Porous body design method and porous body manufacturing method

Abstract

This porous body design method using a computer causes the computer to execute the following processes, each multiple times: a structure generation process for virtually generating, on the computer, the three-dimensional structure of a porous body on the basis of a generation parameter value for generating a porous body; a characteristic prediction / calculation process for predicting or calculating the characteristics of the porous body having the three-dimensional structure generated by the structure generation process; an evaluation process for evaluating the characteristics of the porous body predicted or calculated by the characteristic prediction / calculation process; and an optimization process for searching for the optimal generation parameter value by changing the generation parameter value. In the porous body design method, the three-dimensional structure of the porous body is determined on the basis of a result of evaluating the characteristics of the porous body by the evaluation process.

Description

Porous body design method, porous body manufacturing method

[0001] The present invention relates to a method for designing and manufacturing a porous body.

[0002] Porous materials, which have a large number of pores arranged three-dimensionally within the material, have been widely used in various applications, such as purifying exhaust gases emitted from internal combustion engines such as gasoline and diesel engines. When designing such porous materials, it is necessary to determine the three-dimensional structure of the porous material so that it can achieve properties appropriate for the application.

[0003] Conventional porous body design techniques include, for example, creating several prototypes of porous bodies based on the designer's experience and intuition, measuring the three-dimensional structures and properties of each of these prototypes, and then determining the final three-dimensional structure of the porous body. However, this design technique requires a lot of time and effort to create the prototypes and measure their properties, which limits the scope of the relationship between the three-dimensional structure and properties that can be explored. Therefore, there is a problem in that it is not always possible to determine a three-dimensional structure of a porous body that will provide the desired properties.

[0004] Meanwhile, in recent years, a method using simulation software has also become known, as described in, for example, Patent Document 1. With such simulation software, a three-dimensional model of a porous body is created, and the properties of the porous body can be calculated from this three-dimensional model.

[0005] Japanese Patent No. 6940786

[0006] In the method described in Patent Document 1, a sample is prepared from a partition wall piece cut out from an exhaust gas purification filter, and the sample is photographed using an X-ray CT scanner to obtain continuous tomographic images, which are then loaded into simulation software to create a three-dimensional model, thereby deriving the number of communicating holes in the porous exhaust gas purification filter. Therefore, the exhaust gas purification filter needs to be prepared in advance, which still poses the problem of the time and effort required for prototyping.

[0007] In view of the above problems, an object of the present invention is to provide a useful technique that enables an optimum structure to be determined early in the design of a porous body.

[0008] A porous body design method according to the present invention is a computer-based porous body design method, which comprises the steps of: a structure generation process for virtually generating a three-dimensional structure of the porous body on the computer based on generation parameter values ​​for generating the porous body; a property prediction / calculation process for predicting or calculating the properties of the porous body having the three-dimensional structure generated by the structure generation process; an evaluation process for evaluating the properties of the porous body predicted or calculated by the property prediction / calculation process; and an optimization process for varying the generation parameter values ​​to find optimal values ​​for the generation parameter values, and the three-dimensional structure of the porous body is determined based on the evaluation results of the properties of the porous body obtained by the evaluation process.A porous body manufacturing method according to the present invention manufactures the porous body using three-dimensional shape data based on the three-dimensional structure of the porous body determined by the porous body design method.

[0009] According to the present invention, a useful technique can be provided that enables an optimum structure to be determined early in the design of a porous body.

[0010] Fig. 1 is a block diagram showing the configuration of a porous body design device according to one embodiment of the present invention. Fig. 2 is a flowchart showing the processing flow of a porous body design device according to one embodiment of the present invention. Fig. 3 is a flowchart showing details of an initial learning model generation process. Fig. 4 is a diagram for comparing the three-dimensional structure of a porous body obtained by the porous body design device of this embodiment with the three-dimensional structure of a porous body according to a comparative example. Fig. 5 is a flowchart showing the flow of a manufacturing process for a porous body using a 3D printer.

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the following embodiment, a porous body design device that can be operated by a designer who designs a porous body and can determine the three-dimensional structure of a porous body having desired properties in a relatively short time will be described.

[0012] Fig. 1 is a block diagram showing the configuration of a porous body design apparatus according to one embodiment of the present invention. The porous body design apparatus 100 shown in Fig. 1 is a computer equipped with a control unit 1, a storage unit 2, a memory 3, an operation input device 4, and a display device 5, and is configured by connecting these devices to each other via a bus 6.

[0013] The control unit 1 is configured using, for example, a CPU (Central Processing Unit) and performs various processes and calculations to operate the porous body design apparatus 100. The control unit 1 executes programs stored in the memory unit 2 to realize the functional blocks of a structure generation unit 11, a model creation unit 12, a physical simulation unit 13, an AI calculation unit 14, a property evaluation unit 15, a gene evolution unit 16, a model update unit 17, and a data conversion unit 18. Details of these functional blocks will be described later. Note that some or all of the functions of the control unit 1 may be realized using devices other than a CPU, such as a GPU (Graphic Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit), etc.

[0014] The storage unit 2 is configured using a large-capacity, non-volatile storage device, such as a magnetic storage device such as a hard disk drive (HDD) or a solid state drive (SSD), and stores programs executed by the control unit 1 and various information used in the processing of the control unit 1. The information stored in the storage unit 2 includes generation parameter data 21, porous body structure data 22, learning model data 23, input parameter data 24, characteristic data 25, characteristic evaluation data 26, and three-dimensional shape data 27. Details of these data will be described later.

