Information processing device, characteristic prediction method, and control program

Abstract

In the design and manufacturing process, the characteristics of the final product are predicted early. [Solution] The information processing device 10 includes a prediction unit 120 that predicts the characteristics of an object based on intermediate descriptors, which are data indicating the state and / or structure of the acquired intermediates, in the design and manufacturing process of an object that is a final manufactured product having predetermined characteristics, from raw materials through intermediates.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, a characteristic prediction method, and a control program.

Background Art

[0002] As the production of products such as polymers, composite materials, and biomaterials shifts to complex systems, in order to determine the material composition / process for realizing desired material properties, it is necessary to acquire a large amount of deductive information such as computer simulation data or inductive information such as experimental data, resulting in poor work efficiency.

[0003] In the dynamic estimation system disclosed in Patent Document 1, multivariate information regarding redox potential is acquired using three working electrodes having different materials or surface treatments in a culture solution that is a measurement target system, and the optimal conditions of the measurement target system are estimated using this information.

[0004] In Non-Patent Document 1, in the development of a composite material (thermal diffusion film) in which an inorganic filler is blended with a resin, the concept of Process-Structure-Property Linkage is introduced, and efforts are made to improve the development efficiency.

[0005] In the development and production of a film containing a filler for thermal diffusion, since the mechanism between the composition of the materials used and the process conditions (Process), which are explanatory variables (inputs), and the thermal diffusion characteristics (Property), which are target variables (outputs), is complex, a large amount of data is required for inductive verification by experiments or the like, and not only does deductive verification by simulation or the like require a great deal of time cost, but there is also a possibility that an appropriate solution cannot be found depending on the assumed model.

[0006] To address these challenges, the method disclosed in Non-Patent Document 1 does not directly link the explanatory variable (Process) and the objective variable (Property), but instead uses a "filler particle dispersion structure" (Structure) as a descriptor to connect them. This allows for verifiable relationships between Process and Structure, and between Structure and Property, and by linking these results, the input / output design (Process-Property) of composite materials is made more efficient. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2021-43097 [Non-patent literature]

[0008] [Non-Patent Document 1] Shinji Ozawa (Kaneka Corporation), “Research and Development of Resin / Inorganic Filler Composite Materials,” [online], January 19, 2022, Ultra-Advanced Materials Ultra-High-Speed ​​Development Platform Technology Project (Ultra-Ultra Project) Final Results Presentation Meeting, Ultra-Advanced Materials Ultra-High-Speed ​​Development Platform Technology Project (Ultra-Ultra Project), Internet <URL https: / / www.admat.or.jp / library / 5975666db3de4b020a7803ae / 61ea4ddf3b19fb676e934836.pdf> [Overview of the project] [Problems that the invention aims to solve]

[0009] However, the method described in Non-Patent Document 1 has the problem that the calculation cost increases as the physical properties and structure of the final product become more complex, as the descriptors are deductively derived from the raw materials and process conditions, thus compromising immediacy. Furthermore, depending on the assumed model, the relationship between the descriptor and the target variable may not be valid, making it impossible to derive a solution.

[0010] Furthermore, with the recent improvements in computer performance, the development of AI (artificial intelligence) technologies such as machine learning, and advancements in image processing and analysis technologies, it is becoming increasingly possible to easily recognize information that cannot be perceived by human senses. In this context, the process of quantifying and verbalizing raw material composition, process conditions, or various characteristics solely for the purpose of being recognizable and understandable to humans is highly likely to compromise important and essential information in manufacturing.

[0011] This invention has been made in view of the above circumstances, and aims to improve the efficiency of the design and manufacturing process by acquiring image information of the state and structure of intermediate products between raw materials and the final product that are generated during the design or manufacturing process of materials or products (hereinafter referred to as the design and manufacturing process), and utilizing this as a new descriptor, thereby appropriately predicting the characteristics of the final product. [Means for solving the problem]

[0012] The above objectives of the present invention are achieved by the following means.

