101 calculates statistics for the production site information for each operation, and acquires the calculated statistics as first production site information.

[0043] The statistical quantity may be, for example, the mean value, standard deviation, minimum value, maximum value, median value, mode value, a value at a predetermined percentile, the mean value ±3×standard deviation, or the interquartile range.

[0044] For example, the first pre-processing unit 101 calculates the average value for each of the time series data of the extrusion speed for each operation and the time series data of the resin viscosity for each operation, and acquires the calculated average value as the first production site information.

[0045] Then, the first pre-processing unit 101 creates the first learning data 3011 by associating the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation with the time series data of the quality information for each operation.

[0046] The first preprocessing unit 101 stores the first training data 3011 in the training data storage unit 301 of the training device 300. The training data storage unit 301 is realized by the auxiliary storage unit 13.

[0047] For example, the first pre-processing unit 101 stores in the learning data storage unit 301 first learning data 3011, which corresponds the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation to the time series data of the quality information for each operation.

[0048] The model generation unit 302 of the learning device 300 generates a trained model for inferring productivity information indicating productivity at a production site of the production site information from the production site information, using the first training data 3011. The model generation unit 302 is realized by the processor 11 and the communication unit 14. The model generation unit 302 is an example of a model generation means.

[0049] For example, the model generation unit 302 generates a trained model for diagnosing the quality of an injection-molded product from the average value of time-series data of extrusion speed for one operation of an injection molding machine and the average value of time-series data of resin viscosity for one operation. The model generation unit 302 learns the quality of an injection-molded product based on first training data 3011 created by combining the average value of the time-series data of extrusion speed for each operation and the average value of the time-series data of resin viscosity for each operation with the time-series data of quality information for each operation. Therefore, the model generation unit 302 generates a trained model for diagnosing the quality of an injection-molded product from the extrusion speed and resin viscosity when the injection-molded product is produced.

[0050] The learning algorithm used by the model generation unit 302 may be a known algorithm such as supervised learning, unsupervised learning, reinforcement learning, etc. As an example, a case where a neural network is applied will be described.

[0051] The model generation unit 302 learns productivity information, for example, by supervised learning according to a neural network model. Here, supervised learning refers to a technique in which pairs of input and result (label) data are provided to the learning device 300, and the learning device 300 learns the features of the learning data and infers the result from the input.

[0052] A neural network is composed of an input layer including a plurality of neurons, an intermediate layer (hidden layer) including a plurality of neurons, and an output layer including a plurality of neurons. The intermediate layer may be one layer, or two or more layers.

[0053] For example, in a three-layer neural network as shown in Figure 4, when multiple inputs are input to the input layer (X1-X3), the values ​​are multiplied by weight W1 (w11-w16) and input to the intermediate layer (Y1-Y2), and the result is further multiplied by weight W2 (w21-w26) and output from the output layer (Z1-Z3). This output result changes depending on the values ​​of weights W1 and W2.

[0054] In the present application, the neural network learns the productivity information by so-called supervised learning in accordance with the first learning data 3011, which is a combination of the first production site information and the productivity information, created by the first pre-processing unit 101.

[0055] That is, the neural network learns by inputting the first production site information to the input layer and adjusting the weights W1 and W2 so that the result output from the output layer approaches the productivity information.

[0056] The model generation unit 302 generates and outputs a trained model by performing the above-described learning.

[0057] FIG. 5 shows an example of the relationship between the quality information and the average value of the time-series data of the extrusion speed and the average value of the time-series data of the resin viscosity, which are the first training data 3011. As shown in legend 601 in FIG. 5 , the open circles in FIG. 5 represent the data of the average value of the extrusion speed and the average value of the resin viscosity at which the injection-molded product was determined to be good, while the filled circles in FIG. 5 represent the data of the average value of the extrusion speed and the average value of the resin viscosity at which the injection-molded product was determined to be defective. In FIG. 5 , if the average value of the extrusion speed falls within range 602, the injection-molded product is determined to be good regardless of the average value of the resin viscosity. That is, FIG. 5 shows that, in the first training data 3011, the extrusion speed affects the quality of the injection-molded product, but the resin viscosity does not. Therefore, when the model generation unit 302 generates a trained model using the first training data 3011, a trained model is generated in which the value of the extrusion speed contributes to quality diagnosis, but the value of the resin viscosity does not contribute to quality diagnosis.

