Information processing device, information processing method, and program

Abstract

To provide an information processing device, an information processing method, and a program for predicting the quality of a product acquired through processing including one or more processes with high accuracy.SOLUTION: There are provided an information processing device including at least one processor, an information processing method and a program that predict quality of a product acquired through processing including one or more processes. The processor is configured to: acquire at least one piece of information of chemical information and physical information of an object to be processed and a processing object at two points of different elapsed processing times between before processing and after the processing in at least one process of the process; acquire a calculation value of difference between numerical values acquired from the information respectively at the two points; and predict the quality of the product on the basis of the calculation value by using a learning model undergoing machine learning on the basis of a known data set of explanatory variables and objective variables where the difference is the explanatory variable and the quality of the object is the objective variable.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] The quality of a product is predicted using a machine learning model. For example, Patent Document 1 proposes a machine learning model that learns physical property-related data derived from physical property data representing the physical properties of a product as learning input data in order to improve the accuracy of prediction. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-018354 A Summary of the Invention [Problem to be solved by the invention]

[0004] Various techniques have been studied, including those described in Patent Document 1, but the current situation is that the accuracy of predicting product quality is insufficient.

[0005] The present disclosure has been made in consideration of the above-mentioned circumstances, and the problem that one embodiment of the present disclosure aims to solve is to provide an information processing device that predicts with high accuracy the quality of a product obtained by a process that includes one or more treatments. A problem to be solved by another embodiment of the present disclosure is to provide an information processing method for predicting with high accuracy the quality of a product obtained by a process including one or more treatments. The problem to be solved by another embodiment of the present disclosure is to provide a program for causing a computer to execute information processing for predicting with high accuracy the quality of a product obtained by a process including one or more treatments. [Means for solving the problem]

[0006] The present disclosure includes the following aspects. <1> An information processing device for predicting quality of a product obtained by a process including one or more treatments, At least one processor, the processor comprising: In at least one treatment of the process, at least one of chemical information and physical information of the treated object and the treated object is obtained at two different points in time between before and after the treatment; The difference between the two numerical values ​​obtained from the above information is calculated, The difference is used as an explanatory variable, and the quality of the product is used as a response variable. Using a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable, the quality of the product is predicted based on the calculated value. Information processing device. <2> The processor performs an acquisition of a condition value of a processing condition for at least one operation of the process; The explanatory variables include the above differences as well as the treatment conditions. In predicting the quality of the product, a prediction of the quality of the product is performed based on the calculated value and the condition value. <1> The information processing device described in <3> The processor acquires a spectroscopic spectrum as chemical information in at least one process in the acquisition of the information. <1> or <2> The information processing device described in <4> In acquiring the calculated value, the processor acquires the calculated value of the intensity at the wavenumber or the integrated intensity in the wavenumber region for a characteristic wavenumber or wavenumber region in the optical spectrum that changes due to at least one process. <3> The information processing device described in <5> The processor determines the wave number or the wave number range based on a quantum chemical calculation. <4> The information processing device described in <6> The characteristic wave number or wave number region includes a wave number or wave number region derived from a by-product. <4> or <5> The information processing device described in <7> At least one step of the process is a flow process using a flow path; The processor acquires a state quantity of a flow field as physical information of the flow processing in acquiring the information. <1> ~ <6> 13. An information processing device according to claim 12, <8> The processor acquires a state quantity of a flow field by a computational fluid analysis in acquiring information. <7> The information processing device described in <9> the flow process is a process of mixing a plurality of fluids, The state quantity of the flow field is the mixing ratio of the fluid calculated based on the computational fluid analysis. <8> The information processing device described in <10> 1. An information processing method for predicting the quality of a product obtained by a process including one or more treatments, comprising: In at least one treatment of the process, at least one of chemical information and physical information of the treated object and the treated object is obtained at two different points in time between before and after the treatment; Obtaining the calculated difference between the two numerical values ​​obtained from the above information, and Using the difference as an explanatory variable and the quality of the product as a response variable, a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable is used to predict the quality of the product based on the calculated value. An information processing method comprising: <11> A program for causing a computer to execute information processing for predicting the quality of a product obtained by a process including one or more treatments, In at least one treatment of the process, at least one of chemical information and physical information of the treated object and the treated object is obtained at two different points in time between before and after the treatment; Obtaining the calculated difference between the two numerical values ​​obtained from the above information, and Using the difference as an explanatory variable and the quality of the product as a response variable, a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable is used to predict the quality of the product based on the calculated value. A program for causing a computer to execute information processing, including: Effect of the Invention

[0007] According to one embodiment of the present disclosure, an information processing device is provided that predicts with high accuracy the quality of a product obtained by a process including one or more treatments. According to another embodiment of the present disclosure, there is provided an information processing method for predicting with high accuracy the quality of a product obtained by a process including one or more treatments. According to another embodiment of the present disclosure, there is provided a program for causing a computer to execute information processing for predicting with high accuracy the quality of a product obtained by a process including one or more treatments. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating a process including one processing step. [Diagram 2] FIG. 2 is a diagram illustrating a process including multiple processes. [Diagram 3] FIG. 3 is a diagram showing an example of an infrared spectrum. [Figure 4] FIG. 4 is a diagram showing an example of a mixed state of fluids obtained by the computational fluid analysis. [Diagram 5] FIG. 5 is a diagram showing an example of a mixed state of fluids obtained by the computational fluid analysis. [Figure 6] FIG. 6 is a diagram illustrating an information processing device, a learning device, a flow processing device, an information acquiring device, and a quality evaluation device. [Figure 7] FIG. 7 is a diagram illustrating an example of an outline of processes in the flow processing device, the information acquiring device, and the quality evaluating device. [Figure 8] FIG. 8 is a diagram illustrating an outline of the processing in the information processing device and the learning device. [Figure 9] FIG. 9 is a diagram illustrating an example in which information data, calculated value data, and quality data are associated with each other by a common ID. [Figure 10] FIG. 10 is a diagram illustrating a flow processing apparatus having a processing section with a T-shaped junction. [Figure 11]FIG. 11 is a block diagram illustrating a computer that constitutes the information processing device and the learning device. [Figure 12] FIG. 12 is a block diagram illustrating a processing unit of a CPU of the learning device. [Figure 13] FIG. 13 is a diagram illustrating an example of the learning unit in detail. [Figure 14] FIG. 14 is a block diagram showing a processing section of a CPU of the information processing device. [Figure 15] FIG. 15 is a diagram illustrating an example of an outline of the processes in the flow processing device, the information acquiring device, and the quality evaluating device. [Figure 16] FIG. 16 is a diagram illustrating an outline of the processing in the information processing device and the learning device. [Figure 17] FIG. 17 is a diagram illustrating an example in which information data, calculated value data, processing condition data, and quality data are associated with each other by a common ID. [Figure 18] FIG. 18 is a block diagram illustrating a processing unit of a CPU of the learning device. [Figure 19] FIG. 19 is a diagram illustrating an example of the learning unit in detail. [Figure 20] FIG. 20 is a block diagram showing a processing section of a CPU of the information processing device. [Figure 21] FIG. 21 is a diagram illustrating a flow processing apparatus having a processing section with a cross-shaped junction. [Figure 22] FIG. 22 is a table showing information data of infrared spectroscopy. [Diagram 23] FIG. 23 is a table showing calculation value data obtained based on the information data of the infrared spectrum. [Figure 24] FIG. 24 is a table showing information data of the state quantities of the flow field and calculation value data obtained based on the information data. [Diagram 25] FIG. 25 is a table showing the processing condition data. [Figure 26] FIG. 26 is a table showing the coefficient of determination. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] The information processing device, the information processing method, and the program according to the present disclosure will be described in detail below.

[0010] In this disclosure, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. In addition, in this specification, the same idea as "A and / or B" is also applied when three or more things are expressed by connecting them with "and / or."

[0011] The drawings referred to in the following description are illustrative and schematic, and the present disclosure is not limited to these drawings. The same reference numerals indicate the same components. Also, the reference numerals in the drawings may be omitted.

[0012] <Information processing device> The information processing device according to the present disclosure includes: An information processing device for predicting quality of a product obtained by a process including one or more treatments, At least one processor, the processor comprising: In at least one treatment of the process, at least one piece of chemical information and physical information of the treated object and the treated object is acquired (hereinafter, may be simply referred to as "acquisition of information") at two different points in time between before and after the treatment, Obtain a calculated value of the difference between the two numerical values ​​obtained from the above information (hereinafter, simply referred to as "obtaining the calculated value"); The above difference is used as an explanatory variable, and the quality of the product is used as a dependent variable. Using a trained model that has been machine-learned based on a known data set of the explanatory variables and dependent variables, the quality of the product is predicted based on the above calculated value (hereinafter, sometimes simply referred to as "quality prediction").

[0013] In Patent Document 1, a trained model based on machine learning based on physical property data such as spectroscopic data obtained from the product and the quality of the product is used to predict the quality of the product. However, there is a concern that such a method has low prediction accuracy of the product quality.

[0014] In contrast, the information processing device according to the present disclosure acquires at least one of chemical information and physical information from the processed object and the processed object at two different points in time before and after the processing during the process of producing the product.Then, the information processing device according to the present disclosure predicts the quality of the product using a trained model that is machine-learned based on the calculated value obtained from the acquired information and the quality of the product.This makes it possible to predict the quality of the product with high accuracy.

[0015] [A process that includes one or more operations] The information processing device includes at least one processor and predicts the quality of a product obtained by a process including one or more operations.

[0016] Examples of the process include a process in which a product is obtained by subjecting a starting material to one treatment, as shown in Figure 1. Also, examples of the process include a process in which a product is obtained by subjecting a starting material to a first treatment through an nth treatment (n is an integer of 2 or more), in sequence, as shown in Figure 2. Examples of starting materials include raw materials for the production and intermediates for the production.

[0017] The process shown in Fig. 2 may be a continuous process in which each treatment is performed continuously, or may include a batch process. A continuous process is preferable because it is possible to produce a product while predicting the quality of the product in-line, thereby improving the production speed and simultaneously controlling the quality.

