Information processing apparatus, learning model creating apparatus, information processing method, learning model creating method, and program

Abstract

To perform estimation in consideration of influence between batches.SOLUTION: The information processing device includes a manufacturing condition acquisition unit that acquires n-th manufacturing conditions that are manufacturing conditions of an n-th polymerization step that is a polymerization step of an n-th batch (n is a natural number), and a manufacturing result index estimation unit that estimates a manufacturing result index to be checked in a checking step of each of m batches (m is an integer of m ≥ 0) after the n-th batch based on a first learned model obtained by machine learning of a relationship between the manufacturing conditions of the polymerization step of a predetermined batch and the manufacturing result index to be checked in the checking step of a plurality of batches after the predetermined batch and the n-th manufacturing conditions acquired by the manufacturing condition acquisition unit.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device, a learning model creation device, an information processing method, a learning model creation method, and a program. [Background technology]

[0002] Conventionally, there are manufacturing processes for manufacturing resins and the like through multiple steps using a so-called batch production method. It would be desirable to be able to estimate the manufacturing results of such manufacturing processes under certain manufacturing conditions. For example, Patent Document 1 discloses a technology for estimating the manufacturing results. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-080701 Summary of the Invention [Problem to be solved by the invention]

[0004] In batch production, the manufacturing conditions in the manufacturing process of the previous batch may affect the manufacturing results of the next batch. In other words, in batch production, changes in the manufacturing environment between previous and next batches may affect the manufacturing results. However, in conventional technology, the influence between batches has not been fully considered.

[0005] The present disclosure has been made in consideration of these circumstances, and aims to provide an information processing device, a learning model creation device, an information processing method, a learning model creation method, and a program that can perform estimation taking into account the influence between batches. [Means for solving the problem]

[0006] One aspect of the present disclosure is an information processing device that presents manufacturing conditions for a manufacturing process for manufacturing a melt-moldable fluororesin for each batch, the manufacturing process including a polymerization process and a confirmation process for a manufacturing result index of a resin manufactured through the polymerization process, and that includes: a manufacturing condition acquisition unit that acquires nth manufacturing conditions that are manufacturing conditions for the nth polymerization process, which is the polymerization process for the nth batch (n is a natural number); a first trained model that has been machine-learned to determine the relationship between the manufacturing conditions of the polymerization process for a predetermined batch and the manufacturing result index confirmed in the confirmation process for multiple batches after the predetermined batch; and a manufacturing result index estimation unit that estimates the manufacturing result index confirmed in the confirmation process for m (m is an integer greater than or equal to 0) batches after the nth batch, based on the nth manufacturing conditions acquired by the manufacturing condition acquisition unit.

[0007] One aspect of the present disclosure is a learning model creation device including a processing unit that creates a learning model by machine learning the relationship between the production conditions of the polymerization step for a given batch and the production result indicators confirmed in the confirmation step for batches subsequent to the given batch, using information indicating the production conditions of the polymerization step for each batch as explanatory variables and the production result indicators confirmed in the confirmation step for batches subsequent to the given batch as objective variables, based on a dataset including, for a production process of producing a melt-moldable fluororesin on a batch-by-batch basis, information indicating the production conditions of the polymerization step for each batch as learning data and production result indicators confirmed in the confirmation step for batches subsequent to the given batch, as training data.

[0008] One aspect of the present disclosure is an information processing method executed by an information processing device that presents manufacturing conditions for a manufacturing process that manufactures melt-moldable fluororesin for each batch, the information processing method including a polymerization process and a confirmation process of a manufacturing result index of the resin manufactured through the polymerization process. The information processing method acquires nth manufacturing conditions that are the manufacturing conditions for the nth polymerization process, which is the polymerization process for the nth batch, and estimates manufacturing result indexes that will be confirmed in the confirmation process for each of m batches (m is an integer greater than or equal to 0) after the nth batch based on the acquired nth manufacturing conditions and a first trained model that has machine-learned the relationship between the manufacturing conditions of the polymerization process for a predetermined batch and the manufacturing result indexes that will be confirmed in the confirmation process for multiple batches after the predetermined batch.

[0009] One aspect of the present disclosure is a computer-executed learning model creation method for a manufacturing process that produces a melt-moldable fluororesin on a batch-by-batch basis, the method including a polymerization step and a confirmation step of a manufacturing result indicator of a resin produced through the polymerization step, the method including, based on a dataset including, as learning data, information indicating the manufacturing conditions of the polymerization step for each batch and manufacturing result indicators confirmed in the confirmation step for subsequent batches, as training data, the method creating a learning model by machine learning the relationship between the manufacturing conditions of the polymerization step for a given batch and the manufacturing result indicators confirmed in the confirmation step for subsequent batches, using information indicating the manufacturing conditions of the polymerization step for each batch as explanatory variables and the manufacturing result indicators confirmed in the confirmation step for subsequent batches as objective variables.

[0010] One aspect of the present disclosure is a program that causes a computer to function as the information processing device described above. One aspect of the present disclosure is a program that causes a computer to function as the aforementioned learning model creation device. [Effects of the Invention]

[0011] According to the information processing device, learning model creation device, information processing method, learning model creation method, and program disclosed herein, it is possible to obtain estimation results that take into account the influence between batches in a batch production system. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a diagram illustrating an example of a configuration of an information processing system 10 according to an embodiment of the present invention. [Figure 2] 3A to 3C are diagrams illustrating an example of a manufacturing process according to the present embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of a functional configuration of an information processing device 3 according to the present embodiment. [Figure 4] FIG. 4 is a diagram showing an example of correspondence information 400 according to the present embodiment. [Figure 5] FIG. 2 is a diagram showing an example of the flow of operations of the information processing device 3 of the present embodiment. [Figure 6] FIG. 1 is a diagram illustrating an example of a learning model creation device 200 according to the present embodiment. [Figure 7] 10 is a flowchart showing an example of the operation of the learning model creation device 200 of this embodiment. [Figure 8] FIG. 10 is a diagram illustrating an example of a functional configuration of an information processing device 31 according to a second embodiment. [Figure 9] FIG. 2 is a diagram showing an example of the flow of operations of the information processing device 31 of the present embodiment. [Figure 10] FIG. 2 is a diagram illustrating an example of a learning model creation device 250 according to the present embodiment. [Figure 11] 10 is a flowchart showing an example of the operation of the learning model creation device 250 of this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] [First embodiment] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. FIG. 1 is a diagram showing an example of the configuration of an information processing system 10 according to the present embodiment. The information processing system 10 includes terminal devices 1-1 and 1-2, servers 2-1, 2-2, and 2-3, an information processing device 3, and a terminal device 4. An example of the servers 2-1, 2-2, and 2-3 is a cloud server. Terminal devices 1-1 and 1-2 are installed in, for example, a chemical plant. Terminal device 1-1 acquires real-time data from the chemical plant and transmits it to terminal device 1-2 and server 2-1. Terminal device 1-2 acquires the real-time data transmitted by terminal device 1-1 and transmits it to server 2-2. Server 2-2 acquires the real-time data transmitted by terminal device 1-2 and transmits the acquired real-time data to server 2-3. Server 2-2 creates batch data by collecting a certain amount of the acquired real-time data and transmits it to information processing device 3. Server 2-1 acquires the real-time data transmitted by terminal device 1-1 and transmits the acquired real-time data to information processing device 3.

[0014] The information processing device 3 acquires the real-time data transmitted by the server 2-1 and the batch data transmitted by the server 2-2. The information processing device 3 constructs a trained model by machine learning the acquired batch data. The information processing device 3 performs estimation by inputting the acquired real-time data into the constructed trained model. The information processing device 3 transmits the estimation result to the terminal device 4. The terminal device 4 receives and displays the estimation result transmitted by the information processing device 3. The terminal device 4 receives and displays the real-time data transmitted by the server 2-3. Next, the manufacturing process of a chemical plant will be described.

[0015] [Manufacturing process] FIG. 2 is a diagram showing an example of a manufacturing process according to this embodiment. The manufacturing process according to this embodiment is one in which products are manufactured in so-called batches. A lot may be used instead of or together with a batch, and a semi-finished product may be used instead of or together with a finished product. In the following description, a manufacturing method in which products are manufactured in batches is also simply referred to as a batch production method. In other words, the manufacturing process according to this embodiment employs a batch production method. "Melt-moldable" means that the resin exhibits melt fluidity. "Exhibiting melt fluidity" means that there is a temperature at which the melt flow rate is 0.1 to 1000 g / 10 min at a temperature 20°C or more higher than the melting point of the resin under a load of 49 N. "Melt flow rate" means the melt mass flow rate (MFR) specified in JIS K 7210:1999 (ISO 1133:1997).

[0016] Examples of melt-moldable fluororesin include ETFE (Ethylene Fluoride), which is a copolymer containing tetrafluoroethylene (hereinafter also referred to as "TFE") units and ethylene (hereinafter also referred to as "E") units as essential units, and in which the total proportion of TFE units and E units to the total units of the melt-moldable fluororesin is 80 mol % or more. Examples include copolymers (hereinafter also referred to as "PFA") containing etrafluoroethylene, TFE units, and perfluoro(alkyl vinyl ether) (hereinafter also referred to as "PAVE") units as essential units, with the proportion of PAVE units relative to the total of TFE units and PAVE units being 0.1 mol% or more but less than 20 mol%, and with the total proportion of TFE units and PAVE units relative to the total units of melt-processable fluororesin being 80 mol% or more; copolymers (hereinafter also referred to as "FEP") containing TFE units and hexafluoropropylene (hereinafter also referred to as "HFP") units as essential units, with the total proportion of TFE units and HFP units relative to the total units of melt-processable fluororesin being 80 mol% or more; and polymers (hereinafter also referred to as "PVdF") consisting only of vinylidene fluoride (hereinafter also referred to as "VdF") units, which are essential units. Among these, ETFE, PFA, and FEP are preferred, ETFE and PFA are more preferred, and ETFE is even more preferred.