[0058] The model generation unit 302 stores the trained model generated using the first training data 3011 in the trained model storage unit 303. The trained model storage unit 303 is realized by the auxiliary storage unit 13.

[0059] For example, the model generation unit 302 stores in the trained model storage unit 303 a trained model in which the value of the extrusion speed contributes to quality diagnosis and the value of the resin viscosity does not contribute to quality diagnosis.

[0060] Next, the function of the learning system 1 during inference will be described.

[0061] The second collection unit 202 of the collection device 200 collects production site information to obtain inference data for performing inference using the trained model. The second collection unit 202 is realized by the processor 11 and the communication unit 14. The second collection unit 202 is an example of a second collection means.

[0062] For example, the second collection unit 202 collects production site information including time-series data of extrusion speed, time-series data of resin viscosity, and flag data from the injection molding machine to obtain inference data to be applied to the trained model stored in the trained model storage unit 303. The flag data, like the flag data collected by the first collection unit 201, indicates periods when the injection molding machine is performing an extrusion operation and periods when it is not performing an extrusion operation.

[0063] The second collection unit 202 also collects productivity information indicating productivity at the production site from the devices at the production site.

[0064] For example, similar to the first collection unit 201, the second collection unit 202 collects time-series data of quality information indicating whether an injection-molded product is a good product or a defective product from an injection molding machine.

[0065] The second pre-processing unit 102 of the learning management device 100 acquires second production site information as data for inference by performing pre-processing on the production site information collected by the second collection unit 202 of the collection device 200. The second pre-processing unit 102 is realized by the processor 11 and the communication unit 14. The second pre-processing unit 102 is an example of second pre-processing means.

[0066] The preprocessing performed by the second preprocessing unit 102 is the same as the preprocessing performed by the first preprocessing unit 101. That is, the preprocessing performed by the second preprocessing unit 102 includes a process of extracting the production site information for each operation performed by the equipment. The second preprocessing unit 102 calculates statistics for the production site information for each operation and acquires the calculated statistics as second production site information. The statistics calculated by the second preprocessing unit 102 are the same as the statistics calculated by the first preprocessing unit 101.

[0067] For example, when the second collection unit 202 collects the time-series data of the extrusion speed, the time-series data of the resin viscosity, and the flag data, the second pre-processing unit 102 extracts the time-series data of the extrusion speed and the time-series data of the resin viscosity for each operation in real time based on the flag data. Then, the second pre-processing unit 102 calculates the average value of each of the time-series data of the extrusion speed and the time-series data of the resin viscosity for each operation.

[0068] The second pre-processing unit 102 stores the second production site information, which has been pre-processed, as inference data in the inference data storage unit 401 of the inference device 400. The inference data storage unit 401 is realized by the auxiliary memory unit 13.

[0069] For example, the second pre-processing unit 102 stores the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation in the inference data storage unit 401 in real time as inference data.

[0070] The inference unit 402 of the inference device 400 infers productivity information from the second production site information using the trained model. The inference unit 402 is realized by the processor 11 and the communication unit 14. The inference unit 402 is an example of an inference means.

[0071] For example, the inference unit 402 acquires quality information for each operation by inputting, in real time, the average value of the time-series data of the extrusion speed for each operation and the average value of the time-series data of the resin viscosity for each operation stored in the inference data storage unit 401 into the trained model stored in the trained model storage unit 303 of the learning device 300. For example, if the average value of the time-series data of the extrusion speed for a certain operation is a value included in the range 601 in Figure 6, the inference unit 402 acquires quality information indicating that the injection-molded product produced by that operation is a "good product."

[0072] The second notification unit 403 notifies the productivity information inferred from the second production site information using the trained model. That is, the second notification unit 403 notifies the productivity information inferred by the inference unit 402. The second notification unit 403 is realized by the processor 11, the communication unit 14, and the output unit 16. The second notification unit 403 is an example of a second notification means.