[0018] The term "treatment" refers to an operation in which a substance undergoes one or more chemical or physical changes.

[0019] The treatment includes chemical operations and physical operations, such as chemical reaction, heating, cooling, mixing, separation, production, degassing, compression, concentration, removal, adsorption, desorption, electrolysis, reflux, stretching, coating, drying, swelling, evaporation, condensation, melting, solidification, sublimation, dissolution, recrystallization, plasma treatment, and chromatography.

[0020] Multiple processes may be performed simultaneously and treated as one process, and one process may actually involve multiple processes.

[0021] A treatment can be classified as either a chemical operation or a physical operation, but for example, heating can be a treatment in which heat is applied to an object subjected to the treatment to cause a physical change and at the same time a chemical reaction is promoted to cause a chemical change. Therefore, a treatment does not necessarily have to be classified as either a chemical operation or a physical operation, and may be treated as falling under both a chemical operation and a physical operation. Furthermore, a change caused by a treatment does not necessarily have to be classified as either a chemical change or a physical change, and may be treated as falling under both a chemical change and a physical change.

[0022] The product may be one or more. For example, when multiple compounds are obtained by a chemical reaction, the quality may be predicted for one compound as a product, or the quality may be predicted for multiple compounds as products. Moreover, one quality of the product may be predicted, or multiple qualities may be predicted.

[0023] A specific example of the process is a process including a flow process. More specifically, a process including a first flow process for mixing a plurality of raw materials and a second flow process for heating and polymerizing the mixed raw materials (hereinafter, sometimes referred to as a "polymer synthesis process") can be mentioned. This makes it possible to obtain a polymer as a product. The quality of the product (that is, the polymer) includes, for example, the number average molecular weight, the weight average molecular weight, the molecular weight dispersity, the yield, the purity, the composition, the proportion of surface functional groups, and the like.

[0024] In addition, examples of processes that include flow treatment include purification (concentration) of proteins. More specifically, examples of processes that include flow treatment include a process in which a buffer solution containing a protein is supplied to a column to perform chromatography (hereinafter, sometimes referred to as a "protein purification process"). This allows a purified protein to be obtained as a product. The quality of the product (i.e., purified protein) includes, for example, the concentration and purity of the protein itself, as well as the concentrations of aggregates, DNA, impurities, etc. contained in the purified protein.

[0025] Specific examples of the process include a process that includes a process of continuously treating a film while winding the film from roll to roll. More specifically, a process that includes a process of applying a layer-forming material to a film (hereinafter sometimes referred to as a "layer formation process") can be used, and a film with a layer formed thereon can be obtained as a product. A specific example of the process includes a process that includes a process of applying a plasma treatment to a film (hereinafter sometimes referred to as a "surface modification process") can be used, and a film with a modified surface can be obtained as a product. Product qualities (ie, layered films, surface modified films) include, for example, optical properties, surface roughness, wrinkles, crystallinity, etc. Furthermore, the quality of the film on which the layer is formed may further include, for example, coating streaks, coating unevenness, layer thickness, and the like.

[0026] The processor will be described in more detail below.

[0027] [Getting information] In at least one process, the processor acquires at least one piece of chemical information and physical information (hereinafter sometimes referred to as "specific information") of the treated object and the treated object at two different points in time between before and after the treatment.

[0028] For example, in the process shown in Figure 1, the following four patterns are possible for two points with different processing time between before and after processing. Processing progresses in the chronological order of point a1 → point b1 → point c1 → point d1. (1) a1 point before processing and d1 point after processing (2) a1 point before processing and b1 or c1 point during processing (3) b1 point in process and c1 point in process, (4) b1 or c1 point during processing and d1 point after processing

[0029] The pattern for acquiring the specific information may be one or more. Furthermore, in one pattern, one piece of specific information may be acquired, or multiple pieces of specific information may be acquired.

[0030] For example, when a process includes multiple treatments as shown in FIG. 2, the two points with different processing time lapses between before and after the treatment may have the same pattern as that described above for FIG. 1 for each treatment.

[0031] Regarding the material to be subjected to the treatment, the one that has not been treated is called the "material to be treated" and the one that has been treated is called the "material to be treated." The material to be treated may be the product. Also, the starting material may be the material to be treated.

[0032] Point a1, point a2, . . ., a before processing n At the point, only the workpiece is present. Point b1, point b2, ..., b n point, and the c1 point, c2 point, ..., c n In this respect, the object to be treated and the object to be treated may be mixed. After processing, d1 point, d2 point, . . ., d n At this point, the treated material is present. Depending on the degree of progress of the treatment, the treated material may remain.

[0033] Examples of chemical information include spectroscopic spectra (eg, infrared spectroscopic spectra, Raman spectroscopic spectra, nuclear magnetic resonance spectra, etc.), pH, concentration, and the like.

[0034] Examples of physical information include the degree of mixing of multiple fluids, electrical conductivity, temperature, transmittance, refractive index, turbidity, image, color, viscosity, contact angle, tension, and state quantities of the flow field.

[0035] The chemical information and physical information may be obtained by actual measurement or by simulation. For example, the spectroscopic spectrum, which is chemical information, may be obtained by actual measurement or by quantum chemical calculation. For example, an infrared spectroscopic spectrum as shown in FIG. 3 can be obtained for polystyrene. In addition, as physical information obtained by simulation, for example, state quantities of a flow field, which will be described later, can be mentioned.

[0036] The processor may acquire a spectroscopic spectrum as chemical information in at least one process of the process in acquiring information. The spectroscopic spectrum may be acquired by infrared spectroscopy, Fourier transform infrared spectroscopy, Raman spectroscopy, nuclear magnetic resonance spectroscopy, etc.

[0037] When at least one of the processes is a flow process using a flow path, the processor may acquire a state quantity of a flow field as physical information of the flow process in acquiring information. The "state quantity of a flow field" is an index that indicates the state of a flow field, regardless of the properties of a material inside the flow field.

[0038] Examples of state quantities of a flow field include the mixing state (eg, mixing rate, component ratio, uniformity), flow velocity distribution, residence time distribution, Reynolds number, turbulent energy, temperature distribution, and pressure distribution.

[0039] In acquiring the information, the processor may acquire state quantities of the flow field by computational fluid analysis.

[0040] When the flow process is a process of mixing a plurality of fluids, the state quantity of the flow field may be a mixing ratio of the fluids calculated based on a computational fluid analysis. The "mixing ratio" here refers to the degree of mixing of multiple fluids on a volumetric basis. The mixed state of multiple fluids is simulated using computational fluid analysis, and images are created of regions where the fluids are not mixed and regions where at least two fluids are mixed (mixed regions). The mixing ratio is the ratio of the area of ​​the mixed region to the area of ​​all regions (i.e., the total area).

[0041] For example, regarding the mixing ratio of two fluids (fluid 1, fluid 2), the mixed state of these fluids is simulated by computational fluid analysis, and an area where the two fluids exist without mixing and an area where the two fluids are mixed (mixed area) are visualized. By displaying fluid 1 in white, fluid 2 in black, and further displaying the area where the two fluids are mixed (mixed area) in gray, it is possible to visualize the mixed state as shown in FIG. 4, for example. Then, the mixing ratio can be calculated from the area of ​​the mixed area relative to the area of ​​the total area. Furthermore, when simulating a state in which mixing has progressed further than the mixed state shown in Fig. 4, it is possible to image a mixed state such as that shown in Fig. 5. In the mixed state shown in Fig. 5, fluid 1 and fluid 2 are completely mixed, and the mixing ratio is 1.

[0042] The computational fluid analysis may be performed using general simulation software (open source, commercially available software, etc.). The analysis conditions include the flow velocity of the fluid, the flow rate of the fluid, the pressure of the fluid (defined as the boundary conditions at the inlet and outlet of the analysis region), the concentration of the fluid (defined as the concentration at the inlet of the analysis region when calculating the mixing of fluids, etc.), the shape of the analysis region (equivalent to, for example, the diameter, length, etc. of the pipe through which the fluid flows), the physical property information of the fluid (viscosity, dynamic viscosity coefficient, density, diffusion coefficient), pump pressure, etc.

[0043] [Get calculated value] The processor obtains a calculated value of the difference (hereinafter, sometimes simply referred to as the "difference") between the two points of the numerical value (hereinafter, sometimes referred to as the "information value") obtained from the above information. Hereinafter, the calculated value of the difference will be simply referred to as the "calculated value."

[0044] Examples of information values ​​include intensity and integrated intensity (e.g., peak intensity and integrated intensity at a peak) obtained from a spectroscopic spectrum, chromaticity, image analysis values ​​(brightness, etc.), and other physical property values ​​described above for chemical information and physical information.

[0045] In obtaining the calculated values, the processor may obtain a calculated value of an intensity at the wavenumber or an integrated intensity in the wavenumber region for a characteristic wavenumber or wavenumber region in the spectrum that changes due to at least one process. In this embodiment, the processor may determine the wave number or wave number range based on quantum chemical calculations, thereby allowing the wave number of interest to be determined.

[0046] In addition, when the wavenumber or wavenumber region is determined based on quantum chemical calculations, it is easy to extract the wavenumber or wavenumber region that is thought to be involved in the reaction. Therefore, it is possible to reduce the possibility that the wavenumber or wavenumber region that is correlated with the objective variable despite having a small (or no) contribution to the reaction is taken into account in the machine learning model, and it may be possible to predict the quality with higher accuracy.

[0047] In the aspect of obtaining a calculated value of the intensity at the wavenumber or the integrated intensity at the wavenumber region for a characteristic wavenumber or wavenumber region as described above, sparse modeling may be used to extract wavenumbers having a high contribution rate to prediction accuracy from among a plurality of wavenumbers.

[0048] In the above embodiment, the characteristic wave number or wave number region may include a wave number or wave number region originating from a by-product. In the case of a chemical reaction characterized by the production and amount of a by-product, by selecting the wave number or wave number region originating from the by-product as the characteristic wave number or wave number region, it may be possible to predict the quality with higher accuracy.