[0013] (1) In the design and manufacturing process of an object that is a final product having predetermined characteristics, starting from raw materials and going through intermediates, An information processing device comprising a prediction unit that predicts the characteristics of an object based on an intermediate descriptor, which is data indicating the state and / or structure of the intermediate object obtained.

[0014] (2) The information processing apparatus according to (1) above, wherein the intermediate is a fluid that can move and / or change over time.

[0015] (3) The information processing apparatus described in (1) above, wherein the intermediate has a phase different from that of the object.

[0016] (4) The information processing apparatus described in (1) above, wherein the intermediate descriptor includes image information and / or feature quantities extracted from the image information.

[0017] (5) The image information includes a tomographic image, and the information processing apparatus according to (4) above.

[0018] (6) The intermediate descriptor is data obtained by irradiating the intermediate object with waves of a predetermined frequency and detecting at least any one of the transmitted wave, reflected wave, and scattered wave from the intermediate object, and the information processing apparatus according to (4) above.

[0019] (7) The prediction unit predicts the characteristics of the object from the intermediate descriptor using a first learned model that outputs the characteristics of the object based on the intermediate descriptor, and the information processing apparatus according to (1) above.

[0020] (8) There are a plurality of the intermediate descriptors for one object, and the information processing apparatus according to (1) above.

[0021] (9) The intermediate descriptor is selected from among the plurality of intermediate descriptors according to the characteristics, and the information processing apparatus according to (8) above.

[0022] (10) The intermediate descriptor is selected from among the plurality of intermediate descriptors according to the component size of the object that determines the characteristics, and the information processing apparatus according to (8) above.

[0023] (11) There are a plurality of the intermediate descriptors for one object, and is selected from among the plurality of intermediate descriptors by setting the frequency or signal length of the wave, and the information processing apparatus according to (6) above.

[0024] (12) The information processing apparatus according to (1) above further includes a process parameter adjustment unit that adjusts the process parameters based on the intermediate descriptor using a second learned model that outputs process parameters related to the intermediate object based on the intermediate descriptor.

[0025] (13) In the design and manufacturing process of an object that is a final product having predetermined characteristics through an intermediate object from raw materials, (a) A step of obtaining an intermediate descriptor which is data indicating the state and / or structure of the intermediate, A characteristic prediction method comprising the step (b) predicting the characteristics of the object based on the intermediate descriptor obtained in step (a).

[0026] (14) The characteristic prediction method according to (13) above, wherein in step (b), the characteristics of the object are predicted from the intermediate descriptor using a first trained model that outputs the characteristics of the object based on the intermediate descriptor.

[0027] (15) In the design and manufacturing process of an object that is a final product having predetermined characteristics, starting from raw materials and going through intermediates, (a) A step of obtaining an intermediate descriptor which is data indicating the state and / or structure of the intermediate, A control program for causing a computer to perform a process including the step (b) of predicting the properties of the object based on the intermediate descriptor obtained in step (a). [Effects of the Invention]

[0028] The information processing device according to the present invention includes a prediction unit that predicts the characteristics of a final product based on intermediate descriptors, which are image data or the like that show the state and / or structure of acquired intermediate products, in a design and manufacturing process that goes from raw materials through intermediate products to a final product having predetermined characteristics. This enables the accurate prediction of the characteristics of the final product by utilizing AI technology, image processing technology, etc., and optimizes the parameters of the design and manufacturing process to perform highly efficient design or manufacturing. [Brief explanation of the drawing]

