Design assistance system and design assistance method

Abstract

A design assistance system 1: calculates a value indicating a non-uniformity of a state in a spatial region on the basis of information relating to the state at each position when a process is performed with respect to a raw material distributed in the spatial region; generates a state distribution model 108 that outputs data indicating the non-uniformity of the state after the process on the basis of the value; acquires a value of data representing the non-uniformity of the state after the process by inputting a value of a set process design parameter into the state distribution model 108; acquires a quality parameter of a deliverable by inputting each item of data representing a possible state, identified from the value of the data, into a quality prediction model 105; and, on the basis of each quality parameter, changes the value of the process design parameter such that the quality of the product falls within a predetermined range.

Description

Design support system and design support method

[0001] The present invention relates to a design support system and a design support method.

[0002] INCORPORATION BY REFERENCE This application claims priority to Japanese Patent Application No. 2023-090509, filed May 31, 2023, the contents of which are incorporated herein by reference. In the chemical and materials industries, technical studies are conducted in stages from the start of product development to mass production. Specifically, studies are typically conducted in four phases: (1) laboratory experiments, (2) bench tests, (3) pilot plant tests, and (4) full-scale plant design and operation. In laboratory experiments, beaker-level trial production is conducted while varying the raw material combination and various reaction conditions to determine the product's basic physical properties and yield. In bench tests, small-scale equipment is assembled to produce the product obtained in the laboratory experiments, and various unit operations and processes, such as distillation or adsorption, are designed, and detailed physical properties are investigated and designed. In pilot plant tests, a pilot plant similar in scale to the full-scale plant is constructed, and operating conditions, operation methods, and instrumentation systems are determined. In full-scale plant design, the plant is designed and operated based on the technology established in the pilot plant. Note that bench testing and pilot testing may be integrated into other phases.

[0003] The specifications of the equipment, facilities, manufacturing processes, etc. vary greatly between such phases. For example, if the manufacturing process is a reaction process, the volume of the reactor may increase by several tens to several hundreds of times as the phase progresses.

[0004] As a result, differences in specifications between phases can result in situations where the desired quality (physical properties and yield) cannot be achieved. A specific example is given using a reaction process. First, as the volume of the reactor increases, the surface area per unit volume decreases, reducing the efficiency of heat transfer from the outside. Alternatively, the surface area per unit volume in contact with the agitator blades decreases, potentially reducing agitation efficiency. As a result, heat transfer delays occur and temperature or concentration variations increase. Generally, reaction rates depend on the temperature and concentration of each substance. Therefore, the ratio of by-products and the distribution of physical properties vary depending on the reaction rate. Therefore, operating conditions such as reactor size, agitator blade shape, and reaction temperature time profile must be designed while taking into account the changes in physical properties or yield due to the aforementioned heat transfer delays and temperature or concentration variations. The design work for each specification in response to the progression of phases is collectively referred to as scale-up.

[0005] Scaling up is a time-consuming task. For example, if the reaction conditions devised in lab experiments cannot be reproduced in bench tests, the physical properties or yields must be redesigned in the lab experiments, resulting in significant rework. Furthermore, pilot plant tests can take as long as six months to a year and a half due to the large capacity and long manufacturing time. For this reason, scaling up is a bottleneck in product development for chemical manufacturers and material manufacturers.

[0006] As a technique for efficiently scaling up, for example, Non-Patent Document 1 discloses a method for optimizing plant operating conditions with as few experiments as possible immediately after completing verification in a pilot plant test and starting production in the manufacturing equipment of the actual plant, thereby enabling the production of a desired product. Specifically, the method discloses a method for obtaining good quality with as few experiments as possible by using a combination of transfer learning and Gaussian process regression to build a model (quality prediction model) that predicts the average and variability of quality from the operating conditions, and conducting experiments under conditions that are likely to increase the objective function while taking the variability into account.

[0007] Manabu Kano, Ryosuke Yoshizaki, Bayesian optimization of operating conditions during scale-up of manufacturing facilities, Chemical Engineering, Vol. 80, No. 12, pp. 769-772 (2016).

[0008] Non-Patent Document 1 requires that the number of outputs of the quality prediction model be the same for pilot plants and full-scale plants. In other words, the number of KPIs and their components to be optimized must be the same before and after scale-up. However, when scaling up from laboratory testing, the number of KPIs and their components often differs. For example, in bench equipment, pilot plants, and full-scale plants, yield must be optimized taking into account heat transfer delay and temperature or concentration variations, but these are generally not considered in laboratory testing. Furthermore, while laboratory testing considers quality parameters as univariates, it is possible to optimize multiple statistics related to quality parameters (e.g., mean, maximum, minimum, etc.) at bench, pilot, and full-scale plant scales. In such cases, the method described in Non-Patent Document 1 cannot be applied.

[0009] Another approach that can be considered, different from that described in Non-Patent Document 1, is to perform a first-principles simulation on a plant scale (such as a coupled simulation of fluid and reaction in the case of a reaction process, or a kneading simulation in the case of an extrusion process), precisely estimate the conditions such as concentration and temperature within the process, and estimate the quality based on the results. However, because the above simulation requires an enormous computational load, it is difficult to repeat trials while changing conditions such as equipment specifications and operating conditions, or to perform a simulation of the entire time period from the start to the end of production.

[0010] As such, it has been difficult to build a system that utilizes data obtained from laboratory tests to efficiently understand the quality and yield in bench tests, pilot plants, and full-scale plants, and to obtain products of stable quality in full-scale plants.

[0011] The present invention has been made in view of the above background, and its purpose is to provide a design support system and a design support method that are capable of designing a process for obtaining deliverables of appropriate quality in processes for producing deliverables that are carried out at different scales.

[0012] One aspect of the present invention for solving the above problem includes a storage device for storing a quality prediction model that receives input of data representing a state and data representing a process performed on a raw material in the state, and outputs data representing the quality of a product obtained from the raw material after the process; a state distribution model construction process that acquires state information at each position in the spatial domain when a process is performed on raw materials distributed in a spatial domain, calculates a value representing the non-uniformity of the state in the spatial domain based on each acquired piece of information, and receives input of process design parameters that are data defining a process performed on the raw material present in the spatial domain based on the calculated value representing the non-uniformity and data defining the process, and generates a state distribution model that outputs data representing the non-uniformity of the state in the spatial domain during or after the process; and a process for setting values ​​of the process design parameters and inputting the set values ​​of the process design parameters into the state distribution model. a manufacturing quality prediction process that identifies a plurality of states that can be taken in the spatial domain from the acquired data values ​​that represent the non-uniformity of the state in the spatial domain, and inputs data representing each of the identified states and the value of the set process design parameter into the quality prediction model for each of the states, thereby acquiring quality parameters of a product obtained from the raw material after the process indicated by the value of the set process design parameter; and a parameter search process that changes the value of the set process design parameter based on the acquired quality parameters so that the quality of the product obtained from the raw material falls within a predetermined range.

[0013] According to the present invention, it is possible to design a process for producing deliverables of appropriate quality in processes for producing deliverables carried out at different scales. Configurations and effects other than those described above will become apparent from the following description of the embodiments.

[0014] 1 is a diagram illustrating an example of the configuration of a design support system in this embodiment. FIG. 1 is a diagram illustrating an example of laboratory experiment data. FIG. 2 is a diagram illustrating an example of the configuration of a plant experiment. FIG. 3 is a diagram illustrating an example of plant experiment data obtained by a plant experiment. FIG. 4 is a diagram illustrating an example of CAE simulation data. FIG. 5 is a diagram illustrating an example of a state distribution model database. FIG. 6 is a diagram illustrating an example of hardware provided in an information processing device constituting the design support system. FIG. 7 is a flow diagram illustrating an example of a first process design parameter determination process. FIG. 8 is a diagram illustrating an example of a state distribution model construction screen. FIG. 9 is a flow diagram illustrating details of a state distribution model construction process. FIG. 10 is a flow diagram illustrating details of a state distribution generation process. FIG. 11 is a flow diagram illustrating details of a manufacturing quality prediction process. FIG. 12 is a flow diagram illustrating details of a process search process. FIG. 13 is a diagram illustrating an example of a process search screen. FIG. 14 is a flow diagram illustrating details of a state distribution model regeneration process. FIG. 15 is a diagram illustrating an example of manufacturing performance data. FIG. 16 is a flow diagram illustrating an example of a second process design parameter determination process.

