Learning device, estimation device, method for generating learned model, estimation method, learning program, and estimation program

Abstract

To facilitate data analysis in a lot unit regarding kneading of a rubber material.SOLUTION: Data analysis device 1 comprises: a learning data acquisition unit 701 that acquires lot-unit process data and lot-unit quality data; a learning statistical process data generation unit 702 that generates multiple batch-unit statistical process data based on each of the multiple batch-unit process data included in the lot-unit process data, and generates lot-unit statistical process data based on the multiple batch-unit statistical process data; a learning statistical quality data generation unit 703 that generates lot-unit statistical quality data based on the multiple batch-unit quality data; a learning data generation unit 704 that generates learning data including the lot-unit statistical process data and the lot-unit statistical quality data; and a learning unit 705 that performs learning of a machine learning model 706 based on the learning data.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present invention relates to a learning device, an estimation device, a method for generating a trained model, an estimation method, a learning program, and an estimation program, and in particular to a learning and estimation technique for data generated during a rubber material mixing process. [Background technology]

[0002] Conventionally, in order to improve the quality of a rubber material kneading product obtained in a rubber material kneading process, a technology for performing data analysis using statistical process data showing statistical information generated based on process data showing the kneading state during kneading of one batch of rubber material and quality data showing the quality of the rubber material kneading product has been known. For example, Patent Document 1 describes a technology for analyzing, by machine learning, statistical data generated based on measurement data showing a time series of measurement values ​​showing the operating state of a kneader as the kneading state, and the degree of abnormality in kneading of a kneading batch. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-59058 Summary of the Invention [Problem to be solved by the invention]

[0004] Rubber materials are mixed in batches, but the mixer divides each batch into several parts, and the process data output from the mixer can only be used on a batch-by-batch basis, making it difficult to analyze the rubber material mixing process on a batch-by-lot basis after the fact.

[0005] Furthermore, rubber materials are mixed through a plurality of mixing steps, and between each mixing step, a part or all of the rubber material in each batch of the previous mixing step is combined with a part or all of the rubber material in a batch other than the previous batch to produce a rubber material in the next mixing step, so the relationship between the batches in each mixing step is complex.For this reason, correlations between statistical process data and quality data based on process data obtained on a batch-by-batch basis are difficult to find, and it has not been easy to analyze data on a lot-by-lot basis using machine learning, which assumes that there is some correlation.

[0006] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a learning device, an estimation device, a method for generating a trained model, an estimation method, a learning program, and an estimation program that can facilitate lot-by-lot data analysis of the kneading process of rubber materials. [Means for solving the problem]

[0007] (1) A learning device according to the present invention is a learning device for a manufacturing process in which a rubber material in a unit of a lot is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of the rubber material in each of the plurality of batches are obtained after the last step of the plurality of kneading steps, the learning device comprising: a data acquisition means for acquiring lot unit process data, which is a collection of a plurality of batch unit process data, each indicating a kneading state of the rubber material in a unit of a batch, during kneading of the rubber material in a unit of a batch; and lot unit quality data, which is a collection of a plurality of batch unit quality data, each indicating the quality of the kneaded products of the rubber material in a unit of a batch, and a data acquisition means for acquiring lot unit process data, which is a collection of a plurality of batch unit quality data, each indicating the quality of the kneaded products of the rubber material in a unit of a batch, based on each of the plurality of batch unit process data. The system includes a statistical process data generation means for generating a plurality of batch-unit statistical process data each indicating batch-unit statistical information related to a process, and for generating lot-unit statistical process data indicating lot-unit statistical information related to the manufacturing process based on the plurality of batch-unit statistical process data; a statistical quality data generation means for generating lot-unit statistical quality data indicating lot-unit statistical information related to the quality of the plurality of kneading products based on the plurality of batch-unit quality data; a learning data generation means for generating learning data including the lot-unit statistical process data and the lot-unit statistical quality data; and a learning means for performing learning of a machine learning model based on the learning data.

[0008] (2) In the learning device of (1), the statistical process data generation means may generate a plurality of mixer unit statistical process data indicating statistical information per mixer unit regarding the manufacturing process based on the plurality of batch unit statistical process data, and may generate the lot unit statistical process data based on the plurality of mixer unit statistical process data.

[0009] (3) In the learning device of (2), the statistical process data generation means may divide each of the plurality of batch-unit process data into predetermined time periods according to a predetermined rule, generate the plurality of batch-unit statistical process data for each time period, and generate the lot-unit statistical process data for each time period based on the plurality of batch-unit statistical process data for each time period, and the learning data generation means may generate the learning data including the lot-unit statistical process data for each time period and the lot-unit quality data.

[0010] (4) In the learning device of (1) to (3), the learning device includes a first kneader assigned to a first kneading process, which indicates one of the multiple kneading processes, in which multiple lots of rubber material are divided into multiple batches of rubber material and fed in sequence, and the rubber material of each batch is kneaded in sequence; and a second kneader assigned to a second kneading process, which indicates a process subsequent to the first kneading process, in which, after kneading by the first kneader, the rubber material of each lot is divided into multiple batches of rubber material and fed in sequence, and the rubber material of each batch is kneaded in sequence. a first batch unit process data acquisition means for acquiring a plurality of first batch unit process data indicating a mixing start time and a mixing end time of each batch in the first mixer and an operating state of the first mixer during mixing, based on the manufacturing process data; and a plurality of second batch unit process data acquisition means for acquiring a plurality of second batch unit process data indicating a mixing start time and a mixing end time of each batch in the second mixer and an operating state of the second mixer during mixing, based on the manufacturing process data. a first lot mixing period specifying means for specifying a mixing period of each of the lots in the first mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the first mixer, based on the plurality of first batch unit process data; a second lot mixing period specifying means for specifying a mixing period of each of the lots in the second mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the second mixer, based on the plurality of second batch unit process data; and a lot unit process data generating means for selecting a portion of the plurality of first batch unit process data and a portion of the plurality of second batch unit process data corresponding to each of the lots, based on the mixing period of each of the lots in the first mixer and the mixing period of each of the lots in the second mixer, to generate a plurality of lot unit process data indicating operating states of both the first mixer and the second mixer while the rubber material of each of the lots is being mixed;The system further includes a quality data processing means for processing quality data related to the quality of the kneading product, the quality data processing means including: a quality data acquisition means for acquiring the quality data; a lot unit batch group identification means for identifying a batch group of each lot to be kneaded by a final kneader assigned to the final kneading step based on the kneading period of each lot in a final kneading step indicating the last step of the plurality of kneading steps identified by the manufacturing process data processing means; and a lot unit quality data generation means for dividing the quality data into lot units based on the batch group of each lot to generate a plurality of lot unit quality data, wherein the data acquisition means may acquire the lot unit process data related to at least one lot out of the plurality of lot unit process data from the manufacturing process data processing means, and acquire the lot unit quality data related to at least the one lot out of the plurality of lot unit quality data from the quality data processing means.

[0011] (5) The estimation device according to the present invention is an estimation device for a manufacturing process in which a rubber material in a unit of a lot is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading processes, and kneaded products of the rubber material in each of the plurality of batches are obtained after the last process of the plurality of kneading processes, and includes: a data acquisition means for acquiring lot-unit process data that is a collection of a plurality of batch-unit process data in units of lots, each indicating a kneading state of the rubber material in a batch-unit during kneading; a statistical process data generation means for generating a plurality of batch-unit statistical process data that each indicate statistical information on a batch unit regarding the manufacturing process based on each of the plurality of batch-unit process data, and for generating a lot-unit statistical process data that indicates statistical information on a lot unit regarding the manufacturing process based on the plurality of batch-unit statistical process data; and an estimation means for estimating the quality of the kneaded product in units of a lot based on the output when the lot-unit statistical process data is input into a trained model.

[0012] (6) In the estimation device of (5), the statistical process data generation means may generate a plurality of mixer unit statistical process data indicating statistical information on a mixer unit basis regarding the manufacturing process based on the plurality of batch unit statistical process data, and may generate the lot unit statistical process data based on the plurality of mixer unit statistical process data.