[0015] The memory 3 is configured using a high-speed, volatile storage device such as a DRAM (Dynamic Random Access Memory), and is used as a work area when the control unit 1 executes a program.

[0016] The operation input device 4 is a device for detecting operation input by the user and is configured using, for example, a keyboard or a mouse. The display device 5 is a device for displaying the processing results of the porous body design apparatus 100 on a screen to present them to the user and is configured using, for example, a liquid crystal display or an organic EL display. Note that other computers, smartphones, etc. that can communicate with the porous body design apparatus 100 may also be used as the operation input device 4 or the display device 5.

[0017] The configuration of the porous body design apparatus 100 shown in Fig. 1 may be physically constructed on a single computer, or may be constructed in a distributed manner on multiple computers. The porous body design apparatus 100 may also be realized by a cloud computer installed on a cloud, a virtual machine operating in a virtual environment, or the like.

[0018] Next, we will explain each of the functional blocks in the control unit 1: the structure generation unit 11, the model creation unit 12, the physical simulation unit 13, the AI ​​calculation unit 14, the property evaluation unit 15, the gene evolution unit 16, the model update unit 17, and the data conversion unit 18. In the porous body design system 100, the control unit 1 operates as these functional blocks, making it possible to determine the three-dimensional structure of a porous body having desired properties.

[0019] The structure generator 11 sets values ​​of generation parameters for generating the porous body to be evaluated. The generation parameters are physical quantities that can be adjusted in the manufacturing process when actually producing the porous body, and include, for example, the average and standard deviation of particle diameters of spheres and ellipsoids contained in the base material that serves as the raw material for the porous body, the volume ratio of spheres to ellipsoids in the base material, the amount and particle diameter of a pore-forming material that is mixed with the base material to form voids in the porous body, and the degree of sintering of the base material. The structure generator 11 can set the values ​​of these generation parameters based on, for example, user operations input via the operation input device 4. The values ​​of each generation parameter set by the structure generator 11 are stored in the memory unit 2 as generation parameter data 21.

[0020] Furthermore, the structure generation unit 11 performs a structure generation process to virtually generate a three-dimensional structure of the porous body to be evaluated on the porous body design device 100 based on the values ​​of each generation parameter set as described above. For this process, for example, GeoDict, a microstructure simulation software developed by Math2Market GmbH, can be used. Data representing the three-dimensional structure of the porous body generated by the structure generation unit 11 performing the structure generation process is stored in the storage unit 2 as porous body structure data 22.

[0021] The model creation unit 12 creates a learning model capable of predicting properties corresponding to the three-dimensional structure of the porous body generated by the structure generation unit 11. The learning model is a prediction model that learns the relationship between the three-dimensional structure and properties of the porous body and is also called a surrogate model. For example, a support vector machine (SVM) can be used to generate a learning model using support vector regression that can predict the properties of the porous body from the features of the three-dimensional structure of the porous body. Methods for building learning models using SVM are known and are described, for example, in Tomoki Yasuda, Shinichi Ookawara, Shiro Yoshikawa, and Hideyuki Matsumoto, "Machine learning and data-driven characterization framework for porous materials: Permeability prediction and channeling defect detection," Chemical Engineering Journal 420 (2021) 130069 DOI: 10.1016 / j.cej.2021.130069. Information about the learning model created by the model creation unit 12 is stored in the memory unit 2 as learning model data 23. It should be noted that instead of the porous body generated by the structure generation unit 11, a learning model previously created by the model creation unit 12 for another porous body may be used as the learning model data 23. Alternatively, a learning model acquired from outside may be stored in the storage unit 2 as the learning model data 23.

[0022] Before the model creation unit 12 starts creating a learning model, a predetermined number of samples of the three-dimensional structure of the porous body are generated in advance by the structure generation process of the structure generation unit 11. Various structural features of each sample are then expressed by the values ​​of multiple predetermined types of structural descriptors. Structural descriptors representing the three-dimensional structure of the porous body include, for example, porosity, solid fraction, specific surface area, pore diameter, particle diameter, pore or solid structural uniformity, pore code length, and solid code length. More specifically, the structural uniformity of the pores or solid refers to the variance of values ​​related to the pores or solid calculated for each region by dividing the porous body into multiple regions. Examples of values ​​related to the pores or solid calculated for each region include the above-mentioned porosity, solid fraction, specific surface area, pore diameter, and particle diameter. The pore code length is the continuous length of the pores in a predetermined direction in the porous body, and the solid code length is the continuous length of the solid in a predetermined direction in the porous body. The model creation unit 12 selects, from these structural descriptors, multiple structural descriptors that contribute highly to the properties of the porous body as input parameters for the learning model. Then, using the values ​​of each input parameter for each sample and the properties of the porous body obtained for each sample by the physical simulation described below performed by the physical simulation unit 13, a learning model is created that predicts properties from the input parameters.