[0015] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description and drawings are examples for explaining the present disclosure, and some omissions and simplifications have been made as appropriate for clarity of explanation. The present disclosure can be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

[0016] The design support system according to this embodiment is used when conducting small-scale experiments (for example, small-scale trial production at the beaker level having a spatial region where a small-scale process is carried out, such as a laboratory experiment; hereinafter, small-scale experiments may also be referred to as laboratory experiments) and large-scale tests (large-scale testing or production of a product in a large-scale device having a spatial region where a large-scale process is carried out, such as a bench test, pilot plant test, or actual plant design (operation); hereinafter, large-scale production may also be referred to as plant experiments) for the development and production of a product (an end product, such as a resin or fiber, produced by carrying out a predetermined process (physical or chemical process) on raw materials.

[0017] Specifically, the design support system first constructs a state distribution model that calculates values ​​(in this embodiment, the mean and variance values ​​when the internal states are considered as random variables) representing the nonuniformity of the distribution of internal states (spatial distribution of factors that affect the quality of the deliverable, such as temperature) in a spatial region of a given size where the process is executed, based on data from the results of an actual plant experiment (large-scale test) or data obtained by executing a coupled simulation (hereinafter also referred to as a spatial process model) using a coupled model described below (hereinafter referred to as CAE (Computer Aided Engineering) simulation data). The state distribution model outputs values ​​of the distribution of internal states in the spatial region in the production of products at various scales (under various equipment sizes and operating conditions). However, in this case, only a small amount of data is required to be obtained from the plant experiment or coupled simulation.

[0018] Furthermore, the design support system builds a quality prediction model that predicts the quality (hardness, etc.) of products obtained from raw materials based on actual data from laboratory experiments (small-scale experiments) that were actually conducted on a small scale such that the distribution of internal states as described above does not need to be taken into account.

[0019] The design support system sets the external conditions of the large-scale test (for example, the size or shape of the plant equipment, mixing conditions, or plant operating conditions; hereinafter referred to as process design parameters) and inputs the set process design parameters into a state distribution model, thereby outputting values ​​(in this embodiment, the mean and variance values ​​when the internal states are considered to be random variables) that represent the heterogeneity or range of the distribution of the internal states of the plant due to the large-scale test.The design support system 1 then samples state values ​​(temperature values) based on this heterogeneity value, and inputs each sampled value and the corresponding process design parameter into a quality prediction model, thereby predicting the quality of the product manufactured under the state (temperature) and process design parameters.

[0020] Here, since the original data of the input values ​​to the quality prediction model are random variables as described above, the product quality values ​​output (predicted) from the quality prediction model are also output as values ​​that assume quality variability corresponding to the random variables. This makes it possible to grasp deviations and variability in product quality when scaling up from small-scale experiments to large-scale tests. In other words, the design support system can search for process design parameters that take into account the variability in predicted values ​​of the quality of products manufactured in large-scale tests, so that the quality of the manufactured products falls within a specified range and satisfies plant constraints, etc.

[0021] In the following description, the process performed on the raw material is assumed to be a reaction process, but the present invention can be widely applied to other types of processes such as other physical processes (e.g., mixing processes). For example, the present invention may be applied to continuous manufacturing processes such as chemicals, pharmaceuticals, or steel. Furthermore, the present invention may be applied to processes involving equipment or facilities where it is important to understand the internal state, such as energy plants such as thermal power plants.

[0022] <Configuration of Design Support System> FIG. 1 is a diagram showing an example of the configuration of a design support system 1 according to this embodiment.

[0023] The design support system 1 is connected to laboratory experiment data 21 that stores experimental data from small-scale experiments, plant experiment data 22 that stores experimental data from large-scale tests, and manufacturing performance data 23 that stores data on the manufacturing performance of products actually manufactured on an actual scale in a reactor 30 or the like, which will be described later.

[0024] The design support system 1 also stores a quality prediction model 105 , a response characteristic model 106 , CAE simulation data 107 , a state distribution model 108 , a CAE simulation model 109 , and a state distribution model database 115 .

[0025] 2 is a diagram showing an example of the lab experiment data 21. The lab experiment data 21 is data storing the experimental conditions and experimental results of small-scale experiments. Specifically, the lab experiment data 21 stores, for each small-scale experiment, an experiment ID 211 for identifying the experiment, experimental conditions 212 for the experiment, reaction characteristics 213 for the experiment, and quality 214 of the product obtained by the experiment, in association with each other.

[0026] The experimental conditions 212 and reaction characteristics 213 are data related to the raw materials or the process performed on the raw materials. Specifically, the experimental conditions 212 include the composition of the raw materials used in the experiment (ingredients and their blending), the reaction temperature, and the reaction time. The reaction characteristics 213 include the reaction rate, reaction heat, and solution viscosity measured by the experiment under those experimental conditions. The quality 214 includes the hardness and elongation of the substance obtained by the experiment under those experimental conditions. Note that the data items described here are only examples. The experimental conditions 212 and reaction characteristics 213 may include parameters representing the details of various physical or chemical processes. Furthermore, the quality 214 may include parameters representing various physical or chemical properties of the product.

[0027] The laboratory experiment data 21 is used to generate the quality prediction model 105 and the reaction characteristic model 106 .

[0028] (Configuration of Plant Experiment and Plant Experiment Data) Next, Fig. 3 is a diagram showing an example of the configuration of a plant experiment, and Fig. 4 is a diagram showing an example of plant experiment data 22 obtained by the plant experiment.

[0029] First, as shown in Figure 3, in this plant experiment, thermometers 51 to 55 are installed at various locations inside a vessel (reactor 30) having a spatial region capable of accommodating raw materials. A predetermined operation system or control system information processing (for example, the design support system 1) measures the temperatures of the thermometers 51 to 55 at predetermined time intervals while operating the reactor 30, and registers the measured temperatures in the plant experiment data 22. Alternatively, an operator may register the measured temperatures in the plant experiment data 22 while operating the reactor 30. Preferably, the positions of the thermometers 51 to 55 are set so that the temperature values ​​measured at each position vary.

[0030] Next, as shown in FIG. 4 , the plant experiment data 22 is data in which, for each plant experiment, a plant experiment ID 221, equipment and facility specifications 222 in the plant experiment, the composition of the raw materials in the plant experiment (raw material blend 223), and plant experiment results 224 are associated with one another. The plant experiment results 224 include parameters (temperatures) measured by each sensor (e.g., thermometers 51 to 55) at each time. Note that the time and temperature in the plant experiment results 224 have a different data granularity, but have the same structure as the time and temperature in the simulation results 1074 in the CAE simulation data 107, which will be described later. The plant experiment data 22 is used to generate a state distribution model 108.

[0031] 5 is a diagram showing an example of CAE simulation data 107. The CAE simulation data 107 is a database that records the execution results of a coupled simulation model in which a CAE simulation model 109 and a reaction characteristics model 106 are coupled. Specifically, the CAE simulation data 107 stores, for each coupled simulation, an ID 1071 for identifying the coupled simulation, equipment and facility specifications 1072 that are data that define the reaction process, the composition of raw materials (raw material blend 1073) set as input values ​​in the coupled simulation, and the results 1074 of the coupled simulation, all in association with each other.