[0073] For example, each time the inference unit 402 acquires quality information for each operation of the injection molding machine, the second notification unit 403 displays the acquired quality information on the screen of the inference device 400 or on the screen of another information processing device used by the user of the learning system 1.

[0074] The second preprocessing unit 102 also creates second training data 3012 by associating the second production site information with productivity information indicating productivity at the production site of the second production site information. The second preprocessing unit 102 then stores the created second training data 3012 as training data in the inference data storage unit 401. The second training data 3012 is transmitted to the training data storage unit 301 at any timing set by the user of the training system 1.

[0075] For example, the second pre-processing unit 102 creates second learning data 3012 by associating the average value of the time-series data of the extrusion speed for each operation and the average value of the time-series data of the resin viscosity for each operation with the time-series data of the quality information for each operation. Then, the second pre-processing unit 102 stores the second learning data 3012 in the inference data storage unit 401. For example, as described below, when the instruction unit 104 of the learning management device 100 instructs the model generation unit 302 of the learning device 300 to regenerate a trained model, the second learning data 3012 stored in the inference data storage unit 401 is moved to the learning data storage unit 301 of the learning device 300 by the instruction unit 104 or the model generation unit 302.

[0076] Next, the function of the learning system 1 when re-learning a trained model will be described.

[0077] The comparison unit 103 of the learning management device 100 compares a first trend, which is a trend of the first production site information, with a second trend, which is a trend of the second production site information, and determines whether there is a difference between the first trend and the second trend. The comparison unit 103 is realized by the processor 11 and the communication unit 14. The comparison unit 103 is an example of a comparison means.

[0078] For example, the comparison unit 103 determines whether there is a difference between the first trend and the second trend based on a timing preset by the user of the learning system 1 (for example, 10:00 every day) or based on an operation by the user.

[0079] The first trend is the trend of the first production site information, and is the trend of the first training data 3011 used when generating the trained model stored in the trained model storage unit 303, i.e., the trend of past production site information. The second trend is the trend of the second production site information, and is the trend of the second training data 3012 obtained after generating the trained model stored in the trained model storage unit 303, and is the trend of the current production site information.

[0080] For example, the first trend is represented by the median of the first production site information, and the second trend is represented by the median of the second production site information. If the difference between the median of the first production site information and the median of the second production site information is greater than a value set by the user of the learning system 1, the comparison unit 103 determines that there is a difference between the first trend and the second trend.

[0081] FIG. 6 shows an example of the relationship between the quality information and the average value of the time-series data of extrusion speed and the average value of the time-series data of resin viscosity, which are the first learning data 3011 and the second learning data 3012. "ME1" and "ME2" in FIG. 6 respectively represent the median of the group of average values ​​of the time-series data of extrusion speed, which is the first production site information, and the median of the group of average values ​​of the time-series data of extrusion speed, which is the second production site information. Also, "MR1" and "MR2" in FIG. 6 respectively represent the median of the group of average values ​​of the time-series data of resin viscosity, which is the first production site information, and the median of the group of average values ​​of the time-series data of resin viscosity, which is the second production site information. The difference between the median "ME1" and the median "ME2" is assumed to be smaller than the value set by the user, and the difference between the median "MR1" and the median "MR2" is assumed to be larger than the value set by the user. In this case, the comparison unit 103 determines that there is no difference between the first and second trends in the time series data of the extrusion speed, but that there is a difference between the first and second trends in the time series data of the resin viscosity.

[0082] When the comparison unit 103 determines that there is a difference between the first trend and the second trend, the instruction unit 104 instructs the model generation unit 302 of the learning device 300 to regenerate a trained model using the first training data 3011 and the second training data 3012. In other words, when it is determined that there is a difference between the trend of the past production site information and the trend of the current production site information, the instruction unit 104 instructs to re-learn productivity information using the past and current production site information. The instruction unit 104 is realized by the processor 11 and the communication unit 14. The instruction unit 104 is an example of instruction means.

[0083] For example, when the comparison unit 103 determines that there is a difference between the first trend and the second trend of the resin viscosity time series data, the instruction unit 104 instructs the model generation unit 302 to regenerate the trained model using the first training data 3011 and the second training data 3012.