[0023] The physical simulation unit 13 performs physical calculations based on the three-dimensional structure of the porous body generated by the structure generation unit 11, and performs a physical simulation of the case where a predetermined substance is passed through the porous body. As a result, when a fluid, such as exhaust gas from an internal combustion engine, is passed through the porous body, values ​​such as the permeability of the fluid and the filtration efficiency (capture efficiency) of particulate matter (PM) contained in the fluid can be calculated as the properties of the porous body. In addition, the properties of the porous body to be calculated may include mechanical strength properties, electrochemical properties, thermal conductivity properties, heat exchange properties, electrical conductivity properties, gas adsorption properties, gas purification performance, catalyst coating properties, and removal efficiency of substances captured in the porous body. The results of the calculation of the properties of the porous body by the physical simulation unit 13 are stored in the memory unit 2 as property data 25. The physical simulation unit 13 can be realized, for example, by utilizing the functions of the aforementioned GeoDict.

[0024] The AI ​​calculation unit 14 performs calculations related to the prediction of porous body properties by AI (artificial intelligence) using the learning model created by the model creation unit 12. By inputting the values ​​of each of the above-mentioned input parameters determined for the three-dimensional structure of the porous body into the learning model, the AI ​​calculation unit 14 can perform AI calculations using well-known AI techniques such as support vector regression, and predict the properties of the porous body. The values ​​of each input parameter used in the AI ​​calculation unit 14 and the obtained property prediction results of the porous body are stored in the memory unit 2 as input parameter data 24 and property data 25, respectively.

[0025] The property evaluation unit 15 performs an evaluation process to evaluate the properties of the porous body calculated or predicted by the physical simulation unit 13 or the AI ​​calculation unit 14. The property evaluation unit 15 performs an evaluation process for the properties of the porous body, for example, by plotting each calculated or predicted property value on a graph. Note that other methods may be used as long as they can appropriately evaluate the properties of the porous body. For example, an evaluation score for the properties of the porous body may be calculated from each property value, or each property value may be determined to satisfy a predetermined standard value, and an OK / NG judgment for the properties of the porous body may be made based on the judgment result. The property evaluation results of the porous body by the property evaluation unit 15 are stored in the memory unit 2 as property evaluation data 26.

[0026] The gene evolution unit 16 performs a process to change the values ​​of the generation parameters. The gene evolution unit 16 performs a gene evolution process using a well-known calculation method such as a genetic algorithm, thereby changing the value of each generation parameter represented by the generation parameter data 21 so as to improve the characteristics of the porous body generated by that parameter value. By repeating this gene evolution process by the gene evolution unit 16, the contents of the generation parameter data 21 can be successively updated, and optimal parameter values ​​can be searched for.

[0027] The model update unit 17 performs an update process for the learning model based on the results of the physical simulation performed by the physical simulation unit 13. As will be described later, the physical simulation unit 13 performs a physical simulation based on the three-dimensional structure of the porous body each time the AI ​​calculation unit 14 performs AI calculation a predetermined number of times. When the physical simulation is performed by the physical simulation unit 13 in this manner, the model update unit 17 performs an update process for the learning model using the results and updates the contents of the learning model data 23. This makes it possible to reflect the results of the physical simulation in the learning model and update the learning model data 23 so that the values ​​of the properties calculated by AI calculation from the input parameter values ​​corresponding to the three-dimensional structure of the porous body become more accurate.

[0028] The data conversion unit 18 converts the porous body structure data 22 based on the three-dimensional structure of the porous body finally determined by the structure generation unit 11 into three-dimensional shape data 27. The three-dimensional shape data 27 generated by conversion from the porous body structure data 22 by the data conversion unit 18 is stored in the storage unit 2 and, as necessary, is read from the storage unit 2 and provided to an external device outside the porous body design device 100. For example, by inputting the three-dimensional shape data 27 into a 3D printer, the 3D printer is operated according to the three-dimensional shape data 27 to produce a porous body that reproduces the three-dimensional structure of the porous body represented by the porous body structure data 22. This makes it possible to create a prototype of the porous body designed by the porous body design device 100. The method for producing a porous body using a 3D printer and the three-dimensional shape data 27 will be described in detail below.

[0029] Next, details of the processing performed by the control unit 1 will be described with reference to Fig. 2. Fig. 2 is a flowchart showing the processing flow of a porous body design device according to one embodiment of the present invention. The porous body design device 100 searches for a three-dimensional structure of a porous body having desired properties by repeatedly executing the processing shown in the flowchart of Fig. 2 using the control unit 1 in response to operational inputs from the user. This supports the user in designing a porous body.

[0030] When the user instructs the porous body design apparatus 100 to start designing a porous body via the operation input device 4, in step S10 the control unit 1 sets the initial values ​​0 for the variables i and j. The variable i is a variable for counting the number of times the AI ​​calculation unit 14 has performed AI calculations between the time the physical simulation unit 13 performed one physical simulation and the time the next physical simulation is performed. The variable j is a variable for counting the number of times the gene evolution unit 16 has performed gene evolution processing, i.e., the number of generations of the generation parameters that have been set so far by the gene evolution processing. While the processing shown in the flowchart of FIG. 2 is being executed, the values ​​of the variables i and j are stored in the memory 3.

[0031] In step S20, the control unit 1 performs an initial learning model generation process to set the initial state of the learning model data 23. Here, the structure generation unit 11, model creation unit 12, and physical simulation unit 13 are used to generate a large number of samples of the three-dimensional structure of the porous body, and the relationship between the values ​​of input parameters selected from the structural descriptors of each sample and the properties is determined. Then, a learning model is created based on the determined relationship, thereby setting the initial state of the learning model data 23. Details of the initial learning model generation process will be described later with reference to FIG. 3.