[0032] The equipment and facility specifications 1072 are data relating to the vessel (reactor) in which the process is carried out or the apparatus, device, or facility used in the process (hereinafter, these reactors, devices, devices, and facilities are collectively referred to as "apparatus"), which are set as input values ​​in the coupled simulation. The equipment and facility specifications 1072 include, for example, the shape of the reactor, the radius of the reactor, the height of the reactor, the shape of the agitator blades to be installed in the reactor, the area of ​​the agitator blades, and the heat removal method to be set in the reactor. The raw material blend 1073 includes the components of the raw materials and their blending ratios. In addition, the simulation results 1074 include the flow velocity (three-dimensional flow velocity), temperature, and viscosity at each position (three-dimensional position coordinates) in the reactor at each time.

[0033] 6 is a diagram showing an example of the state distribution model database 115. The state distribution model database 115 stores data sources for generating each state distribution model 108.

[0034] Specifically, the state distribution model database 115 has each of the following data: an ID of the state distribution model 108, a type of data used to generate the state distribution model 108 (learning data source 1152), information specifying the content of the data source (learning data ID 1153), and model data 1154 storing the content of the state distribution model 108. The learning data source 1152 is data indicating, for example, either the CAE simulation data 107 or the plant experiment data 22. The learning data ID 1153 is information specifying a specific data source of the CAE simulation data 107 or the plant experiment data 22 (specifically, the data ID 2111 of the plant experiment data 22 or the data ID 1071 of the CAE simulation data 107).

[0035] Next, as shown in FIG. 1 , the design support system 1 has functional units (programs) including a quality prediction model construction unit 101, a reaction characteristic model construction unit 102, a CAE simulation unit 103, a state distribution model construction unit 104, a state distribution generation unit 110, a manufacturing quality prediction unit 111, a process search unit 112, and a user interface unit 113.

[0036] The quality prediction model construction unit 101 generates a quality prediction model 105 that predicts the quality of a product (deliverable) manufactured based on specified experimental conditions and reaction characteristics of raw materials, based on data from small-scale experiments.

[0037] Specifically, the quality prediction model 105 is a model that receives data representing a state (temperature) and data representing a reaction process to be performed on raw materials in that state (experimental conditions 212 and reaction characteristics 213 of the laboratory experiment data 21), and outputs data representing the quality of a product generated from the raw materials in that reaction process. Note that in this embodiment, it is assumed that the quality prediction model 105 is generated and stored in advance.

[0038] In this embodiment, the quality prediction model 105 is generated based on the laboratory experiment data 21. That is, the input values ​​of the quality prediction model 105 are data corresponding to the experiment conditions 212 and the reaction characteristics 213 of the laboratory experiment data 21. The output value of the quality prediction model 105 is data corresponding to the quality 214 of the laboratory experiment data 21.

[0039] The reaction characteristic model construction unit 102 generates a reaction characteristic model 106. In this embodiment, the reaction characteristic model 106 is a model that calculates changes in characteristics from the start to the end of a reaction in a large-scale test. The reaction characteristic model 106 is a model that uses the composition of raw materials, the reaction temperature, or the reaction time as input values, and the reaction rate, the reaction heat, or the solution viscosity as output values.

[0040] In this embodiment, the reaction characteristic model 106 is generated based on a predetermined machine learning algorithm and the laboratory experiment data 21. The reaction characteristic model 106 is a model trained by a neural network (such as a convolutional neural network (CNN)) that includes, for example, an input layer to which input values ​​are input, one or more intermediate layers (hidden layers) that extract and output feature quantities from the input values, and an output layer that outputs output values ​​from the feature quantities. For example, the reaction characteristic model 106 is generated by machine learning using, as training data, experimental conditions 212 (corresponding to input values) and reaction characteristics 213 (corresponding to output values) of the laboratory experiment data 21.

[0041] In this embodiment, it is assumed that the reaction characteristic model 106 is generated and stored in advance. Note that the reaction characteristic model 106 may be a mathematical formula that uses, for example, a predetermined reaction stoichiometry, chemical descriptors, or molecular weight distribution as input variables, and a reaction rate, reaction heat, viscosity, or the like as output variables.

[0042] The CAE simulation unit 103 stores a CAE simulation model 109. The CAE simulation unit 103 executes the CAE simulation model 109 to generate CAE simulation data 107.

[0043] The CAE simulation model 109 is a model that predicts the behavior of fluids in a reactor in a large-scale test. Specifically, the CAE simulation model 109 includes a governing equation (Navier-Stokes equation) that uses reactor specifications (shape, etc.) as input values ​​and calculates the state (temperature, viscosity, etc.) at each position in the reactor at each time after the material movement process (fluid movement) in the reactor.

[0044] The CAE simulation unit 103 executes a coupled simulation model that combines the CAE simulation model 109 and the reaction characteristics model 106. That is, the coupled simulation model is a model that receives as input data defining the process (flow reaction process) of raw materials distributed in the spatial region of the reactor and outputs the state of the raw materials at each position in the reactor at each time after the start of the flow reaction process.

[0045] In this embodiment, the reaction characteristic model constructing unit 102 generates a coupled simulation model in advance based on the state distribution model database 115 .

[0046] The coupled simulation model may include a different type of simulation model from the models described above. For example, if the process for manufacturing a product includes an extrusion process, a kneading simulation model may be included. Furthermore, in a process that does not involve a chemical reaction (e.g., a steel manufacturing process), the reaction characteristic model construction unit 102 and the reaction characteristic model 106 are not necessary, and the coupled simulation model may be composed of only the CAE simulation model 109.

[0047] While such large-scale coupled simulations can be performed using various commercial simulator software or software such as OSS (Open Source Software), running these software programs requires significant loads and time. For example, to accurately track the progress of reactions and changes in viscosity, fine-grained calculations are required in both time and space. Furthermore, because solutions consistent with the boundary conditions at the reactor walls and agitator blades must always be calculated, there is a risk that the calculations will not converge or that physically meaningless solutions will be obtained under certain conditions. For this reason, such software programs can only be run within a very limited range, and using them alone typically does not provide the information necessary to identify process design parameters. However, in this embodiment, even in such situations (even when only a limited number of simulation results are available), process design parameters can be identified by combining the state distribution model 108 generated based on the results of these software programs with the quality prediction model 105.

[0048] Next, the state distribution model construction unit 104 generates a state distribution model 108 using the CAE simulation data 107 or the plant experiment data 22 .

[0049] The state distribution model 108 receives as input data on the composition of the raw materials (ingredients of the raw materials and their blending ratios) and data (process design parameters) that define the process to be performed on the raw materials present in the reactor, and outputs data that represents the non-uniformity of the state inside the reactor at each time point after the start of the process (during or after the execution of the process).

[0050] Here, the process design parameters in the input values ​​are data that define the manufacturing process. In this embodiment, for example, the process design parameters are equipment specifications (shape, size, etc.) or operating conditions (process method such as stirring method or cooling method, or time profile of the state for smooth reaction, etc.), but are not limited thereto.

[0051] The output value of the state distribution model 108, which is data representing the non-uniformity, is a state parameter whose distribution may change or vary depending on the execution of the process, and serves as an input value for the quality prediction model 105. In this embodiment, the output value of the state distribution model 108 is the temperature inside the reactor, but is not limited to this and may be, for example, the concentration or density of the raw material or product inside the reactor. Furthermore, the data representing the non-uniformity is a probability distribution (mean and variance), but other data representing non-uniformity may also be used. While a typical CAE simulation calculates the state at each position in three-dimensional space, this embodiment expresses the state (variation) at each position as a whole using a probability distribution, which is a parameter that can represent the state at each position, thereby reducing the cost associated with calculating or storing data on the state at each position.

[0052] The state distribution model 108 has a probability density function f(y(t)) of a state (here, temperature) expressed, for example, by the following Equation 1, and further has a function of generating a state y(t) by sampling according to the probability density at time t.

[0053] Here, state y(t) is the state value at time t, σ(t) is the variance of state y(t) at time t, and μ(t) is the mean of state y(t) at time t.