[0084] The first notification unit 105 notifies the result of the comparison by the comparison unit 103. Specifically, when the comparison unit 103 determines that there is a difference between the first trend and the second trend, the first notification unit 105 notifies that there is a difference. The first notification unit 105 is realized by the processor 11, the communication unit 14, and the output unit 16. The first notification unit 105 is an example of a first notification means.

[0085] For example, when the comparison unit 103 determines that there is a difference between the first trend and the second trend of the resin viscosity time-series data, the first notification unit 105 displays a message that there is a difference on the screen of the learning management device 100 or on the screen of another information processing device used by the user of the learning system 1. Note that when the comparison unit 103 determines that there is no difference between the first trend and the second trend, the first notification unit 105 may display a message that there is no difference on the screen of the learning management device 100 or on the screen of another information processing device used by the user of the learning system 1.

[0086] When the model generation unit 302 receives an instruction to regenerate a trained model from the instruction unit 104, it generates a trained model using the first training data 3011 and the second training data 3012 stored in the training data storage unit 301.

[0087] 6 , an injection-molded product is determined to be a good product when the average value of the extrusion speed falls within a range 602 and the average value of the resin viscosity falls within a range 603. That is, from Fig. 6 , it can be seen that both the extrusion speed and the resin viscosity affect the quality of the injection-molded product in the first training data 3011 and the second training data 3012. Therefore, when the model generation unit 302 regenerates the trained model using the first training data 3011 and the second training data 3012, a trained model is generated in which the values ​​of the extrusion speed and the resin viscosity contribute to quality diagnosis.

[0088] In addition, the comparison unit 103 determines whether the second production site information deviates from the frequency range of the first production site information, and if the comparison unit 103 determines that the second production site information deviates, the second notification unit 403 notifies that a change has occurred in the collected production site information.

[0089] The frequency range can be arbitrarily set by the user of the learning system 1, and is, for example, a range of the mean value ±3×standard deviation, or an interquartile range.

[0090] For example, the comparison unit 103 determines whether the average value of the time-series data of resin viscosity for the operation of the injection molding machine collected by the second collection unit 202 deviates from a frequent range determined from the average value of the time-series data of resin viscosity included in the first learning data. If the comparison unit 103 determines that the average value deviates from the frequent range, the second notification unit 403 notifies the user that a sudden change not intended by the user has occurred in the equipment related to measuring the resin viscosity or in the environment in which the resin viscosity is measured.

[0091] Next, the learning process executed by the learning system 1 according to this embodiment will be described with reference to Fig. 7. The learning process in Fig. 7 is a process that is started by an operation by the user of the learning system 1, for example.

[0092] In order to create learning data, the first collection unit 201 of the collection device 200 collects production site information indicating the operating status of equipment at the production site and productivity information indicating productivity at the production site (step S101).

[0093] For example, the first collection unit 201 collects time-series data of extrusion speed, time-series data of resin viscosity, and flag data from the injection molding machine. The first collection unit 201 also collects time-series data of quality information from the injection molding machine.

[0094] The first pre-processing unit 101 of the learning management device 100 executes a process of extracting production site information for each operation from the production site information collected by the first collecting unit 201 of the collecting device 200 (step S102).

[0095] For example, the first pre-processing unit 101 extracts the time series data of the extrusion speed for each operation, with the start point being the point when the flag data switches from OFF to ON and the end point being the point when the flag data switches from ON to OFF.

[0096] The first pre-processing unit 101 calculates statistics for the production site information for each extracted operation, and acquires the calculated statistics as first production site information (step S103).

[0097] For example, the first pre-processing unit 101 calculates the average value for each of the time series data of the extrusion speed for each operation and the time series data of the extrusion speed for each operation, and acquires the calculated average value as the first production site information.

[0098] The first preprocessing unit 101 creates first learning data 3011 by associating the first production site information with productivity information indicating the productivity at the production site of the first production site information (step S104).

[0099] For example, the first pre-processing unit 101 creates first learning data 3011 by associating the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation with the time series data of the quality information for each operation, and stores the first learning data 3011 in the learning data storage unit 301.

[0100] The model generation unit 302 of the learning device 300 uses the first learning data 3011 to generate a trained model for inferring productivity information at the production site of the production site information from the production site information (step S105).