[0032] In step S30, the control unit 1 sets initial values ​​of the generation parameters of the porous body using the structure generation unit 11. Here, for example, the initial values ​​of the generation parameters of the porous body are set by randomly varying the values ​​of each generation parameter used when generating a sample of the three-dimensional structure of the porous body in the initial learning model generation process of step S20 within a predetermined range or according to a predetermined rule. In step S30, the values ​​of each generation parameter are randomly varied by a predetermined population number (number of individuals in each generation) P, and set as the initial values ​​of the generation parameters. For example, if P = 100, 100 combinations of generation parameter values ​​are set as the initial values ​​of the generation parameters. Note that the initial values ​​of each generation parameter set here are used as parent generations for the first-generation generation parameters in the gene evolution process of step S150, described below. Therefore, in the following description, the initial values ​​of the generation parameters are referred to as the zeroth-generation generation parameters. After the initial values ​​of each generation parameter set in this way are stored in the memory unit 2 as generation parameter data 21, the process proceeds to step S40.

[0033] In step S40, the control unit 1 causes the structure generation unit 11 to generate a three-dimensional structure of the porous body whose properties are to be predicted based on the initial values ​​of the generation parameters set in step S30 (when j = 0) or the jth generation generation parameters set in step S150 (when j ≥ 1), which will be described later. As described above, simulation software such as GeoDict is used to virtually generate a three-dimensional structure of the porous body according to each generation parameter value represented by the generation parameter data 21. This allows the generation of three-dimensional structures of the porous body whose properties are to be predicted, with the same number of populations as the preset population number P.

[0034] In step S50, the control unit 1 calculates input parameter values ​​corresponding to each three-dimensional structure generated in step S40 using the structure generation unit 11. As described above, the control unit 1 calculates a combination of input parameter values ​​for the three-dimensional structure of each porous body whose properties are to be predicted generated in step S40 using structural descriptors previously selected by the structure generation unit 11 as having a high degree of contribution to the properties of the porous body, such as porosity, solid fraction, specific surface area, pore diameter, particle diameter, pore or solid structural uniformity, pore chord length, and solid chord length, as input parameters of the learning model. After the input parameter values ​​calculated for each porous body whose properties are to be predicted are stored in the storage unit 2 as input parameter data 24, the process proceeds to step S60.

[0035] In step S60, the control unit 1 determines whether the value of the current variable i is a predetermined value M. If i=M, the process proceeds to step S90, and if i<M, the process proceeds to step S70. The value of M represents the set number of times that the AI ​​calculation is performed before a physical simulation is performed to update the learning model, and can be set to any number greater than or equal to 1, for example, M=10.

[0036] In step S70, the control unit 1 causes the AI ​​calculation unit 14 to perform property prediction by AI calculation using a learning model based on the input parameter values ​​calculated in step S50. Here, the learning model data 23 is read to obtain a learning model, and input into this learning model the values ​​of the input parameters determined from the three-dimensional structure of each porous body whose properties are to be predicted, thereby performing AI calculations related to the prediction of the properties of the porous body. This determines a predicted property value for each combination of input parameters, and performs a property prediction for each porous body whose properties are to be predicted. After the predicted property values ​​obtained for each porous body whose properties are to be predicted are stored in the memory unit 2 as property data 25, the process proceeds to step S80.

[0037] In step S80, the control unit 1 adds 1 to the value of the variable i, and the process proceeds to step S120.

[0038] In step S90, the control unit 1 performs a physical simulation based on each of the three-dimensional structures generated in step S40, thereby calculating the properties of each porous body whose properties are to be predicted, and calculating the property values. After storing the calculation results of the property values ​​obtained for each porous body whose properties are to be predicted in this way as property data 25 in the storage unit 2, the process proceeds to step S100.

[0039] In step S100, the control unit 1 causes the model update unit 17 to update the learning model represented by the learning model data 23 based on the results of the physical simulation performed in step S90. Here, the input parameter data 24 stored in step S50 and the property data 25 stored in step S90 for the porous bodies that have been the subject of the physical simulations performed up to that point are read from the storage unit 2, and the contents of the learning model (weight values) are updated based on the relationship between the input parameter values ​​and property values ​​for each porous body whose properties are to be predicted in these data. Once the contents of the updated learning model have been recorded in the learning model data 23, the process proceeds to step S110.

[0040] In step S110, the control unit 1 resets the value of the variable i to 0, and proceeds to step S120.

[0041] In step S120, the control unit 1 causes the property evaluation unit 15 to plot the property values ​​predicted or calculated in step S70 or S90 on a graph for each porous body whose properties are to be predicted, and evaluate the property prediction / calculation results. Here, for example, a two-dimensional graph is prepared, with one axis representing the fluid permeability (one of the properties of the porous body) and the other axis representing the filtration efficiency (another property of the porous body) on the other axis. The property values ​​predicted or calculated so far for each porous body whose properties are to be predicted are plotted on this two-dimensional graph. This allows evaluation of the property prediction / calculation results so far for each porous body whose properties are to be predicted based on the position of each point plotted on the two-dimensional graph. Note that, as long as the property prediction / calculation results can be evaluated, it is not necessary to actually plot the points on the graph; simply retaining data indicating the position of the points on the graph is sufficient. Alternatively, as described above, the property prediction / calculation results may be evaluated using methods other than plotting on a graph.