[0049] For example, in the infrared spectrum of polystyrene shown in Figure 3, the characteristic wavenumbers are determined by quantum chemical calculations, and the peak intensities of peaks A to N corresponding to the determined wavenumbers can be obtained. Peaks A to C are derived from aromatic ring CH stretching, peaks D and E are derived from aliphatic CH stretching, peaks F to I are derived from aromatic ring monosubstitutions, peaks J and K are derived from aromatic ring C=C stretching, peak L is derived from CH bending, peak M is derived from CH out-of-plane bending, and peak N is derived from aromatic ring bending. In addition to the polystyrene product, each peak may also contain peak components derived from by-products and unreacted reactants.

[0050] [Quality Prediction] The processor uses the difference as an explanatory variable and the product quality as a response variable, and predicts the product quality based on the calculated value using a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable.

[0051] Details of the trained model will be described later.

[0052] [Get condition value] The processor may acquire condition values ​​of processing conditions for at least one operation of the process (hereinafter, may be simply referred to as "acquiring condition values"). In this embodiment, the explanatory variables include processing conditions in addition to the difference, and in predicting quality, the quality of the product is predicted based on the calculated values ​​and the condition values. This may enable prediction of quality with higher accuracy.

[0053] In the case of flow processing, examples of the condition values ​​include the flow rate of the fluid, the concentration of the fluid, the diameter and length of the pipe through which the fluid flows, the reaction temperature, the viscosity of the fluid, and the pump pressure.

[0054] The condition value may be an actual measurement value or a set value.

[0055] [Example of polymer synthesis process] The above polymer synthesis process will be taken as an example for explanation. In the first flow process in which a plurality of raw materials are mixed, a plurality of raw materials (processing objects) are present at point a1 before the process. At points b1 and c1 during the first flow process, a portion of the multiple raw materials is mixed. That is, unmixed raw materials (material to be processed) and mixed raw materials (material to be processed) are mixed. The mixing ratios at points b1 and c1 may be different. At point d1 after the first flow treatment, the mixed raw material (treated material) is present.

[0056] For example, the mixing ratio may be acquired at point b1 during the first flow process and at point c1 during the first flow process (acquisition of information). A numerical value obtained from the mixing ratio, for example, a calculated value of the difference between the above two points is acquired from the mixing ratio itself (acquisition of calculated value).

[0057] In the second flow treatment in which the mixed raw materials are heated and polymerized, the mixed raw materials (processing target) are present at point a2 before the treatment. At point b2 during the second flow treatment and at point c2 during the second flow treatment, a part of the mixed raw material (substance to be treated) is polymerized to generate a polymer (substance to be treated and product). At point b2 and point c2, the degree of progress of polymerization may differ. At point d2 after the second flow treatment, a polymer (treatment product and product) is present.

[0058] For example, an infrared spectrum may be obtained at a point a2 before the second flow process and at a point d2 after the second flow process (obtaining information). A value obtained from the infrared spectrum, for example, an integrated intensity at a peak, is used to calculate the difference between the above two points (obtaining a calculated value).

[0059] For example, number average molecular weight may be selected as a product (polymer) quality.

[0060] A trained model is constructed by machine learning based on a known data set of explanatory variables and objective variables, with at least one of the difference obtained from the mixing ratio and the difference obtained from the integrated intensity at the peak as the explanatory variable and the number average molecular weight as the objective variable. Using the trained model, the number average molecular weight of the polymer is predicted based on the calculated value (quality prediction).

[0061] The processing conditions of the first flow process include, for example, the flow rate of the raw material, the concentration of the raw material, and the diameter and length of the pipe through which the fluid flows, and the processing conditions of the second flow process include the reaction temperature. These may be acquired as condition values ​​of the processing conditions (acquisition of condition values).

[0062] When the condition values ​​of the processing conditions are obtained, a trained model is constructed by machine learning based on a known data set of explanatory variables and objective variables, with at least one of the differences and the processing conditions as explanatory variables and the number average molecular weight as objective variable. Using the trained model, the number average molecular weight of the polymer is predicted based on the calculated value and the condition values ​​(quality prediction).

[0063] [Example of protein purification process] The above protein purification process will be taken as an example. In a flow process in which a buffer solution containing a protein is supplied to a column to perform chromatography, an untreated buffer solution (subject to be treated) exists at point a1 before the treatment. At a point b1 during the flow treatment and at a point c1 during the flow treatment, impurities are separated from a portion of the buffer solution. At point d1 after the flow treatment, a buffer solution (treated product) containing purified protein with few impurities is present.

[0064] For example, a spectrum may be obtained at a point a1 before the flow processing and at a point d1 after the flow processing (obtaining information). A value obtained from the spectrum, for example, an integrated intensity at a peak, is used to calculate the difference between the above two points (obtaining a calculated value).

[0065] For example, purity may be selected as a quality of the product (purified protein).

[0066] Using the difference as the explanatory variable and the purity as the objective variable, a trained model is constructed by machine learning based on a known data set of explanatory variables and objective variables. Using the trained model, the purity of the purified protein is predicted based on the calculated value (quality prediction).

[0067] An example of the process conditions for the flow process is the pump pressure, which may be acquired as a condition value of the process conditions (acquisition of the condition value).

[0068] When the condition values ​​of the processing conditions are obtained, a trained model is constructed by machine learning based on a known data set of explanatory variables and objective variables, with the difference and processing conditions as explanatory variables and the purity as objective variable. Using the trained model, the purity of the purified protein is predicted based on the calculated value and the condition values ​​(quality prediction).

[0069] [Layer formation process] The above-mentioned layer formation process will be taken as an example for explanation. In a process of applying a layer-forming material to a film, at point a1 before the process, there is a film (processing target) to which the layer-forming material has not been applied. At point b1 during processing and at point c1 during processing, a layer is formed in a part of the film. At point d1 after processing, a film (processed product) with layers formed over the entire surface is present.

[0070] For example, the transmittance may be acquired at point b1 during processing and at point c1 during processing (acquisition of information). A calculated value of the difference between the above two points is acquired from a numerical value obtained from the transmittance, for example, the transmittance itself (acquisition of a calculated value). In this case, for example, the layer thickness may be selected as the quality of the product (film on which a layer is formed).

[0071] Using the difference as an explanatory variable and the layer thickness as a response variable, a trained model is constructed by machine learning based on a known data set of explanatory variables and response variables. Using the trained model, the layer thickness of the film on which the layers are formed is predicted based on the calculated value (quality prediction).

[0072] Also, for example, the tension of the film may be acquired at point b1 during processing and at point c1 during processing (acquisition of information). A calculated value of the difference between the above two points is acquired from a numerical value obtained from the tension, for example, the tension itself (acquisition of a calculated value). In this case, for example, the presence or absence of wrinkles may be selected as the quality of the product (film on which layers are formed).

[0073] Using the difference as the explanatory variable and the presence or absence of wrinkles as the objective variable, a trained model is constructed by machine learning based on a known data set of explanatory variables and objective variables. Using the trained model, the presence or absence of wrinkles in the film on which the layers are formed is predicted based on the calculated value (quality prediction).

[0074] [Surface modification process] The surface modification process will be described as an example. In a process in which a film is subjected to a plasma treatment, at point a1 before the treatment, there is a film (subject to be treated) that has not been subjected to the plasma treatment. At point b1 during processing, and at point c1 during processing, a portion of the film is subjected to plasma treatment. At point d1 after the treatment, there exists a film (treated product) that has been entirely subjected to plasma treatment.

[0075] For example, the temperature may be acquired at point b1 during processing and at point c1 during processing (acquisition of information). A calculated value of the difference between the above two points is acquired from a numerical value obtained from the temperature, for example, the temperature itself (acquisition of a calculated value).

[0076] For example, contact angle may be selected as a quality of the product (plasma treated film).

[0077] Using the difference as an explanatory variable and the contact angle as a response variable, a trained model is constructed by machine learning based on a known data set of explanatory variables and response variables. Using the trained model, the contact angle of the plasma-treated film is predicted based on the calculated value (quality prediction).

[0078] Hereinafter, the information processing device according to the present disclosure will be described in more detail with reference to a first embodiment and a second embodiment, taking a polymer synthesis process as an example.

[0079] [First embodiment] 6, an information processing device 11 is connected to a learning device 10 that constructs a trained model via a network 12 so that they can communicate with each other. The learning device 10 and the information processing device 11 are, for example, desktop personal computers. The network 12 is, for example, a LAN (Local Area Network) or a WAN (Wide Area Network) such as the Internet or a public communication network. A flow processing device 13, an information acquisition device 14, and a quality evaluation device 15 are also connected to the network 12.

[0080] 7, the flow processing device 13 produces a product PR from raw material RM by a process including a flow process, according to processing condition data PCD. The information acquisition device 14 acquires specific information at two different points in the processing time between before and after the processing in at least one of the processes, and outputs the acquired information data IFD. The quality evaluation device 15 evaluates the quality of the product PR, and outputs quality data QD that is the evaluation result. The information data IFD from the information acquisition device 14 and the quality data QD from the quality evaluation device 15 are each transmitted to the learning device 10.

[0081] In Fig. 8, the learning device 10 acquires information data IFD from an information acquisition device 14 and quality data QD from a quality evaluation device 15. The learning device 10 derives differential calculated value data (hereinafter abbreviated as calculated value data) CD from the information data IFD. The calculated value data CD is set as learning input data IDL. The information data IFD, the calculated value data CD derived from the information data IFD, and the quality data QD are associated with each other by a common ID (Identification Data) assigned to one product PR, as shown in Fig. 9. The calculated value data CD and the quality data QD constitute a known data set.

[0082] The learning device 10 has a machine learning model M. The machine learning model M is a model for predicting the quality of a product PR. The machine learning model M outputs learning output data ODL corresponding to learning input data IDL.

[0083] The machine learning model M may be one that uses linear regression, Gaussian process regression, support vector regression, decision tree, regression tree, ensemble method, bagging method, boosting method, gradient boosting method, etc. In addition, there may be one that uses a simple perceptron, a multilayer perceptron, a deep neural network, a convolutional neural network, a deep belief network, a recurrent neural network, a probabilistic neural network, etc. There is no particular restriction on which of the above-mentioned machine learning models M is used, and any machine learning model M of any method may be selected.