[0029] [Figure 1] This figure shows the configuration of the design and manufacturing process, including the information processing device 10, etc., according to the first embodiment. [Figure 2] This is a schematic diagram to conceptually explain intermediate descriptors. [Figure 3A]As an example of multiple intermediate descriptors, this is a schematic diagram to conceptually explain intermediate descriptors that depend on the size of the constituent elements of the final product. [Figure 3B] As another example of multiple intermediate descriptors, this is a schematic diagram to conceptually explain intermediate descriptors corresponding to multiple characteristics of the final product. [Figure 4] This figure shows the configuration of the design and manufacturing process, including the information processing device 10b, etc., according to the second embodiment. [Figure 5] This figure shows the configuration of the design and manufacturing process, including the information processing device 10c, etc., according to the third embodiment. [Figure 6A] This figure shows an example configuration in which ultrasonic images of fluids are acquired by an ultrasonic transmitting and receiving unit. [Figure 6B] This figure shows examples of ultrasound images (tomographic images) obtained from three types of fluids. [Figure 7] This flowchart shows the characteristic prediction processing and process parameter adjustment processing performed by the information processing device 10 or 10b. [Figure 8] This is a flowchart showing the processing of the machine learning method for the first trained model. [Figure 9] This is a flowchart showing the processing of the machine learning method for the second trained model. [Figure 10] This flowchart shows the characteristic prediction processing and process parameter adjustment processing performed by the information processing device 10c. [Figure 11] This is a flowchart showing the processing of the machine learning method for the third trained model. [Figure 12] This diagram shows the configuration of the design and manufacturing process in Example 1. [Figure 13] This diagram shows the configuration of the design and manufacturing process in Example 2. [Figure 14] This figure shows the configuration of the design and manufacturing process in Example 3. [Figure 15] This figure shows the configuration of the design and manufacturing process in Example 4. [Figure 16] This figure shows the configuration of the design and manufacturing process in Example 5. [Modes for carrying out the invention]

[0030] Embodiments of the present invention will be described below with reference to the attached drawings. However, the scope of the present invention is not limited to the disclosed embodiments. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant descriptions are omitted. Also, the dimensional ratios in the drawings are exaggerated for illustrative purposes and may differ from the actual ratios.

[0031] Figure 1 is a diagram illustrating the schematic configuration of a design and manufacturing process including an information processing device 10 according to the first embodiment. The information processing device 10 consists of a standalone personal computer located on the premises of a factory or the like where the design and manufacturing process 50 is provided, or an on-premise server or a cloud server using a commercial cloud service. The information processing device 10 includes a CPU, RAM, storage unit, communication interface, etc. The information processing device 10 acquires data from the design and manufacturing process 50 via a separately provided acquisition unit 100 and a characteristic acquisition unit 102 (both described later), which are intermediate descriptors that indicate the state and / or structure of the intermediate product (hereinafter referred to as the state of the intermediate product, etc.) and data related to the characteristics and structure to be realized of the final product (hereinafter referred to as characteristics, etc. data), respectively. Based on these, it predicts the characteristics, etc. of the final product and adjusts parameters such as the composition of raw materials in the design and manufacturing process 50 and the processing conditions of the process from raw materials to the final product. Here, the characteristics, etc. of the final product include at least one of the physical properties, quality, and function of the product. Furthermore, in addition to quantifiable properties such as mechanical properties, physical properties, thermal properties, electrical properties, and flammability, at least one of the properties that are currently difficult to quantify, such as taste, texture, and feel, may also be included. In the following, when "information processing device, etc." is used, it refers to the acquisition unit 100, the property acquisition unit 102, and the information processing device 10.

[0032] Here, an intermediate is a substance obtained as a result of the first process when the final product is obtained from raw materials through a first process and a second process. It is a substance generated in an intermediate step in the production of the final product and is different from the raw materials and the final product. This intermediate includes not only compounds synthesized during the production of chemical products as generally defined, but also a wide range of substances that are generated or output in the series of processes for producing the final product. For example, intermediates include culture media containing cells and microorganisms in bio-manufacturing, resins in a molten state after being mixed with fillers in the production of composite fiber-reinforced resin materials, and fluids during flow synthesis in the production of chemical compounds. Note that the first and second processes may each consist of not only a single process but also multiple processes.

[0033] Furthermore, the intermediate substance is a substance (hereinafter referred to as a fluid) that can move and / or change over time. Here, "moving over time" means that the intermediate substance changes its position due to external force or its own force. Also, "changing over time" means that the appearance and internal state of the intermediate substance (flow velocity distribution, density distribution, concentration distribution, dispersion state of contained substances, number and size of contained substances or foreign matter, etc.) changes.