[0042] In step S125, the control unit 1 selects parent generation parameters for the next generation based on the evaluation results of the property prediction / calculation results obtained in step S120. Here, for example, the control unit 1 relatively evaluates the property values ​​predicted or calculated in steps S70 or S90 from the positions of each point plotted on the graph in step S120. Then, it selects a suitable percentage of these property values, for example, P (where P is the population number) corresponding to the top 50%, and selects the porous body generation parameters corresponding to these property values ​​as the parent generation parameters for the next generation, i.e., the parent generation parameters for the (j+1)th generation. However, when step S125 is first performed, i.e., when j = 0, there are no plotted points from the previous generation. Therefore, in this case, all of the initial values ​​of the generation parameters set in step S30 can be selected as the parent generation parameters for the first generation.

[0043] In step S130, the control unit 1 determines whether the current value of variable j is a predetermined value N. If j=N, the process proceeds to step S140, and if j<N, the process proceeds to step S150. The value of N represents the set number of times the gene evolution unit 16 will execute the gene evolution process, and can be set to any number greater than or equal to 1, for example, N=100.

[0044] In step S140, the control unit 1 determines the final three-dimensional structure of the porous body based on the evaluation results of the porous body properties obtained using the generation parameters of each generation. Here, for example, the control unit 1 evaluates the property values ​​predicted or calculated in step S70 or S90 relative to each other based on the positions of the points plotted on the graph in the previous step S120, and identifies the property value with the highest evaluation. The specific method for this will be described later with reference to FIG. 4.

[0045] When the most highly evaluated characteristic value is identified in the processing of step S140, the control unit 1 determines the three-dimensional structure of the porous body corresponding to that characteristic value as the final three-dimensional structure of the porous body, and terminates the processing shown in the flowchart of Figure 2.

[0046] In step S150, the control unit 1 sets next-candidate generation parameters using the gene evolution unit 16. Here, gene evolution processing is performed based on the generation parameter values ​​selected as next-generation parent generation parameters in the immediately preceding step S125, and child generation parameters for those parent generation parameters are set to set the next-candidate generation parameters. Specifically, using a well-known calculation method such as a genetic algorithm, the jth-generation generation parameters selected in step S125 are multiplied together, and new generation parameters corresponding to the child are set as (j+1)th-generation generation parameters. By repeating this operation for multiple combinations of parent generation parameters, any number of next-generation child generation parameters are generated. For example, P pairs of jth-generation parent generation parameters (P is the population number) are set, and P (j+1)th-generation child generation parameters are created from these pairs as next-candidate generation parameters. After recording the new (j+1)th-generation generation parameter values ​​obtained by the gene evolution processing in the generation parameter data 21, the process proceeds to step S160.

[0047] In step S160, the control unit 1 adds 1 to the value of the variable j, and returns the process to step S40.

[0048] After returning from step S160 to step S40, the control unit 1 repeats the process from step S40 onward. This generates a new three-dimensional structure of the porous body to be subjected to property prediction based on the generation parameters after the gene evolution process, and predicts or calculates property values ​​using AI calculations or physical simulations. The obtained property prediction / calculation results are then plotted on a graph, and in addition to the evaluation results for the property values ​​of the porous body using each generation parameter of the previous generation obtained in the previous process, new evaluation results for the property values ​​of the porous body using each generation parameter of the next generation are obtained. From the obtained evaluation results, the next-generation parent generation parameters to be used in the gene evolution process are selected. This series of processes realizes an optimization process in which the generation parameters of the porous body are changed to search for optimal generation parameters.

[0049] FIG. 3 is a flowchart showing the details of the initial learning model generation process executed in step S20 of FIG.

[0050] In step S210, the control unit 1 sets initial generation parameters for the porous body using the structure generation unit 11. Here, the initial generation parameters are set according to, for example, a combination of predetermined generation parameter values.

[0051] In step S220, the control unit 1 generates a three-dimensional structure of the porous body to be used for generating the initial learning model based on the initial generation parameters set in step S210 using the structure generation unit 11. Here, the three-dimensional structure for generating the initial learning model is generated by processing similar to step S40 in Figure 2.

[0052] In step S230, the control unit 1 calculates values ​​of structural descriptors corresponding to the three-dimensional structure generated in step S220 using the structure generation unit 11. Here, the control unit 1 calculates values ​​of various predetermined structural descriptors, such as porosity, solid fraction, specific surface area, pore diameter, particle diameter, structural uniformity of the pores or solid, pore chord length, and solid chord length, for the three-dimensional structure of the porous body generated in step S220.

[0053] In step S240, the control unit 1 performs a physical simulation based on the three-dimensional structure generated in step S220 using the physical simulation unit 13. Here, by processing similar to step S90 in Fig. 2, the properties of the porous body having the three-dimensional structure for generating the initial learning model are calculated by physical simulation, and the property values ​​obtained as the simulation results are obtained.

[0054] In step S250, the control unit 1 determines whether each of the processes in steps S210 to S240 has been performed for a predetermined number of samples. If the processes in steps S210 to S240 have been performed for the predetermined number of porous bodies and the characteristic values ​​for each sample have been calculated, the process proceeds to step S260. On the other hand, if the number of three-dimensional structures for which characteristic values ​​have been calculated through the processes in steps S210 to S240 is less than the predetermined number of samples and the characteristic values ​​acquired so far do not satisfy the number of samples, the process returns to step S210 and the processes from step S210 onwards are performed again.