[0101] For example, the model generation unit 302 generates a trained model for diagnosing the quality of an injection-molded product from the average value of the time series data of the extrusion speed of one operation of the injection molding machine and the average value of the time series data of the resin viscosity of one operation, and stores the trained model in the trained model storage unit 303.

[0102] Next, the inference process executed by the learning system 1 according to this embodiment will be described with reference to Fig. 8. The inference process of Fig. 8 is a process that is started by an operation by the user of the learning system 1, for example.

[0103] The second collection unit 202 of the collection device 200 collects production site information to obtain inference data for making inferences using the trained model, and collects productivity information to create training data (step S201).

[0104] For example, the second collection unit 202 collects time-series data of extrusion speed, time-series data of resin viscosity, and flag data from the injection molding machine. The second collection unit 202 also collects time-series data of quality information from the injection molding machine.

[0105] The second pre-processing unit 102 of the learning management device 100 executes a process of extracting production site information for each operation from the production site information collected by the second collection unit 202 of the collection device 200 (step S202).

[0106] For example, when the second collection unit 202 collects time series data of the extrusion speed, the second pre-processing unit 102 extracts the time series data of the extrusion speed for each operation in real time, with the start point being the point when the flag data switches from OFF to ON and the end point being the point when the flag data switches from ON to OFF.

[0107] The second pre-processing unit 102 calculates statistics for the production site information for each extracted operation, and acquires the calculated statistics as second production site information (step S203).

[0108] For example, the second pre-processing unit 102 calculates the average value for each of the time series data of the extrusion speed for each operation and the time series data of the resin viscosity for each operation, and stores the calculated average values ​​in the inference data storage unit 401 of the inference device 400 in real time as second production site information.

[0109] The second pre-processing unit 102 creates second learning data 3012 by associating the second production site information with productivity information indicating the productivity at the production site of the second production site information (step S204).

[0110] For example, the second pre-processing unit 102 creates the second learning data 3012 by associating the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation with the time series data of the quality information for each operation.

[0111] The inference unit 402 of the inference device 400 uses the trained model to infer productivity information from the second production site information (step S205).

[0112] For example, the inference unit 402 inputs, in real time, the average value of the time series data of the extrusion speed for each operation and the average value of the time series data of the resin viscosity for each operation stored in the inference data storage unit 401 into the learned model stored in the learned model storage unit 303 of the learning device 300, and obtains quality information for each operation.

[0113] The second notification unit 403 notifies the result of the inference by the inference unit 402, that is, the inferred productivity information (step S206).

[0114] For example, the second notification unit 403 displays the quality information for each operation of the injection molding machine acquired by the inference unit 402 on the screen of the inference device 400 or on the screen of another information processing device used by the user of the learning system 1.

[0115] Next, the relearning process executed by the learning system 1 according to this embodiment will be described with reference to Fig. 9. The relearning process in Fig. 9 is executed, for example, by an operation by a user of the learning system 1 or at preset periodic timing.

[0116] The comparison unit 103 of the learning management device 100 determines whether there is a difference between the first trend, which is the trend of the first production site information, and the second trend, which is the trend of the second production site information (step S301). If the comparison unit 103 determines that there is a difference between the first trend and the second trend (step S301; YES), the instruction unit 104 of the learning management device 100 instructs the model generation unit 302 of the learning device 300 to perform re-learning (step S302). That is, the instruction unit 104 instructs the model generation unit 302 to regenerate a trained model using the first training data 3011 and the second training data 3012. On the other hand, if the comparison unit 103 determines that there is no difference between the first trend and the second trend (step S301; NO), the processing of FIG. 9 ends.