[0034] Intermediates are, for example, compounds in the liquid or gas phase, or mixtures in the liquid or gas phase. An intermediate is, for example, a solution obtained by dissolving or melting one raw material in another liquid (raw material). Another example is a culture medium used to cultivate microorganisms. Furthermore, when the final product is solid or liquid, the intermediate is a liquid or gas phase different from that of the final product.

[0035] The acquisition unit 100 is composed of various sensors and measures the state of the intermediate product, etc. The characteristic acquisition unit 102 is composed of various sensors and measures the characteristics of the final product, etc. The information processing device 10 stores the state of the intermediate product, etc. acquired by the acquisition unit 100 and the characteristics of the final product, etc. acquired by the characteristic acquisition unit 102, and based on this data, predicts the characteristics of the final product, etc. and adjusts parameters related to the composition of raw materials and processing conditions of the process from raw materials to the final product.

[0036] The design and manufacturing process 50 includes an input unit 51 into which raw materials are fed, a first process unit 52 that performs a first process on the input raw materials to produce an intermediate, and a second process unit 53 that performs a second process on the intermediate to produce a final product. In the design and manufacturing process 50, processes such as cultivation, brewing, kneading, and doping to produce high-concentration dissolved substances are carried out, resulting in the production of a fluid as an intermediate. The information processing device 10 then applies a real-time observation modality, mainly using waves (ultrasound, etc.), to the manufacturing process using this intermediate to acquire the dynamic state and / or structure of the fluid as image data, which is used as an intermediate descriptor.

[0037] The design and manufacturing process 50 can be applied to various industries, such as "1. Bio-manufacturing," "2. Composite material development," "3. Chemical synthesis manufacturing," "4. Food manufacturing," "5. Cosmetics manufacturing," and "6. Pharmaceutical manufacturing," as described below (hereinafter also referred to as applicable industries 1 to 6). In other words, the applicable industries in which the information processing device, etc. according to this embodiment is preferably used, and the following are included as intermediates for obtaining intermediate descriptors used to predict the characteristics of the final product in those cases.

[0038] 1. Bio-manufacturing Bio-manufacturing is an initiative that aims to efficiently produce materials from natural or biologically derived resources (biomass) by utilizing biochemical processes, such as photosynthesis, instead of conventional chemical processes. Compared to conventional chemical synthesis, the production processes and products of materials, fuels, and pharmaceuticals using such biotechnology, or the cultivation of small and dynamic useful species like microorganisms, are more complex. Furthermore, the physical properties and processes are not always deductively understood, making it difficult to judge the quality of the final product. In addition, in order to produce (reproduce) good products without loss, it is considered necessary to manage and visualize the state of the fluids that are intermediate products in the design and manufacturing process. In this embodiment, in order to achieve appropriate and rapid process control, the structure and state of the intermediate fluids are used as intermediate descriptors and are acquired and processed by non-destructive, real-time visualization or multidimensional data acquisition.

[0039] 2. Composite Material Development Lightweight plastics and resins are increasingly being introduced into areas where strength and rigidity are required, areas where metals have traditionally been used. Furthermore, the properties required of materials are becoming more sophisticated and complex, and accordingly, the materials themselves are becoming more complex. For example, in composite materials, which are increasingly being introduced into structures because they can achieve both lightness and high rigidity, the conventional method involved stacking and compressing long-fiber prepregs. However, due to issues with manufacturing costs and the increasing complexity of product shapes, there is a shift towards injection molding, where pellets made by kneading thermoplastic resin and short-fiber fillers are heated and melted. However, in this method, it is known that when the injected resin branches and rejoins in the mold, the temperature drops, and it solidifies without being sufficiently mixed, resulting in weld lines, which significantly reduce strength. Also, if the filler is not uniformly dispersed, this also contributes to strength reduction. There are methods such as the short-shot method to address these problems process-wise, but they are time-consuming and labor-intensive, making them inefficient. Therefore, there is a need for a method that accurately controls the state of the molten material poured into the mold and reliably improves the quality of the final product. In this embodiment, data related to intermediate materials in a fluid state that affects quality is converted into intermediate descriptors, and the results are promptly fed back.