[0055] In step S260, the control unit 1 causes the structure generation unit 11 to select, from the structural descriptors of each sample calculated in step S230, structural descriptors that are highly correlated with the characteristic values ​​of the porous body calculated by the physical simulation in step S240 as input parameters of the learning model. Here, for example, the correlation between the structural descriptor value calculated for each sample and the characteristic value is calculated for each structural descriptor, and a predetermined number of structural descriptors are selected in descending order of the obtained correlation values ​​as input parameters of the learning model. This allows structural descriptors that contribute most to the characteristics of the porous body to be selected as input parameters of the learning model.

[0056] In step S270, the control unit 1 causes the model creation unit 12 to create an initial training model based on the input parameters selected in step S260 and the characteristic values ​​of the porous body determined in step S240. Here, the initial training model can be created from the relationship between the input parameters and characteristic values ​​of each sample by performing model training processing using, for example, a well-known method.

[0057] If the initial learning model is created in step S270, the control unit 1 records the contents of the initial learning model in the learning model data 23 and ends the initial learning model generation process shown in the flowchart of Fig. 3. Thereafter, the process proceeds to step S30 in Fig. 2.

[0058] FIG. 4 is a comparative explanatory diagram of the three-dimensional structure of a porous body obtained by the porous body design system 100 of this embodiment and the three-dimensional structure of a porous body according to a comparative example. In the graph of FIG. 4, the horizontal axis represents fluid permeability, one of the properties of the porous body, and the vertical axis represents filtration efficiency, another property of the porous body. Note that in the graph of FIG. 4, the permeability values ​​on the horizontal axis are normalized so that the minimum value is 0 and the maximum value is 1. In this graph of FIG. 4, each point represented by point 43 (each point indicated by a ● marker; hereinafter referred to as "example characteristic points") represents an example of characteristic values ​​for the three-dimensional structure of the porous body according to this embodiment. These example characteristic points were obtained by repeatedly calculating and optimizing the characteristic values ​​according to the flowchart of FIG. 2. On the other hand, each point represented by point 41 (each point indicated by a ■ marker; hereinafter referred to as "physics calculation characteristic points") represents an example of characteristic values ​​calculated by previously creating the same number of three-dimensional structures of porous bodies as the example characteristic points and performing physical simulations on each of the three-dimensional structures, as a comparative example. Furthermore, each point represented by point 42 (each point marked with an x, hereinafter referred to as "AI calculation characteristic point") indicates an example of a characteristic prediction value obtained solely by AI calculation using the initial learning model, as another comparative example. These AI calculation characteristic points were obtained by always executing the process of step S70 in the flowchart of FIG. 2, regardless of the value of variable i.

[0059] Comparing the plotted points on the graph in Figure 4, the AI-calculated characteristic points, including point 42, and the example characteristic points, including point 43, are located generally further to the upper right than the physically calculated characteristic points, including point 41. In other words, it can be seen that the porous bodies corresponding to these characteristic points have three-dimensional structures with better properties than the porous bodies corresponding to the physically calculated characteristic points. As will be described later, combining physical simulation and optimization processing requires performing physical simulations on each of the many three-dimensional structures of porous bodies obtained during the optimization processing, which is not practical because it requires a huge amount of computational time. On the other hand, the porous body design system 100 of this embodiment can predict the characteristic values ​​of the many three-dimensional structures of porous bodies obtained during the optimization processing using AI calculations in a significantly shorter time than physical simulations. Therefore, it is possible to determine the three-dimensional structures of porous bodies that have good properties, such as the AI-calculated characteristic points and the example characteristic points, within a realistic computational time.

[0060] Note that each AI calculation characteristic point shown in Figure 4 represents the predicted results of characteristic values ​​obtained when AI calculations are repeatedly performed using the initial learning model generated in the initial learning model generation process of Figure 3. Therefore, there is a high possibility that the deviation from the true characteristic values ​​for the three-dimensional structure of the porous body obtained during the optimization process will be large, and the three-dimensional structure of the porous body that truly exhibits good characteristics may not necessarily represent the three-dimensional structure of the porous body that exhibits good characteristics. Furthermore, deviation from the true characteristic values ​​may make it difficult to accurately evaluate the relative superiority of the characteristic values ​​of each three-dimensional structure. On the other hand, the porous body design device 100 of this embodiment performs physical simulations at predetermined intervals as described above, and updates the learning model based on the results. Therefore, the deviation from the true characteristic values ​​for the three-dimensional structure of the porous body obtained during the optimization process is small. Note that each example characteristic point shown in Figure 4 is a value after the physical simulation has been updated. Therefore, the example characteristic points can more reliably determine the three-dimensional structure of the porous body that exhibits good characteristics in terms of the combination of filtration efficiency and permeability than the AI ​​calculation characteristic points.

[0061] As described above, the porous body design device 100 of this embodiment performs the processing shown in the flowchart of FIG. 2, thereby making it possible to determine a three-dimensional structure of a porous body that can provide better characteristics than the comparative example.

[0062] In the flowchart of Figure 2, the characteristic values ​​predicted or calculated by AI calculation or physical simulation are evaluated, for example, by plotting them on a graph in step S120, and then the genetic evolution process in step S150 are repeated to finally obtain the example characteristic points shown in Figure 4. In step S140, the point exhibiting the best characteristics among the example characteristic points, for example, the point with normalized transmittance and filtration efficiency values ​​both equal to or greater than 0.2 and located farthest from the origin of the graph, is selected as the most highly evaluated characteristic value. The three-dimensional structure corresponding to this characteristic value is then identified, and the final three-dimensional structure of the porous body is determined.