[0084] An example of the ensemble method is the random forest. As is well known, the random forest uses randomly sampled training data and randomly selected explanatory variables to create multiple groups of decision trees with low correlation, and improves the accuracy of prediction by integrating and averaging their prediction results. In this case, the control parameters of the machine learning model M include the number of explanatory variables to be selected and the number of branches of the decision tree.

[0085] Deep neural networks have a relatively large number of control parameters that can be flexibly combined, allowing them to demonstrate high predictive performance for a variety of data structures. Control parameters include the number of network layers and nodes, the type of activation function, the dropout rate, mini-batch size, number of epochs, learning rate, etc.

[0086] There are multiple execution frameworks for such machine learning models M, and an appropriate one can be selected from them. For example, you can choose from Tensorflow, CNTK (Cognitive Toolkit), Theano, Caffe, mxnet, Keras, PyTorch, Chainer, Scikit-learn, Caret, MATLAB, etc.

[0087] The quality data QD is data for checking the quality data QD against the learning output data ODL. The higher the prediction accuracy of the machine learning model M, the smaller the difference between the quality data QD and the learning output data ODL. For this reason, the learning device 10 compares the learning output data ODL with the quality data QD that has the same ID as the learning input data IDL, and evaluates the prediction accuracy of the machine learning model M. Then, the machine learning model M is updated according to the evaluation result. The learning device 10 inputs the learning input data IDL to the machine learning model M, outputs the learning output data ODL from the machine learning model M, evaluates the prediction accuracy of the machine learning model M, and updates the machine learning model M while changing the learning input data IDL and the quality data QD. Then, this series of processes is repeated until the prediction accuracy of the machine learning model M reaches a preset level. The learning device 10 transmits the machine learning model M whose prediction accuracy has reached a preset level to the information processing device 11 as a trained model TM to be used in actual operation.

[0088] The information processing device 11 receives the trained model TM from the learning device 10. The information processing device 11 provides the trained model TM with calculation value data for prediction CDF, which is calculation value data of a product PR whose quality is unknown. The calculation value data for prediction CDF is data derived from the information data for prediction IFDF, which is information data IFD of the product PR whose quality is unknown, in the same manner as the calculation value data CD. The trained model TM outputs quality prediction data QFD according to the calculation value data for prediction CDF.

[0089] In the first embodiment, the flow treatment device 13 performs an anionic polymerization reaction of polystyrene, which is the product PR, by a process including two flow treatments (a first flow treatment and a second flow treatment) using a flow path.

[0090] The first flow process is a process for mixing a first raw material RM1 (a solution of polystyryllithium dissolved in a solvent) with a second raw material RM2 (aqueous methanol solution), and obtaining a mixture of the first raw material RM1 and the second raw material RM2 (hereinafter sometimes referred to as a "raw material mixture"). The second flow process is a process for anionic polymerization of the raw material mixture, and obtaining polystyrene as a product PR.

[0091] Polystyryllithium produces polystyrene, which is the product PR, through an anionic polymerization reaction. Tetrahydrofuran is used as the solvent. Small amounts of toluene and hexane are also mixed into the solution. The raw material for the flow treatment may be a mixture of reactants such as polystyryllithium and other substances, as in the first raw material RM1, or may be composed of only reactants. Methanol is also used as a terminator for anionic polymerization reactions.

[0092] In FIG. 10, flow processing device 13 includes a first material supply unit 20, a second material supply unit 21, a processing section 22, a temperature control unit 23, a recovery and disposal section 24, a setting unit 25, a system controller 26, and the like.

[0093] The first raw material supply unit 20 is connected to the upstream end of the processing section 22 via a pipe (not shown). The first raw material supply unit 20 supplies the first raw material RM1 to the processing section 22. The first raw material supply unit 20 has a pump for sending the first raw material RM1 to the processing section 22. By controlling the rotation speed of this pump, the flow rate of the first raw material RM1 sent from the first raw material supply unit 20 to the processing section 22 is adjusted.

[0094] The second raw material supply unit 21, like the first raw material supply unit 20, is connected to the upstream end of the processing section 22 via a pipe (not shown). The second raw material supply unit 21 supplies the second raw material RM2 to the processing section 22. Like the first raw material supply unit 20, the second raw material supply unit 21 also has a pump for sending the second raw material RM2 to the processing section 22. The flow rate of the second raw material RM2 sent from the second raw material supply unit 21 to the processing section 22 is adjusted by controlling the rotation speed of this pump.

[0095] The processing section 22 is a section for carrying out a first flow treatment (mixing treatment) and a second flow treatment (anionic polymerization reaction treatment). The processing section 22 has a confluence 30 and a reaction section 31. The first flow treatment is carried out in the confluence 30, and the second flow treatment is carried out in the reaction section 31.

[0096] The junction 30 is composed of a first pipe section 32, a second pipe section 33, and a third pipe section 34. The first pipe section 32 and the second pipe section 33 are connected in a straight line, and the third pipe section 34 intersects with the first pipe section 32 and the second pipe section 33 at a right angle. In other words, the junction 30 is T-shaped.

[0097] The first pipe section 32 is connected to the first raw material supply section 20, and the second pipe section 33 is connected to the second raw material supply section 21. The third pipe section 34 is connected to the reaction section 31. The first raw material RM1 is supplied to the first pipe section 32 from the first raw material supply section 20, and the second pipe section 33 is supplied with the second raw material RM2 from the second raw material supply section 21. The first raw material RM1 and the second raw material RM2 join together in the third pipe section 34 and are sent to the reaction section 31 in a mixed state.

[0098] The first pipe section 32 is provided with a first flow rate sensor 35 for detecting the flow rate of the first raw material RM1 passing through the first pipe section 32. The second pipe section 33 is provided with a second flow rate sensor 36 for detecting the flow rate of the second raw material RM2 passing through the second pipe section 33. Furthermore, the third pipe section 34 is provided with a third flow rate sensor 37 for detecting the flow rate of the raw material mixture passing through the third pipe section 34.

[0099] The reaction section 31 is a long and thin tube formed by linearly connecting multiple straight tubes having the same inner diameter. The length L of the reaction section 31 can be changed by changing the number of connected straight tubes and / or the length of the straight tubes. The inner diameter Φ of the reaction section 31 can also be changed by changing the inner diameter of the connected straight tubes.

[0100] The inside of the reaction section 31 is a flow path through which the raw material mixture flows, and is the place where the second flow treatment is performed. When the raw material mixture passes through the reaction section 31, the anionic polymerization reaction proceeds, and a polystyrene solution is obtained. The second flow treatment may also proceed slightly in the third tube section 34 of the confluence section 30, but the length of the third tube section 34 is very short compared to the length L of the reaction section 31. For this reason, the third tube section 34 is ignored, and the length L of the reaction section 31 is regarded as the reaction path length, which is the length of the place where the second flow treatment is performed. Similarly, the inner diameter Φ of the reaction section 31 is regarded as the reaction path diameter, which is the diameter of the place where the second flow treatment is performed.

[0101] The temperature adjusting unit 23 includes a heater and / or a cooler, and adjusts the temperature (hereinafter referred to as the reaction temperature) inside the reaction unit 31. At the downstream end of the reaction unit 31, a temperature sensor 38 is provided for detecting the reaction temperature.

[0102] The recovery and disposal section 24 is a section for recovering polystyrene, which is the product PR, and disposing of waste resulting from failed reactions. The recovery and disposal section 24 has a recovery section 40 and a disposal section 41. The recovery section 40 and the disposal section 41 are connected to the downstream end of the reaction section 31 by a three-way valve 42. This three-way valve 42 makes it possible to switch between the recovery line connecting the reaction section 31 and the recovery section 40 and the disposal line connecting the reaction section 31 and the disposal section 41.

[0103] The recovery unit 40 precipitates polystyrene from the polystyrene solution. The recovery unit 40 filters the precipitated polystyrene from the solution and collects it. The collected polystyrene is then dried. More specifically, the recovery unit 40 has a container equipped with a stirrer, contains methanol in the container, and mixes the polystyrene solution into the stirred methanol to precipitate polystyrene. The recovery unit 40 also has a thermostatic chamber with a pressure reduction function, and dries the methanol by heating the thermostatic chamber under reduced pressure.

[0104] The waste section 41 is a tank for storing waste. Here, the waste refers to waste that is sent out from the reaction section 31 when the flow rate of the first raw material RM1, the flow rate of the second raw material RM2, the flow rate of the raw material mixture, the reaction temperature, etc. are disturbed for some reason and production cannot be performed under the initially planned processing conditions.

[0105] The setting unit 25 accepts settings of processing conditions for the product PR by the operator of the flow processing device 13. The processing conditions accepted by the setting unit 25 are registered in the system controller 26 as processing condition data PCD.

[0106] The system controller 26 comprehensively controls the overall operation of the flow processing device 13. The system controller 26 is connected to the first raw material supply unit 20, the second raw material supply unit 21, the temperature adjustment unit 23, the first flow rate sensor 35, the second flow rate sensor 36, the third flow rate sensor 37, the temperature sensor 38, and the three-way valve 42.

[0107] The system controller 26 controls the rotation speed of the pump of the first raw material supply unit 20 in accordance with the flow rate of the first raw material RM1 detected by the first flow rate sensor 35, thereby adjusting the flow rate of the first raw material RM1. Similarly, the system controller 26 controls the rotation speed of the pump of the second raw material supply unit 21 in accordance with the flow rate of the second raw material RM2 detected by the second flow rate sensor 36, thereby adjusting the flow rate of the second raw material RM2. In addition, the system controller 26 drives the temperature adjustment unit 23 in accordance with the reaction temperature detected by the temperature sensor 38. Furthermore, the system controller 26 controls the three-way valve 42 to switch between the recovery line and the waste line described above.

[0108] Examples of processing condition data PCD include the concentration (unit: mol / l) and flow rate (unit: ml / min) of the first raw material RM1, the concentration (unit: mol / l) and flow rate (unit: ml / min) of the second raw material RM2, the reaction path diameter Φ (unit: mm), the reaction path length L (unit: m), and the reaction temperature (unit: °C).