[0017] From the viewpoints of heat resistance, mechanical properties, chemical resistance, etc., ETFE, PFA, and FEP may contain units based on other monomers in addition to the above units. Examples of other monomers include fluoroalkylethylene, propylene, vinyl chloride, vinylidene chloride, vinyl fluoride, a monomer having an oxygen polar group, a monomer having a fluorine atom and a nitrile group, and a monomer having a fluorine atom and multiple vinyl groups. In the case of ETFE, other monomer units may include PAVE units, HFP units, and VdF units. In the case of PFA, other monomer units may include E units, HFP units, and VdF units. In the case of FEP, other monomer units may include E units, PAVE units, and VdF units. In the case of ETFE, it is preferable that it contains a fluoroalkylethylene unit. Examples of the fluoroalkylethylene include CH2=CH(CF2)2F, CH2=CH(CF2)3F, CH2=CH(CF2)4F, CH2=CF(CF2)3H, and CH2=CF(CF2)4H, with CH2=CH(CF2)2F and CH2=CH(CF2)4F being preferred.

[0018] As the PAVE and other monomers, the monomers described in WO 2020 / 196779 can be used. When the melt-processable fluororesin contains units based on other monomers, the ratio of the units based on other monomers to the total units of the melt-processable fluororesin is preferably 0.01 to 20.0 mol%, more preferably 0.05 to 15.0 mol%, and even more preferably 0.1 to 10.0 mol%. The melt-processable fluororesin does not have to contain units based on other monomers.

[0019] The manufacturing process of this embodiment includes a first process PC1, a second process PC2 which is a manufacturing process subsequent to the first process PC1, and a third process PC3. As an example, the first step PC1 includes a polymerization step PC11. The second step PC2 includes a granulation step PC21 and a drying step PC22. The third step PC3 includes a confirmation step PC31. In the confirmation step PC31, the quality of the resin produced through the first step PC1 and the second step PC2, such as fluidity and production efficiency, e.g., yield, is confirmed. The indicator of fluidity confirmed in the confirmation step PC31 is also referred to as the fluidity indicator. The indicator of yield confirmed in the confirmation step PC31 is also referred to as the yield.

[0020] That is, the manufacturing process of this embodiment produces resin batch by batch, and includes a first process PC1 including a polymerization process PC11, a second process PC2 which is a manufacturing process after the first process PC1, and a third process PC3 which includes a confirmation process PC31 of a fluidity indicator 442 which indicates the fluidity of the resin manufactured through the first process PC1 and the second process PC2. The first step PC1 may include other steps in addition to the polymerization step PC11. The second step PC2 may include other steps in addition to the granulation step PC21 and the drying step PC22. The third step PC3 may include other steps in addition to the confirmation step PC31.

[0021] In the following description, a manufacturing process subsequent to a certain manufacturing process will be simply referred to as a “subsequent process,” and a manufacturing process prior to a certain manufacturing process will be simply referred to as a “previous process.” For example, the second process PC2 and the third process PC3 are subsequent processes of the first process PC1. The information processing device 3 of this embodiment presents manufacturing conditions PM1 for the first step PC1 (particularly, the polymerization step PC11) among these manufacturing steps.

[0022] In the figure, of the multiple batches in the manufacturing process, the first batch BT1 to the third batch BT3 are shown as examples, and the batches after the third batch BT3 are not shown. Each of the first batch BT1 to the third batch BT3 includes a polymerization step PC11, a granulation step PC21, a drying step PC22, and a verification step PC31. The polymerization step PC11 of the first batch BT1 is also referred to as the first polymerization step PC111. Similarly, the granulation step PC21 of the first batch BT1 is also referred to as the second granulation step PC212, the drying step PC22 as the first drying step PC221, and the verification step PC31 as the third verification step PC313. The steps of the second batch BT2 and the third batch BT3 are the same as those of the first batch BT1, and therefore their explanation will be omitted.

[0023] In one example of this embodiment, the first batch BT1 to the third batch BT3 are adjacent batches on the time axis, i.e., the batch immediately following the first batch BT1 is the second batch BT2, and the batch immediately following the second batch BT2 is the third batch BT3. In the following description, batches adjacent to each other on the time axis will also be referred to as the nth batch BT(n), the n+1th batch BT(n+1), the n+2nd batch BT(n+2), etc., where n is a natural number.

[0024] Any batch after the n-th batch BT(n) will also be referred to as the m-th batch BT(m), where m is an integer greater than or equal to 0. In particular, in relation to the nth batch BT(n) and the n+1th batch BT(n+1), the nth batch BT(n) is also called the "previous batch" and the n+1st batch BT(n+1) is also called the "next batch."

[0025] The nth batch BT(n) starts at time t1S and ends at time t1E. The n+1th batch BT(n+1) starts at time t2S and ends at time t2E. The n+2nd batch BT(n+2) starts at time t3S and ends at time t3E.

[0026] The first process PC1 of the nth batch BT(n), for example, the first polymerization process PC111 shown in the figure, ends at time t11E. Between time t11E when the first polymerization process PC111 ends and time t2S when the second polymerization process PC112 starts, maintenance work is performed on the manufacturing equipment used in the polymerization process PC11, such as light cleaning, heavy cleaning, replenishing or replacing consumables, and repairing or replacing broken parts. In the following description, maintenance work on manufacturing equipment is also referred to as maintenance work.

[0027] In one example of this embodiment, the time t2S when the (n+1)th batch BT(n+1) starts is earlier than the time t1E when the nth batch BT(n) ends. In other words, in the batch production method of this embodiment, the next batch starts before the previous batch ends. In the following description, a batch production method configured so that the next batch is started before the previous batch is finished is also referred to as a pipeline method. On the other hand, a batch production method in which the next batch is started after the previous batch is finished is also called a non-pipeline method. The pipeline system has a shorter idle time for manufacturing equipment than the non-pipeline system, so the pipeline system can improve the operating efficiency of manufacturing equipment compared to the non-pipeline system.

[0028] As described above, in the confirmation process PC31 for each batch, manufacturing result indicators such as resin production efficiency, quality, etc. are confirmed. An example of resin production efficiency is yield, and an example of quality is resin fluidity (Melt Flow Rate (MFR)) and the presence or absence of contamination. In one example of this embodiment, these production result indices are dominated to a greater extent by production conditions PM1 in the first process PC1, particularly the polymerization process PC11, such as raw material index 420 and polymerization condition index 430, than by production condition indices in the second process PC2, which is the subsequent process. In other words, the production condition PM1 in the polymerization process PC11 has a dominant influence on these production result indices. Therefore, in order to stabilize or improve the yield and quality of each batch, it is desirable to feed back the production result index of the previous batch to the production condition PM1 of the polymerization step PC11 of the next batch.

[0029] On the other hand, if the above-mentioned pipeline method is adopted, the polymerization process PC11 of the next batch, for example, the second polymerization process PC112, will be started before the manufacturing result indicator is obtained in the confirmation process PC31 of the previous batch, for example, the first confirmation process PC311. 2, the time t2S when the second polymerization process PC112 of the (n+1)th batch BT(n+1) is started is earlier than the time t1E when the production result index of the nth batch BT(n) is obtained. In the example shown in the same figure, the production result index of the nth batch BT(n) is fed back to the production condition PM1 of the third polymerization process PC113, which is started at time t3S after time t1E.

[0030] In other words, when the pipeline method is adopted, the production result index confirmed in the confirmation step PC31 of the previous batch cannot be fed back to the production condition PM1 of the polymerization step PC11 of the next batch. It is preferable to be able to feed back the production result index of the previous batch to the production conditions PM1 of the polymerization step PC11 of the next batch, since this will stabilize or improve the production efficiency and quality of the next batch. The information processing device 3 of this embodiment provides a function that can feed back the production result index of the previous batch to the production condition PM1 of the polymerization process PC11 of the next batch. The specific functional configuration of the information processing device 3 will be described below.

[0031] [Functional configuration of information processing device 3] FIG. 3 is a diagram showing an example of the functional configuration of the information processing device 3 of this embodiment. The information processing device 3 includes, as its software or hardware functional units, a manufacturing condition acquisition unit 310, a manufacturing information provision unit 330, an estimation unit 331, a result acquisition unit 340, and a presentation unit 350. The estimation unit 331 includes, as its software or hardware functional units, a manufacturing result index estimation unit 332, a target value estimation unit 334, and a manufacturing condition estimation unit 336. An example of the information processing device 3 is a cloud server. Here, the explanation will be given assuming that the nth batch BT(n) is the previous batch and the (n+1)th batch BT(n+1) is the next batch.