[0054] Furthermore, the time evolution of σ(t) and μ(t) is determined by the variance, mean, and process design parameters (external inputs) at past times before time t, as shown in Equation 2.

[0055] Here, θ(t) is an external input at time t and corresponds to a process design parameter. In this embodiment, θ(t) is a vector consisting of equipment specification data (fixed value data independent of time t) and process manipulated variables (variable values ​​dependent on time t). d is the order.

[0056] Although Equation 1 and Equation 2 are equations for the case where the state y is a single variable, similar descriptions are possible for the case where the state y is a multivariate.

[0057] Next, the state distribution generation unit 110 inputs the raw material composition (raw material blend, etc.) and the values ​​of the process design parameters into the state distribution model 108 generated by the state distribution model construction unit 104, and outputs data (i.e., mean and variance) representing the distribution (non-uniformity) of the state inside the reactor at each time after the start of the process.

[0058] The manufacturing quality prediction unit 111 calculates (samples) multiple specific state values ​​that can be taken inside the reactor from the data representing the distribution (non-uniformity) of states inside the reactor output by the state distribution generation unit 110, and inputs the state data, the composition of the raw materials, and the process design parameters into the quality prediction model 105, thereby predicting the quality of the product to be produced after the process is completed.

[0059] In this embodiment, the manufacturing quality prediction unit 111 randomly samples temperature data based on the mean and variance, and inputs the sampled temperature data into the quality prediction model 105 to predict quality (hardness and elongation), thereby identifying the quality variation (histogram or probability density function) in the target process.

[0060] The process search unit 112 searches for process design parameter contents and raw material composition such that the quality predicted by the manufacturing quality prediction unit 111 falls within a predetermined range. That is, the process search unit 112 can search for process design parameter contents and raw material composition by performing optimization to search for the probability (proportion) at which the quality of the deliverable falls within a predetermined range, taking into account the variation in quality and adding predetermined KPIs or constraints.

[0061] The user interface unit 113 has a function (input function) of receiving parameters required for a series of processes performed by the design support system 1 from the user 40, and a function (output function) of providing calculation results, control results, etc. to the user 40.

[0062] Specifically, the input function accepts information input from the user 40 using, for example, a keyboard, a touch panel, a voice input device, a gaze detection device, etc. The output function outputs information via, for example, a monitor display, a printer, a voice synthesizer, etc.

[0063] The user interface unit 113 may be provided in an information processing device different from the design support system 1. This information processing device is, for example, a personal computer such as a laptop, notebook, tablet, or desktop computer operated by the user 40, a smartphone, or a wearable terminal such as a goggle or wristwatch.

[0064] FIG. 7 illustrates an example of hardware included in an information processing device constituting the design support system 1. The information processing device 200 includes, for example, a processor 201 (control device), a memory 202, an external storage device 203, a communication device 204, an output device 205, an input device 206, and a read / write device 207, which are interconnected via a communication line 208. The processor 201 is not limited to a central processing unit (CPU), but may be any other device with a computing function. The external storage device 203 is a device that rewritably stores relatively large amounts of data, such as a hard disk device, a flash memory device, a magneto-optical disk device, or an optical disk device. The communication device 204 is configured as a network interface card (NIC) or the like, and communicates with external devices via a communication network CN (a wireless or wired communication network such as the Internet, a local area network (LAN), a wide area network (WAN), or a dedicated line). The output device 205 is a device used by the user interface unit 113, such as a monitor display or a printer. The input device 206 is a device used by the user interface unit 113, and may be, for example, a keyboard, a pointing device, or a touch panel. The read / write device 207 reads and writes information from and to a storage medium MM that non-temporarily stores computer programs. The communication line 208 may be a system bus that connects devices within a single computer, or a communication network that connects multiple computers. In other words, the design support system 1 can be realized by providing the main functions of the design support system 1 on multiple computers and connecting these computers to each other via a communication network.

[0065] Various data stored in the design support system 1 or used for processing can be used by the processor 201 reading it from the memory 202 or the external storage device 203. The functions of each functional unit of the design support system 1 can be realized by the processor 201 loading a predetermined computer program stored in the external storage device 203 into the memory 202 and executing it.

[0066] The above-mentioned predetermined computer program may be stored (downloaded) into the external storage device 203 from the storage medium MM via the read / write device 207 or from a network via the communication device 204, and then loaded onto the memory 202 and executed by the processor 201. The computer program may also be directly loaded onto the memory 202 from the storage medium MM via the read / write device 207 or from a network via the communication device 204, and then executed by the processor 201.

[0067] In the following, the case where the design support system 1 is configured by one information processing device will be described, but all or part of these functions may be distributed across one or more computers, such as a cloud, and the functions of the design support system 1 may be realized by communicating with each other via a network. Next, the processing performed by the design support system 1 will be described.

[0068] First, a process for determining process design parameters (hereinafter referred to as a first process design parameter determination process) when generating the state distribution model 108 based on the CAE simulation model 109 will be described.

[0069] <First Process Design Parameter Determination Processing> FIG. 8 is a flowchart illustrating an example of the first process design parameter determination processing.

[0070] The design support system 1 sets fixed values ​​of process design parameters assuming a plant experiment (S801). Specifically, the user interface unit 113 may accept input of fixed values ​​to be set in a large-scale test from the user 40, or may receive fixed values ​​from a database prepared in advance.

[0071] Here, in this embodiment, the user interface unit 113 sets the reactor shape to "cylinder", the stirring blade shape to "propeller", and the cooling method to "jacket" as fixed values ​​of the process design parameters.

[0072] Then, the design support system 1 sets a plurality of combinations of various values ​​based on the fixed values ​​of the process design parameters, and executes coupled simulation using the set plurality of combinations as input values ​​(S802).

[0073] Specifically, first, the CAE simulation unit 103 receives input from the user 40 of various values ​​of process design parameters to be set in the large-scale test.

[0074] For example, the CAE simulation unit 103 receives input of the reactor radius r, reactor height h, and nominal values ​​r0, h0, and S0 of the stirring blade area S from the user 40, and generates conditions in which r, h, and S are changed based on these input values. It is assumed that each parameter will not change significantly from the nominal value, and generates, for example, (r, h, S) = (0.25r0, 0.25h0, 0.25S0), (0.5r0, 0.5h0, 0.5S0), (r, h, S), (2r0, 2h0, 2S0), and (4r0, 4h0, 4S0).

[0075] The design support system 1 then executes the coupled simulation by inputting the values ​​set above into the coupled simulation models (the CAE simulation model 109 and the reaction characteristics model 106). The design support system 1 stores the data output by the coupled simulation models in the CAE simulation data 107.

[0076] Next, the design support system 1 executes a state distribution model construction process S803 for generating a state distribution model 108 based on the CAE simulation data 107 stored in S802. Details of the state distribution model construction process S803 will be described later.

[0077] Thereafter, the design support system 1 sets candidate values ​​(external inputs θ0, θ1, ...) of the process design parameters in order to search for the process design parameters (S804). For example, the design support system 1 may set a value obtained by adding a predetermined increase or decrease value to the nominal value of the condition value of the process design parameter (a value obtained by adding noise), or may accept an input of an initial value from the user 40.

[0078] Then, the design support system 1 executes a state distribution generation process S805 in which the state distribution (average and variance) in the reactor is calculated by inputting the candidate values ​​of the process design parameters set in S804, etc., into the state distribution model 108 generated in S803 and executing the state distribution model 108. The state distribution generation process S805 will be described in detail later.

[0079] (State Distribution Model Construction Screen) FIG. 9 is a diagram showing an example of a state distribution model construction screen 1200 displayed by the design support system 1. As shown in FIG.

[0080] The state distribution model construction screen 1200 has a prerequisites input field 1201 that accepts input of the composition of the raw materials and fixed values ​​of the process design parameters from the user, a design parameter input field 1202 that accepts input of combinations of values ​​of the process design parameters from the user, and a state distribution model construction specification field 1203 that is specified by the user when generating the state distribution model 108 based on the data input in the prerequisites input field 1201 and the design parameter input field 1202.