[0013] (7) In the estimation device of (6), the statistical process data generation means may divide each of the plurality of batch-unit process data into predetermined time periods according to a predetermined rule, generate the plurality of batch-unit statistical process data for each time period, and generate the lot-unit statistical process data for each time period based on the plurality of batch-unit statistical process data for each time period, and the estimation means may estimate the lot-unit quality of the kneading product based on an output when the lot-unit statistical process data for each time period is input into a trained model.

[0014] (8) In the estimation device of (5) to (7), the estimation device includes a first kneader assigned to a first kneading process, which indicates one of the plurality of kneading processes, in which a plurality of lots of the rubber material are divided into a plurality of batches of the rubber material and sequentially fed, and the rubber material of each batch is kneaded in turn, and a second kneader assigned to a second kneading process, which indicates a process subsequent to the first kneading process, in which, after kneading by the first kneader, the rubber material of each lot is divided into a plurality of batches of the rubber material and respectively fed, and the rubber material of each batch is kneaded in turn, a manufacturing process data processing means for processing manufacturing process data acquired from each of the first and second batch unit process data acquiring means for acquiring, based on the manufacturing process data, a plurality of first batch unit process data indicating the start time and end time of mixing of each batch in the first mixer and the operating state of the first mixer during mixing; and a plurality of second batch unit process data acquiring means for acquiring, based on the manufacturing process data, a plurality of second batch unit process data indicating the start time and end time of mixing of each batch in the second mixer and the operating state of the second mixer during mixing. a first lot mixing period specifying means for specifying a mixing period of each of the lots in the first mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the first mixer, based on the plurality of first batch unit process data; and a second batch unit process data specifying means for specifying a mixing period of each of the lots in the second mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the second mixer, based on the plurality of second batch unit process data. and lot unit process data generation means for selecting a portion of the plurality of first batch unit process data and a portion of the plurality of second batch unit process data corresponding to each of the lots based on the kneading period of each of the lots in the first kneader and the kneading period of each of the lots in the second kneader, and generating a plurality of the lot unit process data indicating operating states of both the first kneader and the second kneader while the rubber material of each of the lots is being kneaded,The data acquisition means may acquire the lot-unit process data relating to at least one lot from the plurality of lot-unit process data from the manufacturing process data processing means.

[0015] (9) A method for generating a trained model according to the present invention is a method for generating a trained model for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last step of the plurality of kneading steps, the method comprising: a data acquisition step for acquiring lot unit process data, which is a collection of a plurality of batch unit process data, which respectively indicate the kneading state of the rubber material in a batch unit during kneading; and lot unit quality data, which is a collection of a plurality of batch unit quality data, which respectively indicate the quality of the kneaded products of the rubber material in a batch unit; and based on each of the plurality of batch unit process data, The method includes a statistical process data generation step of generating a plurality of batch-unit statistical process data each indicating statistical information on a batch-unit basis regarding the manufacturing process, and generating lot-unit statistical process data indicating statistical information on a lot-unit basis regarding the manufacturing process based on the plurality of batch-unit statistical process data; a statistical quality data generation step of generating lot-unit statistical quality data indicating statistical information on a lot-unit basis regarding the quality of the plurality of kneading products based on the plurality of batch-unit quality data; a learning data generation step of generating learning data including the lot-unit statistical process data and the lot-unit statistical quality data; and a learning step of performing learning of a machine learning model based on the learning data.

[0016] (10) An estimation method according to the present invention is an estimation method for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading processes, and kneaded products of the rubber material in each of the plurality of batches are obtained after the last kneading process of the plurality of kneading processes, and includes: a data acquisition step of acquiring lot-unit process data, which is a collection of a plurality of batch-unit process data indicative of a kneading state of the rubber material in a batch unit, on a lot-by-lot basis; a statistical process data generation step of generating a plurality of batch-unit statistical process data indicative of batch-unit statistical information about the manufacturing process based on each of the plurality of batch-unit process data, and generating lot-unit statistical process data indicative of lot-unit statistical information about the manufacturing process based on the plurality of batch-unit statistical process data; and an estimation step of estimating the quality of the kneaded product in a lot unit based on the output when the lot-unit statistical process data is input into a trained model.

[0017] (11) A learning program according to the present invention is a learning program for a manufacturing process in which a rubber material in a unit of a lot is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of the respective batches of the rubber material are obtained after the last step of the plurality of kneading steps, the learning program including: data acquisition means for acquiring lot unit process data, which is a collection of a plurality of batch unit process data, each indicating a kneading state of the rubber material in a unit of a batch, during kneading of the rubber material in a unit of a batch; and lot unit quality data, which is a collection of a plurality of batch unit quality data, each indicating the quality of the kneaded products of the rubber material in a unit of a batch, The program causes a computer to function as a statistical process data generating means for generating a plurality of batch-unit statistical process data each indicating statistical information on a batch-unit basis regarding a manufacturing process, and for generating lot-unit statistical process data indicating statistical information on a lot-unit basis regarding the manufacturing process based on the plurality of batch-unit statistical process data, a statistical quality data generating means for generating lot-unit statistical quality data indicating statistical information on a lot-unit basis regarding the quality of the plurality of mixing deliverables based on the plurality of batch-unit quality data, a learning data generating means for generating learning data including the lot-unit statistical process data and the lot-unit statistical quality data, and a learning means for executing learning of a machine learning model based on the learning data. This program may be stored in a computer-readable information storage medium such as a magneto-optical disk or a semiconductor memory.

[0018] (12) An estimation program according to the present invention is an estimation program for a manufacturing process in which a rubber material in a unit lot is divided into multiple batches of the rubber material and kneaded in each of multiple kneading steps, and kneaded products of the rubber material in each batch are obtained after the last kneading step of the multiple kneading steps, the program causing a computer to function as: a data acquisition means for acquiring lot-unit process data that collects multiple batch-unit process data in units of lots, each indicating a kneading state of the rubber material in each batch during kneading; a statistical process data generation means for generating multiple batch-unit statistical process data that indicate batch-unit statistical information about the manufacturing process based on each of the multiple batch-unit process data, and generating multiple batch-unit statistical process data that indicate lot-unit statistical information about the manufacturing process based on the multiple batch-unit statistical process data; and an estimation means for estimating the quality of the kneaded products in units of lots based on an output when the lot-unit statistical process data is input to a trained model. This program may be stored in a computer-readable information storage medium such as a magneto-optical disk or a semiconductor memory. [Effects of the Invention]

[0019] According to the present invention, data analysis for each lot regarding the kneading of rubber materials can be easily performed. [Brief explanation of the drawings]

[0020] [Figure 1] 1 is a block diagram showing a hardware configuration of a data analysis device according to a first embodiment. [Figure 2] 1A to 1C are diagrams illustrating an example of a kneading process for a rubber material. [Figure 3] FIG. 2 is a functional block diagram showing an example of functions realized by the data analysis device according to the first embodiment. [Figure 4] FIG. 2 is a functional block diagram showing an example of a function related to processing of manufacturing process data among functions realized by the data analysis device according to the first embodiment. [Figure 5]FIG. 10 is a diagram illustrating an example of manufacturing process data including a plurality of batch unit process data. [Figure 6A] FIG. 10 is a diagram illustrating a process for identifying the kneading period of each lot. [Figure 6B] FIG. 10 is a diagram illustrating a process for identifying the kneading period of each lot. [Figure 6C] FIG. 10 is a diagram illustrating a process for identifying the kneading period of each lot. [Figure 6D] FIG. 10 is a diagram illustrating a process for identifying the kneading period of each lot. [Figure 7] FIG. 2 is a functional block diagram showing an example of a function related to quality data processing among functions realized by the data analysis device according to the first embodiment. [Figure 8] FIG. 10 is a diagram illustrating an example of quality data. [Figure 9] FIG. 2 is a functional block diagram showing an example of functions relating to data analysis among functions realized by the data analysis device according to the first embodiment. [Figure 10] FIG. 4 is a diagram illustrating a process for generating lot-unit statistical process data according to the first embodiment. [Figure 11] FIG. 10 is a diagram illustrating a process for generating lot-unit statistical process data according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, a learning device, an estimation device, a trained model generation method, an estimation method, a learning program, and an estimation program according to the present invention will be described in detail with reference to the drawings.