[0032] The manufacturing condition acquisition unit 310 acquires the manufacturing conditions PM1 for the nth batch BT(n), i.e., the previous batch. In the following description, the manufacturing conditions PM1 for the nth batch BT(n) are also referred to as the nth manufacturing conditions PM1-(n). Because the manufacturing conditions PM1 for each batch are process data derived from the plant, they are preprocessed. For example, missing data is optionally or automatically complemented using a general-purpose approximation method using previous and subsequent data. In addition, defective data columns are deleted, and character string information may be encoded. The production conditions PM1 include a raw material index 420 and a polymerization condition index 430. The raw material indicators 420 include, for example, the amount of other monomers, the amount of polymerization initiator, such as perbutyl pivalate (PBPV), the amount of chain transfer agent, such as methanol, the composition of the recovered gas and the composition of the recovered solvent recovered in the granulation process PC21, and the like. The polymerization condition index 430 includes the polymerization initiation pressure, polymerization pressure, polymerization time, the total amount of essential unit monomers supplied, inert analysis values, operating conditions of the agitator of the polymerization tank, etc. The total amount of essential unit monomers supplied is, for example, the total amount of E and TFE supplied in the case of ETFE.

[0033] That is, the manufacturing condition acquisition unit 310 acquires the nth manufacturing condition PM1-(n), which is the manufacturing condition PM1 of the nth polymerization process PC11-(n), which is the polymerization process PC11 of the nth batch, from the manufacturing conditions PM1 which include the resin raw material indicator 420 and the polymerization condition indicator 430 in the polymerization process PC11.

[0034] The nth manufacturing condition PM1-(n) may be provided by a user operating an input device (not shown), or may be provided from another external device such as a manufacturing management device without the intervention of a user. That is, the manufacturing condition acquisition unit 310 may acquire the nth manufacturing condition PM1-(n) via a user, or may acquire the nth manufacturing condition PM1-(n) from another external device without the intervention of a user.

[0035] The manufacturing information providing unit 330 inputs the n-th manufacturing condition PM1-(n) acquired by the manufacturing condition acquiring unit 310 to the manufacturing result index estimating unit 332. The manufacturing result index estimation unit 332 has a first trained model 332a. The manufacturing result index estimation unit 332 acquires the nth manufacturing condition PM1-(n) from the manufacturing information provision unit 330. Based on the acquired nth manufacturing condition PM1-(n) and the first trained model 332a, the manufacturing result index estimation unit 332 estimates a manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n+1) to (n+m) subsequent to the nth batch BT(n). The manufacturing result index estimation unit 332 inputs the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of the estimated batch BT(n+1) to batch (n+m) to the target value estimation unit 334.

[0036] The first trained model 332a was created by machine learning the relationship between the manufacturing conditions for each batch and the manufacturing result indicators confirmed in the confirmation process for subsequent batches, using the information indicating the manufacturing conditions for each batch as an explanatory variable and the manufacturing result indicators confirmed in the confirmation process for subsequent batches as a target variable, based on a first training dataset that includes information indicating the manufacturing conditions for each batch as training data and manufacturing result indicators confirmed in the confirmation process for subsequent batches as training data.

[0037] An example of correspondence information indicating the relationship between the manufacturing conditions for each batch and the manufacturing result indicators confirmed in the confirmation process for the subsequent batch will be described with reference to FIG. 4 is a diagram showing an example of correspondence information 400 of this embodiment. In the correspondence information 400, a batch number 410, a raw material index 420, a polymerization condition index 430, and a production result index 440 are associated with each other for each batch. The raw material indicators 420 include a resin raw material amount 421, an initiator amount 422, and a chain transfer agent amount 423. The polymerization condition indicators 430 include a polymerization initiation pressure 431, a polymerization pressure 432, and a polymerization time 433. The raw material index 420 and the polymerization condition index 430 are examples of the production conditions PM1. The manufacturing result indicators 440 include a yield 441 and a liquidity indicator 442 .

[0038] In the following description, the production condition PM1 for which the batch number 410 is k, i.e., the kth batch, will also be referred to as the kth production condition PM1-(k). Furthermore, the production result index 440 for which the batch number 410 is k+1, i.e., the k+1th batch, will also be referred to as the k+1th production result index 440-(k+1). The liquidity index 442 for which the batch number 410 is k+1, i.e., the k+1th batch, will also be referred to as the k+1th liquidity index 442-(k+1). Returning to FIG. 3, the description will continue.

[0039] The first trained model 332a acquires a manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to the (n+m)th confirmation processes PC31-(n+m) of m batches BT(n+1) to (n+m) subsequent to the nth batch BT(n), for the nth manufacturing condition PM1-(n) input by the manufacturing result indicator estimation unit 332. The first trained model 332a outputs the manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to the (n+m)th confirmation processes PC31-(n+m) of the acquired batches BT(n+1) to (n+m) to the manufacturing result indicator estimation unit 332.

[0040] The target value estimation unit 334 has a second trained model 334a. The target value estimation unit 334 acquires from the manufacturing result indicator estimation unit 332 manufacturing result indicators 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to (n+m)th confirmation processes PC31-(n+m) of m batches BT(n+1) to (n+m). The target value estimation unit 334 estimates target values ​​of manufacturing result indicators of the (n+m+1) batch BT(n+m+1) based on the acquired manufacturing result indicators 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to (n+m)th confirmation processes PC31-(n+m) of m batches BT(n+1) to (n+m) and the second trained model 334a. The target value estimation unit 334 inputs information indicating the estimated target value of the manufacturing result index to the manufacturing condition estimation unit 336.

[0041] The second trained model 334a was created by machine learning the relationship between the manufacturing result indexes for each batch and the manufacturing result indexes for subsequent batches, using the information indicating the manufacturing result indexes confirmed in the confirmation process for each batch as an explanatory variable and the manufacturing result indexes confirmed in the confirmation process for subsequent batches as a target variable, based on a second training dataset that includes, as training data, information indicating the manufacturing result indexes confirmed in the confirmation process for each batch and the target value of the manufacturing result indexes confirmed in the confirmation process for subsequent batches.

[0042] The second trained model 334a acquires the manufacturing result index confirmed in the confirmation step of the batch (n+m+1) that comes after the batch BT(n+m), for the manufacturing result index 440 confirmed in each of the (n+1)th confirmation steps PC31-(n+1) to (n+m)th confirmation steps PC31-(n+m) of the batches BT(n+1) to (n+m) input by the target value estimation unit 334. The second trained model 334a outputs the manufacturing result index 440 confirmed in the confirmation step of the acquired batch (n+m+1) to the target value estimation unit 334.

[0043] The manufacturing condition estimation unit 336 has a third trained model 336a. The manufacturing condition estimation unit 336 acquires a manufacturing result indicator 440 confirmed in the confirmation process of batch (n+m+1) from the target value estimation unit 334. The manufacturing condition estimation unit 336 estimates the manufacturing conditions of the (n+m+1) batch BT(n+m+1) based on the acquired manufacturing result indicator 440 confirmed in the confirmation process of batch (n+m+1) and the third trained model 336a. The manufacturing condition estimation unit 336 inputs information indicating the estimated (n+m+1)th manufacturing conditions PM1-(n+m+1) of the (n+m+1) batch BT(n+m+1) to the result acquisition unit 340.

[0044] The third trained model 336a was created by machine learning the relationship between the manufacturing result indicators of a batch and the manufacturing conditions of that batch, using the information indicating the manufacturing result indicators confirmed in the batch-by-batch confirmation process as explanatory variables and the information indicating the manufacturing conditions of each batch as target variables, based on a dataset that includes, as learning data, information indicating the manufacturing result indicators confirmed in the batch-by-batch confirmation process and information indicating the manufacturing conditions of each batch as training data. The third trained model 336a acquires the manufacturing conditions PM1 of the (n+m+1) batch BT(n+m+1) for the manufacturing result index 440 confirmed in the confirmation process of the batch (n+m+1) input by the manufacturing condition estimation unit 336. The third trained model 336a outputs the acquired manufacturing conditions PM1 of the (n+m+1) batch BT(n+m+1) to the manufacturing condition estimation unit 336.

[0045] The result acquisition unit 340 acquires the manufacturing conditions PM1 output by the manufacturing condition estimation unit 336 as the estimated result of the (n+m+1) manufacturing conditions PM1-(n+m+1) of the polymerization process PC11 of the (n+m+1) batch BT(n+m+1). That is, the result acquisition unit 340 receives the nth manufacturing condition PM1-(n) of the manufacturing condition PM1 to the manufacturing result index estimation unit 332, and acquires the (n+m+1)th manufacturing condition PM1-(n+m+1) output from the manufacturing condition estimation unit 336 as the estimated result of the manufacturing condition PM1 of the polymerization process PC11 of the (n+m+1)th batch BT(n+m+1).

[0046] The presentation unit 350 outputs the estimation result acquired by the result acquisition unit 340 to a display device (not shown). The display device displays the estimation result output by the presentation unit 350. That is, the presentation unit 350 presents the estimation result acquired by the result acquisition unit 340. The flow of operations of each functional unit of the information processing device 3 described above will be described with reference to FIG.

[0047] [Operation flow of information processing device 3] FIG. 5 is a diagram showing an example of the flow of operations of the information processing device 3 of this embodiment. (Step S1-1) The manufacturing condition acquisition unit 310 acquires the manufacturing conditions PM1 of the polymerization process PC11 of the n-th batch BT(n), that is, the n-th manufacturing conditions PM1-(n). (Step S2-1) The manufacturing result index estimation unit 332 estimates the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batch BT(n+1) to batch (n+m) based on the nth manufacturing condition PM1-(n) and the first trained model 332a.

[0048] (Step S3-1) The target value estimation unit 334 estimates the target value of the manufacturing result index of the (n+m+1) batch BT(n+m+1) based on the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batches BT(n+1) to (n+m) and the second trained model 334a.