[0109] The system controller 26 adjusts the flow rate of the first raw material RM1 by controlling the pump rotation speed of the first raw material supply unit 20 so that the flow rate of the first raw material RM1 detected by the first flow rate sensor 35 coincides with the flow rate of the first raw material RM1 registered in the processing condition data PCD. Similarly, the system controller 26 adjusts the flow rate of the second raw material RM2 by controlling the pump rotation speed of the second raw material supply unit 21 so that the flow rate of the second raw material RM2 detected by the second flow rate sensor 36 coincides with the flow rate of the second raw material RM2 registered in the processing condition data PCD.

[0110] Furthermore, the system controller 26 drives the temperature adjusting unit 23 so that the reaction temperature detected by the temperature sensor 38 coincides with the reaction temperature registered in the processing condition data PCD.

[0111] When the deviation between the values ​​detected by the first flow velocity sensor 35, the second flow velocity sensor 36, and the temperature sensor 38 (and the first flow velocity sensor 51, the second flow velocity sensor 52, and the third temperature sensor 53 described below) and the values ​​registered in the processing condition data PCD exceeds a preset range, the system controller 26 controls the three-way valve 42 to switch to the waste line and guide the waste to the waste section 41. Note that if the reaction fails and waste is produced, naturally the information data IFD and the quality data QD are not output. For this reason, the processing condition data PCD in the case where waste is produced is discarded without being transmitted to the learning device 10.

[0112] The information acquisition device 14 acquires specific information (i.e., at least one of chemical information and physical information of the processed object and the processed object) at two different points in the processing time between before and after the processing in at least one of the first flow processing and the second flow processing, and outputs the acquired information data IFD.

[0113] Furthermore, the information acquiring device 14 may acquire specific information, for example, at two points in the first flow process where the processing time elapses differently between before and after the processing, i.e., at point b1 of the first pipe section 32 and point d1 of the third pipe section 34 shown in Fig. 10. The specific information may be, for example, a state quantity (physical information) of the flow field.

[0114] For example, in the second flow process, the information acquiring device 14 may acquire specific information at two different points in the course of processing time between before and after the process, i.e., points a2 and d2 of the reaction section 31 shown in Fig. 10. The specific information may be, for example, an infrared spectrum (chemical information).

[0115] The quality evaluation device 15 may output, for example, the number average molecular weight of the product PR as the quality data QD.

[0116] 11, the computers that make up the learning device 10 and information processing device 11 (hereinafter, these may be collectively referred to as "information processing devices, etc.") have the same basic configuration, and include a storage device 60, a memory 61, a CPU (Central Processing Unit, processor) 62, a communication unit 63, a display 64, and an input device 65. These are interconnected via a bus line 66.

[0117] The storage device 60 is a hard disk drive built into a computer constituting the information processing device 11 or the like, or connected via a cable or network. Alternatively, the storage device 60 is a disk array in which a plurality of hard disk drives are connected in series. The storage device 60 stores control programs such as an operating system, various application programs, and various data associated with these programs. Note that a solid state drive may be used instead of or in addition to the hard disk drive.

[0118] The memory 61 is a work memory for executing processes by the CPU 62. The CPU 62 loads programs stored in the storage device 60 into the memory 61 and executes processes according to the programs, thereby providing overall control over each part of the computer.

[0119] The communication unit 63 is a network interface that controls the transmission of various information via the network 12. The display 64 displays various screens. A computer constituting the learning device 10, etc., accepts input of operation instructions from the input device 65 via the various screens. The input device 65 is a keyboard, a mouse, a touch panel, etc.

[0120] In the following description, each unit of the learning device 10 is distinguished by adding the subscript "A" to it, and each unit of the information processing device 11 is distinguished by adding the subscript "B" to it.

[0121] 12, a first operating program 70 is stored in the storage device 60A of the study device 10. The first operating program 70 is an application program for causing a computer to function as the study device 10.

[0122] The storage device 60A also stores information data IFD from the information acquisition device 14 and quality data QD from the quality evaluation device 15. Calculated value data CD derived from the information data IFD and a machine learning model M are also stored. A plurality of sets of the information data IFD, the calculated value data CD, and the quality data QD are stored.

[0123] When the first operating program 70 is launched, the CPU 62A of the computer constituting the learning device 10 works in cooperation with the memory 61, etc. to function as a first read / write (hereinafter abbreviated as RW (Read Write)) control unit 75, a first derivation unit 76, a learning unit 77, and a transmission control unit 78.

[0124] The first RW control unit 75 controls reading of various data in the storage device 60A and storage of various data in the storage device 60A. The first RW control unit 75 reads information data IFD from the storage device 60A and outputs the information data IFD to the first derivation unit 76. The first RW control unit 75 also stores the calculated value data CD from the first derivation unit 76 in the storage device 60A. When the information data IFD is a spectroscopic spectrum, the first derivation unit 76 may determine the wavenumber or wavenumber region of the peak based on a quantum chemical calculation. At that time, a wavenumber having a high contribution rate to the prediction accuracy may be extracted from among the multiple wavenumbers by sparse modeling.

[0125] The first RW control unit 75 reads the calculated value data CD and the quality data QD from the storage device 60A, and outputs them to the learning unit 77. The first RW control unit 75 also reads the machine learning model M from the storage device 60A, and outputs the machine learning model M to either the learning unit 77 or the transmission control unit 78. The first RW control unit 75 also stores the machine learning model M from the learning unit 77 in the storage device 60A.

[0126] The first derivation unit 76 receives information data IFD from the first RW control unit 75. The first derivation unit 76 derives calculated value data CD from the information data IFD. The first derivation unit 76 assigns the same ID as the information data IFD to the derived calculated value data CD, and outputs the calculated value data CD to the first RW control unit 75. The first derivation unit 76 derives the calculated value data CD every time new information data IFD is transmitted from the information acquisition device 14.

[0127] The learning unit 77 receives the learning input data IDL, the quality data QD, and the machine learning model M from the first RW control unit 75. The learning unit 77 provides the learning input data IDL to the machine learning model M to learn it, and outputs a trained model TM.

[0128] The transmission control unit 78 receives the machine learning model M from the first RW control unit 75. The machine learning model M that the transmission control unit 78 receives from the first RW control unit 75 is the trained model TM. The transmission control unit 78 controls the transmission of the trained model TM to the information processing device 11.

[0129] 13, the learning unit 77 has a first processing unit 85, an evaluation unit 86, and an update unit 87. The first processing unit 85 provides the learning input data IDL (calculated value data CD) to the machine learning model M, and causes the machine learning model M to output the learning output data ODL. The learning output data ODL is, for example, the number average molecular weight, the same as the quality data QD. The first processing unit 85 outputs the learning output data ODL to the evaluation unit 86.

[0130] The evaluation unit 86 receives the learning output data ODL from the first processing unit 85. The evaluation unit 86 compares the learning output data ODL with the quality data QD, and evaluates the accuracy of the prediction of the machine learning model M. The evaluation unit 86 outputs the evaluation result to the update unit 87.

[0131] The evaluation unit 86 evaluates the accuracy of prediction of the machine learning model M using, for example, a loss function. The loss function is a function that represents the degree of difference between the learning output data ODL and the quality data QD. The closer the calculated value of the loss function is to 0, the higher the accuracy of prediction of the machine learning model M.

[0132] The update unit 87 updates the machine learning model M according to the evaluation result from the evaluation unit 86. For example, the update unit 87 changes the values ​​of various parameters of the machine learning model M by a stochastic gradient descent method involving a learning coefficient or the like. The learning coefficient indicates the range of change in the values ​​of various parameters of the machine learning model M. In other words, the larger the learning coefficient is, the larger the range of change in the values ​​of the various parameters becomes, and the greater the degree of update of the machine learning model M becomes.

[0133] The input of the learning input data IDL to the machine learning model M by the first processing unit 85 and the output of the learning output data ODL to the evaluation unit 86, the evaluation of the prediction accuracy by the evaluation unit 86, and the update of the machine learning model M by the update unit 87 are repeated until the prediction accuracy reaches a preset level. The machine learning model M whose prediction accuracy has reached the preset level is stored in the storage device 60A by the first RW control unit 75 as a trained model TM.

[0134] 14, a second operating program 110 is stored in a storage device 60B of the information processing device 11. The second operating program 110 is an application program for causing a computer to function as the information processing device 11.

[0135] The storage device 60B also stores the trained model TM from the learning device 10 and the prediction information data IFDF from the information acquisition device 14. In addition, the prediction information data IFDF in the storage device 60B is information data of a product PR whose quality is unknown and whose quality is predicted using the trained model TM.

[0136] The storage device 60B also stores calculation value data CDF for prediction derived from the calculation information data IFDF.

[0137] When the second operating program 110 is started, the CPU 62B of the computer constituting the information processing device 11 functions as a second RW control unit 115, a second derivation unit 116, a second processing unit 117, and a display control unit 118 in cooperation with the memory 61, etc.

[0138] The second RW control unit 115, like the first RW control unit 75 of the learning device 10, controls reading of various data in the storage device 60B and storage of various data in the storage device 60B. The second RW control unit 115 reads the prediction information data IFDF from the storage device 60B (acquisition of information) and outputs the prediction information data IFDF to the second derivation unit 116. In addition, the second RW control unit 115 stores the prediction calculation value data CDF from the second derivation unit 116 in the storage device 60B. When the prediction information data IFDF is a spectroscopic spectrum, the second derivation unit 116 may determine the wave number of the peak based on a quantum chemical calculation. At that time, a peak having a high contribution rate to the prediction accuracy may be extracted from the multiple peaks by sparse modeling.

[0139] The second RW control unit 115 reads the trained model TM from the storage device 60B and outputs the trained model TM to the second processing unit 117. By reading the trained model TM from the storage device 60B, the second RW control unit 115 acquires the trained model TM.

[0140] The second RW control unit 115 reads out the calculation value data CDF for calculation from the storage device 60B (acquires calculated values), and outputs these to the second processing unit 117. The second RW control unit 115 acquires the calculation value data CD for calculation by reading out the calculation value data CDF for calculation from the storage device 60B.