[0040] 3.Chemical compound production The flow method is attracting attention as a chemical synthesis method. The flow method involves pressurizing a container and proceeding with synthesis while flowing a small amount of reaction solution, making it a highly efficient and scalable manufacturing method. The flow method is also considered an effective method for producing complex chemical products using natural resources and other materials through future biotechnology. However, even with this method, as products and materials become more complex and composite as described above, it is becoming increasingly difficult to easily judge their quality. Furthermore, in order to improve manufacturing efficiency, it is necessary to observe the intermediate fluid without waiting for the evaluation of the final product and to quickly feed the results back into the manufacturing process. However, currently, the main means of monitoring the state of fluids are thermometers, pressure gauges, or visual inspection using cameras, and it is not possible to obtain sufficient data to deal with increasingly complex products and materials. In this embodiment, data on intermediates in a flowing state that affects manufacturing quality is acquired as multidimensional data of intermediate descriptors and processed rapidly.

[0041] 4. Food manufacturing In recent years, with the aim of realizing a sustainable society, the development of artificial food ingredients has been progressing, and some alternative meats are already appearing on the market. Food ingredients are directly related to human senses, and there are still many evaluation indicators that cannot be quantified, such as taste, texture, mouthfeel, and swallowability. Furthermore, considering the wide variety of base materials, the number of parameters that must be considered during manufacturing becomes astronomical, resulting in a complex system that cannot be chemically explained. Therefore, the production of artificial food ingredients relies on human evaluation through actual consumption, and faces many challenges in terms of stability, reproducibility, and efficiency. For example, protein-based food ingredients (dairy beverages, sausages, collagen, etc.) are produced by selecting microorganisms, transforming them, carrying out cell proliferation and fermentation in a reactor, recovering proteins, and mixing and synthesizing multiple proteins. However, managing the growth state of microorganisms in the culture medium is extremely important. Currently, various sensors are installed to manage and control the state inside the reactor, but conventional devices have difficulty accurately grasping the multivariate data of the state of microorganisms in the culture medium. In this embodiment, in order to accurately manufacture food products, which are difficult to quantify in terms of quality, based on the complex material of living organisms, the state of the fluid (intermediate) and its contents is captured as multidimensional data using intermediate descriptors and quickly fed back into the manufacturing process.

[0042] 5. Cosmetics manufacturing Cosmetics are products that come into direct contact with human skin, and therefore, high quality is especially important. With the increasing diversity of society in recent years, demand for cosmetics has expanded, and instead of mass production and mass consumption, a wide variety of products are being developed. Meanwhile, corporate responsibility for products is being questioned more than ever before, and the concepts of quality assurance and safety are also changing. Specifically, there is a growing tendency to emphasize the "results" of product characteristics rather than the "reasons" such as principles and mechanisms, and the need for individual product inspections, rather than random sampling inspections after mass production, is increasing. Cosmetics are a kind of complex system in which various elements are intertwined. However, evaluation and measurement of the final product have limitations in the parameters that can be obtained, making it difficult to explain its quality quantitatively or numerically. As a result, judging the quality of cosmetics has to rely on vague indicators such as the five senses (sight, smell, touch, etc.) and past experience and sensibilities, which is extremely inefficient. On the other hand, many cosmetics are liquids (fluids) as intermediate materials in their design and manufacturing processes. Evaluating the state of this intermediate product before commercialization is thought to yield many parameters related to the characteristics of the final product. However, conventional sensors such as thermometers, viscometers, or cameras cannot acquire sufficient quantity and quality of data to address the complexity of product quality. In this embodiment, multidimensional and multivariate data is rapidly acquired and processed in large quantities as intermediate descriptors for the intermediate product in a flowing state.