[0049] (Step S4-1) The manufacturing condition estimation unit 336 estimates the manufacturing conditions of (n+m+1) batch BT(n+m+1) of (n+m+1) batch BT(n+m+1) based on the manufacturing result index 440 confirmed in the confirmation process of (n+m+1) batch BT(n+m+1) and the third trained model 336a. (Step S5-1) The presentation unit 350 outputs the estimation result of the manufacturing condition PM1 estimated in step S4-1, that is, the (n+m+1)th manufacturing condition PM1-(n+m+1).

[0050] In the above-described embodiment, the manufacturing result index estimation unit 332 estimates the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n+1) to (n+m) after the nth batch BT(n) based on the acquired nth manufacturing condition PM1-(n) and the first trained model 332a, but this example is not limiting. For example, the manufacturing result index estimation unit 332 may be configured to estimate the manufacturing result index 440 confirmed in each of the (n)th confirmation process PC31-(n) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n) to batch (n+m) after the nth batch BT(n) based on the acquired nth manufacturing condition PM1-(n) and the first trained model 332a. In this case, the target value estimation unit 334 acquires from the manufacturing result index estimation unit 332 the manufacturing result index 440 confirmed in each of the (n)th confirmation processes PC31-(n) to (n+m)th confirmation processes PC31-(n+m) of m batches BT(n) to (n+m), and estimates the target value of the manufacturing result index of the (n+m+1) batch BT(n+m+1) based on the manufacturing result index 440 confirmed in each of the (n)th confirmation processes PC31-(n) to (n+m)th confirmation processes PC31-(n+m) of the acquired m batches BT(n) to (n+m) and the second trained model 334a.

[0051] In the above-described embodiment, the estimation unit 331 may acquire the nth manufacturing conditions PM1-(n) from the manufacturing information providing unit 330, and acquire the manufacturing conditions PM1 for the (n+m+1) batch BT(n+m+1) based on the acquired nth manufacturing conditions PM1-(n) and the learning model. In this case, the learning model is created by machine learning the manufacturing conditions for each batch and the manufacturing conditions for subsequent batches, using the information indicating the manufacturing conditions for each batch as explanatory variables and the manufacturing conditions for subsequent batches as objective variables, based on a learning dataset that includes information indicating the manufacturing conditions for each batch as learning data and the manufacturing conditions PM1 for subsequent batches as training data.

[0052] As described above, the information processing system 10 of this embodiment estimates the (n+m+1)th production conditions PM1-(n+m+1) of the (n+m+1)th batch BT(n+m+1) based on the nth production conditions PM1-(n) of the polymerization process PC11 of the nth batch BT(n) in a pipelined production process. Therefore, according to the information processing system 10 of this embodiment, the production result indicator confirmed in the confirmation process PC31 of the (n+m)th batch BT(n+m) can be fed back to the (n+m+1)th production conditions PM1-(n+m+1) of the polymerization process PC11 of the (n+m+1)th batch before the (n+m) batch is completed. The information processing system 10 configured in this manner can detect quality fluctuations in the manufactured product, for example, in the resin, at an early stage and reflect them in the (n+m+1)th batch at an early stage. In other words, the information processing system 10 can make timely recipe changes.

[0053] According to the information processing system 10, it is possible to use the results of learning a large amount of information efficiently or with high accuracy through machine learning, and therefore it is possible to make the fluidity index 442 for the next batch, for example, the manufacturing result index 440 such as MFR, more stable or closer to the target value. The information processing system 10 also includes the amount 423 of the chain transfer agent as a manufacturing condition PM1. When methanol is used as the chain transfer agent in the resin polymerization step PC11, a slight change in the concentration of methanol can have a significant effect on the flowability index 442, for example, the manufacturing result index 440 such as MFR. In such a case, the production conditions PM1 in the polymerization step PC11, for example, the amount of methanol as a chain transfer agent, have a greater effect on the fluidity index 442, for example, MFR, than various production conditions in the subsequent steps.

[0054] According to the information processing system 10 of this embodiment, by estimating the (n+m+1)th production condition PM1-(n+m+1) taking into account the amount 423 of the chain transfer agent, the production result indicators 440, such as the fluidity indicator 442, confirmed in the confirmation process of the batch (n+m+1), for example, the MFR, can be made more stable or closer to the target value. Preferred examples of the chain transfer agent include alcohols such as methanol, ethanol, 2,2,2-trifluoroethanol, 2,2,3,3-tetrafluoropropanol, 1,1,1,3,3,3-hexafluoroisopropanol, and 2,2,3,3,3-pentafluoroisopropanol; hydrocarbons such as n-pentane, n-hexane, and cyclohexane; and hydrofluorocarbons such as CFH.

[0055] [Variations] The information processing device 3 may estimate the (n+m+1)th production condition PM1-(n+m+1) taking into account the maintenance index PM2. Here, the maintenance index PM2 is an index that indicates the maintenance status of manufacturing equipment. For example, the maintenance index PM2 includes the implementation status of maintenance activities such as equipment cleaning, such as the number of batches after heavy cleaning work, and the replacement status of consumables such as packing, filters, belts, valves, desiccants, and oil, such as the number of batches after valve and packing replacement.

[0056] That is, the manufacturing conditions PM1 include a maintenance index PM2 that indicates the maintenance status of the manufacturing equipment. In this case, the first trained model 332a is created by machine learning the relationship between the manufacturing conditions including the maintenance index PM2 for each batch and the manufacturing result index confirmed in the confirmation process of the batch after the batch. The third trained model 336a is created by machine learning the relationship between the manufacturing result index of the batch and the manufacturing conditions including the maintenance index PM2 for that batch. The manufacturing condition acquisition unit 310 acquires the nth manufacturing condition PM1-(n) including the maintenance index PM2. The manufacturing information provision unit 330 inputs the nth manufacturing condition PM1-(n) including the maintenance index PM2 to the manufacturing result index estimation unit 332. That is, the manufacturing information provision unit 330 inputs the nth manufacturing condition PM1-(n) taking the maintenance index PM2 into consideration to the manufacturing result index estimation unit 332.

[0057] The manufacturing result index estimation unit 332 estimates the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batch BT(n+1) to batch (n+m) based on the nth manufacturing condition PM1-(n) including the maintenance index PM2 acquired from the manufacturing information provision unit 330 and the first trained model 332a. The target value estimation unit 334 estimates the target value of the manufacturing result index of the (n+m+1) batch BT(n+m+1) based on the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batches BT(n+1) to (n+m) obtained from the manufacturing result index estimation unit 332 and the second trained model 334a.

[0058] The manufacturing condition estimation unit 336 estimates the (n+m+1)th manufacturing condition PM1-(n+m+1) that includes the maintenance indicator PM2 of the (n+m+1) batch BT(n+m+1) based on the manufacturing result indicator 440 confirmed in the confirmation process of batch (n+m+1) obtained from the target value estimation unit 334 and the third trained model 336a. As a result, the result acquisition unit 340 acquires the estimation result of the (n+m+1)th production condition PM1-(n+m+1) taking into account the maintenance index PM2 of the (n+m+1) batch BT(n+m+1). The presentation unit 350 outputs the estimation result of the (n+m+1)th production condition PM1-(n+m+1) taking into account the maintenance index PM2.

[0059] Generally, when maintenance of manufacturing equipment is performed, such as deep cleaning or replacement of consumables, the manufacturing environment may change significantly between the previous batch and the next batch, or the manufacturing environment may change discontinuously. Furthermore, the manufacturing environment may change gradually with each batch since the maintenance of the manufacturing equipment. The information processing system 10 configured as in this modified example can obtain an estimated result of the manufacturing condition PM1 that takes into account the maintenance status of the manufacturing equipment. Therefore, even if the manufacturing environment changes due to maintenance of the manufacturing equipment, the information processing system 10 can stabilize the fluidity index 442 (e.g., MFR) for the next batch or make it close to the target value.

[0060] [Learning model creation device] A method for creating a trained model will be described. The trained model is created by a training model creation device. That is, the training model creation device creates the trained model. Note that the information processing device 3 may include the training model creation device. That is, the information processing device 3 may create the trained model.

[0061] 6 is a diagram showing an example of a learning model creation device 200 according to this embodiment. The learning model creation device 200 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The learning model creation device 200 creates a first trained model 332a by machine learning the relationship between the manufacturing conditions for each batch and the manufacturing result indicators confirmed in the confirmation process of subsequent batches, based on the first training dataset, with information indicating the manufacturing conditions for each batch as explanatory variables and manufacturing result indicators confirmed in the confirmation process of subsequent batches as objective variables. The first training dataset includes information indicating the manufacturing conditions for each batch as training data and includes manufacturing result indicators confirmed in the confirmation process of subsequent batches as training data.

[0062] The learning model creation device 200 creates a second trained model 334a by machine learning the relationship between the manufacturing result index for each batch and the manufacturing result index for subsequent batches, based on the second training dataset, using information indicating the manufacturing result index confirmed in the confirmation process for each batch as an explanatory variable and the manufacturing result index confirmed in the confirmation process for subsequent batches as a target variable. The second training dataset includes information indicating the manufacturing result index confirmed in the confirmation process for each batch as training data, and includes target values ​​of the manufacturing result index confirmed in the confirmation process for subsequent batches as training data.

[0063] The learning model creation device 200 creates a third trained model 336a by machine learning the relationship between the manufacturing result indicators of a batch and the manufacturing conditions of that batch, based on the third training dataset, using information indicating the manufacturing result indicators confirmed in the confirmation process for each batch as explanatory variables and information indicating the manufacturing conditions for that batch as target variables. The third training dataset includes information indicating the manufacturing result indicators confirmed in the confirmation process for each batch as training data, and information indicating the manufacturing conditions for that batch as training data.