[0081] The state distribution model construction screen 1200 also has a design parameter setting field 1208 that receives, from the user, input of a combination of values ​​of process design parameters to be used as input values ​​for the generated state distribution model 108, a state distribution model execution specification field 1204 that is specified by the user when executing the state distribution model 108 based on the input values ​​input in the design parameter setting field 1208, a graph display field 1205 that displays a graph representing the execution result of the state distribution model 108, and a quality prediction display field 1206, which will be described later.

[0082] The graph displayed in the graph display field 1205 has time (elapsed time from the start of the reaction) on the horizontal axis and the state inside the reactor (here, temperature) on the vertical axis. This graph includes a first graph 1209 that represents the range of state values ​​and a second graph 1210 that represents changes in the average value within that range. This graph also includes a time series graph (sample path graph 1211) of a virtual state (temperature) generated by connecting values ​​1212 sampled at each time from the range represented by the first graph 1209. Details of the sampled values ​​1212 and the sample path will be described later.

[0083] Furthermore, a state limit value 1215 (in this figure, an upper limit value of temperature) is displayed in the graph display field 1205. The limit value 1215 will be described in detail later.

[0084] 8, the design support system 1 executes a manufacturing quality prediction process S806 for predicting the quality of each product to be manufactured by inputting information on each state sampled from the distribution of states output in S805 into the quality prediction model 105. The manufacturing quality prediction process S806 will be described in detail later.

[0085] Then, the design support system 1 executes a process search process S807 in which the quality of the process design parameters and the composition of raw materials set in S804 are evaluated based on the product qualities (quality distribution) predicted in the manufacturing quality prediction process S806. Specifically, the process search unit 112 inputs the quality distribution, the process design parameters, and the composition of raw materials into a predetermined objective function for evaluating the quality of the process design parameters, etc., and acquires the values. The process search process S807 will be described in detail later.

[0086] If the change in the value of the objective function acquired in S807 becomes equal to or less than a predetermined value (if the difference between the value acquired in S807 and the value acquired previously in S807 becomes equal to or less than a predetermined value) (S808: YES), the process of S809 is executed, and if the change in the value of the objective function acquired in S807 exceeds the predetermined value (S808: NO), the process search unit 112 repeats the process of S804 to set different process design parameters.

[0087] Although an example has been shown in which the optimization process from S804 to S808 is performed using the so-called gradient method, the optimization problem may be solved using other optimization methods, such as the greedy method or random search.

[0088] In S809, the design support system 1 executes a state distribution model regeneration process S809 for updating the state distribution model 108 based on the process design parameters changed in the process search process S807. Details of the state distribution model regeneration process S809 will be described later. This completes the first process design parameter determination process.

[0089] As a result, it is possible to search for design parameters of a process that is likely to achieve the target quality in actual manufacturing. Next, details of the state distribution model construction process S803, the state distribution generation process S805, the manufacturing quality prediction process S806, and the process search process S807 will be described.

[0090] <State Distribution Model Construction Process> FIG. 10 is a flowchart illustrating the state distribution model construction process S803 in detail.

[0091] The CAE simulation unit 103 extracts one target process design parameter that meets the conditions set in S801 from the CAE simulation data 107 (S901).

[0092] In this embodiment, the CAE simulation unit 103 extracts data of a data ID in which the reactor shape is cylindrical, the stirring blade shape is propeller, and the cooling method is jacket from the CAE simulation data 107 .

[0093] Next, the CAE simulation unit 103 acquires the simulation results 1074 (spatial distribution of temperature at each time) from the CAE simulation data 107 relating to the coupled simulation performed based on the process design parameters extracted in S901 and the set raw material configuration (raw material composition) (S902).

[0094] The CAE simulation unit 103 randomly samples multiple positions from the simulation results 1074 obtained in S902 and obtains time series of temperatures at the multiple positions, thereby generating multiple virtual time series of temperatures (sample paths) and storing the generated sample paths as learning data (S903).

[0095] The CAE simulation unit 103 may perform sampling automatically, or may accept a designation of data to be sampled (sampled values ​​1212) from the user 40 on the state distribution model construction screen 1200. The number of data to be sampled at each time may be constant, or may be a number according to the variance.

[0096] Based on the learning data stored in S903, the CAE simulation unit 103 calculates the state value y(t) at each time point t as shown in Equation 3-5. ０ )...y(t ｎ－１ ) (S904). This identification can be performed using a probabilistic identification technique such as Gaussian process regression.

[0097] Here, N Ｎ is an N-dimensional normal distribution, Equation 4 is the mean at each time point, and Equation 5 is the covariance matrix between time points. Also, μ is the mean, and V is the variance.

[0098] The CAE simulation unit 103 checks whether all target process design parameters have been extracted in S901 (S905). If all target process design parameters have been extracted in S901 (S905: YES), the CAE simulation unit 103 executes the process of S906, and if there are process design parameters that have not been extracted in S901 (S905: NO), the CAE simulation unit 103 repeats the process of S902 to select those process design parameters.

[0099] In S906, the CAE simulation unit 103 identifies the state distribution model 108 shown in Equation 6 based on the stochastic process model generated by the above processing, the state (mean μ(t), variance V(t, t)) obtained in S902, and the process design parameters (external input data) (S906). This completes the state distribution model construction processing S803.

[0100] The state distribution model 108 can be identified by a general system identification method, such as ARX model identification or Hammerstein-Wiener model identification, etc. This makes it possible to understand how the state distribution changes when the process design parameters change.

[0101] Here, the component (first component) related to the mean μ(t) in Equation 6 can also be identified by determining the structure of a differential equation from a theoretical equation for reaction heat based on the energy balance law or a reaction rate equation, and estimating its parameters using the mean μ(t), variance V(t, t), and external input data. In this case, the range of values ​​that the mean μ(t) can take is limited, so it is expected that it will be easier to predict the state distribution for process design parameters in unknown areas (i.e., areas where CAE simulation has not been performed).

[0102] <State Distribution Generation Processing> Next, FIG. 11 is a flowchart illustrating details of the state distribution generation processing S805.

[0103] The state distribution generation unit 110 acquires the process design parameters acquired in S804 as external inputs (S1001). The state distribution generation unit 110 also sets initial values ​​for the state (temperature) distribution (mean μ(t) and variance V(t, t)) (S1002). The initial value for the state (temperature) mean is set to, for example, the initial condition (initial temperature, etc.) at the time of product manufacturing. The initial value for the state (temperature) variance is set to zero because there is no unevenness.

[0104] The state distribution generation unit 110 calculates the mean and variance of the state (temperature) at each time by integrating the state distribution model shown in Equation 6 in the time direction based on the initial values ​​of the external input and state (temperature) distribution set in S1001 and S1002, respectively (S1003). This completes the state distribution generation process S805.

[0105] <Production Quality Prediction Processing> FIG. 12 is a flowchart illustrating the details of the production quality prediction processing S806.

[0106] First, the manufacturing quality prediction unit 111 sets the number of samplings Ns (S1101). The number of samplings is the number of times to sample the state value at each time. The manufacturing quality prediction unit 111 sets the number of samplings Ns, for example, by reading values ​​recorded in a setting file created in advance by the user 40.

[0107] In addition, the manufacturing quality prediction unit 111 calculates the total number Nt of times (sampling times) at which quality prediction is performed and the sample times (T １ , T ２ , ...T Ｎｔ Specifically, the product quality prediction unit 111 specifies a range of possible sample time points (a time range from the start of the reaction to the end of the reaction) based on the reaction time in the laboratory experiment data 21. Then, the product quality prediction unit 111 sets Nt by receiving input of each sample time point within this range from the user 40.