[0043] 6. Pharmaceutical manufacturing (biopharmaceuticals) Pharmaceuticals directly affect the human body and are therefore naturally required to be of high quality. In recent years, the role of pharmaceuticals has become increasingly important due to the aging populations of developed countries, global population growth, and the expansion of global pandemics. Meanwhile, the wave of biotechnology has also reached the pharmaceutical field, and efforts are underway to produce more effective pharmaceuticals from animal cells and other sources using genetic engineering and other technologies, instead of conventional chemical synthesis methods. However, compared to conventional pharmaceuticals made from small molecules, the manufacturing process of biopharmaceuticals, which involves microorganisms, has the problem that a far greater number of parameters must be monitored and measured, as some mechanisms are not yet fully understood. Furthermore, from the perspective of ensuring the quality of pharmaceuticals, sampling inspection of the final product has fundamental challenges in terms of safety and efficiency, and this is expected to become an even bigger problem in the age of biopharmaceuticals. On the other hand, in many biopharmaceutical processes, fluids (culture media) are used as intermediates before reaching the final product. Therefore, by evaluating the state of these intermediates, it is highly likely that many parameters related to the characteristics of the pharmaceutical can be efficiently obtained and, in some cases, fed back into the upstream process, thereby enabling the stable production of high-quality pharmaceuticals. In this embodiment, multidimensional and multivariate data is rapidly acquired and processed as intermediate descriptors for intermediate materials in a flowing state.

[0044] (Acquisition part 100) The acquisition unit 100 acquires data indicating the state of the intermediate product as an intermediate descriptor. The acquisition unit 100 may be equipped with a sensor or measuring instrument. Alternatively, the acquisition unit 100 may be configured as a communication interface and acquire measurement data from sensors or the like that are pre-installed in the design and manufacturing process 50. For example, the acquisition unit 100 acquires the state of the intermediate product for the above-mentioned applicable industries 1 to 6 as an intermediate descriptor.

[0045] (Characteristic acquisition unit 102) The characteristic acquisition unit 102 acquires data indicating the characteristics of the final product. The characteristic acquisition unit 102 may be equipped with sensors or measuring instruments. Alternatively, the characteristic acquisition unit 102 may be configured as a communication interface and acquire measurement data from sensors or the like that are pre-installed in the design and manufacturing process 50.

[0046] (Information processing device 10) The information processing device 10 comprises a processing unit 110, a prediction unit 120, a first process parameter adjustment unit 131, a first storage unit 141, and a second storage unit 142.

[0047] The processing unit 110 performs predetermined pre-processing on the intermediate descriptors acquired by the acquisition unit 100 in order to make them correspond to the data to be input to the prediction unit 120. Here, the predetermined pre-processing includes any of the following: digital conversion, up / down sampling, normalization, data size adjustment, filtering, image creation, statistical analysis, feature extraction, etc.

[0048] The prediction unit 120 includes a first pre-trained model and uses this first pre-trained model to output (predict) the characteristics of the final product based on the input intermediate descriptor. The method for training the first pre-trained model will be described later.

[0049] The first process parameter adjustment unit 131 includes a second trained model and, using the second trained model's demodulated data, outputs or adjusts the composition of raw materials in the input unit 51 and the processing conditions in the first process unit 52 based on the input intermediate descriptor. Furthermore, the first process parameter adjustment unit 131 may also adjust the processing conditions of the second process unit 53. Hereinafter, these compositions and processing conditions will be referred to as process parameters. Process parameters include at least one of the following: the composition, concentration, particle size, content, size, shape, etc., of the raw materials for the final product, and environmental conditions such as temperature, humidity, pressure, pH, applied current, applied voltage, and flow rate in the process of processing them.

[0050] The first memory unit 141 stores multiple sets of data, such as intermediate descriptors acquired by the acquisition unit 100 and characteristics of the final product acquired by the characteristic acquisition unit 102. This is training sample data (training data) and is used to train the first trained model used in the prediction unit 120. The training method for the first trained model will be described later with reference to Figure 8.

[0051] The second memory unit 142 stores multiple sets of intermediate descriptors acquired by the acquisition unit 100 and process parameter data input to the input unit 51 and / or the first process unit 52. This is training sample data (training data) and is used to train the second trained model used by the first process parameter adjustment unit 131. The training method for the second trained model will be described later with reference to Figure 9.

[0052] (Intermediate descriptor) The following examples of intermediate descriptors will be explained with reference to Figures 2 to 3B.