[0022] [1. First embodiment] First, a first embodiment, which is an example of an embodiment of a data analysis device according to the present invention, will be described.

[0023] [1-1. Hardware configuration for realizing data analysis according to the first embodiment] 1 is a block diagram showing the hardware configuration of a data analysis device according to the first embodiment. As shown in Fig. 1, the data analysis device 1 includes a processing device 11, a storage device 12, a display device 13, and an input device 14.

[0024] The processing device 11 includes, for example, a CPU, a GPU, etc., and operates according to a program stored in the storage device 12. As the processing device 11, an FPGA may be used.

[0025] The storage device 12 may be, for example, a ROM, RAM, HDD, SSD, etc., and stores programs executed by the processing device 11, various data to be processed, etc. The storage device 12 includes a main storage device and an auxiliary storage device. For example, the main storage device is a volatile memory such as RAM, and the auxiliary storage device is a non-volatile memory such as ROM, EEPROM, flash memory, or hard disk.

[0026] The display device 13 is, for example, a liquid crystal display or an organic EL display, and displays data stored in the storage device 12, processing results of the processing device 11, and the like.

[0027] The input device 14 is a user interface such as a keyboard or a mouse, and receives an operation input from the operator and inputs a signal indicating the content of the operation input to the processing device 11.

[0028] The data analysis device 1 having the above configuration is used to analyze the mixing process of rubber materials. For example, the data analysis device 1 may be connected to multiple rubber mixers and physical property measuring devices for the mixed products of the rubber material via a communication network, and manufacturing process data generated in the mixers and quality data indicating the measurement results of the physical properties of the mixed products of the rubber material are provided via the communication network. The data analysis device 1 synthesizes data for each lot from the manufacturing process data and quality data, enabling analysis of data for each lot. Note that one or more personal computers, server computers, etc. can be used as the processing device 11 and storage device 12 described above.

[0029] [1-2. Rubber material mixing process (manufacturing process)] Fig. 2 is a diagram showing an example of a rubber material kneading process. Here, the rubber material kneading process includes a plurality of kneading steps, and as shown in Fig. 2, a case will be described in which the kneading process includes four kneading steps: a first kneading step 21, a second kneading step 22, a third kneading step 23, and a fourth kneading step 24.

[0030] In the rubber material mixing process, a rubber material in a lot unit is divided into multiple batches of rubber material and mixed in each of multiple mixing steps, and mixed products of each of the multiple batches of rubber material are obtained after the last step of the multiple mixing steps. Note that a lot is a unit of rubber material with the same specifications.

[0031] For example, one mixer is assigned to each mixing step. In this example, the first mixing step 21 is assigned to a first mixer 211, the second mixing step 22 is assigned to a second mixer 221, the third mixing step 23 is assigned to a third mixer 231, and the fourth mixing step 24 is assigned to a fourth mixer 241.

[0032] A plurality of lots of rubber material are divided into a plurality of batches and fed into the first kneader 211 in turn, and the first kneader 211 kneads the rubber material in the batches in turn.

[0033] In the example shown in the figure, first, the rubber material of lot L1 is divided into eight batches, batch 2101, batch 2102, ..., batch 2108, and sequentially fed into the first kneader 211. After the rubber material of lot L1 has been fed, the rubber material of batches from lot L2 onwards (batch 2109, ...) are sequentially fed. The number of batches for each lot may be the same or different for each lot.

[0034] The first kneader 211 kneads the rubber material of batch 2101, batch 2102, ..., batch 2108 of lot L1 in that order. After kneading all the batches of rubber material of lot L1, the first kneader 211 kneads the rubber material of batches from lot L2 onwards (batch 2109, ...) in that order.

[0035] After being mixed by the first mixer 211, the rubber material of each lot is divided into a plurality of batches and fed into the second mixer 221, and the second mixer 221 mixes the rubber material of each batch in order.

[0036] For example, the rubber material of lot L1 mixed by the first mixer 211 is divided into 10 batches, namely batch 2201, batch 2202, batch 2203, ..., batch 2210, and is sequentially fed into the second mixer 221. After all the rubber materials of lot L1 have been fed, the rubber materials of lot L2 and subsequent batches (batch 2211, ...) are sequentially fed. Each batch fed into the second mixer 221 may be made from only one of the multiple batches mixed by the first mixer 211. Alternatively, each batch may be made by combining rubber materials of multiple batches. For example, the rubber material of batch 2201 is a portion of the rubber material of batch 2101. Furthermore, for example, the rubber material of batch 2202 is a combination of a portion of the rubber material of batch 2101 and a portion of the rubber material of batch 2102. The rubber material fed into the second kneader 221 may be the rubber material kneaded by the first kneader 211 to which a new raw material has been added.

[0037] The second kneader 221 kneads the rubber material of batch 2201, batch 2202, ..., batch 2110 of lot L1 in sequence, and after kneading all the rubber material of batches of lot L1, it kneads the rubber material of batches from lot L2 onwards (batch 2211, ...) in sequence.

[0038] The third kneader 231 and the fourth kneader 241 are used and operate in the same manner as the first kneader 211 and the second kneader 221. For example, the rubber material of lot L1 kneaded by the second kneader 221 is divided into 11 batches, batch 2301 to batch 2311, and sequentially fed into the third kneader 231. After all the rubber material of lot L1 has been fed, the rubber material of lot L2 and subsequent batches (batch 2312, ...) are sequentially fed into the fourth kneader 241. Furthermore, the rubber material of lot L1 kneaded by the third kneader 231 is divided into 16 batches, batch 2401 to batch 2416, and sequentially fed into the fourth kneader 241. After all the rubber material of lot L1 has been fed, the rubber material of lot L2 and subsequent batches (batch 2417, ...) are sequentially fed into the fourth kneader 241. Here, the fourth mixer is a final mixer assigned to the final mixer step, which is the last step among the mixer steps. The mixer product of each batch mixed by the fourth mixer 241, which is the final mixer, is measured for data relating to the quality of the mixer product, such as viscosity.

[0039] In the first embodiment, the rubber material of lot L1 is divided into 8 batches and fed into the first kneader 211, 10 batches into the second kneader 221, 11 batches into the third kneader 231, and 16 batches into the fourth kneader 241, but the number of batches to be kneaded in each kneader may be any number.

[0040] [1-3. Functions realized by the data analysis device according to the first embodiment] 3 is a functional block diagram showing an example of functions realized by the data analysis device according to the first embodiment. As shown in FIG. 3, the data analysis device 1 includes a manufacturing process data processing unit 3, a quality data processing unit 5, and a data analysis unit 7. These functions are realized by a processing device 11 executing a program stored in a storage device 12.

[0041] [1-3-1. Manufacturing process data processing section] First, the manufacturing process data processing unit 3 will be described in detail below. Fig. 4 is a functional block diagram showing an example of functions related to processing of manufacturing process data among the functions realized by the data analysis device according to the first embodiment. As shown in Fig. 4, the manufacturing process data processing unit 3 includes a first batch unit process data acquisition unit 311, a second batch unit process data acquisition unit 321, a third batch unit process data acquisition unit 331, a fourth batch unit process data acquisition unit 341, a first lot mixing period determination unit 312, a second lot mixing period determination unit 322, a third lot mixing period determination unit 332, a fourth lot mixing period determination unit 342, and a lot unit process data generation unit 300.