[0108] For example, the manufacturing quality prediction unit 111 receives input of values ​​1212 at each time from the user 40 on a sample path graph 1211 displayed on the state distribution model construction screen 1200 (in the case of the same figure, Nt=4).

[0109] Then, the production quality prediction unit 111 sets the index i for counting sampling to 1 (S1103), and executes the process of S1104.

[0110] In S1104, the manufacturing quality prediction unit 111 calculates the product quality at the sample time t=T １ , T ２ , ..., T ＮｔFor each time point, a state value y(t)(T １ , T ２ , ..., T Ｎｔ ) (here, the temperature value) is extracted (sampled).

[0111] Next, the product quality prediction unit 111 sets the index j for the sampling time point to 1 (S1105), and executes the process of S1106.

[0112] In S1106, the manufacturing quality prediction unit 111 calculates the sample time T ｊ The state value y(t ｊ ) and the preset composition of raw materials are input to the quality prediction model 105, and the quality prediction model 105 is executed to calculate (predict) the quality Q of the product. Note that Q is calculated (predicted) by [q １ , q ２ , ..., q ｋ ] Ｔ and q ｋ is the value of quality k (e.g., hardness, elongation) at the end of the process.

[0113] The production quality prediction unit 111 stores the calculated quality Q (S1107).

[0114] The production quality prediction unit 111 repeats the processes of S1106 and S1107 for each time (j=1, . . . , Nt) (S1108, S1111) to calculate and store the quality Q corresponding to each sample path.

[0115] Thereafter, the manufacturing quality prediction unit 111 repeats the processes of S1104 to S1108 for each sample (i=1, . . . , Ns) (S1109, S1112) to calculate and store the quality Q for each sample.

[0116] Thereafter, the product quality prediction unit 111 outputs the distribution of the quality Q calculated by the above process in the form of a histogram, a probability density function, or the like (S1110).

[0117] For example, the manufacturing quality prediction unit 111 displays a graph 1213 relating to hardness and a graph 1214 relating to elongation in the quality prediction display field 1206 of the state distribution model construction screen 1200. The graph 1213 relating to hardness is a graph in which the horizontal axis represents the hardness value and the vertical axis represents the probability (probability density) of the occurrence of that hardness value. The graph 1214 relating to elongation is a graph in which the horizontal axis represents the elongation value and the vertical axis represents the probability (probability density) of the occurrence of that elongation value.

[0118] <Process Search Processing> Next, FIG. 13 is a flowchart illustrating the process search processing S807 in detail.

[0119] The process search unit 112 acquires data indicating the range of acceptable products (the range of required product quality) (S1301).

[0120] For example, the process search unit 112 receives input of the upper limit and lower limit of the range of non-defective products from the user 40. In this embodiment, the process search unit 112 calculates the hardness q １ and elongation q ２ The upper limit of (q １，ｍａｘ , q ２，ｍａｘ ) and the lower limit (q ２，ｍａｘ , q ２，ｍｉｎ ) are accepted from the user 40.

[0121] The process search unit 112 calculates each quality q based on the quality Q calculated in the manufacturing quality prediction process S806 and the range of non-defective products acquired in S1301. １ , q ２ , ... is the probability (proportion) that the range falls outside the range of good products, １ ), P(q ２ ), ...) are calculated (S1302). In this embodiment, the process search unit 112 calculates the defect rate using the following Equation 7.

[0122] This makes it possible to search for process design parameters that take into account the quality distribution according to the variability of the state within the process.

[0123] Based on the defect rate calculated in S1302, the process search unit 112 generates an objective function whose value decreases as the defect rate decreases, and a constraint condition equation related to the objective function. In this embodiment, the process search unit 112 generates an objective function related to each defect rate shown in the following Formula 8, and a constraint condition equation related to the state of the device shown in Formula 9.

[0124] where a and b are constants. θ is an external input, Tmin is the minimum temperature, Tmax is the maximum temperature, and T is the temperature. Note that T may be the average temperature, the maximum temperature (used for comparison with Tmax), or the minimum temperature (used for comparison with Tmin). Note that the process search unit 112 may automatically generate the objective function and the constraint condition equation, or may receive input of the objective function and the constraint condition equation from the user 40.

[0125] The process search unit 112 searches for the external input θ so that the value of the objective function is minimized while satisfying the constraint condition equation. ０ , θ １ , ... is changed (S1303). Note that the change of the external input θ in the search calculation can be performed by, for example, storing the calculated value of the objective function and changing the external input θ ０ , θ １ , ..., and apply the external input θ in the direction with the steepest gradient. ０ , θ １ , ... can be changed.

[0126] The objective function and constraint condition formulas described here are merely examples, and any formulas may be used as long as they optimize explanatory variables (quality or defect rate) under constraint conditions.

[0127] In addition, although the case where the reactor shape, agitator blade shape, cooling method, raw material combination, and blending ratio are fixed and the process design parameters are searched for has been described, these may also be search targets. In this case, the reactor shape, agitator shape, cooling method, and raw material combination are generally discrete variables, so the optimization problem shown in S1303 becomes a mixed integer programming problem. Therefore, a solver different from that shown in this embodiment is used for the optimization.

[0128] (Process Search Screen) FIG. 14 is a diagram showing an example of a process search screen 1400 displayed on the design support system 1 in the process search processing S807.

[0129] The process search screen 1400 includes a good product range input section 1401 that receives input of the good product range (upper and lower limits) for each type of quality from the user 40, an objective function definition section 1402 that displays the objective function or attaches input of the objective function from the user 40, a constraint condition definition section 1403 that displays information required for generating the constraint condition formula (e.g., the range (upper and lower limits) of the state (temperature)) or receives input of information required for generating the constraint condition formula from the user 40, and a process search execution section 1404 that is selected by the user when executing a search calculation.

[0130] The process search screen 1400 also includes a graph display section 1405 that displays a graph of the distribution of each quality of the product calculated during the search calculation process, a defect rate display section 1406 that displays the defect rate calculated during the search calculation process, and a process design parameter display section 1407 that displays the values ​​of the process design parameters calculated by the search calculation.

[0131] Graph display section 1405 displays a graph 1408 for each type of product quality, with the horizontal axis representing the quality value and the vertical axis representing the probability (probability density) of that quality value. Graph 1408 also displays the range (upper and lower limits) of each quality.

[0132] <State Distribution Model Regeneration Processing S809> FIG. 15 is a flowchart illustrating the state distribution model regeneration processing S809 in detail.

[0133] After the process search process S807 is executed, the user refers to the process search screen 1400 to check the predicted quality of the manufactured product, and determines the process design parameters (e.g., specifications of the actual plant equipment) to be set in the large-scale test, and then performs test production or actual production of the product (S1501).

[0134] Then, the process search unit 112 records the manufactured product and the adopted process design parameters in the following manufacturing performance data 23 (S1502). For example, the process search unit 112 may accept data input from the user 40, or may automatically acquire data from a predetermined experiment management system (for example, a DCS (Distributed Control System) that operates the reactor 30).

[0135] 16 is a diagram showing an example of the manufacturing performance data 23. The manufacturing performance data 23 stores a data ID 231, a product type 232 which is the type of raw material of the product, a model 233 which is data on the specifications of the equipment used to manufacture the product, operation data 234 which is data on the operating conditions used in manufacturing the product, and a quality 235 of the manufactured product in association with each other.

[0136] The product type 232 is data on the combination of the composition (mixture) of raw materials. The operation data 234 is, for example, time-series data on a target value given to a control device, an operation variable given to a process by a control device, or a controlled variable to be controlled in a process. In this embodiment, the target value is the target temperature of the reactor, the operation variable is the jacket cold water temperature for temperature control, and the controlled variable is the temperature (i.e., state) inside the reactor.

[0137] The quality 235 is not necessarily specified for each production run of a product, but is data only for the inspected product. Generally, there is variation in the quality inspection values, so it is possible to store a range.