[0141] The second derivation unit 116 receives the calculation information data IFDF from the second RW control unit 115. The second derivation unit 116 derives calculation value data CDF for calculation from the calculation information data IFDF.

[0142] The second processing unit 117 receives the calculation value data for prediction CDF and the trained model TM from the second RW control unit 115. The second processing unit 117 provides the calculation value data for prediction CDF to the trained model TM to predict quality (quality prediction). The second processing unit 117 outputs quality prediction data QFD, which is the quality prediction result by the trained model TM, to the display control unit 118. The quality prediction data QFD is, for example, number average molecular weight, like the quality data QD.

[0143] The display control unit 118 controls the display of the quality prediction data QFD and the like on the display 64B.

[0144] In this way, the quality of the product obtained from a process involving one or more treatments can be predicted. It will be made available for viewing.

[0145] [Second embodiment] In the second embodiment, the quality of the product is predicted using the processing condition data in addition to the calculated value data. The second embodiment will be described in detail below, but the other parts are the same as those described in the first embodiment. Also, the flow processing device, information acquisition device, and quality evaluation device in the second embodiment are the same as those in the first embodiment.

[0146] In Fig. 15, the flow processing device 13 produces a product PR from a raw material RM according to processing condition data PCD by a process including a flow process. The information acquisition device 14 acquires specific information at two different points in the processing time between before and after the processing in at least one of the processes, and outputs the acquired information data IFD. The quality evaluation device 15 evaluates the quality of the product PR, and outputs quality data QD that is the evaluation result. The processing condition data PCD from the flow processing device 13, the information acquisition device 14, and the quality data QD from the quality evaluation device 15 are each transmitted to the learning device 10C.

[0147] In FIG. 16, the learning device 10C acquires processing condition data PCD from the flow processing device 13, information data IFD from the information acquisition device 14, and quality data QD from the quality evaluation device 15. The learning device 10C derives differential calculated value data CD from the information data IFD. The processing condition data PCD and this calculated value data CD are set as learning input data IDL. The processing condition data PCD, information data IFD, calculated value data CD derived from the information data IFD, and quality data QD are associated with each other by a common ID assigned to one product PR, as shown in FIG. 17. The processing condition data PCD, calculated value data CD, and quality data QD constitute a known data set.

[0148] The information processing device 11C receives the trained model TM from the learning device 10C. The information processing device 11C provides the trained model TM with prediction processing condition data PCDF, which is processing condition data for a product PR whose quality is unknown, and prediction calculation value data CDF, which is calculation value data for a product PR whose quality is unknown. The prediction calculation value data CDF is data derived from prediction information data IFDF, which is information data IFD for a product PR whose quality is unknown, in the same manner as the calculation value data CD. The trained model TM outputs quality prediction data QFD according to the prediction processing condition data PCDF and the prediction calculation value data CDF.

[0149] 18, a storage device 60AC of a learning device 10C stores processing condition data PCD from the flow processing device 13, information data IFD from the information acquisition device 14, and quality data QD from the quality evaluation device 15. Calculated value data CD derived from the information data IFD and a machine learning model M are also stored. A plurality of sets of processing condition data PCD, information data IFD, calculated value data CD, and quality data QD are stored.

[0150] The first RW control unit 75C reads out the calculated value data CD, the processing condition data PCD, and the quality data QD from the storage device 60AC, and outputs them to the learning unit 77C. The first RW control unit 75C also reads out the machine learning model M from the storage device 60AC, and outputs the machine learning model M to either the learning unit 77C or the transmission control unit 78C. The first RW control unit 75C also stores the machine learning model M from the learning unit 77C in the storage device 60AC.

[0151] As shown in FIG. 19, the learning unit 77C has a first processing unit 85C, an evaluation unit 86C, and an update unit 87C. The first processing unit 85C provides the learning input data IDL (processing condition data PCD, calculated value data CD) to the machine learning model M, and causes the machine learning model M to output the learning output data ODL. The learning output data ODL is, for example, the number average molecular weight, the same as the quality data QD. The first processing unit 85C outputs the learning output data ODL to the evaluation unit 86C. The machine learning model M, whose prediction accuracy has reached a preset level, is stored in the storage device 60AC as a trained model TM by the first RW control unit 75C.

[0152] 20, the storage device 60BC of the information processing device 11C stores the trained model TM from the learning device 10C and the prediction information data IFDF from the information acquisition device 14. The storage device 60BC also stores the prediction processing condition data PCDF. The prediction processing condition data PCDF may be input by an operator via the input device 65B. More specifically, an input screen in which an input box for each item of the prediction processing condition data PCDF is displayed on the display 64B, and the prediction processing condition data PCDF may be input via the input screen. The prediction processing condition data PCDF and the prediction information data IFDF are processing condition data and information data of a product PR whose quality is unknown and whose quality is predicted using the trained model TM.

[0153] The storage device 60BC also stores prediction calculation value data CDF derived from the prediction information data IFDF.

[0154] When the second operating program 110C is started, a CPU 62BC of a computer constituting the information processing device 11C cooperates with the memory 61 and the like to function as a second RW control unit 115C, a second derivation unit 116C, a second processing unit 117C, and a display control unit 118C.

[0155] The second RW control unit 115C, like the first RW control unit 75C of the learning device 10C, controls reading of various data in the storage device 60BC and storage of various data in the storage device 60BC. The second RW control unit 115C reads prediction information data IFDF from the storage device 60BC (acquisition of information) and outputs the prediction information data IFDF to the second derivation unit 116C. In addition, the second RW control unit 115C stores prediction calculation value data CDF from the second derivation unit 116C in the storage device 60BC. An autoencoder may also be stored in the storage device 60BC, and calculation value data may be derived from information data such as an image using the autoencoder.

[0156] The second RW control unit 115C reads the trained model TM from the storage device 60BC and outputs the trained model TM to the second processing unit 117C. By reading the trained model TM from the storage device 60BC, the second RW control unit 115C acquires the trained model TM.

[0157] The second RW control unit 115C reads out the calculation value data CDF for prediction and the processing condition data PCDF for prediction from the storage device 60BC (obtaining the calculation value, obtaining the condition value), and outputs these to the second processing unit 117C. By reading out the calculation value data CDF for prediction and the processing condition data PCDF for prediction from the storage device 60BC, the second RW control unit 115C obtains the calculation value data CDF for prediction and the processing condition data PCDF for prediction.

[0158] The second derivation unit 116C receives the calculation information data IFDF from the second RW control unit 115C. The second derivation unit 116C derives calculation value data CDF for calculation from the calculation information data IFDF.

[0159] The second processing unit 117C receives the process condition data for prediction PCDF, the calculated value data for prediction CDF, and the trained model TM from the second RW control unit 115C. The second processing unit 117C provides the process condition data for prediction PCDF and the calculated value data for prediction CDF to the trained model TM to predict quality (quality prediction). The second processing unit 117C outputs quality prediction data QFD, which is the quality prediction result by the trained model TM, to the display control unit 118C. The quality prediction data QFD is, for example, number average molecular weight, like the quality data QD.

[0160] In this manner, it is possible to predict the quality of a product obtained by a process including one or more treatments. In the second embodiment, the quality of the product is predicted using the treatment condition data in addition to the calculated value data, so that it may be possible to predict the quality with higher accuracy than in the first embodiment.

[0161] In each of the above embodiments, the processing conditions received by the setting unit 25 of the flow processing device 13 are used as the processing condition data PCD, but this is not limited to this. Actual measurements detected by the first flow velocity sensor 35, the second flow velocity sensor 36, the third flow velocity sensor 37, and the temperature sensor 38 may also be used as the processing condition data PCD.

[0162] In each of the above embodiments, the quality prediction display screen 120 has been exemplified as an output form of the quality prediction data QFD, but this is not limiting. Instead of or in addition to the quality prediction display screen 120, a form in which the quality prediction data QFD is printed out on a paper medium, or a form in which the quality prediction data QFD is output as a data file may be adopted.

[0163] The hardware configuration of the computers that constitute the learning devices 10 and 10C and the information processing devices 11 and 11C can be modified in various ways. For example, the learning device 10 and the information processing device 11 may be integrated into one computer. Also, at least one of the learning device 10 and the information processing device 11 may be configured as multiple computers separated as hardware for the purpose of improving processing capacity and reliability. For example, the function of the first derivation unit 76 of the learning device 10 and the function of the learning unit 77 are distributed and assigned to two computers. In this case, the learning device 10 is configured from two computers. The above also applies to the learning device 10C and the information processing device 11C.

[0164] In this way, the hardware configuration of the computers of the learning devices 10 and 10C and the information processing devices 11 and 11C can be appropriately changed according to the required performance such as processing power, safety, reliability, etc. Furthermore, not limited to hardware, application programs such as the first operating programs 70 and 70C and the second operating programs 110 and 110C can also be duplicated or stored in a distributed manner in multiple storage devices for the purpose of ensuring safety and reliability.

[0165] In each of the above embodiments, for example, the hardware structure of the processing units that perform various processes, such as the first RW control units 75 and 75C, the first derivation units 76 and 76C, the learning units 77 and 77C, the transmission control units 78 and 78C, the second RW control units 115 and 115C, the second derivation units 116 and 116C, the second processing units 117 and 117C, and the display control unit 118, can use the various processors shown below. As described above, the various processors include CPUs 62A, 62AC, 62B and 62BC, which are general-purpose processors that execute software (first operating programs 70 and 70C, second operating programs 110 and 110C) and function as various processing units, as well as dedicated electrical circuits that are processors having a circuit configuration specially designed for executing specific processes, such as programmable logic devices (PLDs) that are processors whose circuit configuration can be changed after manufacture, such as field programmable gate arrays (FPGAs), and application specific integrated circuits (ASICs).

[0166] A single processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs and / or a combination of a CPU and an FPGA). Also, multiple processing units may be configured with a single processor.

[0167] As an example of configuring multiple processing units with one processor, first, there is a form in which one processor is configured with a combination of one or more CPUs and software, and this processor functions as multiple processing units, as represented by computers such as client and server. Second, there is a form in which a processor is used to realize the functions of the entire system including multiple processing units with one IC (Integrated Circuit) chip, as represented by a system on chip (SoC). In this way, the various processing units are configured using one or more of the above various processors as a hardware structure.