[0042] [First batch unit process data acquisition section] The first batch unit process data acquisition unit 311 acquires batch unit process data, which is process data for each batch, from the first mixer 211. Hereinafter, the batch unit process data acquired from the nth mixer will be referred to as nth batch unit process data. In other words, the first batch unit process data acquisition unit 311 acquires multiple pieces of first batch unit process data. The first batch unit process data indicates the mixing start time and mixing end time of one batch in the first mixer 211, as well as the operating state of the first mixer 211 as the mixing state during mixing of the rubber material of that batch.

[0043] FIG. 5 shows an example of manufacturing process data acquired from the first kneader 211. The manufacturing process data 41 records the operating status of the first kneader 211 at predetermined sampling intervals (here, 1 second). In the figure, each row shows data acquired from the first kneader 211 during one sampling, including the sampling date and time, batch number, and operating status of the first kneader 211. This operating status includes the temperature of the kneading chamber, the pressure applied to the kneaded rubber, the power consumption of the first kneader 211, and the rotation speed of the kneading blade. Based on this manufacturing process data 41, the first batch unit process data acquisition unit 311 acquires a group of records with a common batch number as batch unit process data. For example, records acquired between "2021 / 11 / 9 11:50:46" and "2021 / 11 / 9 11:56:42" are compiled into a single file to generate batch unit process data 4101. Batch unit process data 4102 is generated in a similar manner. Each batch unit process data includes the sampling date and time of the first record as the mixing start date and time of each batch, and the sampling date and time of the last record as the mixing end date and time of each batch. In this way, the first batch unit process data acquisition unit 311 acquires a plurality of first batch unit process data.

[0044] [2nd batch unit process data acquisition unit to 4th batch unit process data acquisition unit] The second batch unit process data acquisition unit 321 to the fourth batch unit process data acquisition unit 341 also have the same functions as the first batch unit process data acquisition unit 311. That is, the nth batch unit process data acquisition unit acquires manufacturing process data similar to that shown in Fig. 5 from the nth mixer, and generates a plurality of nth batch unit process data therefrom (n = 1 to 4).

[0045] [First lot mixing period determination department] The first lot mixing period specifying unit 312 specifies the mixing period of each lot in the first mixer 211 based on the plurality of first batch unit process data acquired by the first batch unit process data acquiring unit 311. The mixing period of each lot is the period from the mixing start time of the first batch belonging to the same lot to the mixing end time of the last batch.

[0046] 6A to 6D are diagrams illustrating the process of identifying the mixing period for each lot. The first lot mixing period identifying unit 312 generates the timetable 51 shown in FIG. 6A from a large number of first batch unit process data acquired by the first batch unit process data acquiring unit 311. In the timetable 51, each row indicates time information for one first batch unit process data. This time information includes the mixing start date and time, mixing end date and time, and lead time for each batch in the first mixer 211. The lead time for a given batch is the difference between the mixing end time of that batch in the first mixer 211 and the mixing start time of the next batch. The first lot mixing period identifying unit 312 calculates the lead time for each batch and identifies the mixing period for each lot from that calculation. Specifically, taking into account that the operation of the first mixer 211 is temporarily interrupted when transitioning from one lot to the next, the lead time threshold is set to, for example, two hours, and if the lead time is two hours or more, the batch is determined to be the last batch of the given lot. The next batch is then determined to be the first batch of the next lot. In the example shown in the figure, there is a lead time of more than two hours between the batch that finished mixing at 14:01:21 on "2021 / 11 / 9" and the batch that started mixing at 3:36:55 on "2021 / 11 / 10." The former is determined to be the last batch of a lot, and the latter is determined to be the first batch of the next lot. This identifies the row range in the timetable 51 as the mixing period for one lot. As an example, the timetable 51 shows mixing period L11 for a certain lot.

[0047] In addition to the length of the lead time, a condition for the number of batches to be included in one lot may be set, and the mixing period for each lot may be determined by further considering this condition. For example, a batch that satisfies the requirement of 8 or more batches and has a lead time of 2 hours or more may be determined to be the last batch of a certain lot.

[0048] [2nd Lot Mixing Period Identification Department ~ 4th Lot Mixing Period Identification Department] The second lot mixing period specifying unit 322, the third lot mixing period specifying unit 332, and the fourth lot mixing period specifying unit 342 also have the same functions as the first lot mixing period specifying unit 312. That is, the second lot mixing period specifying unit 322 generates a timetable 52 shown in FIG. 6B and specifies the mixing period of each lot from the lead time of each batch. Here, the timetable 52 shows the mixing period L12 of a certain lot. The third lot mixing period specifying unit 332 generates a timetable 53 shown in FIG. 6C and specifies the mixing period of each lot from the lead time of each batch. The timetable 53 shows the mixing period L13 of a certain lot. Furthermore, the fourth lot mixing period specifying unit 342 generates a timetable 54 shown in FIG. 6D and specifies the mixing period of each lot from the lead time of each batch. Here, the timetable 54 shows the mixing period L14 of a certain lot.

[0049] [Lot unit process data generation section] The lot-unit process data generation unit 300 selects a portion of the first batch unit process data corresponding to a certain lot, a portion of the second batch unit process data corresponding to a certain lot, a portion of the third batch unit process data corresponding to a certain lot, and a portion of the fourth batch unit process data corresponding to a certain lot based on the mixing period of each lot in the first kneader 211, the second kneader 221, the third kneader 231, and the fourth kneader 241, and connects these portions to generate process data for a certain lot, i.e., lot-unit process data. The lot-unit process data includes all of the operating states of the first kneader 211, the second kneader 221, the third kneader 231, and the fourth kneader 241 during mixing of one lot of rubber material. In other words, the lot-unit process data is data obtained by collecting batch-unit process data for each kneader on a lot-by-lot basis.

[0050] At this time, the lot-unit process data generation unit 300 compares the end date and time of the mixing period of each lot in the first mixer 211 with the start date and time of the mixing period of each lot in the second mixer 221, which will be responsible for the next mixing process, and links the mixing periods with the closest dates and times. For example, the end date and time of mixing period L11 in the timetable 51 shown in FIG. 6A, "2021 / 11 / 9" and "14:01:21," are closest to the start date and time of mixing period L12 in the timetable 52 shown in FIG. 6B, "2021 / 11 / 9" and "14:39:13," so the lot-unit process data generation unit 300 links these mixing periods L11 and L12. Similarly, mixing period L12 (FIG. 6B) and mixing period L13 (FIG. 6C), and mixing periods L13 (FIG. 6C) and L14 (FIG. 6D) are also linked. Thereafter, the lot unit process data generation unit 300 concatenates the first batch unit process data in the kneading period L11, the second batch unit process data in the kneading period L12, the third batch unit process data in the kneading period L13, and the fourth batch unit process data in the kneading period L14, thereby generating lot unit process data.

[0051] [1-3-2. Quality Data Processing Section] Next, the quality data processing unit 5 will be described in detail below. Fig. 7 is a functional block diagram showing an example of functions related to quality data processing among the functions realized by the data analysis device according to the first embodiment. As shown in Fig. 7, the quality data processing unit 5 includes a quality data acquisition unit 501, a lot-unit batch group identification unit 502, and a lot-unit quality data generation unit 503.

[0052] [Quality Data Acquisition Department] The quality data acquisition unit 501 acquires quality data. The quality data is data related to the quality of a rubber material kneading product. The quality data includes multiple batch-unit quality data each indicating the quality of a batch-unit rubber material kneading product obtained by kneading in a final kneading process, which is the last process among multiple kneading processes. As described in the description of the rubber material kneading process, the quality of the rubber material kneading product refers to the physical property values ​​of the rubber material kneading product, such as viscosity.

[0053] FIG. 8 is a diagram showing an example of quality data. As shown in FIG. 8, in the quality data 60, each row shows batch-unit quality data for each batch. Note that batch 1 shown in FIG. 8 corresponds to the batch on the first row of the timetable 54 for the final kneading process shown in FIG. 6D (the batch that started kneading at "2021 / 11 / 9" "9:24:35"). Similarly, batch n in FIG. 8 corresponds to the batch on the nth row of the timetable 54 in FIG. 6D. Therefore, batches 14 to 29 in FIG. 8 correspond to the batches on the 14th to 29th rows of the timetable 54 in FIG. 6D. In other words, batches 14 to 29 in FIG. 8 correspond to the batches in the kneading period L14 shown in FIG. 6D.