[0117] For example, the comparison unit 103 determines whether the difference between the median "ME1" of the group of average values ​​of the time-series data of the extrusion speed, which is the first production site information, and the median "ME2" of the group of average values ​​of the time-series data of the extrusion speed, which is the second production site information, is greater than a value set by the user. The comparison unit 103 also determines whether the difference between the median "MR1" of the group of average values ​​of the time-series data of the resin viscosity, which is the first production site information, and the median "MR2" of the group of average values ​​of the time-series data of the resin viscosity, which is the second production site information, is greater than a value set by the user. When the comparison unit 103 determines that the difference between the median "ME1" and the median "ME2" is equal to or less than the value set by the user but the difference between the median "MR1" and the median "MR2" is greater than the value set by the user, the instruction unit 104 instructs the model generation unit 302 to regenerate the trained model. On the other hand, if the comparison unit 103 determines that both the difference between the median value "ME1" and the median value "ME2" and the difference between the median value "MR1" and the median value "MR2" are less than or equal to the values ​​set by the user, it terminates the processing of Figure 9.

[0118] The first notification unit 105 of the learning management device 100 notifies that it has been determined that there is a difference between the first trend and the second trend (Step S303).

[0119] For example, the first notification unit 105 displays on the screen of the learning management device 100 or on the screen of another information processing device used by the user of the learning system 1 that it has been determined that there is a difference between the first trend and the second trend of the resin viscosity time series data.

[0120] The model generation unit 302 of the learning device 300 regenerates the trained model using the first training data 3011 and the second training data 3012 stored in the training data storage unit 301 (step S304).

[0121] For example, the model generation unit 302 generates a new trained model for diagnosing the quality of an injection-molded product from the average value of the time series data of the extrusion speed of one operation of the injection molding machine and the average value of the time series data of the resin viscosity of one operation, and stores the trained model in the trained model storage unit 303.

[0122] According to this embodiment, when a change occurs between the past and present in the trends of production site information indicating the operating status of equipment at the production site, i.e., when a change over time occurs at the production site, a trained model is regenerated based on past data before the change over time and current data after the change over time. This allows automatic re-learning based on the occurrence of a change over time at the production site. Alternatively, a user can optionally execute re-learning based on the occurrence of a change over time at the production site. An accurate trained model that reflects the situation at the production site can be generated, and accuracy can be maintained in diagnosing production site productivity, such as quality prediction and detection of signs of abnormality.

[0123] Furthermore, according to this embodiment, new learning data can be automatically created using current production site information while inference is being performed, thereby reducing the burden on the user of performing re-learning.

[0124] Furthermore, according to this embodiment, differences between past and present trends are determined based on production site information extracted for each equipment operation, making it possible to accurately detect changes over time at the production site based on the production site information for each operation. Furthermore, because the production site information extracted for each equipment operation is used for learning, it is possible to generate a trained model that eliminates noise outside of the operation. This enables accurate productivity assessment, such as quality prediction and abnormality sign detection.

[0125] Furthermore, according to this embodiment, statistics are calculated for the production site information extracted for each operation, and the calculated statistics are used to create learning data, which reduces the processing speed and the size of the accumulated data compared to when the entire production site information extracted for each operation is used.

[0126] Furthermore, according to this embodiment, if the information indicating the current situation at the production site or its trend differs from the trend of information indicating the past situation at the production site, it is possible to notify that there may have been a change in the trend of the collected production site information. This makes it possible to quickly detect changes in the current situation at the production site that could lead to the occurrence of abnormalities at the production site, such as deterioration in quality and a drop in the operating rate, and to reduce quality losses, operating losses, etc., and prevent a decline in productivity.

[0127] (Modifications) Although the embodiments of the present disclosure have been described above, various modifications and applications are possible when implementing the present disclosure.

[0128] In the above embodiment, the learning system 1 includes four devices: the learning management device 100, the collection device 200, the learning device 300, and the inference device 400. However, the configuration of the learning system 1 is not limited to this. Any configuration is possible as long as each functional unit of the learning system 1 is realized. For example, the learning device 300 and the inference device 400 may be built into equipment at the production site or may exist on a cloud server. Furthermore, for example, the first notification unit 105 may be realized in the collection device 200, the learning device 300, or the inference device 400, and the second notification unit 403 may be realized in the learning management device 100, the collection device 200, or the learning device 300.

[0129] In the above embodiment, the first pre-processing unit 101 and the second pre-processing unit 102 extract the production site information for each operation, but this is not limiting. The first pre-processing unit 101 may further extract predetermined intervals from the production site information for each operation, calculate statistics for the extracted intervals, and acquire the calculated statistics as the first production site information. The second pre-processing unit 102 may further extract predetermined intervals from the production site information for each operation, calculate statistics for each extracted interval, and acquire the calculated statistics as the second production site information.