[0064] Furthermore, the learning model creation device 200 may create a learning model by machine learning the manufacturing conditions for each batch and the manufacturing conditions for subsequent batches, based on the learning dataset, with information indicating the manufacturing conditions for each batch as explanatory variables and the manufacturing conditions for subsequent batches as objective variables. The learning dataset includes information indicating the manufacturing conditions for each batch as learning data, and includes the manufacturing conditions PM1 for subsequent batches as training data.

[0065] The information indicating the manufacturing result indicators and the information indicating the manufacturing conditions for each batch, which are confirmed in the confirmation process for each batch, are process data originating from the plant, and therefore are pre-processed. For example, missing data is optionally or automatically complemented using a general-purpose approximation method using previous and subsequent data. Also, defective data columns are deleted. Character string information may be encoded.

[0066] That is, the learning model creation device 200 trains learning models such as a first learning model, a second learning model, a third learning model, and a learning model using training data such as a first learning dataset, a second learning dataset, a third learning dataset, and a learning dataset, and creates learned models. Here, the first learning model is a model that serves as the basis for the first trained model, the second learning model is a model that serves as the basis for the second trained model, the third learning model is a model that serves as the basis for the third trained model, and the learning model is a model that serves as the basis for the trained models.

[0067] For example, the learning model creation device 200 constructs a learned model using algorithms such as a convolution neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM), a random forest, a support vector machine (SVM), or a neural network. An input sample is data that is input to an input layer when training a learning model. An output sample is data such as training data that serves as a correct answer for comparison with an output value output from an output layer when training a learning model.

[0068] The learning model creation device 200 includes an input unit 202 , a receiving unit 204 , a processing unit 206 , an output unit 208 , and a storage unit 210 . The input unit 202 inputs information. As an example, the input unit 202 may have an operation unit such as a keyboard and a mouse. In this case, the input unit 202 inputs information according to an operation performed by a user on the operation unit. As another example, the input unit 202 may input information from an external device. The external device may be, for example, a portable storage medium. Training datasets such as a first training dataset, a second training dataset, and a third training dataset are input to the input unit 202.

[0069] The receiving unit 204 acquires a training dataset from the input unit 202 and accepts the acquired training dataset. The training dataset includes input samples and output samples, and the input samples and output samples are paired. The training dataset is made up of a plurality of pairs.

[0070] Processing unit 206 includes learning model 207, which includes a first learning model, a second learning model, and a third learning model. For all pairs of the first learning dataset, processing unit 206 inputs an input sample to the input layer of the first learning model, calculates the error between the output value output from the output layer and the output sample corresponding to the input sample, and changes the parameters of the first learning model so that the error is as small as possible. For all pairs of the second learning dataset, the processing unit 206 inputs an input sample into the input layer of the second learning model, calculates the error between the output value output from the output layer and the output sample corresponding to the input sample, and changes the parameters of the second learning model so that the error is as small as possible. For all pairs of the third learning dataset, the processing unit 206 inputs an input sample into the input layer of the third learning model, calculates the error between the output value output from the output layer and the output sample corresponding to the input sample, and changes the parameters of the third learning model so that the error is as small as possible.

[0071] The processing unit 206 creates a first trained model, a second trained model, and a third trained model by changing the parameters of the first training model, the second training model, and the third training model. Here, the output sample is an example of training data. The training model 207 is trained by changing the parameters of the training model 207. The first trained model, the second trained model, and the third trained model created as described above are received by the information processing device 3 from the output unit 208 via a network or a medium, and the manufacturing result index estimation unit 332 acquires the first trained model, the target value estimation unit 334 acquires the second trained model, and the manufacturing condition estimation unit 336 acquires the third trained model. The trained models are received by the information processing device 3 from the output unit 208 via a network or a medium, and are acquired by the estimation unit 331. When the learning model creation device 200 is included in the information processing device 3, the manufacturing result index estimation unit 332 may acquire a first trained model from the learning model creation device 200, the target value estimation unit 334 may acquire a second trained model from the learning model creation device 200, and the manufacturing condition estimation unit 336 may acquire a third trained model from the learning model creation device 200. In addition, the estimation unit 331 may acquire a trained model from the learning model creation device 200.

[0072] All or part of the input unit 202, the reception unit 204, the processing unit 206 and the output unit 208 are functional units (hereinafter referred to as software functional units) that are realized, for example, by a processor such as a CPU executing a program stored in the memory unit 210. All or part of the input unit 202, the reception unit 204, the processing unit 206, and the output unit 208 may be realized by hardware such as an LSI, an ASIC, or an FPGA, or may be realized by a combination of a software function unit and hardware, or may be realized by a cloud-based processing device.

[0073] (Operation of the learning model creation device 200) FIG. 7 is a flowchart showing an example of the operation of the learning model creation device 200 of this embodiment. (Step S1-2) The input unit 202 acquires a learning dataset. (Step S2-2) The receiving unit 204 acquires the training data set from the input unit 202 and accepts the acquired training data set.

[0074] (Step S3-2) The processing unit 206 acquires the learning dataset from the receiving unit 204. The processing unit 206 trains the parameters of the learning model 207 based on the acquired learning dataset, and creates the learning model. (Step S4-2) The output unit 208 acquires the learning model 207 from the processing unit 206. The output unit 208 outputs the acquired learning model 207 as a trained model.

[0075] The learning method for creating a trained model is not limited to supervised learning, but may be any of unsupervised learning, semi-supervised learning, reinforcement learning, or deep learning, or may be a combination of these learning methods; any learning method for machine learning is acceptable. Various learning approaches can also be adopted. Regarding the method of providing learning data, either batch learning or online learning can be applied. In batch learning, a large amount of learning data is processed at once and the learning model is updated all at once. On the other hand, in online learning, data is provided sequentially and the learning model is updated sequentially. Because each method provides optimal results under different circumstances and data environments, their use is determined by the situation. The learning model creation device 200 may incorporate an automatic learning model selection algorithm or automatically generate the optimal learning model each time learning is performed. Appropriate combinations of these methods include automatically selecting the optimal network structure for the data using network architecture search (NAS) and searching for the optimal model using meta-learning. Bayesian optimization, such as hyperparameter optimization and multi-armed bandit algorithms, mechanically selects the optimal model and automates the entire learning process. This allows the maintenance of the optimal model even in dynamic data environments.

[0076] In the above-described embodiment, a case has been described in which the manufacturing conditions for each batch and the manufacturing result index are extracted as features. For example, a case has been described in which an index of the resin raw material, an index of the resin polymerization conditions, a maintenance index showing the maintenance status of the manufacturing equipment, a manufacturing result index, etc. are extracted as features. In addition to these, feature generation using engineering methods, mechanical feature generation methods, and learning models and feature generation using images and videos can also be applied. Specifically, information other than that derived from the plant may be added as a feature. Examples of information other than that derived from the plant include information on the raw material lot and the raw material composition. Additionally, features may be generated using an engineering method that uses domain knowledge, and mechanical feature generation methods such as principal component analysis, clustering, natural language processing, dimensional processing, transfer learning, decomposition, Fourier transform, cross features, lag features, graph-based features, and integration with external data may be used to generate features.

[0077] In addition to selection based on domain knowledge, explanatory variables may be automatically selected by mechanically selecting features using correlation coefficients or machine learning models. General-purpose selection methods, such as filter methods, wrapper methods, embedding methods, dimensionality reduction methods, combination methods, driven selection, mutual information, and cross-validation, may be used as feature selection methods. In this embodiment, as an example, selection is performed using features calculated by a random forest, which is one of the embedding methods. Furthermore, feature selection may be updated to optimal features as needed depending on the situation by performing a schedule at any time or periodically.

[0078] Furthermore, the learning model creation device 200 may create a learning model as follows: The learning model creation device 200 predicts a production result index 440, such as a fluidity index, for the (n+m)th batch based on the learning results for n batches. The learning model creation device 200 assumes the predicted production result index 440 to be an actual value and adds it to learning, thereby predicting the production result index 440 for the (n+m+1)th batch. The learning model creation device 200 assumes the predicted manufacturing result index 440 of the (n+m+1)th batch to be an actual value and adds it to learning to estimate the process recipe for the (n+m+2)th batch. Hyperparameter tuning, Bayesian optimization, genetic algorithm, reinforcement learning, simulated annealing, neural architecture search, etc. may be applied.

[0079] The learning model may be evaluated using the mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), coefficient of determination (R-squared), root mean squared error (RMSE), logarithmic mean squared error (RMSLE), F1 score, accuracy, precision, recall, receiver operator characteristic (ROC) curve, area under the ROC curve (AUC), Cohen's Kappa, mean log loss, mean squared log error (MSLE), mean squared error (MPE), etc. between the results inferred using the learning model and the actual measured values, and accuracy verification may be performed by confirming that these indices converge, optimize, or improve.

[0080] As described above, the learning model creation device 200 of this embodiment creates the first learning model, the second learning model, and the third learning model based on the first learning dataset, the second learning dataset, and the third learning dataset, respectively. Also, the learning model creation device 200 creates the learning model based on the training datasets. By configuring in this manner, the learning model creation device 200 can create a learned model. Therefore, by using the learned model created by the learning model creation device 200, the information processing device 3 can shorten the time required to determine the (n+m+1)th manufacturing condition PM1-(n+m+1) for the (n+m+1) batch BT(n+m+1) and can improve the accuracy of determining the (n+m+1)th manufacturing condition PM1-(n+m+1) for the (n+m+1) batch BT(n+m+1).