[0138] 14, the range of hardness among the product qualities deviates from the non-defective range set in S1301. In this case, there is a possibility that the state distribution model 108 generated in the state distribution model regeneration process S809 is unreliable.

[0139] Then, as shown in FIG. 15, the process searching unit 112 modifies the process design parameters based on a specification from the user 40 or automatically (S1503).

[0140] Specifically, first, the process search unit 112 adds the manufacturing performance data 23 and the process design parameters linked thereto to the plant experiment data 22, and re-learns the state distribution model 108 based on the plant experiment data 22 to which the data has been added, to regenerate a new state distribution model 108. The process search unit 112 executes the process search processing S807 again based on the new state distribution model 108.

[0141] As a result, the accuracy of the state distribution model 108 and the prediction of product quality improves as actual manufacturing is repeated, allowing the user 40 to make an appropriate design without repeating trial-and-error simulations.

[0142] Next, a process for determining process design parameters (hereinafter referred to as second process design parameter determination process) when the state distribution model 108 is generated based on the plant experiment data 22 instead of the CAE simulation model 109 will be described.

[0143] <Second Process Design Parameter Determination Processing> FIG. 17 is a flowchart illustrating an example of the second process design parameter determination processing.

[0144] The process of S1701 is similar to S801 of the first process design parameter determination process.

[0145] Then, the design support system 1 acquires data on the results of the plant experiment (measurement data from the thermometers 51 to 55) from the reactor 30, and stores the acquired data in the plant experiment data 22 (S1702).

[0146] The state distribution model constructing unit 104 generates the state distribution model 108 based on the plant experiment data 22 in the same manner as in S803 (S1703). The subsequent processes of S1704 to S1708 are the same as the processes of S804 to S808.

[0147] The state distribution model 108 may be generated based on both the CAE simulation model 109 and the plant experiment data 22. In this case, the time granularity of both data is made to match.

[0148] As described above, the design support system 1 of this embodiment calculates values ​​(mean and variance) representing the heterogeneity of the state in the reactor based on the temperatures at each position inside the reactor when a process is performed on raw materials distributed in the reactor (specifically, the spatial domain of the reactor 30 used in the experiment or the spatial domain set in the coupled simulation). Then, based on the calculated values ​​representing the heterogeneity and data defining the process, the design support system 1 generates a state distribution model 108 to which process design parameters are input and which outputs data (mean and variance) representing the heterogeneity of the state in the reactor after the process. The design support system 1 then inputs the values ​​of the set process design parameters into the state distribution model 108 to obtain values ​​of data representing the heterogeneity of the state in the reactor during or after the process is performed. Based on the values ​​of the data (mean and variance) representing the heterogeneity, the design support system 1 samples a plurality of states that the reactor can take from the values ​​of the data representing the heterogeneity. The quality prediction model 105 inputs the data of the sampled states and the values ​​of the process design parameters for each state, thereby obtaining quality parameters (hardness and elongation) of the product. Then, the design support system 1 changes the values ​​of the process design parameters based on each quality parameter so that the quality of the product falls within a predetermined range.

[0149] That is, the design support system 1 predicts the quality of a product by inputting process design parameters and values ​​of each state data sampled from random variables into the quality prediction model 105 and executing the model, and can determine appropriate values ​​of the process design parameters based on the quality data. Here, the use of sampling based on random variables makes it possible to easily and appropriately represent (represent) the state distribution in a large process space, such as that assumed in a large-scale test. Then, the quality of the product based on this state distribution is predicted by the quality prediction model 105, which is based on, for example, a small-scale experiment, and the process design parameters can be determined based on this prediction.

[0150] As described above, the design support system 1 of this embodiment allows for process design to obtain deliverables of appropriate quality in processes for generating deliverables that are carried out at different scales. For example, it is possible to design a process that keeps quality-related KPIs within a predetermined range while limiting the number of experiments or simulations performed in each process or the number of re-experiments in small-scale processes, leading to a reduction in lead time in scale-up operations.

[0151] In addition, the design support system 1 of this embodiment generates a state distribution model 108 into which at least one of the composition of the raw materials, the specifications of the equipment used in the process, or the operating conditions of the equipment used in the process is input.

[0152] This makes it possible to generate a state distribution model 108 that can appropriately predict changes in the distribution of states (such as temperature) in the process.

[0153] Furthermore, the design support system 1 of this embodiment stores a quality prediction model 105 relating to a process that is carried out on a smaller scale than the process relating to the process design parameters in the state distribution model 108 .

[0154] This makes it possible to appropriately predict the quality of the manufactured product at each experimental level, even when scaling up from small-scale experiments such as laboratory experiments to large-scale tests such as plant experiments in product manufacturing.

[0155] Furthermore, the design support system 1 of this embodiment calculates the variation in quality of the deliverables (probability density) and the probability that the quality falls within a predetermined range (defect rate) based on the quality parameters of each state acquired through execution of the state distribution model 108, and outputs information on the calculated probability distribution and defect rate (graph display section 1405 of the process search screen 1400).

[0156] This allows accurate evaluation of the overall quality of products manufactured in large quantities.

[0157] Furthermore, the design support system 1 of this embodiment changes the values ​​of the process design parameters based on the quality parameters of each state acquired through execution of the state distribution model 108 so that the quality of the deliverable obtained from the raw materials falls within a predetermined range and satisfies predetermined constraints (operating conditions, constraint conditions), and outputs information regarding the range of the quality of the deliverable when the values ​​of the process design parameters are changed (graph display section 1405 or defect rate display section 1406 of the process search screen 1400).

[0158] This allows the user 40 to understand the quality of the product that will be manufactured while satisfying the necessary constraints.

[0159] Furthermore, the design support system 1 of this embodiment acquires the state of each position in the reactor during or after the execution of the process by inputting the contents of the process performed on the raw materials into the coupled simulation model, calculates values ​​(average and variance) representing the heterogeneity based on the acquired state of each position inside the reactor, and generates a state distribution model 108 based on the calculated values ​​representing the heterogeneity.

[0160] In this way, by generating the state distribution model 108 based on the execution results of a coupled simulation assuming a large-scale test or the like, the state distribution model 108 can be generated easily.

[0161] Furthermore, the design support system 1 of this embodiment calculates values ​​(mean and variance) representing heterogeneity based on the state of each position in the reactor during or after the execution of a predetermined process (plant experiment), and generates a state distribution model 108 based on the calculated values ​​representing heterogeneity.

[0162] In this way, by generating the state distribution model 108 based on an actually conducted large-scale test (plant experiment), it is possible to generate a state distribution model 108 with appropriate content.

[0163] Furthermore, the design support system 1 of this embodiment generates the state distribution model 108 by Gaussian process regression.

[0164] This makes it possible to generate a state distribution model 108 that calculates a probability distribution that is expected to actually occur.

[0165] Furthermore, the design support system 1 of this embodiment acquires quality data of the product after test manufacturing or actual manufacturing is performed using the changed process design parameters, generates new process parameters based on the acquired quality data, and regenerates the state distribution model 108 based on the generated parameters.

[0166] In this way, by regenerating the state distribution model 108 based on the manufacturing results of the actual product, it is possible to improve the prediction accuracy of the state of the state distribution model 108.

[0167] Furthermore, the design support system 1 of this embodiment further changes the values ​​of the process design parameters that have been set previously, based on the regenerated state distribution model 108 .

[0168] This allows the accuracy of the state distribution model 108 and the prediction of product quality to improve as the user 40, for example, repeats actual product manufacturing, making it possible to design appropriate process design parameters without repeating trial-and-error simulations.

[0169] Furthermore, the design support system 1 of this embodiment processes processes including reaction processes to be performed on raw materials.

[0170] This allows for consideration of the various reaction processes that may produce the product.

[0171] Furthermore, the design support system 1 of this embodiment receives data defining the reaction process of raw materials distributed in a reactor, and generates a coupled simulation model including a reaction characteristics model 106 that outputs the state of the raw materials at each position in the reactor during or after the execution of the reaction process.