[0130] The predetermined interval is arbitrarily set by the user of the learning system 1. For example, the predetermined interval is the beginning, middle, and end intervals in the production site information of one operation, or a predetermined period of time from the point in time when flag data switches from OFF to ON.

[0131] With this configuration, the production site information extracted for each operation can be further subdivided and used as learning data and inference data, which allows for a more accurate understanding of the characteristics of equipment operations and improves the accuracy of productivity diagnosis, such as detecting changes over time at the production site, predicting quality, and detecting signs of abnormalities.

[0132] In the above embodiment, the case where supervised learning is applied to the learning algorithm used by the model generation unit 302 has been described, but this is not limiting. In addition to supervised learning, reinforcement learning, unsupervised learning, semi-supervised learning, or the like can also be applied to the learning algorithm.

[0133] The model generation unit 302 may also learn productivity information according to learning data created for multiple devices. The model generation unit 302 may collect learning data from multiple devices used in the same area, or may learn productivity information using learning data collected from multiple devices operating independently in different areas. It is also possible to add or remove devices from which learning data is collected during the process. Furthermore, the learning device 300 that has learned productivity information about a certain device may be applied to another device, and the productivity information about the other device may be re-learned and updated.

[0134] Furthermore, the learning algorithm used in the model generation unit 302 may be deep learning, which learns to extract the features themselves, or machine learning may be performed according to other known methods, such as genetic programming, functional logic programming, or support vector machines.

[0135] In addition, by applying an operating program that specifies the operation of the learning management device 100 of the above embodiment to an existing personal computer or information terminal device, it is possible to make the personal computer or information terminal device function as the learning management device 100 of the embodiment.

[0136] Furthermore, the method of distribution of such a program is arbitrary; for example, it may be stored on a computer-readable recording medium such as a CD-ROM (Compact Disk Read-Only Memory), a DVD (Digital Versatile Disk), or a memory card and distributed, or it may be distributed via a communication network such as the Internet.

[0137] The present disclosure allows various embodiments and modifications without departing from the broad spirit and scope of the present disclosure. Furthermore, the above-described embodiments are intended to explain the present disclosure and do not limit the scope of the present disclosure. That is, the scope of the present disclosure is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of equivalent disclosures are considered to be within the scope of the present disclosure.

[0138] According to the present disclosure, it is possible to provide a learning management program, a learning management device, and a learning system that are capable of performing re-learning in a timely manner based on the occurrence of changes over time in the production site.

[0139] 1 Learning system, 10 Information processing device, 11 Processor, 12 Main memory unit, 13 Auxiliary memory unit, 14 Communication unit, 15 Input unit, 16 Output unit, 17 Bus, 100 Learning management device, 101 First preprocessing unit, 102 Second preprocessing unit, 103 Comparison unit, 104 Instruction unit, 105 First notification unit, 200 Collection device, 201 First collection unit, 202 Second collection unit, 300 Learning device, 301 Learning data storage unit, 3011 First learning data, 3012 Second learning data, 302 Model generation unit, 303 Learned model storage unit, 400 Inference device, 401 Inference data storage unit, 402 Inference unit, 403 Second notification unit, 501 to 504 Rectangle, 601 Legend, 602, 603 Range.

Claims

1. A computer comprising: a first preprocessing means for acquiring first production site information by performing preprocessing on production site information indicating the operating status of equipment at a production site, collected in order to create learning data, and correlating the first production site information with productivity information indicating the productivity at the production site of the first production site information, thereby creating first learning data used for generating a trained model for inferring from the production site information, the productivity information indicating the productivity at the production site of the production site information of the production site information; a second preprocessing means for acquiring second production site information by performing the preprocessing on production site information collected in order to acquire inference data for making inference using the trained model, and correlating the second production site information with productivity information indicating the productivity at the production site of the second production site information, thereby creating second learning data; a comparison means for comparing a first trend which is a trend of the first production site information with a second trend which is a trend of the second production site information, and determining whether or not there is a difference between the first trend and the second trend; an instruction means for instructing regeneration of the trained model using the first learning data and the second learning data when the comparison means determines that there is a difference between the first trend and the second trend.