[0081] [Second embodiment] This embodiment differs from the first embodiment in that an information processing device 31 is provided instead of the information processing device 3 in the first embodiment described above, or in addition to the information processing device 3. The other configurations are the same as those in the first embodiment, and therefore description thereof will be omitted. FIG. 8 is a diagram showing an example of the functional configuration of an information processing device 31 according to the second embodiment. The information processing device 31 has, as its functional parts, a manufacturing condition acquisition part 311, a first manufacturing information provision part 321, a second manufacturing information provision part 322, a manufacturing result index estimation part 333, a manufacturing condition estimation part 337, a first result acquisition part 341, a second result acquisition part 342, an index presentation part 351, and a result presentation part 352.

[0082] The manufacturing condition acquisition unit 311 acquires the n-th manufacturing condition PM1-(n), which is the manufacturing condition PM1 of the n-th polymerization step PC11-(n). The first manufacturing information providing unit 321 inputs the n-th manufacturing condition PM1-(n) acquired by the manufacturing condition acquiring unit 311 to the manufacturing result index estimating unit 333 as manufacturing information. The manufacturing result index estimation unit 333 has a fourth trained model 333a. The manufacturing result index estimation unit 333 acquires the nth manufacturing condition PM1-(n) from the first manufacturing information provision unit 321. Based on the acquired nth manufacturing condition PM1-(n) and the fourth trained model 333a, the manufacturing result index estimation unit 333 estimates a manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of m (m is a natural number) batches BT(n) subsequent to the nth batch BT(n).

[0083] The manufacturing result index estimation unit 333 inputs the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of the estimated batch BT(n+1) to batch (n+m) to the first result acquisition unit 341. The fourth trained model 333a was created by machine learning the relationship between the manufacturing conditions for each batch and the manufacturing result indicators confirmed in the confirmation process for batches subsequent to that batch, using the information indicating the manufacturing conditions for each batch as an explanatory variable and the manufacturing result indicators confirmed in the confirmation process for batches subsequent to that batch as a target variable, based on a fourth training dataset that includes information indicating the manufacturing conditions for each batch as training data and manufacturing result indicators confirmed in the confirmation process for batches subsequent to that batch as training data.

[0084] The fourth trained model 333a acquires a manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n+1) to (n+m) subsequent to the nth batch BT(n), for the nth manufacturing condition PM1-(n) input by the manufacturing result indicator estimation unit 333. The fourth trained model 333a outputs the manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of the acquired m batches BT(n+1) to (n+m), to the manufacturing result indicator estimation unit 333.

[0085] The first result acquisition unit 341 acquires from the manufacturing result index estimation unit 333 the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m), which are manufacturing result indexes 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to the (n+m)th confirmation processes PC31-(n+m) of batch BT(n+1) to batch (n+m). The index presentation unit 351 presents the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m) acquired by the first result acquisition unit 341 by outputting them to a display device (not shown).

[0086] The second manufacturing information providing unit 322 inputs the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m) acquired by the first result acquisition unit 341 and the nth manufacturing condition PM1-(n) acquired by the manufacturing condition acquisition unit 311 to the manufacturing condition estimation unit 337. The manufacturing condition estimation unit 337 has a fifth trained model 337a. The manufacturing condition estimation unit 337 acquires the n-th manufacturing condition PM1-(n) and the (n+1)-th manufacturing result index 440-(n+1) to the (n+m)-th manufacturing result index 440-(n+m) from the second manufacturing information provision unit 322. The manufacturing condition estimation unit 337 estimates the manufacturing conditions PM1 of the (n+m+1) batch, i.e., the (n+m+1)th manufacturing conditions PM1-(n+m+1), based on the acquired nth manufacturing conditions PM1-(n) and the (n+1)th to (n+m)th manufacturing result indices 440-(n+1) to 440-(n+m), and the fifth trained model 337a. The manufacturing condition estimation unit 337 inputs information indicating the estimated (n+m+1)th manufacturing conditions PM1-(n+m+1) of the (n+m+1) batch BT(n+m+1) to the second result acquisition unit 342.

[0087] The fifth trained model 337a is based on a dataset including, as learning data, information indicating the n-th manufacturing condition PM1-(n) and the manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to the (n+m)th confirmation processes PC31-(n+m) of batches BT(n+1) to (n+m) subsequent to the n-th batch BT(n), and information indicating the (n+m+1)-th manufacturing condition PM1-(n+m+1) is included as training data. The manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation steps PC31-(n+1) to (n+m) of the subsequent batches BT(n+1) to (n+m) is used as an explanatory variable, and the (n+m+1)th manufacturing condition PM1-(n+m+1) is used as a target variable, and the relationship between the manufacturing condition PM1 of the batch and the manufacturing result indicator confirmed in the confirmation step after that batch, and the manufacturing conditions of the batch after the batch in which the manufacturing result indicator is confirmed is created by machine learning. The fifth trained model 337a acquires the manufacturing conditions of the (n+m+1) batch BT(n+m+1) based on the nth manufacturing conditions PM1-(n) input by the manufacturing condition estimation unit 337 and the manufacturing result indicators 440 confirmed in each of the (n+1)th confirmation steps PC31-(n+1) to (n+m) of batches BT(n+1) to (n+m) after the nth batch BT(n). The fifth trained model 337a outputs the acquired (n+m+1)th manufacturing conditions PM1-(n+m+1) of the (n+m+1) batch BT(n+m+1) to the manufacturing condition estimation unit 337.

[0088] The second result acquisition unit 342 acquires the (n+m+1)th manufacturing condition PM1-(n+m+1) output from the manufacturing condition estimation unit 337 as the estimated result of the manufacturing condition PM1 of the polymerization process PC11 of the (n+m+1)th batch BT(n+m+1). The result presenting unit 352 presents the estimation result acquired by the second result acquiring unit 342 by outputting it to a display device (not shown). The flow of operations of each functional unit of the information processing device 31 described above will be described with reference to FIG.

[0089] [Operation flow of information processing device 31] FIG. 9 is a diagram showing an example of the flow of operations of the information processing device 31 of this embodiment. (Step S1-3) The manufacturing condition acquisition unit 311 acquires the manufacturing conditions PM1 of the polymerization process PC11 of the nth batch BT(n), that is, the nth manufacturing conditions PM1-(n). (Step S2-3) The first manufacturing information providing unit 321 inputs the n-th manufacturing condition PM1-(n) acquired in step S1-3 to the manufacturing result index estimating unit 333. The manufacturing result index estimation unit 333 inputs the nth manufacturing condition PM1-(n) obtained from the first manufacturing information provision unit 321 into the fourth trained model 333a, and obtains the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batch BT(n+1) to batch (n+m) output by the fourth trained model 333a.

[0090] (Step S3-3) The first result acquisition unit 341 acquires the manufacturing result index 440 from the manufacturing result index estimation unit 333 as the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m), which are the manufacturing result indexes 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to the (n+m)th confirmation processes PC31-(n+m) of batch BT(n+1) to batch (n+m). (Step S4-3) The second manufacturing information providing unit 322 inputs the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m) acquired by the first result acquiring unit 341 and the nth manufacturing condition PM1-(n) acquired by the manufacturing condition acquiring unit 311 to the manufacturing condition estimating unit 337. The manufacturing condition estimating unit 337 acquires the nth manufacturing condition PM1-(n) and the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m) from the second manufacturing information providing unit 322. The manufacturing condition estimation unit 337 inputs the acquired nth manufacturing condition PM1-(n) and the (n+1)th manufacturing result index 440-(n+1) to the (n+m)th manufacturing result index 440-(n+m) into the fifth trained model 337a, and acquires the (n+m+1)th manufacturing condition PM1-(n+m+1) output by the fifth trained model 337a.

[0091] (Step S5-3) The second result acquisition unit 342 acquires the (n+m+1)th production condition PM1-(n+m+1) output from the production condition estimation unit 337 as the estimated result of the production condition PM1 of the polymerization process PC11 of the (n+m+1)th batch BT(n+m+1). (Step S6-3) The result presenting unit 352 presents the estimation result acquired by the second result acquiring unit 342 by outputting it to a display device (not shown).

[0092] In the above-described embodiment, the manufacturing result index estimation unit 333 estimates the manufacturing result index 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n+1) to (n+m) after the nth batch BT(n) based on the acquired nth manufacturing condition PM1-(n) and the fourth trained model 333a, but this example is not limiting. For example, the manufacturing result index estimation unit 333 may be configured to estimate the manufacturing result index 440 confirmed in each of the (n)th confirmation process PC31-(n) to the (n+m)th confirmation process PC31-(n+m) of m batches BT(n) to batch (n+m) after the nth batch BT(n) based on the acquired nth manufacturing condition PM1-(n) and the fourth trained model 333a. In this case, the manufacturing condition estimation unit 337 acquires the (n)th manufacturing result index 440-(n) to the (n+m)th manufacturing result index 440-(n+m) from the second manufacturing information providing unit 322, and estimates the manufacturing condition PM1 of the (n+m+1) batch, i.e., the (n+m+1)th manufacturing condition PM1-(n+m+1), based on the acquired (n)th manufacturing result index 440-(n) to the (n+m)th manufacturing result index 440-(n+m) and the fifth trained model 337a.