[0054] [Lot unit batch group identification section] The lot unit batch group identification unit 502 identifies the batch group of each lot to be mixed by the final mixer assigned to the final mix process, based on the mixing period of each lot in the final mix process, which indicates the last process among multiple mix processes, as identified by the manufacturing process data processing unit 3.

[0055] For example, the lot unit batch group specifying unit 502 specifies the batch group 610 (FIG. 8) in the kneading period L14 (FIG. 6D) based on the kneading period L14 specified by the fourth lot kneading period specifying unit 342 of the manufacturing process data processing unit 3.

[0056] [Lot-based quality data generation section] The lot-based quality data generating unit 503 divides the quality data 60 into lots based on the batch groups 610 of each lot identified by the lot-based batch group identifying unit 502, and generates multiple lot-based quality data. Once the batch groups of each lot in the final kneading process are identified, it is possible to determine which batch belongs to which lot, and therefore which batch in the quality data belongs to which lot. Therefore, it is possible to divide the quality data 60 into lots. In other words, the lot-based quality data is data obtained by collecting the batch-based quality data of each batch on a lot-by-lot basis.

[0057] [1-3-3. Data Analysis Section] Next, the data analysis unit 7 will be described in detail below. Fig. 9 is a functional block diagram showing an example of functions related to data analysis among the functions realized by the data analysis device according to the first embodiment. As shown in Fig. 9, the data analysis unit 7 includes a learning data acquisition unit 701, a learning statistics process data generation unit 702, a learning statistics quality data generation unit 703, a learning data generation unit 704, a learning unit 705, a machine learning model 706, a target data acquisition unit 707, a target statistics process data generation unit 708, and an estimation unit 709.

[0058] The functions of the learning data acquisition unit 701, learning statistics process data generation unit 702, learning statistics quality data generation unit 703, learning data generation unit 704, learning unit 705, and machine learning model 706 in the data analysis unit 7 correspond to the functions of a learning device. The functions of the manufacturing process data processing unit 3 and the quality data processing unit 5 may also be included in the functions of the learning device.

[0059] The functions of the machine learning model 706, the target data acquisition unit 707, the target statistical process data generation unit 708, and the estimation unit 709 in the data analysis unit 7 correspond to the functions of an estimation device. The functions of the manufacturing process data processing unit 3 may also be included in the functions of the estimation device.

[0060] [Learning data acquisition section] The learning data acquiring unit 701 acquires lot-based process data and lot-based quality data. More specifically, the learning data acquiring unit 701 may acquire lot-based process data relating to at least one lot among the plurality of lot-based process data from the manufacturing process data processing unit 3, and may acquire lot-based quality data relating to at least the one lot among the plurality of lot-based quality data from the quality data processing unit 5.

[0061] [Learning Statistics Process Data Generation Unit] The learning statistical process data generating unit 702 generates a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data.

[0062] Then, the learning statistical process data generating unit 702 generates a plurality of mixer unit statistical process data indicating statistical information for each mixer regarding the manufacturing process based on the plurality of batch unit statistical process data.

[0063] Then, the learning statistical process data generating unit 702 generates lot-unit statistical process data indicating lot-unit statistical information related to the manufacturing process based on the plurality of kneader-unit statistical process data.

[0064] FIG. 10 is a diagram illustrating the process of generating lot-unit statistical process data according to the first embodiment. As shown in FIG. 10, the lot-unit process data 8 acquired by the learning data acquisition unit 701 includes a plurality of batch-unit process data. For convenience, FIG. 10 shows a portion of the lot-unit process data, i.e., batch-unit process data 4101 to 4103. As described in the description of the manufacturing process processing unit, the lot-unit process data 8 includes batch-unit process data for each batch obtained in batch units from each kneading machine in each kneading process while kneading one lot of rubber material. The batch-unit process data includes sampling data for each sampling time. For example, in the batch-unit process data 4101 shown in FIG. 10, each row corresponds to a piece of sampling data. The flow up to generating lot-unit statistical process data will be described below.

[0065] First, the statistical process data generating unit 702 calculates batch-unit process-related statistics, which are statistics for each batch, based on a group of process-related values, which are groups formed by collecting, for each batch, measurement values ​​that indicate the operating state of the mixer, which are indicated in each sampling data of each batch-unit process data included in the lot-unit process data. Then, the statistical process data generating unit 702 generates, for each batch, batch-unit statistical process data that indicates the batch-unit process-related statistics.

[0066] For example, the batch unit process data 4101 includes measured values ​​of the temperature of the kneading chamber at each sampling time, and the statistical process data generation unit 702 calculates batch unit process-related statistics (e.g., batch unit average value) by taking statistics of a group of the measured values. Then, the statistical process data generation unit 702 generates batch unit statistical process data 8101 indicating the batch unit process-related statistics.

[0067] Similarly, the statistical process data generation unit 702 generates each batch unit statistical process data based on each batch unit process data, such as batch unit statistical process data 8102 based on the batch unit process data 4102, batch unit statistical process data 8103 based on the batch unit process data 4103, and so on.

[0068] Next, the statistical process data generating unit 702 calculates mixer unit process related statistics, which are statistics for each mixer, based on a group formed by collecting batch unit process related statistics for each batch for each mixer.Then, the statistical process data generating unit 702 generates mixer unit statistical process data indicating the mixer unit process related statistics for each mixer.

[0069] For example, statistics of a group of batch unit average values ​​obtained by collecting batch unit average values ​​of the temperatures of the kneading chamber for each batch (batches 2101 to 2108 in FIG. 2) handled by the first kneader 211 are further taken, and kneader unit process-related statistics (e.g., kneader unit average value) for the first kneader 211 are calculated. Then, the statistical process data generation unit 702 generates kneader unit statistical process data 810 for the first kneader 211 indicating the kneader unit process-related statistics for the first kneader 211.

[0070] Similarly, the statistical process data generation unit 702 generates kneader unit statistical process data for each kneader, such as kneader unit statistical process data 820 for the second kneader 221, kneader unit statistical process data 830 for the third kneader 231, and kneader unit statistical process data 840 for the fourth kneader 241.

[0071] Thereafter, the statistical process data generating unit 702 calculates lot-unit process-related statistics, which are statistics for each lot, based on a group formed by collecting the mixer-unit process-related statistics for each lot of the mixers.Then, the statistical process data generating unit 702 generates lot-unit statistical process data for each lot that indicates the lot-unit process-related statistics.

[0072] For example, the statistics of a group of mixer-based average values ​​obtained by collecting the mixer-based average values ​​of the temperatures in the mixer chambers for each mixer (the first mixer 211 to the fourth mixer 241 in FIG. 2) are further taken to calculate lot-based process-related statistics (for example, lot-based average values). Then, the statistical process data generation unit 702 generates lot-based statistical process data 850 for lot L1 indicating the lot-based process-related statistics.

[0073] The statistical process data generation unit 702 may calculate batch-unit process-related statistics based on not only the group of measured values ​​of the temperature of the kneading chamber but also other types of numerical value groups included in the batch-unit process data (e.g., the pressure applied to the kneaded rubber, the power consumption of the kneading machine, the rotation speed of the kneading blade, etc.). The statistical process data generation unit 702 may calculate various batch-unit process-related statistics based on multiple types of numerical value groups, not limited to one type of numerical value group. The designer may decide which type of numerical value group to use, and whether to use one type of numerical value group or multiple types of numerical value groups. When multiple types of numerical value groups are used, the statistical process data generation unit 702 generates batch-unit statistical process data indicating various batch-unit process-related statistics. The same applies to kneader-unit process-related statistics and lot-unit process-related statistics.