[0172] This makes it possible to generate a coupled simulation model that takes into account the various reaction processes that can be used to manufacture a product.

[0173] Furthermore, the quality prediction model 105 of this embodiment is a model to which data representing the composition of the raw materials (such as the blending ratio) is further input, and the design support system 1 obtains the quality parameters of the product by further inputting the data representing the composition of the raw materials into the quality prediction model 105, and changes the process design parameters and the composition of the raw materials.

[0174] In this way, by making the composition of the raw materials a parameter of each model, it is possible to optimize not only the process design parameters but also the composition of the raw materials with respect to product quality.

[0175] The present invention is not limited to the above-described embodiments, and can be implemented using any components within the scope of the present invention. The above-described embodiments and modifications are merely examples, and the present invention is not limited to these contents as long as the characteristics of the invention are not impaired. Furthermore, although various embodiments and modifications have been described above, the present invention is not limited to these contents. Other aspects conceivable within the scope of the technical idea of ​​the present invention are also included within the scope of the present invention.

[0176] For example, part of the hardware provided in each device of this embodiment may be provided in another device.

[0177] Furthermore, each program of each device may be provided in another device, a program may consist of multiple programs, or multiple programs may be integrated into one program.

[0178] Furthermore, operating conditions other than those described in this embodiment may be used as process design parameters. For example, target values ​​(such as the target temperature of the reactor) in the production performance data 23 may be used as process design parameters. Operating conditions are generally defined as a time series, but the state distribution model 108 defined by Equations 6 to 8 can accept external inputs for each time. Therefore, various operating conditions can be included in the process design parameters, and process design parameters can be searched for.

[0179] 1 Design support system, 105 Quality prediction model, 106 Reaction characteristic model, 108 State distribution model

Claims

1. A storage device for storing a quality prediction model in which data representing a state and data representing a process performed on raw materials in the state are input, and data representing the quality of an end product obtained from the raw materials after the process is output; a state distribution model construction process for acquiring state information for each position in the spatial domain when a process is performed on raw materials distributed in the spatial domain, calculating a value representing the non-uniformity of the state in the spatial domain based on each acquired information, and generating a state distribution model in which process design parameters, which are data defining a process performed on the raw materials present in the spatial domain, are input based on the calculated value representing the non-uniformity and data defining the process, and data representing the non-uniformity of the state in the spatial domain during or after execution of the process is output; and a state distribution model execution process for setting values ​​of the process design parameters and inputting the values ​​of the set process design parameters into the state distribution model, thereby acquiring values ​​of data representing the non-uniformity of the state in the spatial domain related to the process design parameters for which the values ​​of the set process design parameters were set, during or after execution of a process indicated by the values ​​of the set process design parameters. a control device that executes a manufacturing quality prediction process that identifies a plurality of possible states in the spatial domain from the acquired data values ​​that represent the heterogeneity of states in the spatial domain, and acquires quality parameters of a product obtained from the raw materials after a process indicated by the values ​​of the set process design parameters by inputting each of the identified states and the values ​​of the set process design parameters into the quality prediction model for each of the states; and a parameter search process that changes the values ​​of the set process design parameters based on the acquired quality parameters so that the quality of the product obtained from the raw materials falls within a predetermined range.

2. The design support system according to claim 1, wherein the control device generates a state distribution model in which at least one of the composition of the raw materials, the specifications of the equipment used in the process, or the operating conditions of the equipment used in the process is input as the process design parameters in the state distribution model construction process.

3. The design support system according to claim 1, wherein said storage device stores a quality prediction model relating to a process carried out on a smaller scale than a process relating to said process design parameters in said state distribution model.

4. The design support system according to claim 1, wherein the control device, in the parameter search process, calculates the variability in the quality of the deliverable and the probability that the quality is within a specified range based on the acquired quality parameters for each state, and outputs information on the calculated variability and probability.

5. The design support system of claim 1, wherein the control device, in the parameter search process, changes the value of the set process design parameter based on each quality parameter of each acquired state so that the quality of the end product obtained from the raw material falls within a specified range and satisfies specified constraints, and outputs information regarding the range of quality of the end product when the value of the process design parameter is changed.

6. The design support system of claim 1, wherein the storage device stores a spatial process model in which data defining a process performed on raw materials distributed in a spatial domain is input and the state of each position in the spatial domain during and after execution of the process is output, and the control device executes: a spatial process model execution unit that sets the contents of a process performed on raw materials distributed in a spatial domain and acquires the state of each position in the spatial domain during and after execution of the set process by inputting the contents of the set process into the spatial process model; and a state distribution model construction process that calculates a value representing the non-uniformity based on the acquired states of each position in the spatial domain and generates the state distribution model based on the calculated value representing the non-uniformity and data representing the contents of the process input into the spatial process model.

7. The design support system of claim 1, further comprising a vessel having a specified spatial area and capable of performing a specified process on raw material under specified conditions, wherein the control device, in the state distribution model construction process, calculates a value representing the non-uniformity based on the state of each position in the spatial area of ​​the vessel during or after the execution of the specified process, and generates the state distribution model based on the calculated value representing the non-uniformity and data representing the contents of the specified process.

8. The design support system according to claim 1, wherein the control device generates the state distribution model by Gaussian process regression in the state distribution model construction process.

9. The design support system of claim 1, wherein the control device, in the state distribution model construction process, acquires quality data of the deliverable obtained by performing the process indicated by the changed process design parameters on the raw material, generates new data defining the performed process based on the acquired quality data, and regenerates the state distribution model based on the generated new data and the calculated value representing the non-uniformity.

10. The design support system according to claim 9, wherein the control device further changes values ​​of the set process design parameters based on the regenerated state distribution model.

11. The design support system according to claim 1, wherein the process includes a reaction process performed on the raw material.

12. The design support system according to claim 6, wherein the control device receives data defining a reaction process of raw materials distributed in a spatial domain, and executes a spatial process model construction process to generate a spatial process model that outputs the state of the raw materials at each position in the spatial domain during and after the reaction process is being executed.

13. The design support system of claim 1, wherein the storage device stores a quality prediction model into which data representing the composition of the raw materials is further input, and the control device, in the manufacturing quality prediction process, obtains quality parameters of the deliverable for each of the states by further inputting data representing the composition of the raw materials into the quality prediction model for each of the states, and in the parameter search process, changes the values ​​of the set process design parameters and the composition of the raw materials.

14. A design support method using an information processing device including a control device and a storage device for storing a quality prediction model in which data representing a state and data representing a process performed on raw materials in the state are input, and data representing the quality of a product obtained from the raw materials after the process is output, wherein the control device performs a state distribution model construction process in which: the control device acquires state information for each position in a spatial domain when a process is performed on raw materials distributed in the spatial domain, calculates a value representing the non-uniformity of the state in the spatial domain based on each acquired information, and generates a state distribution model in which process design parameters, which are data defining a process performed on the raw materials present in the spatial domain, are input, and data representing the non-uniformity of the state in the spatial domain after the process is output, based on the calculated value representing the non-uniformity and the data defining the process; and a state distribution model execution process in which: the control device sets values ​​of the process design parameters, and inputs the values ​​of the set process design parameters into the state distribution model, thereby acquiring values ​​of data representing the non-uniformity of the state in the spatial domain related to the process design parameters for which the values ​​of the set process design parameters have been set, after the process indicated by the values ​​of the set process design parameters. a manufacturing quality prediction process for identifying a plurality of possible states in the spatial domain from the acquired data values ​​representing the heterogeneity of states in the spatial domain, and inputting each piece of data representing each of the identified states and the value of the set process design parameter into the quality prediction model for each of the states, thereby acquiring quality parameters of a product obtained from the raw materials after the process indicated by the value of the set process design parameter, for each of the states; and a parameter search process for changing the value of the set process design parameter based on each of the acquired quality parameters so that the quality of the product obtained from the raw materials falls within a specified range.

Images