[0053] Figure 2 is a schematic diagram for conceptually explaining intermediate descriptors. The explanatory variables are process parameters (composition of raw materials and process conditions, etc.). The dependent variable is the characteristics of the final product (characteristics of the final product, material properties, etc.). Depending on the composition of the raw materials and process conditions, intermediate products are obtained, and intermediate descriptors corresponding to the state of the intermediate products are obtained. There is not limited to one intermediate descriptor for a single final product; there may be multiple intermediate descriptors. Furthermore, intermediate descriptors can be scaled between the explanatory variables and the dependent variable (as will be explained later using Figure 3A).

[0054] Intermediate descriptors are selected that have a predetermined relationship with the characteristics of the final product. This predetermined relationship means that, through statistical analysis or machine learning, the intermediate descriptor and the characteristics have a certain level of correlation or coefficient of determination.

[0055] Intermediate descriptors are preferably acquired by a measurement means (hereinafter referred to as modality) that uses a predetermined wave. The predetermined wave includes at least one of acoustic waves and electromagnetic waves (microwaves, light waves, X-rays, etc.), with ultrasound (elastic waves) being particularly preferred. When ultrasound is used, this wave may be used in combination with means for transporting the intermediate object.

[0056] Furthermore, the intermediate descriptor is preferably image information obtained by photographing an intermediate object using a predetermined wave, and in particular, an ultrasonic image obtained by irradiating it with ultrasound. This image information includes an image of noise and / or distortion caused by the propagation of the wave within the intermediate object. Alternatively, instead of the image information itself, or in addition to it, feature quantities extracted from the image information may be used as the intermediate descriptor. Feature quantities refer to information extracted by applying statistical analysis such as frequency analysis or principal component analysis to the image information. Additionally, the image information used as the intermediate descriptor may include tomographic images. A tomographic image is an image in two or three-dimensional space of changes in intensity information or phase information of reflective or transmitted components that occur due to structural or compositional discontinuities or non-uniformities within the intermediate object when a predetermined wave propagates within the intermediate object, along with positional information. Furthermore, the image information used as the intermediate descriptor may include not only still images but also moving images that change over time.

[0057] The state of intermediate materials affects the characteristics of the final product. Depending on the characteristics of the final product, an appropriate intermediate descriptor may be selected, particularly depending on the characteristics or structure to be achieved. Here, "characteristics to be achieved" refers to the required or noteworthy characteristics of the final product. Similarly, "structure to be achieved" refers to the required or noteworthy structure of the final product. Hereafter, the characteristics and structure to be achieved of the final product will be collectively referred to simply as "characteristics to be achieved."

[0058] Figure 3A is a schematic diagram illustrating an example of multiple intermediate descriptors. As shown in Figure 3A, there are multiple intermediate descriptors corresponding to constituent size or unit structure (hereinafter referred to as constituent size, etc.) that determine or are major factors in the properties of the final product, and an appropriate intermediate descriptor can be selected from among them depending on the properties to be realized. Constituent size, etc. that are major factors in determining the properties of the final product include, for example, interatomic distance, intermolecular distance, unit cell, particle size of contained substances, aggregation size, element size, etc.

[0059] In this case, the modality for obtaining intermediate descriptors can be appropriately selected based on the constituent size of the intermediate and / or final product, by choosing the frequency (wavelength) or wave number (wave duration or signal length), or by selecting modalities with different frequencies or wave numbers. For example, ultrasound, electromagnetic waves (microwaves, light waves, X-rays, etc.) can be used. Specifically, for example, ultrasound can be used if the constituent size is several tens of micrometers or larger, microwaves or millimeter waves can be used if it is on the order of several micrometers, and X-rays or light waves can be used if it is on an even smaller order. Furthermore, the selection of these modalities can be appropriately changed depending on the properties of the intermediate (frequency dependence of transmittance and reflectance, etc.) and the surrounding environment.