[0093] As described above, the information processing system 10 of this embodiment includes an information processing device 31. In a pipeline manufacturing process, the information processing device 31 estimates a manufacturing result indicator 440 to be confirmed in each of the (n+1)th confirmation steps PC31-(n+1) to the (n+m)th confirmation steps PC31-(n+m) of batches BT(n+1) to (n+m) based on the nth manufacturing conditions PM1-(n) of the polymerization step PC11 of the nth batch BT(n). In other words, after the polymerization step PC11 of the nth batch BT(n) is completed and before the batch is completed, for example, before the second step PC2 is started, the information processing device 31 can estimate a manufacturing result indicator 440 to be confirmed in each of the (n+1)th confirmation steps PC31-(n+1) to the (n+m)th confirmation step PC31-(n+m) of batches BT(n+1) to (n+m). Therefore, according to the information processing system 10 of this embodiment, the manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batch BT(n+1) to batch (n+m) can be provided at a stage before the (n+m)th batch is completed.

[0094] In addition, the information processing device 31 of this embodiment estimates the (n+m+1)th manufacturing condition PM1-(n+m+1) of batch (n+m+1) based on the manufacturing result indicator 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of the estimated batches BT(n+1) to (n+m). In other words, the information processing device 31 estimates the manufacturing result index 440 to be confirmed in the (n+m+1)th confirmation process PC31-(n+m+1) of batch BT(n+m+1) based on the manufacturing result index 440 confirmed in each of the (n+1)th confirmation processes PC31-(n+1) to (n+m) of batches BT(n+1) to (n+m), and estimates the (n+m+1)th manufacturing condition PM1-(n+m+1) of batch (n+m+1) based on the estimated manufacturing result index.

[0095] Therefore, according to the information processing device 31 of this embodiment, the manufacturing result indicator confirmed in the confirmation process PC31 of the (n+m)th batch BT(n+m) can be fed back to the (n+m+1)th manufacturing conditions PM1-(n+m+1) of the polymerization process PC11 of the (n+m+1)th batch before the (n+m)th batch is completed. The information processing system 10 configured in this manner can detect quality fluctuations in manufactured products (e.g., resin) early and reflect them in the next batch promptly. In other words, the information processing system 10 allows recipe changes to be made in a timely manner.

[0096] The information processing device 31 of this embodiment also includes an index presenting unit 351. The index presenting unit 351 presents manufacturing result indexes 440 confirmed in each of the (n+1)th confirmation process PC31-(n+1) to the (n+m)th confirmation process PC31-(n+m) of batches BT(n+1) to BT(n+m) that the information processing device 31 is using for estimation. According to the information processing system 10 configured in this manner, it is possible to present to the user the estimated production result index 440 at the time of the yield of the previous batch at the end of the polymerization process PC11 of the previous batch. Therefore, it is possible to quickly detect fluctuations in the quality of the product being manufactured (e.g., resin) and reflect them in the next batch. In other words, the information processing system 10 allows timely recipe changes.

[0097] [Learning model creation device] A method for creating a trained model will be described. The trained model is created by a training model creation device. That is, the training model creation device creates the trained model. Note that the information processing device 31 may include the training model creation device. That is, the information processing device 31 may create the trained model.

[0098] 10 is a diagram showing an example of a learning model creation device 250 according to this embodiment. The learning model creation device 250 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The learning model creation device 250 creates a fourth trained model 333a by machine learning the relationship between the manufacturing conditions for each batch and the manufacturing result indicators confirmed in the confirmation process of subsequent batches, using information indicating the manufacturing conditions for each batch as explanatory variables and manufacturing result indicators confirmed in the confirmation process of subsequent batches as target variables, based on the fourth training dataset. The fourth training dataset includes information indicating the manufacturing conditions for each batch as training data, and includes manufacturing result indicators confirmed in the confirmation process of subsequent batches as training data.

[0099] The learning model creation device 250 creates a fifth trained model 337a by machine learning the relationship between the information indicating the manufacturing conditions for each batch, the manufacturing result indicators for each subsequent batch, and the manufacturing conditions for the batch after the batch for which the manufacturing result indicator is confirmed, using information indicating the manufacturing conditions for each batch and information indicating the manufacturing result indicators confirmed in the confirmation process for each subsequent batch as explanatory variables and information indicating the manufacturing conditions for the batch after the batch for which the manufacturing result indicator is confirmed as the objective variable, based on the fifth training dataset. The fifth training dataset includes, as training data, information indicating the manufacturing conditions for each batch and information indicating the manufacturing result indicators confirmed in the confirmation process for each subsequent batch, and includes, as training data, information indicating the manufacturing conditions for the batch after the batch for which the manufacturing result indicator is confirmed. The information indicating the manufacturing conditions for each batch, the information indicating the manufacturing result indicators confirmed in the batch confirmation process, and the information indicating the manufacturing conditions for each batch are process data originating from the plant, and therefore undergo preprocessing. For example, missing data is optionally or automatically complemented using a general-purpose approximation method using the preceding and following data. In addition, defective data columns are deleted, and character string information is encoded.

[0100] That is, the learning model creation device 250 trains learning models such as the fourth learning model and the fifth learning model using training data such as the fourth learning dataset and the fifth learning dataset, and creates learned models. Here, the fourth learning model is a model that serves as the basis for the fourth trained model, and the fifth learning model is a model that serves as the basis for the fifth trained model.

[0101] For example, the learning model creation device 250 constructs a learned model using algorithms such as CNN, RNN, LSTM, random forest, SVM, and neural network. An input sample is data input to an input layer when training a learning model. An output sample is data such as training data that serves as a correct answer for comparison with an output value output from an output layer when training a learning model.

[0102] All or part of the input unit 252, the reception unit 254, the processing unit 256 and the output unit 258 are functional units (hereinafter referred to as software functional units) that are realized, for example, by a processor such as a CPU executing a program stored in the memory unit 260. In addition, all or part of the input unit 252, the reception unit 254, the processing unit 256, and the output unit 258 may be realized by hardware such as an LSI, an ASIC, or an FPGA, or may be realized by a combination of software functional units and hardware.

[0103] (Operation of the learning model creation device 250) FIG. 11 is a flowchart showing an example of the operation of the learning model creation device 250 of this embodiment. (Step S1-4) The input unit 252 acquires a learning dataset. (Step S2-4) The receiving unit 254 acquires the training data set from the input unit 252 and accepts the acquired training data set.

[0104] (Step S3-4) The processing unit 256 acquires the learning dataset from the receiving unit 254. The processing unit 256 trains the parameters of the learning model 257 based on the acquired learning dataset, and creates the learning model. (Step S4-4) The output unit 258 acquires the learning model 257 from the processing unit 256. The output unit 258 outputs the acquired learning model 257 as a trained model.

[0105] The learning method for creating a trained model is not limited to supervised learning, but may be any of unsupervised learning, semi-supervised learning, reinforcement learning, or deep learning, or may be a combination of these learning methods; any learning method for machine learning is acceptable. Various learning approaches can also be adopted. Regarding the method of providing training data, either batch learning or online learning can be applied. In batch learning, a large amount of training data is processed at once and the training model is updated all at once. On the other hand, in online learning, data is provided sequentially and the training model is updated sequentially. Because each provides optimal results under different circumstances and data environments, their use is determined by the situation. The learning model creation device 250 may combine an automatic learning model selection algorithm or automatically generate the optimal learning model each time training is performed. Appropriate combinations of these algorithms include, for example, automatically selecting the optimal network structure for the data using network architecture search (NAS) or searching for the optimal model using meta-learning. Bayesian optimization, such as hyperparameter optimization and multi-armed bandit algorithms, mechanically selecting the optimal model and automating the entire training process, can be applied. This allows for the maintenance of an optimal model even in dynamic data environments.

[0106] In the above-described embodiment, a case has been described in which the manufacturing conditions for each batch and the manufacturing result index are extracted as features. For example, a case has been described in which an index of the resin raw material, an index of the resin polymerization conditions, a maintenance index showing the maintenance status of the manufacturing equipment, a manufacturing result index, etc. are extracted as features. In addition to these, feature generation using engineering methods, mechanical feature generation methods, and learning models and feature generation using images and videos can also be applied. Specifically, information other than that derived from the plant may be added as a feature. Examples of information other than that derived from the plant include information on the raw material lot and the raw material composition. Additionally, features may be generated using an engineering method that uses domain knowledge, and mechanical feature generation methods such as principal component analysis, clustering, natural language processing, dimensional processing, transfer learning, decomposition, Fourier transform, cross features, lag features, graph-based features, and integration with external data may be used to generate features.

[0107] In addition to selection based on domain knowledge, explanatory variables may be automatically selected by mechanically selecting features using correlation coefficients or machine learning models. General-purpose selection methods, such as filter methods, wrapper methods, embedding methods, dimensionality reduction methods, combination methods, driven selection, mutual information, and cross-validation, may be used as feature selection methods. In this embodiment, as an example, selection is performed using features calculated by a random forest, which is one of the embedding methods. Furthermore, feature selection may be updated to optimal features as needed depending on the situation by performing a schedule at any time or periodically.

[0108] Furthermore, the learning model creation device 250 may create a learning model as follows: The learning model creation device 250 predicts a production result index 440, such as a fluidity index, for the (n+m)th batch based on the learning results for n batches. The learning model creation device 250 assumes the predicted production result index 440 to be an actual value and adds it to learning, thereby predicting the production result index 440 for the (n+m+1)th batch. The learning model creation device 250 assumes the predicted manufacturing result index 440 of the (n+m+1)th batch to be an actual value and adds it to learning to estimate the process recipe for the (n+m+2)th batch. Hyperparameter tuning, Bayesian optimization, genetic algorithm, reinforcement learning, simulated annealing, neural architecture search, etc. may be applied. The learning model may be evaluated using the mean squared error, mean absolute error, mean absolute percentage error, coefficient of determination, root mean squared error, logarithmic mean squared error, F1 score, accuracy rate, precision, recall, receiver operating characteristic curve (ROC curve), area under the ROC curve (AUC), Cohen's kappa, mean logarithmic loss, mean squared logarithmic error, mean squared error, etc. between the results inferred using the learning model and the actual measured values, and accuracy verification may be performed by confirming that these indicators converge, optimize, or improve.