[0168] Furthermore, more specifically, the hardware structure of these various processors can be an electric circuit that combines circuit elements such as semiconductor elements.

[0169] Furthermore, the quality prediction by the information processing device can also be suitably used in a search device configured to search for processing conditions so as to obtain a desired quality based on the prediction result.

[0170] [Variations] The processing section 22 may be replaced with a processing section 45 shown in Fig. 21. In Fig. 21, the same components as those in Fig. 10 are given the same reference numerals, and the description thereof will be omitted.

[0171] The confluence 46 of the processing section 45 shown in Fig. 21 is composed of a first pipe section 47, a second pipe section 48, a third pipe section 49, and a fourth pipe section 50. The first pipe section 47 and the second pipe section 48 are connected in a straight line. Similarly, the third pipe section 49 and the fourth pipe section 50 are connected in a straight line. The first pipe section 47 and the second pipe section 48 intersect with the third pipe section 49 and the fourth pipe section 50 at right angles. In other words, the confluence 46 is cross-shaped.

[0172] The first pipe section 47 and the second pipe section 48 are connected to the first raw material supply section 20, and the third pipe section 49 is connected to the second raw material supply section 21. The fourth pipe section 50 is connected to the reaction section 31. The first raw material RM1 is supplied to the first pipe section 47 and the second pipe section 48 from the first raw material supply section 20, and the second raw material RM2 is supplied to the third pipe section 49 from the second raw material supply section 21. The first raw material RM1 and the second raw material RM2 join together in the fourth pipe section 50 and are sent to the reaction section 31 in a mixed state.

[0173] The first pipe section 47 and the second pipe section 48 are provided with a first flow velocity sensor 51 and a second flow velocity sensor 52 for detecting the flow velocity of the first raw material RM1 passing through the first pipe section 47 and the second pipe section 48. Furthermore, the third pipe section 49 is provided with a third flow velocity sensor 53 for detecting the flow velocity of the second raw material RM2 passing through the third pipe section 49. Furthermore, the fourth pipe section 50 is provided with a fourth flow velocity sensor 54 for detecting the flow velocity of the raw material mixture passing through the fourth pipe section 50.

[0174] In this case, the system controller 26 controls the rotation speed of the pump of the first raw material supply unit 20 to adjust the flow rate of the first raw material RM1 in accordance with the average value of the flow rate of the first raw material RM1 detected by the first flow rate sensor 51 and the flow rate of the first raw material RM1 detected by the second flow rate sensor 52. The system controller 26 also controls the rotation speed of the pump of the second raw material supply unit 21 in accordance with the flow rate of the second raw material RM2 detected by the third flow rate sensor 53 to adjust the flow rate of the second raw material RM2.

[0175] When the processing section 45 is used, the system controller 26 adjusts the flow rate of the first raw material RM1 by controlling the rotation speed of the pump of the first raw material supply unit 20 so that the average value of the flow rate of the first raw material RM1 detected by the first flow rate sensor 51 and the flow rate of the first raw material RM1 detected by the second flow rate sensor 52 matches the flow rate of the first raw material RM1 registered in the processing condition data PCD. Similarly, the system controller 26 adjusts the flow rate of the second raw material RM2 by controlling the rotation speed of the pump of the second raw material supply unit 21 so that the flow rate of the second raw material RM2 detected by the third flow rate sensor 53 matches the flow rate of the second raw material RM2 registered in the processing condition data PCD.

[0176] <Information processing method> The information processing method according to the present disclosure includes: 1. An information processing method for predicting the quality of a product obtained by a process including one or more treatments, comprising: In at least one treatment of the process, at least one of chemical information and physical information of the treated object and the treated object is obtained at two different points in time between before and after the treatment; Obtaining the calculated difference between the two numerical values ​​obtained from the above information, and Using the difference as an explanatory variable and the quality of the product as a response variable, a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable is used to predict the quality of the product based on the calculated value. Includes.

[0177] Each configuration of the information processing method is derived from the above-mentioned configuration of the information processing device.

[0178] <Program> The program according to the present disclosure is A program for causing a computer to execute information processing for predicting the quality of a product obtained by a process including one or more treatments, In at least one treatment of the process, at least one of chemical information and physical information of the treated object and the treated object is obtained at two different points in time between before and after the treatment; Obtaining the calculated difference between the two numerical values ​​obtained from the above information, and Using the difference as an explanatory variable and the quality of the product as a response variable, a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable is used to predict the quality of the product based on the calculated value. Includes.

[0179] Each configuration of the information processing method is derived from the above-mentioned configuration of the information processing device. EXAMPLES

[0180] The present disclosure will be described in more detail below with reference to examples, but the present disclosure is not limited to these examples.

[0181] In order to demonstrate the high accuracy of quality prediction by the information processing according to the present disclosure, a polymer synthesis process using the flow processing device 13 shown in Fig. 10 was assumed, including comparative examples, and the number average molecular weight (quality) of polystyrene (product) was predicted. The data used in the quality prediction is described in the following (1) to (3).

[0182] (1) Infrared Spectroscopy As chemical information at point a2 before the second flow treatment and point d2 after the second flow treatment, infrared spectra as shown in FIG. 3 were obtained. The infrared spectra were obtained at level numbers 1 to 10 as shown in FIG. 22, and the treatment conditions for each level were different as shown in FIG. 25. The difference in the integrated intensity at the peak (hereinafter, sometimes simply referred to as "integral intensity"), which is a numerical value obtained from the infrared spectrum, was used as an explanatory variable, and calculation value data of the difference was obtained as shown in FIG. 23. At that time, the characteristic wavenumber in this reaction was determined by quantum chemical calculation, and peaks A to N in the infrared spectrum corresponding to the determined wavenumber were specified. In addition, these wavenumbers included wavenumbers derived from by-products, that is, these peaks included peaks derived from by-products. Peaks A to C are due to aromatic ring CH stretching, peaks D and E are due to aliphatic CH stretching, peaks F to I are due to aromatic ring monosubstitutions, peaks J and K are due to aromatic ring C=C stretching, peak L is due to CH bending, peak M is due to CH out-of-plane bending, and peak N is due to aromatic ring bending. Note that data on peaks D to M are omitted in Figures 22 and 23.

[0183] (2) Mixing ratio (quantity of state of the flow field) At point b1 during the first flow processing and point d1 after the first flow processing, the mixing ratio (state quantity of the flow field) was acquired as physical information. The numerical value obtained from the mixing ratio, here the difference of the mixing ratio itself, was used as an explanatory variable, and the calculated value data of the difference was acquired as shown in Figure 24. The level numbers in Figure 24 correspond to the level numbers in Figures 22 and 23.

[0184] The mixing ratio was obtained by simulating the mixed state of the first raw material RM1 (a solution in which polystyryllithium is dissolved in a solvent) and the second raw material RM2 (methanol) by computational fluid analysis. Specifically, an area in which the two raw materials (first raw material RM1, second raw material RM2) exist without mixing, and an area in which the two raw materials are mixed (mixed area) were visualized. The first raw material RM1 was displayed in white, the second raw material RM2 in black, and the mixed area in gray to visualize the mixed state. The mixing ratio was then calculated from the area of ​​the mixed area relative to the area of ​​the entire area.

[0185] (3) Processing conditions As shown in FIG. 25, processing conditions corresponding to level numbers 1 to 10 shown in FIGS. 22 and 23 were prepared.

[0186] [Example 1] Using the information processing device 11 described in the first embodiment, the number average molecular weight (quality) of polystyrene (product) obtained in a polymer synthesis process was predicted. The learning device 10, the flow processing device 13, and other configurations are also as described in the first embodiment. As the machine learning algorithm, the Gradient Boosting Regressor of the scikit-learn library in Python was used.

[0187] The differences between level numbers 2 to 10 shown in Figure 23 were used as explanatory variables, and the number average molecular weight was used as the response variable. Using a trained model TM that was machine-learned based on a known data set of explanatory variables and response variables, the number average molecular weight of polystyrene of level number 1 was predicted based on the calculated value described in level number 1. For level number 2, the number average molecular weight of polystyrene was predicted in the same manner as above based on the known data set obtained from levels other than level number 2. The same operation was repeated to predict the number average molecular weights of polystyrene for level numbers 3 to 10.

[0188] The coefficient of determination was calculated from the predicted values ​​of the number average molecular weights of level numbers 1 to 10 and the values ​​of the number average molecular weights shown in Figure 23. The results are shown in Figure 26.

[0189] [Comparative Example 1] The number average molecular weights of standard numbers 1 to 10 were predicted and the coefficient of determination was calculated in the same manner as in Example 1, except that the integrated intensity at point d2 shown in Fig. 22 was used as the explanatory variable instead of the difference. The results are shown in Fig. 26.

[0190] [Example 2] Using the information processing device 11C described in the second embodiment, the number average molecular weight (quality) of polystyrene (product) obtained in the polymer synthesis process was predicted. The learning device 10C, the flow processing device 13, and other configurations are also as described in the second embodiment. The machine learning algorithm is the same as that in Example 1.

[0191] The differences between level numbers 2 to 10 shown in Figure 23 and the processing conditions for level numbers 2 to 10 shown in Figure 25 were used as explanatory variables, and the number average molecular weight was used as the response variable.The number average molecular weight of polystyrene of level number 1 was predicted based on the calculated values ​​and condition values ​​described in level number 1 using a trained model TM that was machine-learned based on a known data set of explanatory variables and response variables. For level number 2, the number average molecular weight of polystyrene was predicted in the same manner as above based on the known data set obtained from levels other than level number 2. The same operation was repeated to predict the number average molecular weights of polystyrene for level numbers 3 to 10.

[0192] The coefficient of determination was calculated from the predicted values ​​of the number average molecular weights of level numbers 1 to 10 and the values ​​of the number average molecular weights shown in Figure 23. The results are shown in Figure 26.