[0063] For example, when the point showing the best characteristics among the physically calculated characteristic points, the AI ​​calculated characteristic points, and the example characteristic points is selected from the graph of FIG. 4 , the characteristic values ​​of each point are compared with the search time and number of searched individuals required to search for each of these points from the physically calculated characteristic points, the AI ​​calculated characteristic points, and the example characteristic points. The comparison results are, for example, as follows. In the following description, the physically calculated characteristic points are calculated for 524 different three-dimensional structures of porous bodies, and based on 500 of these calculation results, the optimization process and the AI ​​calculation or physical simulation are repeated 10 times until the total number of searched individuals reaches 5,000. In this example, the characteristic value and search time of each selected point of the AI ​​calculated characteristic points and the example characteristic points are normalized, with the characteristic value and search time of each selected point being 100. Here, the larger the characteristic value, the better the characteristics, and the larger the search time, the longer the processing time. (1) Physical calculation characteristic points: characteristic value = 100, search time = 100 (total time = 100), number of individuals = 524 (2) AI calculation characteristic points: characteristic value = 147, search time per run = 6 (total time = 56), number of individuals searched = 5000 (3) Example characteristic points: characteristic value = 152, search time per run = 15 (total time = 150), number of individuals searched = 5000

[0064] The above comparison results show that example characteristic points can obtain better properties than physically calculated characteristic points and machine learning characteristic points, even in a relatively short search time. Since physically calculated characteristic points represent characteristic values ​​calculated by physical simulation, similar to AI-calculated characteristic points and example characteristic points, if 5,000 physically calculated characteristic points are calculated by performing physical simulations on 5,000 different three-dimensional structures of porous bodies and a point with good properties is selected from among them, it is possible to obtain better properties than example characteristic points. However, this would require approximately 10 times the search time shown in the example of physically calculated characteristic points above, i.e., a total time of 1,000 hours after normalization, which is not realistic. Therefore, it can be seen that by adopting a porous body design method using the porous body design apparatus 100 of this embodiment, characteristic values ​​can be obtained for a large number of search individuals within a realistic time frame, and a three-dimensional structure of a porous body having the desired properties can be determined within a relatively short period of time.

[0065] Next, the method for manufacturing a porous body by a 3D printer using the three-dimensional shape data 27 will be described in detail with reference to Fig. 5. Fig. 5 is a flowchart showing the flow of the manufacturing process for a porous body using a 3D printer.

[0066] When the user instructs the porous body design apparatus 100 to start manufacturing a porous body via the operation input device 4, in step S310 the control unit 1 generates three-dimensional shape data 27 using the data conversion unit 18. Here, the porous body structure data 22 is read from the storage unit 2 and subjected to a predetermined data conversion process to convert the porous body structure data 22 into three-dimensional shape data 27. Note that the porous body structure data 22 may be used as the three-dimensional shape data 27 as is. In this case, the control unit 1 does not need to have the data conversion unit 18.

[0067] In step S320, the three-dimensional shape data 27 created in step S310 is input to a 3D printer (not shown). Here, the three-dimensional shape data 27 is input to the 3D printer by, for example, connecting the porous body design apparatus 100 and the 3D printer wirelessly or via a wire, and transmitting the three-dimensional shape data 27 from the porous body design apparatus 100 to the 3D printer. Alternatively, the three-dimensional shape data 27 may be input to the 3D printer by transferring the three-dimensional shape data 27 from the porous body design apparatus 100 to the 3D printer via a storage medium such as a USB memory. In addition, the three-dimensional shape data 27 can be input to the 3D printer by any other method.

[0068] In step S330, the 3D printer starts injecting material to form a porous body based on the three-dimensional shape data 27 input in step S320. Then, in step S340, the head of the 3D printer is moved to the coordinate position represented by the three-dimensional shape data 27 input in step S320, and then material is injected.

[0069] In step S350, it is determined whether or not material has been injected from the 3D printer for all three-dimensional structures of the porous body represented by the three-dimensional shape data 27 input in step S320. If there are any three-dimensional structure portions for which material has not yet been injected, the process returns to step S340 to continue injecting material, and if material has been injected for all three-dimensional structures, the process proceeds to step S360.

[0070] In step S360, the injection of material in the 3D printer is completed. This completes the production of a porous body using the 3D printer, and a porous body that reproduces the three-dimensional structure designed by the porous body design device 100 can be created. After the processing of step S360 has been executed, the processing shown in the flowchart of FIG. 5 is terminated.

[0071] Although the above describes an example of a method for manufacturing a porous body using a 3D printer and the three-dimensional shape data 27, a porous body that reproduces the three-dimensional structure designed by the porous body design device 100 may be manufactured by other methods. Alternatively, a porous body may be manufactured using something other than a 3D printer. As long as a porous body that reproduces the three-dimensional structure designed by the porous body design device 100 can be appropriately manufactured using the three-dimensional shape data 27, any method can be selected from a variety of well-known methods and used to manufacture the porous body.

[0072] According to the embodiment of the present invention described above, the following advantageous effects are achieved.