[0074] The types of the statistics are not limited to the average value, but may be median, maximum value, minimum value, mode, range, variance, standard deviation, skewness, kurtosis, integrated value, slope, etc. The batch-unit process-related statistics, mixer-unit process-related statistics, and lot-unit process-related statistics do not all need to be the same type of statistics; they may be different. The lot-unit statistical process data 850 indicating multiple types of statistics may be generated using not only one type of statistics, but also multiple types of statistics. For example, the lot-unit statistical process data 850 indicating the average, maximum, and variance of the temperatures of the mixer chamber may be generated. The designer may decide which type of statistics to use, and whether to use one type or multiple types of statistics. When multiple types of statistics are used, the statistical process data generation unit 702 generates batch-unit statistical process data indicating various batch-unit process-related statistics. The same applies to the mixer-unit process-related statistics and the lot-unit process-related statistics.

[0075] [Learning Statistics Quality Data Generation Department] The learning statistical quality data generating unit 703 generates lot-unit statistical quality data indicating lot-unit statistical information regarding the quality of a plurality of kneading products, based on a plurality of batch-unit quality data.

[0076] The lot-unit quality data acquired by the learning data acquisition unit 701 includes multiple batch-unit quality data for one lot.

[0077] The statistical quality data generating unit 703 calculates lot-unit quality-related statistics, which are statistics for each lot, based on the quality-related numerical values, which are numerical values ​​indicating the quality of the kneaded product of the rubber material, included in each batch-unit quality data.The statistical quality data generating unit 703 then generates lot-unit statistical quality data, which indicates the lot-unit quality-related statistical values, for each lot.

[0078] For example, the batch-unit quality data includes measured values ​​of viscosity of a kneaded product of a rubber material, and the statistical quality data generating unit 703 calculates lot-unit quality-related statistics (e.g., lot-unit average values) by taking statistics of a group of the measured values. Then, the statistical quality data generating unit 703 generates lot-unit statistical quality data indicating the lot-unit quality-related statistics.

[0079] The statistical quality data generating unit 703 may calculate lot-unit quality-related statistics based on not only the group of viscosity measurement values ​​but also other types of numerical group included in the batch-unit quality data. The statistical quality data generating unit 703 may calculate various lot-unit quality-related statistics based on multiple types of numerical group, not just one type of numerical group. The designer may decide which type of numerical group to use, and whether to use one type of numerical group or multiple types of numerical groups. When multiple types of numerical groups are used, the statistical quality data generating unit 703 generates lot-unit statistical quality data indicating various lot-unit quality-related statistics.

[0080] Furthermore, the type of the statistical quantity is not limited to the average value, but may be a median, a maximum value, a minimum value, a mode, a range, a variance, a standard deviation, a skewness, a kurtosis, an integrated value, a slope, or the like. Furthermore, not only one type of statistical quantity but multiple types of statistical quantities may be used to generate lot-unit statistical quality data indicating multiple types of statistical quantities. For example, lot-unit statistical process data 850 indicating the average, maximum, and variance of the temperatures of the kneading chamber may be generated. The type of statistical quantity to be used, and whether to use one type of statistical quantity or multiple types of statistical quantities, may be determined appropriately by the designer. When multiple types of statistical quantities are used, the statistical quality data generation unit 703 generates lot-unit statistical quality data indicating various lot-unit quality-related statistics.

[0081] [Learning data generation section] The learning data generating unit 704 generates learning data including lot-based statistical process data and lot-based statistical quality data.

[0082] [Study Department] The learning unit 705 executes learning of the machine learning model based on the learning data.

[0083] [Machine learning model] The machine learning model 706 is a machine learning model that receives lot-based statistical process data as input and lot-based quality data as output.

[0084] For example, the machine learning model 706 may be trained using a plurality of training data. The training data may include, for example, training input data and teacher data. The training input data included in the training data may be, for example, lot-based statistical process data input to the machine learning model 706 during training of the machine learning model 706. The teacher data included in the training data may be, for example, lot-based statistical quality data corresponding to the lot-based statistical process data included in the training data.

[0085] The pairs of learning input data and teacher data included in the training data may be obtained during the kneading process of the rubber material.

[0086] Furthermore, an output may be identified when learning input data included in the training data is input to the machine learning model 706. Then, an error (comparison result) between the output and the teacher data included in the training data may be identified. Then, based on the identified error, the parameter values ​​of the machine learning model 706 may be updated by, for example, backpropagation.

[0087] The above process may then be performed on multiple sets of training data to allow the machine learning model 706 to learn.

[0088] The machine learning model 706 may be a machine learning model that implements machine learning such as Adaboost, random forest, neural network, support vector machine (SVM), nearest neighbor classifier, etc.

[0089] Furthermore, the machine learning model 706 does not necessarily have to be included in the data analysis device 1, and may exist outside the data analysis device 1. In this case, for example, the control unit 11 may further include a communication unit (not shown), and the learning unit 705 and the estimation unit 709 (described later) may input and output data to the external machine learning model 706 via the communication unit.

[0090] [Target data acquisition section] The target data acquisition unit 707 acquires lot-unit process data, which is a collection of a plurality of batch-unit process data each indicating a mixing state during mixing of the rubber material in batch units, on a lot-unit basis.

[0091] [Target Statistical Process Data Generation Unit] The target statistical process data generation unit 708 is similar to the learning statistical process data generation unit 702. Note that the processing content executed by the target statistical process data generation unit 708 is the same as the processing content executed by the learning statistical process data generation unit 702 in learning the machine learning model 706.

[0092] [Estimation part] The estimation unit 709 estimates the quality of the kneading product on a lot-by-lot basis based on the output when the lot-by-lot statistical process data is input to the trained machine learning model 706 (trained model).

[0093] [1-4. Summary of the first embodiment] As described above, the learning device (data analysis device 1) includes a learning data acquisition unit 701 that acquires lot-unit process data and lot-unit quality data, a learning statistical process data generation unit 702 that generates multiple batch-unit statistical process data based on each of multiple batch-unit process data included in the lot-unit process data and generates lot-unit statistical process data based on the multiple batch-unit statistical process data, a learning statistical quality data generation unit 703 that generates lot-unit statistical quality data based on the multiple batch-unit quality data, a learning data generation unit 704 that generates learning data including the lot-unit statistical process data and the lot-unit statistical quality data, and a learning unit 705 that performs training of a machine learning model 706 based on the training data. This learning device makes it easy to perform lot-unit data analysis of a rubber material mixing process.

[0094] The estimation device (data analysis device 1) also includes a target data acquisition unit 707 that acquires lot-unit process data, a target statistical process data generation unit 708 that generates a plurality of batch-unit statistical process data based on each of a plurality of batch-unit process data included in the lot-unit process data and generates lot-unit statistical process data based on the plurality of batch-unit statistical process data, and an estimation unit 709 that estimates the quality of a kneading resultant product on a lot-by-lot basis based on an output when the lot-unit statistical process data is input to a trained machine learning model 706. This estimation device makes it possible to easily perform lot-by-lot data analysis of the kneading process of a rubber material.

[0095] [2. Second Embodiment] Next, a second embodiment of the data analysis device 1 will be described. In the second embodiment, a case will be described in which the kneading period for each batch is divided into predetermined time periods and data analysis is performed. In the following description of the second embodiment, explanations of the same points as in the first embodiment will be omitted.

[0096] [Learning Statistics Process Data Generation Unit] The learning statistical process data generation unit 702 divides each of the plurality of batch unit process data into predetermined time periods according to a predetermined rule, generates a plurality of batch unit statistical process data for each time period, and generates lot unit statistical process data for each time period based on the plurality of batch unit statistical process data for each time period.

[0097] Fig. 11 is a diagram illustrating a process for generating lot unit statistical process data according to the second embodiment. As shown in Fig. 11, each batch unit process data includes sampling data indicating the operating state of the mixer at each sampling time.

[0098] 11, the learning statistical process data generation unit 702 divides each batch of process data into predetermined time periods according to a predetermined rule based on each sampling time. Here, the time period is divided into three time periods: an early time period 80a, a middle time period 80b, and a final time period 80c. However, the number of time periods into which the data is divided can be determined as appropriate by the designer. Furthermore, the manner in which the data is divided can also be determined as appropriate by the designer.