[0193] [Comparative Example 2] The number average molecular weights of standard numbers 1 to 10 were predicted and the coefficient of determination was calculated in the same manner as in Example 2, except that the integrated intensity at point d2 shown in Figure 22 was used as the explanatory variable instead of the difference. The results are shown in Figure 26.

[0194] [Example 3] Using the information processing device 11C described in the second embodiment, the number average molecular weight (quality) of polystyrene (product) obtained in the polymer synthesis process was predicted. The learning device 10C, the flow processing device 13, and other configurations are also as described in the second embodiment. The machine learning algorithm is the same as that in Example 1.

[0195] The differences between level numbers 2 to 10 shown in Figure 24 and the processing conditions for level numbers 2 to 10 shown in Figure 25 were used as explanatory variables, and the number average molecular weight was used as the response variable.The number average molecular weight of polystyrene of level number 1 was predicted based on the calculated values ​​and condition values ​​described in level number 1 using a trained model TM that was machine-learned based on a known data set of explanatory variables and response variables. For level number 2, the number average molecular weight of polystyrene was predicted in the same manner as above based on the known data set obtained from levels other than level number 2. The same operation was repeated to predict the number average molecular weights of polystyrene for level numbers 3 to 10.

[0196] The coefficient of determination was calculated from the predicted values ​​of the number average molecular weights of Level No. 1 to Level No. 10 and the values ​​of the number average molecular weights shown in Figure 24. The results are shown in Figure 26.

[0197] [Comparative Example 3] The number average molecular weights of level numbers 1 to 10 were predicted and the coefficient of determination was calculated in the same manner as in Example 3, except that the mixing ratio at point b1 shown in Figure 24 was used as the explanatory variable instead of the difference. The results are shown in Figure 26.

[0198] As can be seen from Example 1 and Comparative Example 1, Example 1, in which the calculated value of the difference in integrated intensity was used as the explanatory variable, had a higher coefficient of determination and higher quality prediction accuracy than Comparative Example 1, in which the integrated intensity itself was used as the explanatory variable. The same is true for the comparison between Example 2 and Comparative Example 2, and the comparison between Example 3 and Comparative Example 3, and Example 2 and Example 3 have high prediction accuracy.

[0199] Furthermore, as can be seen from Examples 1 and 2, Example 2, in which the processing conditions were used as explanatory variables in addition to the calculated value of the integrated intensity difference, had a higher coefficient of determination than Example 1 and was more accurate in predicting quality.

[0200] The technology of the present disclosure can be appropriately combined with the various embodiments and various modified examples described above. In addition, it is not limited to the above embodiments, and various configurations can be adopted without departing from the gist of the technology. Furthermore, the technology of the present disclosure extends to a storage medium that non-temporarily stores a program, in addition to a program.

[0201] The above description and illustrations are detailed descriptions of the parts related to the technology of the present disclosure, and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, function, action, and effect is an example of the configuration, function, action, and effect of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above description and illustrations, within the scope of the gist of the technology of the present disclosure. In addition, in order to avoid confusion and to facilitate understanding of the parts related to the technology of the present disclosure, the above description and illustrations omit explanations of technical common sense that do not require explanation in order to enable the implementation of the technology of the present disclosure. [Explanation of symbols]

[0202] 10, 10C Learning Device 11, 11C Information processing device 12 Network 13 Flow Processing Device 14 Information acquisition device 15 Quality evaluation equipment 20 1st raw material supply section 21 2nd raw material supply section 22, 45 Reaction Section 23 Temperature control section 24 Collection and Disposal Section 25 Setting section 26 System Controller 30, 46 Junction 31 Reaction section 32~34 1st pipe section ~ 3rd pipe section 35~37 1st flow velocity sensor ~ 3rd flow velocity sensor 38 Temperature Sensor 40 Collection Department 41 Disposal Department 42 Three-way valve 47~50 1st pipe section~4th pipe section 51 to 54 First flow velocity sensor to fourth flow velocity sensor 60, 60A, 60AC, 60B, 60BC Storage Devices 61 Memory 62, 62A, 62AC, 62B, 62BC CPU 63 Communications Department 64, 64B Display 65, 65B Input Devices 66 Bus Line 70, 70C First operating program (operating program of learning device) 75, 75C First read / write control unit (first RW control unit) 76, 76C 1st derivation part 77, 77C Learning Section 78, 78C Transmission control section 85, 85C First processing section 86, 86C Evaluation Section 87, 87C update section 110, 110C Second operating program 115, 115C second read / write control unit (second RW control unit) 116, 116C second derived part 117, 117C Second Processing Section 118, 118C Display control unit a1, a n , b1, b n , c1, c n , d1, d n Differences in the processing time between before and after processing CD Calculated value data (calculated value data of difference) CDF forecast calculation data (calculated value data for difference forecast) IDL training input data IFD Information Data IFDF Prediction Information Data L reaction path length M Machine learning model ODL training output data PCD Processing Condition Data Processing condition data for PCDF prediction PR Products QD Quality Data QFD quality prediction data RM raw material RM1, RM2 1st raw material, 2nd raw material TM pre-trained model Φ Reaction path diameter

Claims

1. An information processing device for predicting quality of a product obtained by a process including one or more treatments, At least one processor, the processor comprising: In at least one of the processes, at least one of chemical information and physical information of the object to be treated and the object to be treated is obtained at two different points in time between before the treatment and after the treatment; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; The difference is set as an explanatory variable, and the quality of the product is set as a response variable. Using a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable, a prediction of the quality of the product is performed based on the calculated value. In the acquisition of the information, a spectroscopic spectrum is acquired as the chemical information in at least one of the processes; In obtaining the calculated value, for a characteristic wavenumber or wavenumber region in the spectrum that changes due to the at least one of the processes, the calculated value of the intensity at the wavenumber or the integrated intensity in the wavenumber region is obtained. Information processing device.

2. The processor acquires a condition value of a processing condition for at least one of the processes; the explanatory variables include the processing conditions in addition to the difference, The information processing apparatus according to claim 1 , wherein the prediction of the quality of the product is performed based on the calculated value and the condition value.

3. The information processing device according to claim 1 , wherein the processor determines the wave number or the wave number region based on a quantum chemical calculation.

4. The information processing device according to any one of claims 1 to 3, wherein the characteristic wave number or wave number region includes a wave number or wave number region derived from a by-product.

5. At least one of the processes is a flow process using a flow path; 5. The information processing apparatus according to claim 1, wherein the processor, in acquiring the information, acquires a state quantity of a flow field as the physical information of the flow process.

6. An information processing device for predicting quality of a product obtained by a process including one or more treatments, At least one processor, the processor comprising: In at least one of the processes, at least one of chemical information and physical information of the object to be treated and the object to be treated is obtained at two different points in time between before the treatment and after the treatment; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; The difference is set as an explanatory variable, and the quality of the product is set as a response variable. Using a trained model that has been machine-learned based on a known data set of the explanatory variables and the response variable, a prediction of the quality of the product is performed based on the calculated value. At least one of the processes is a flow process using a flow path; The processor acquires a state quantity of a flow field as the physical information of the flow processing in acquiring the information. Information processing device.

7. The processor acquires a condition value of a processing condition for at least one of the processes; the explanatory variables include the processing conditions in addition to the difference, The information processing apparatus according to claim 6 , wherein the prediction of the quality of the product is performed based on the calculated value and the condition value.

8. An information processing method in which a processor included in an information processing device predicts quality of a product obtained by a process including one or more operations, comprising: The processor, acquiring at least one of chemical information and physical information of the object to be treated and the object to be treated at two different points in time between before the treatment and after the treatment in at least one of the processes; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; and predicting the quality of the product based on the calculated value using a trained model that is machine-learned based on a known data set of the explanatory variables and the target variable, with the difference as an explanatory variable and the quality of the product as a target variable; Includes performing, The processor, In the acquisition of the information, a spectroscopic spectrum is acquired as the chemical information in at least one of the processes; In obtaining the calculated value, for a characteristic wavenumber or wavenumber region in the spectrum that changes due to the at least one of the processes, the calculated value of the intensity at the wavenumber or the integrated intensity in the wavenumber region is obtained. Information processing methods.

9. An information processing method in which a processor included in an information processing device predicts quality of a product obtained by a process including one or more operations, comprising: The processor, acquiring at least one of chemical information and physical information of the object to be treated and the object to be treated at two different points in time between before the treatment and after the treatment in at least one of the processes; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; and predicting the quality of the product based on the calculated value using a trained model that is machine-learned based on a known data set of the explanatory variables and the target variable, with the difference as an explanatory variable and the quality of the product as a target variable; [0023] At least one of the processes is a flow process using a flow path; The processor acquires a state quantity of a flow field as the physical information of the flow processing in acquiring the information. Information processing methods.

10. A program for causing a computer to execute information processing for predicting the quality of a product obtained by a process including one or more treatments, comprising: acquiring at least one of chemical information and physical information of the object to be treated and the object to be treated at two different points in time between before the treatment and after the treatment in at least one of the processes; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; and predicting the quality of the product based on the calculated value using a trained model that is machine-learned based on a known data set of the explanatory variables and the target variable, with the difference as an explanatory variable and the quality of the product as a target variable; Including, In the acquisition of the information, a spectroscopic spectrum is acquired as the chemical information in at least one of the processes; In obtaining the calculated value, for a characteristic wavenumber or wavenumber region in the spectrum that changes due to the at least one of the processes, the calculated value of the intensity at the wavenumber or the integrated intensity in the wavenumber region is obtained. A program for causing a computer to carry out information processing.

11. A program for causing a computer to execute information processing for predicting the quality of a product obtained by a process including one or more treatments, acquiring at least one of chemical information and physical information of the object to be treated and the object to be treated at two different points in time between before the treatment and after the treatment in at least one of the processes; Obtaining a calculated value of the difference between the two numerical values ​​obtained from the information; and predicting the quality of the product based on the calculated value using a trained model that is machine-learned based on a known data set of the explanatory variables and the target variable, with the difference as an explanatory variable and the quality of the product as a target variable; Including, At least one of the processes is a flow process using a flow path; In the information acquisition, a state quantity of a flow field is acquired as the physical information of the flow processing. A program for causing a computer to carry out information processing.

Images