[0073] (1) A method for designing a porous body using the porous body design device 100 involves having the porous body design device 100 execute the following steps multiple times: a structure generation process (step S40) that virtually generates a three-dimensional structure of the porous body on the porous body design device 100 based on generation parameter values ​​for generating the porous body; a property prediction / calculation process (steps S70, S90) that predicts or calculates the properties of the porous body having the three-dimensional structure generated by the structure generation process; an evaluation process (step S120) that evaluates the properties of the porous body predicted or calculated by the property prediction / calculation process; and an optimization process (steps S125, S150) that varies the generation parameter values ​​to search for optimal generation parameter values. The three-dimensional structure of the porous body is then determined (step S140) based on the evaluation results of the porous body properties obtained by the evaluation process. This provides a useful technology that enables early determination of an optimal structure in the design of a porous body.

[0074] (2) In the property prediction / calculation process, at least one of a physical simulation (step S90) that calculates the properties of the porous body by physical calculation based on the three-dimensional structure generated by the structure generation process, and an AI calculation (step S70) that predicts the properties of the porous body using a learning model of the three-dimensional structure is performed. In this way, physical simulation and AI calculation can be performed as appropriate to determine a three-dimensional structure of the porous body that will provide better properties.

[0075] (3) In the porous body design method, an update process (step S100) is performed in which the learning model is updated based on the results of a physical simulation by the porous body design device 100. This allows the learning model to be optimized so that more accurate property prediction values ​​can be obtained compared to when AI calculations using the learning model are performed alone.

[0076] (4) In the property prediction / calculation process, a physical simulation is performed in step S90 every time the AI ​​calculation in step S70 is performed a predetermined number M times (step S60: Yes). Then, after the physical simulation in step S90 is performed in the property prediction / calculation process, the porous body design system 100 is caused to perform the update process in step S100. This makes it possible to adjust the update frequency of the learning model by arbitrarily changing the value of M, and thus makes it possible to determine the optimal structure in a short time with high accuracy.

[0077] (5) It is also possible to manufacture a porous body (steps S330 to S360) using three-dimensional shape data based on the three-dimensional structure of the porous body determined by the porous body design method described above. In this way, a prototype of the porous body designed by the porous body design system 100 can be quickly and easily produced.

[0078] The present invention is not limited to the above-described embodiment, and can be implemented using any components without departing from the spirit of the present invention.

[0079] In the above embodiment, one of the structural descriptors representing the three-dimensional structure of the porous body is selected as an input parameter of the learning model, and the relationship between this input parameter and the properties of the porous body is represented by the learning model. However, instead of the structural descriptor, a generation parameter may be used as an input parameter of the learning model. In this case, all of the multiple generation parameters that exist for the generation of the porous body may be used as input parameters, or one of the multiple generation parameters that contributes most to the properties of the porous body may be selected as an input parameter. In this way, the properties of the porous body can be directly predicted from the generation parameters using machine learning without performing the process of generating the three-dimensional structure of the porous body and calculating the input parameters (the process of steps S40 and S50 in FIG. 2), thereby further shortening the processing time.

[0080] The above-described embodiments and modifications are merely examples, and the present invention is not limited to these details as long as the features of the invention are not impaired. Furthermore, although various embodiments and modifications have been described above, the present invention is not limited to these details. Other aspects that can be considered within the scope of the technical idea of ​​the present invention are also included within the scope of the present invention.

[0081] 1...control unit, 2...storage unit, 3...memory, 4...operation input device, 5...display device, 6...bus, 11...structure generation unit, 12...model creation unit, 13...physical simulation unit, 14...AI calculation unit, 15...characteristic evaluation unit, 16...gene evolution unit, 17...model update unit, 18...data conversion unit, 21...generated parameter data, 22...porous body structure data, 23...learning model data, 24...input parameter data, 25...characteristic data, 26...characteristic evaluation data, 27...three-dimensional shape data, 100...porous body design device

Claims

1. A method for designing a porous body using a computer, comprising the steps of: a structure generation process for virtually generating a three-dimensional structure of the porous body on the computer based on generation parameter values ​​for generating the porous body; a property prediction / calculation process for predicting or calculating properties of the porous body having the three-dimensional structure generated by the structure generation process; an evaluation process for evaluating the properties of the porous body predicted or calculated by the property prediction / calculation process; an optimization process for searching for an optimal value of the generation parameter by changing the generation parameter value; determining a three-dimensional structure of the porous body based on the evaluation results of the properties of the porous body by the evaluation process; The property prediction / calculation process is a porous body design method in which a physical simulation is performed to calculate a property value for a specified property among multiple properties possessed by the porous body by physical calculation based on the three-dimensional structure generated by the structure generation process, and an AI calculation is performed to input input parameters corresponding to the three-dimensional structure into a learning model of the three-dimensional structure to predict the property value of the porous body, thereby predicting or calculating the properties of the porous body.

2. The porous body design method according to claim 1, A porous body design method that causes the computer to execute an update process for updating the learning model based on the results of the physical simulation.

3. The porous body design method according to claim 2, In the characteristic prediction / calculation process, the physical simulation is performed every time the AI ​​calculation is executed a predetermined number of times, The porous body design method further comprises causing the computer to execute the update process when the physical simulation is performed in the property prediction / calculation process.

4. A porous body manufacturing method, comprising the steps of: manufacturing the porous body by using three-dimensional shape data based on the three-dimensional structure of the porous body determined by the porous body design method according to any one of claims 1 to 3.