[0099] For example, the learning statistical process data generation unit 702 divides each of the multiple batch-unit process data (4101, 4102, 4103, ...) into time zones 80a, 80b, and 80c, and generates batch-unit statistical process data for time zone 80a (8101a, 8102a, 8103a, ...), batch-unit statistical process data for time zone 80b (8101b, 8102b, 8103b, ...), and batch-unit statistical process data for time zone 80c (8101c, 8102c, 8103c, ...).

[0100] Then, the learning statistical process data generating unit 702 generates kneader-unit statistical process data (810a, 820a, 830a, 840a) for the time period 80a based on the batch-unit statistical process data for the time period 80a. The same applies to the time periods 80b and 80c.

[0101] Then, the learning statistical process data generating unit 702 generates lot-unit statistical process data 850a for the time period 80a based on the mixer-unit statistical process data for the time period 80a. The same applies to the time periods 80b and 80c.

[0102] [Learning data generation section] The learning data generating unit 704 generates the learning data including the lot-unit statistical process data for each time period and the lot-unit quality data.

[0103] For example, the learning data generating unit 704 generates learning data including lot-based statistical process data 850a, lot-based statistical process data 850b, lot-based statistical process data 850c, and lot-based quality data.

[0104] Alternatively, the learning data generating unit 704 may generate learning data including lot-based statistical process data for some time periods, rather than all of the lot-based statistical process data for each time period, and lot-based quality data.

[0105] For example, the learning data generating unit 704 may generate learning data including lot-based statistical process data 850a and lot-based quality data.

[0106] Alternatively, the learning data generating unit 704 may generate learning data including the lot-based statistical process data 850a, the lot-based statistical process data 850b, and the lot-based quality data.

[0107] [Target Statistical Process Data Generation Unit] The target data statistical process data generation unit 708 is similar to the learning statistical process data generation unit 702. Note that the processing content executed by the target statistical process data generation unit 708 is the same as the processing content executed by the learning statistical process data generation unit 702 in learning the machine learning model 706.

[0108] [Estimation part] The estimation unit 709 estimates the quality of each lot of the kneading product based on the output when the lot-based statistical process data for each time period is input to the trained machine learning model 706 (trained model).

[0109] For example, the estimation unit 709 may input the lot-based statistical process data 850a, the lot-based statistical process data 850b, and the lot-based statistical process data 850c to the trained machine learning model 706 (trained model).

[0110] Alternatively, if the machine learning model 706 is trained using training data including lot-based statistical process data for a certain time period and lot-based quality data, the estimation unit 709 may input the lot-based statistical process data for that certain time period to the trained machine learning model 706 (trained model).

[0111] [3. Modifications] The present disclosure is not limited to the above-described embodiments, and can be modified as appropriate without departing from the spirit of the present disclosure.

[0112] For example, multiple mixers may be assigned to each mixing step. In this case, a timetable may be generated for each mixer in each mixing step, and the mixing period of each lot may be specified for each mixer. Mixers whose mixing periods are close to each other (for example, mixers whose mixing start dates and times are within a predetermined time) may be grouped together as a mixing period group for the same lot. In this case, in the mixing period group for a certain lot in a certain mixing step, the earliest mixing start date and time may be set as the mixing start date and time of the lot, and the latest mixing end date and time may be set as the mixing end date and time of the lot, and the mixing periods of the mixing steps may be linked together as described in the first embodiment.

[0113] Furthermore, for example, the learning statistical process data generating unit 702 may generate lot-unit process data directly based on a plurality of batch-unit statistical process data, without generating kneader-unit statistical process data. In this case, the learning statistical process data generating unit 702 may calculate lot-unit process-related statistics based on a group formed by collecting batch-unit process-related statistics for one lot. The same applies to the target statistical process data generating unit 708.

[0114] Alternatively, for example, the learning statistical process data generating unit 702 may generate lot-unit statistical process data indicating each mixer-unit process-related statistical quantity without calculating lot-unit process-related statistical quantities. The same applies to the target statistical process data generating unit 708. [Explanation of symbols]

[0115] 1 data analysis device, 11 processing device, 12 storage device, 13 display device, 14 input device, 3 manufacturing process data processing unit, 5 quality data processing unit, 7 data analysis unit, 21 first kneading process, 22 second kneading process, 23 third kneading process, 24 fourth kneading process, 211 first kneader, 221 second kneader, 231 third kneader, 241 fourth kneader, 2101 to 2109 batches, 2201 to 2211 batches, 2301 to 2312 batches, 2401 to 2417 batches, L1 first lot, L2 second lot, 311 first batch unit process data acquisition unit, 312 first lot kneading period determination unit, 321 second batch unit process data acquisition unit, 322 second lot kneading period determination unit, 331 third batch unit process data acquisition unit, 332 Third lot mixing period specification unit, 341 Fourth batch unit process data acquisition unit, 342 Fourth lot mixing period specification unit, 300 Lot unit process data generation unit, 41 Manufacturing process data, 4101 to 4103 Batch unit process data, 51 to 54 Timetable, L11 to L14 Mixing period, 501 Quality data acquisition unit, 502 Lot unit batch group specification unit, 503 Lot unit quality data generation unit, 60 Quality data, 610 Batch group, 701 Learning data acquisition unit, 702 Learning statistics process data generation unit, 703 Learning statistics quality data generation unit, 704 Learning data generation unit, 705 Learning unit, 706 Machine learning model, 707 Target data acquisition unit, 708 Target statistics process data generation unit, 709 Estimation unit, 8 Lot unit process data, 8101 to 8103 Batch unit statistical process data, 810 to 840 Statistical process data per kneader, 850 statistical process data per lot, 8101a to 8103a statistical process data per batch, 810a to 840a statistical process data per kneader, 850a statistical process data per lot, 8101b to 8103b statistical process data per batch, 810b to 840b statistical process data per kneader, 850b statistical process data per lot, 8101c to 8103c statistical process data per batch, 810c to 840c statistical process data per kneader, 850c statistical process data per lot.

Claims

1. A learning device for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading processes, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last process of the plurality of kneading processes, a data acquisition means for acquiring lot unit process data, which is a collection of a plurality of batch unit process data, each indicating a kneading state during kneading of the rubber material in a batch unit, on a lot unit basis, and lot unit quality data, which is a collection of a plurality of batch unit quality data, each indicating a quality of the kneading result of the rubber material in a batch unit, on a lot unit basis; a statistical process data generating means for generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and for generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; a statistical quality data generating means for generating lot-unit statistical quality data indicating lot-unit statistical information regarding the quality of the plurality of kneading products based on the plurality of batch-unit quality data; a training data generating means for generating training data including the lot-unit statistical process data and the lot-unit statistical quality data; a learning means for executing learning of a machine learning model based on the learning data; A learning device including:

2. the statistical process data generating means generates a plurality of kneader unit statistical process data indicating statistical information for each kneader regarding the manufacturing process based on the plurality of batch unit statistical process data, and generates the lot unit statistical process data based on the plurality of kneader unit statistical process data. The learning device according to claim 1 .

3. the statistical process data generating means divides each of the plurality of batch unit process data into predetermined time periods according to a predetermined rule, generates the plurality of batch unit statistical process data for each time period, and generates the lot unit statistical process data for each time period based on the plurality of batch unit statistical process data for each time period; the learning data generating means generates the learning data including the lot-unit statistical process data for each time period and the lot-unit quality data. The learning device according to claim 2 .