[0109] As described above, the learning model creation device 250 of this embodiment creates a learning model based on the fourth learning data set and the fifth learning data set. By configuring in this manner, the learning model creation device 250 can create a learned model. Therefore, by using the learned model created by the learning model creation device 250, the information processing device 31 can shorten the time required to determine the (n+m+1)th manufacturing condition PM1-(n+m+1) for the (n+m+1) batch BT(n+m+1) and can improve the accuracy of determining the (n+m+1)th manufacturing condition PM1-(n+m+1) for the (n+m+1) batch BT(n+m+1).

[0110] Furthermore, the information processing system 10 may be configured by appropriately combining the functions of the information processing device 3 of the first embodiment and the information processing device 31 of the second embodiment. According to the information processing device 31 configured in this manner, it is possible to use the results of learning large amounts of information efficiently or with high accuracy through machine learning, and therefore it is possible to make the liquidity index 442 in the next batch, for example, MFR, more stable or closer to the target value.

[0111] A program for realizing the functions of any of the components in any of the above-described devices may be recorded on a computer-readable recording medium, and the program may be read and executed by a computer system. Note that the term "computer system" here includes an operating system and hardware such as peripheral devices. In addition, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, CDs (Compact Discs)-ROMs (Read Only Memory), and storage devices such as hard disks built into computer systems. Furthermore, the term "computer-readable recording medium" also includes a medium that stores a program for a certain period of time, such as a volatile memory within a computer system that serves as a server or client when the program is transmitted via a network such as the Internet or a communication line such as a telephone line. Such volatile memory may be, for example, RAM (Random Access Memory). The recording medium may also be, for example, a non-transitory recording medium.

[0112] The above program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network such as the Internet or a communication line such as a telephone line. The above program may also be one that realizes part of the above-mentioned functions. Furthermore, the above program may be a so-called differential file that can realize the above-mentioned functions in combination with a program already recorded in a computer system. A differential file may also be called a differential program.

[0113] Furthermore, the functions of any of the components in any of the above-described devices may be implemented by a processor. For example, each process in the embodiments may be implemented by a processor operating based on information such as a program and a computer-readable recording medium storing information such as the program. Here, the functions of each unit of the processor may be implemented by, for example, individual hardware, or may be implemented by integrated hardware. For example, the processor may include hardware, and the hardware may include at least one of a circuit for processing digital signals and a circuit for processing analog signals. For example, the processor may be configured using one or more circuit devices mounted on a circuit board, or one or both of one or more circuit elements. An integrated circuit (IC) or the like may be used as the circuit device, and a resistor or a capacitor may be used as the circuit element.

[0114] Here, the processor may be, for example, a CPU. However, the processor is not limited to a CPU, and various types of processors such as a GPU (Graphics Processing Unit) or a DSP (Digital Signal Processor) may be used. The processor may also be, for example, a hardware circuit such as an ASIC (Application Specific Integrated Circuit). The processor may also be, for example, composed of multiple CPUs, or may be, for example, composed of a hardware circuit such as a multiple ASIC. The processor may also be, for example, composed of a combination of multiple CPUs and a hardware circuit such as a multiple ASIC. The processor may also include, for example, one or more of an amplifier circuit or a filter circuit that processes analog signals.

[0115] The embodiments of this disclosure have been described in detail above with reference to the drawings, but the specific configuration is not limited to this embodiment, and includes designs within the scope that do not deviate from the gist of this disclosure. [Explanation of symbols]

[0116] 1-1, 1-2... terminal device, 2-1, 2-2, 2-3... server, 3... information processing device, 4... terminal device, 310, 311... manufacturing condition acquisition unit, 330... manufacturing information provision unit, 332... manufacturing result index estimation unit, 332a... first learned model, 334... target value estimation unit, 334a... second learned model, 336... manufacturing condition estimation unit, 336a... third learned model, 340... result acquisition unit, 350... presentation unit, 321... first manufacturing information provision unit, 322... second manufacturing information Provision unit, 333... manufacturing result index estimation unit, 333a... fourth learned model, 337... manufacturing condition estimation unit, 337a... fifth learned model, 341... first result acquisition unit, 342... second result acquisition unit, 351... index presentation unit, 352... result presentation unit, 200, 250... learning model creation device, 202, 252... input unit, 204, 254... reception unit, 206, 256... processing unit, 207, 257... learning model, 208, 258... output unit, 210, 260... memory unit,

Claims

1. 1. An information processing device that presents manufacturing conditions for a manufacturing process for manufacturing a melt-moldable fluororesin for each batch, the manufacturing process including a polymerization process and a confirmation process for a manufacturing result index of a resin manufactured through the polymerization process, a manufacturing condition acquisition unit that acquires n-th manufacturing conditions, which are manufacturing conditions for an n-th polymerization step that is the polymerization step of the n-th batch (n is a natural number); a production result index estimation unit that estimates production result indexes that will be confirmed in the confirmation steps for each of m batches (m is an integer greater than or equal to 0) after the nth batch, based on a first trained model that has been machine-learned to understand the relationship between the production conditions of the polymerization step for a predetermined batch and the production result indexes that will be confirmed in the confirmation steps for a plurality of batches after the predetermined batch, and the nth production condition acquired by the production condition acquisition unit; An information processing device comprising:

2. a target value estimation unit that estimates a target value of the manufacturing result index of the (n+m+1)th batch based on a second trained model that has been machine-learned to understand the relationship between the manufacturing result index of each batch and the manufacturing result index of subsequent batches, and the manufacturing result index estimated by the manufacturing result index estimation unit; a manufacturing condition estimation unit that estimates manufacturing conditions of the (n+m+1)th batch based on a third trained model that has been machine-learned to understand the relationship between a manufacturing result index of a batch and a manufacturing condition of the batch, and the target value of the manufacturing result index of the (n+m+1)th batch estimated by the target value estimation unit; The information processing device according to claim 1 , further comprising:

3. a manufacturing condition estimation unit that estimates manufacturing conditions of the (n+m+1)th batch based on a fifth trained model that has been machine-learned to understand the relationship between a manufacturing result index of a batch and a manufacturing condition of the batch, and the manufacturing result index estimated by the manufacturing result index estimation unit; The information processing device according to claim 1 , comprising:

4. The manufacturing conditions include an index of raw materials of the resin in the polymerization step and an index of polymerization conditions of the resin; The information processing device according to claim 1 .

5. The manufacturing conditions include a maintenance index indicating the maintenance status of the manufacturing equipment. The information processing device according to claim 1 .

6. a processing unit that creates a learning model by machine learning the relationship between the production conditions of the polymerization step for a predetermined batch and the production result indicators confirmed in the confirmation step for batches subsequent to the predetermined batch, using information indicating the production conditions of the polymerization step for each batch as explanatory variables and the production result indicators confirmed in the confirmation step for batches subsequent to the predetermined batch as objective variables, based on a dataset including, for a production process of producing a melt-moldable fluororesin for each batch, a polymerization step and a confirmation step of a production result indicator of a resin produced through the polymerization step; A learning model creation device comprising:

7. The learning model creation device according to claim 6 , wherein the manufacturing conditions include an index of raw materials of the resin in the polymerization process and an index of polymerization conditions of the resin.

8. The learning model creation device according to claim 6 , wherein the manufacturing conditions include a maintenance index indicating a maintenance status of manufacturing equipment.

9. The learning model creation device according to claim 6 , wherein the manufacturing conditions include one or both of an index generated using domain knowledge and an index generated using a transfer learning technique.

10. The learning model creation device according to claim 6 , wherein the manufacturing conditions include one or both of an index selected using a correlation coefficient and an index selected using a trained model.

11. 1. An information processing method executed by an information processing device that presents manufacturing conditions for a manufacturing process for manufacturing a melt-moldable fluororesin for each batch, the method including: a polymerization process; and a confirmation process for a manufacturing result index of a resin manufactured through the polymerization process, the information processing method comprising: acquire n-th production conditions, which are production conditions for an n-th polymerization step, which is the polymerization step for the n-th batch (n is a natural number); An information processing method for estimating a production result index to be confirmed in the confirmation process for each of m batches (m is an integer greater than or equal to 0) subsequent to the nth batch, based on a first trained model that has been machine-learned to determine the relationship between the production conditions of the polymerization process for a predetermined batch and the production result index to be confirmed in the confirmation process for multiple batches subsequent to the predetermined batch, and the acquired nth production condition.

12. A computer-executed learning model creation method for a manufacturing process for manufacturing a melt-moldable fluororesin on a batch-by-batch basis, the method comprising: a polymerization process; and a confirmation process for a production result indicator of a resin manufactured through the polymerization process, the method comprising: based on a dataset including, as learning data, information indicating the production conditions of the polymerization process for each batch; and production result indicators confirmed in the confirmation process for subsequent batches as training data, the method creates a learning model by machine learning the relationship between the production conditions of the polymerization process for a given batch and the production result indicators confirmed in the confirmation process for subsequent batches, using information indicating the production conditions of the polymerization process for each batch as explanatory variables and the production result indicators confirmed in the confirmation process for subsequent batches as objective variables.

13. A program that causes a computer to function as the information processing device according to any one of claims 1 to 3.

14. A program that causes a computer to function as the learning model creation device according to any one of claims 6 to 10.

Images