2. The learning management program according to claim 1, wherein the pre-processing includes a process of extracting the production site information for each operation performed by the device.

3. The learning management program of claim 2, wherein the pre-processing is a process of extracting the production site information for each operation based on a change in the value of flag data indicating a period during which the equipment is performing an operation and a period during which the operation is not performing, a change in the value of the production site information relative to a predetermined value, a change in the file in which the production site information is stored, or a predetermined interval.

4. A learning management program as described in claim 2 or 3, wherein the first pre-processing means calculates statistics for the production site information for each operation and acquires the calculated statistics as the first production site information, and the second pre-processing means calculates statistics for the production site information for each operation and acquires the calculated statistics as the second production site information.

5. A learning management program as described in any one of claims 2 to 4, wherein the first pre-processing means further extracts predetermined intervals from the production site information for each operation, calculates statistics for each of the extracted intervals, and acquires the calculated statistics as the first production site information, and the second pre-processing means further extracts predetermined intervals from the production site information for each operation, calculates statistics for each of the extracted intervals, and acquires the calculated statistics as the second production site information.

6. The learning management program according to claim 4 or 5, wherein the statistical quantity is a mean value, a standard deviation, a minimum value, a maximum value, a median value, a mode value, a value at a predetermined percentile, a mean value ±3 x standard deviation, or an interquartile range.

7. A learning management program as described in any one of claims 1 to 6, which causes the computer to function as a first notification means that, when the comparison means determines that there is a difference between the first trend and the second trend, notifies the user that said difference has been determined.

8. A learning management program described in any one of claims 1 to 7, which causes the computer to function as a second notification means for notifying productivity information inferred from the second production site information using the trained model.

9. The learning management program according to claim 8, wherein the comparison means determines whether the second production site information deviates from the frequency range of the first production site information, and the second notification means, when the comparison means determines that the second production site information deviates, notifies that there may be a change in the trend of the collected production site information.

10. A first preprocessing means for acquiring first production site information by performing preprocessing on production site information indicating the operating status of equipment at a production site, collected in order to create learning data, and for creating first learning data used in generating a trained model for inferring productivity information indicating productivity at the production site of the production site information from the production site information, by associating the first production site information with productivity information indicating productivity at the production site of the first production site information; a second preprocessing means for acquiring second production site information by performing the preprocessing on production site information collected in order to acquire inference data for making inference using the trained model, and for creating second learning data by associating the second production site information with productivity information indicating productivity at the production site of the second production site information; a comparison means for comparing a first trend which is a trend of the first production site information with a second trend which is a trend of the second production site information, and determining whether or not there is a difference between the first trend and the second trend; an instruction means for instructing regeneration of the trained model using the first learning data and the second learning data when the comparison means determines that there is a difference between the first trend and the second trend.

11. A first collection means for collecting production site information indicating the operating status of equipment at a production site in order to create learning data; a first preprocessing means for acquiring first production site information by performing preprocessing on the production site information collected by the first collection means, and creating first learning data by associating the first production site information with productivity information indicating productivity at the production site of the first production site information; a model generation means for generating a trained model using the first learning data to infer productivity information indicating productivity at the production site of the production site information from the production site information; a second collection means for collecting the production site information in order to acquire inference data for making inference using the trained model; a second preprocessing means for acquiring second production site information as the inference data by performing preprocessing on the production site information collected by the second collection means, and creating second learning data by associating the second production site information with productivity information indicating productivity at the production site of the second production site information; and an inference means for inferring the productivity information from the second production site information using the trained model. A learning system comprising: a comparison means for comparing a first trend which is a trend of the first production site information with a second trend which is a trend of the second production site information, and determining whether or not there is a difference between the first trend and the second trend; and an instruction means for instructing the model generation means to regenerate the trained model using the first learning data and the second learning data when the comparison means determines that there is a difference between the first trend and the second trend, wherein the model generation means regenerates the trained model using the first learning data and the second learning data when the instruction means receives an instruction to regenerate the trained model.

Images