4. The learning device a first kneader assigned to a first kneading process representing one of the plurality of kneading processes, in which a plurality of lots of the rubber material are divided into a plurality of batches of the rubber material and sequentially fed, and the batches of the rubber material are sequentially kneaded; a second kneader assigned to a second kneading process, which is a process subsequent to the first kneading process, in which, after kneading by the first kneader, the rubber material of each lot is divided into a plurality of batches of the rubber material and fed to the second kneading process, and the rubber material of each batch is kneaded in order; a manufacturing process data processing means for processing the manufacturing process data acquired from each of the a first batch unit process data acquisition means for acquiring a plurality of first batch unit process data, each indicating a mixing start time and a mixing end time of each batch in the first mixer and an operating state of the first mixer during mixing, based on the manufacturing process data; a second batch unit process data acquisition means for acquiring a plurality of second batch unit process data, each indicating a mixing start time and a mixing end time of each batch in the second mixer and an operating state of the second mixer during mixing, based on the manufacturing process data; a first lot mixing period specifying means for specifying a mixing period of each of the lots in the first mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the first mixer, based on the plurality of first batch unit process data; a second lot mixing period specifying means for specifying a mixing period of each of the lots in the second mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the second mixer, based on the plurality of second batch unit process data; a lot unit process data generating means for selecting a portion of the plurality of first batch unit process data and a portion of the plurality of second batch unit process data corresponding to each lot based on a mixing period of each lot in the first mixer and a mixing period of each lot in the second mixer, and generating a plurality of lot unit process data indicating operation states of both the first mixer and the second mixer while the rubber material of each lot is being mixed; a manufacturing process data processing means including: A quality data processing means for processing quality data relating to the quality of the kneading product, quality data acquisition means for acquiring the quality data; a batch group specifying means for specifying a batch group of each lot to be kneaded by a final kneader assigned to the final kneading step, based on the kneading period of each lot in a final kneading step that indicates the last step of the plurality of kneading steps, which is specified by the manufacturing process data processing means; a lot-unit quality data generating means for dividing the quality data into lot units based on the batch group of each lot, and generating a plurality of lot-unit quality data; quality data processing means including: Further comprising: the data acquisition means acquires the lot-unit process data relating to at least one lot from the plurality of lot-unit process data from the manufacturing process data processing means, and acquires the lot-unit quality data relating to at least the one lot from the plurality of lot-unit quality data from the quality data processing means. The learning device according to any one of claims 1 to 3.

5. An estimation device for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last step of the plurality of kneading steps, a data acquisition means for acquiring lot-unit process data that is a collection of a plurality of batch-unit process data that respectively indicate a mixing state during mixing of the rubber material in batch units; a statistical process data generating means for generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and for generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; an estimation means for estimating the quality of the kneading product on a lot-by-lot basis based on an output when the lot-by-lot statistical process data is input into a trained model; An estimation device comprising:

6. the statistical process data generating means generates a plurality of kneader unit statistical process data indicating statistical information for each kneader regarding the manufacturing process based on the plurality of batch unit statistical process data, and generates the lot unit statistical process data based on the plurality of kneader unit statistical process data. The estimation device according to claim 5 .

7. the statistical process data generating means divides each of the plurality of batch unit process data into predetermined time periods according to a predetermined rule, generates the plurality of batch unit statistical process data for each time period, and generates the lot unit statistical process data for each time period based on the plurality of batch unit statistical process data for each time period; The estimation means estimates the quality of the kneading product on a lot-by-lot basis based on an output when the lot-by-lot statistical process data for each time period is input to a trained model. The estimation device according to claim 6 .

8. The estimation device includes: a first kneader assigned to a first kneading process representing one of the plurality of kneading processes, in which a plurality of lots of the rubber material are divided into a plurality of batches of the rubber material and sequentially fed, and the batches of the rubber material are sequentially kneaded; a second kneader assigned to a second kneading process, which is a process subsequent to the first kneading process, in which, after kneading by the first kneader, the rubber material of each lot is divided into a plurality of batches of the rubber material and fed to the second kneading process, and the rubber material of each batch is kneaded in order; a manufacturing process data processing means for processing the manufacturing process data acquired from each of the a first batch unit process data acquisition means for acquiring a plurality of first batch unit process data, each indicating a mixing start time and a mixing end time of each batch in the first mixer and an operating state of the first mixer during mixing, based on the manufacturing process data; a second batch unit process data acquisition means for acquiring a plurality of second batch unit process data, each indicating a mixing start time and a mixing end time of each batch in the second mixer and an operating state of the second mixer during mixing, based on the manufacturing process data; a first lot mixing period specifying means for specifying a mixing period of each of the lots in the first mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the first mixer, based on the plurality of first batch unit process data; a second lot mixing period specifying means for specifying a mixing period of each of the lots in the second mixer, including a mixing start time of the first batch and a mixing end time of the last batch of each of the lots in the second mixer, based on the plurality of second batch unit process data; a lot unit process data generating means for selecting a portion of the plurality of first batch unit process data and a portion of the plurality of second batch unit process data corresponding to each lot based on a mixing period of each lot in the first mixer and a mixing period of each lot in the second mixer, and generating a plurality of lot unit process data indicating operation states of both the first mixer and the second mixer while the rubber material of each lot is being mixed; The manufacturing process data processing means further includes: the data acquisition means acquires the lot-unit process data relating to at least one lot from the plurality of lot-unit process data from the manufacturing process data processing means; The estimation device according to any one of claims 5 to 7.

9. A method for generating a trained model for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading processes, and kneading products of each of the plurality of batches of the rubber material are obtained after the last process of the plurality of kneading processes, a data acquisition step of acquiring lot unit process data, which is a collection of a plurality of batch unit process data, each indicating a kneading state during kneading of the rubber material in a batch unit, on a lot unit basis, and lot unit quality data, which is a collection of a plurality of batch unit quality data, each indicating a quality of the kneading resultant of the rubber material in a batch unit, on a lot unit basis; a statistical process data generating step of generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; a statistical quality data generating step of generating lot-unit statistical quality data indicating lot-unit statistical information regarding the quality of the plurality of kneading products based on the plurality of batch-unit quality data; a learning data generating step of generating learning data including the lot-unit statistical process data and the lot-unit statistical quality data; a learning step of executing learning of a machine learning model based on the learning data; How to generate a trained model including:

10. 1. An estimation method for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last step of the plurality of kneading steps, a data acquisition step of acquiring lot-unit process data obtained by collecting a plurality of batch-unit process data, each of which indicates a kneading state during kneading of the rubber material in batch units, in lot units; a statistical process data generating step of generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; an estimation step of estimating the quality of the kneading product in units of lots based on an output when the lot-unit statistical process data is input into a trained model; Estimation methods including:

11. A learning program for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last step of the plurality of kneading steps, a data acquisition means for acquiring lot unit process data, which is a collection of a plurality of batch unit process data, each indicating a kneading state during kneading of the rubber material in a batch unit, in a lot unit; and lot unit quality data, which is a collection of a plurality of batch unit quality data, each indicating a quality of the kneading resultant of the rubber material in a batch unit, in a lot unit; a statistical process data generating means for generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and for generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; a statistical quality data generating means for generating lot-unit statistical quality data indicating lot-unit statistical information regarding the quality of the plurality of kneading products based on the plurality of batch-unit quality data; a training data generating means for generating training data including the lot-unit statistical process data and the lot-unit statistical quality data; a learning means for executing learning of a machine learning model based on the learning data; A learning program for making computers function as.

12. An estimation program for a manufacturing process in which a rubber material in a lot unit is divided into a plurality of batches of the rubber material and kneaded in each of a plurality of kneading steps, and kneaded products of each of the plurality of batches of the rubber material are obtained after the last step of the plurality of kneading steps, a data acquisition means for acquiring lot-unit process data that is a collection of a plurality of batch-unit process data that respectively indicate the mixing state of the rubber material during mixing of the batch unit; a statistical process data generating means for generating a plurality of batch unit statistical process data each indicating batch unit statistical information related to the manufacturing process based on each of the plurality of batch unit process data, and for generating lot unit statistical process data indicating lot unit statistical information related to the manufacturing process based on the plurality of batch unit statistical process data; an estimation means for estimating the quality of the kneading product on a lot-by-lot basis based on an output when the lot-by-lot statistical process data is input into a trained model; An estimation program for making a computer function as a.

Images