[0060] Figure 3B is a schematic diagram illustrating another example of multiple intermediate descriptors. As shown in Figure 3B, when there are multiple properties to be realized in the final product, multiple intermediate descriptors can be provided corresponding to these properties, and an appropriate one can be selected from among them according to the properties to be realized. Figure 3B shows a case where the final product is a composite material composed of multiple materials. For example, if the properties to be realized in the final product relate to elastic properties or structural information, ultrasound can be used as the modality for acquiring intermediate descriptors. If the properties to be realized relate to conductivity or dielectric properties, microwaves can be used, and if they relate to optical properties such as refractive index or interatomic forces, light waves or X-rays can be used.

[0061] (Second embodiment) Figure 4 is a diagram showing the configuration of the design and manufacturing process including the information processing device 10b according to the second embodiment. As shown in Figure 4, the information processing device, etc. according to the second embodiment differs from the first embodiment (Figure 1) in that it includes a second process parameter adjustment unit 132 instead of the first process parameter adjustment unit 131, and a third storage unit 143 instead of the second storage unit 142. However, the connections and operations of the other blocks are the same as in Figure 1. Therefore, the same numbers are used for parts that are the same as in Figure 1 and their explanations are omitted, while new numbers are used only for parts that differ from Figure 1, and their contents are explained below.

[0062] In Figure 4, the second process parameter adjustment unit 132 uses the third trained model to input the characteristics of the final product predicted by the prediction unit 120, and outputs or adjusts the process parameters.

[0063] The third memory unit 143 stores multiple sets of data, including the characteristics of the object and process parameters. This is training sample data (training data) and is used to train the third pre-trained model used in the second process parameter adjustment unit 132. The training method for the third pre-trained model will be described later.

[0064] (Third embodiment) Figure 5 is a diagram showing the configuration of the design and manufacturing process including the information processing device 10c according to the third embodiment. As shown in Figure 5, the information processing device according to the third embodiment differs from the first embodiment (Figure 1) in that it includes an ultrasonic transmitting / receiving unit 111 as the acquisition unit 100 and an image data generation unit 112 as the processing unit 110, but the connections and operations of the other blocks are the same as in Figure 1. Therefore, the same numbers are used for parts that are the same as in Figure 1 and their explanations are omitted, and new numbers are used only for parts that differ from Figure 1, and their contents are explained below.

[0065] The ultrasonic transmitting / receiving unit 111 includes, for example, an ultrasonic probe (such as a piezoelectric element), and obtains a received signal by irradiating an intermediate liquid with ultrasonic waves and receiving transmitted and / or reflected waves from the intermediate. If the intermediate is autonomously flowing, the ultrasonic waves may be irradiated while fixing its position and / or direction, and if the intermediate is stationary, the ultrasonic waves may be irradiated while moving its position and / or direction relative to the intermediate. The image data generation unit 112 generates an ultrasonic image based on the strength information and phase information of the received signal. The ultrasonic image includes an ultrasonic tomographic image representing the cross-sectional information of the intermediate object, and may be a still image or a moving image. The image data generation unit 112 may also generate quantitative data such as feature quantities from this ultrasonic image.

[0066] Figure 6A shows an example configuration for acquiring ultrasonic images of liquid-phase intermediates (fluids) using an ultrasonic transmitting / receiving unit 111 (ultrasonic probe). Figure 6B shows examples of ultrasonic images (tomographic images) obtained by irradiating three types of fluids A to C with different compositions using the configuration shown in Figure 6A. Here, fluids A to C are milk products, for example, when a dairy product such as yogurt is the final product, serving as intermediates. Also, while Figure 6B shows a still image at a certain point in time, actual ultrasonic images are moving images that change over time. By observing this figure, information can be obtained such as the fact that fluid A contains large particles but has a low particle density, and that it has a fast downward flow rate including the particles, indicating low viscosity. Information can also be obtained such as the fact that fluid C has fine particles, a high particle density, and a slow flow rate, indicating high viscosity, and that fluid B exhibits an intermediate state between fluids A and C in terms of particle size and flow rate. Furthermore, considering the recent improvements in computing power, the development of AI technologies such as machine learning, and the advancements in image processing and analysis technologies, it is highly likely that by applying these technologies, it will be possible to obtain not only information that humans can recognize or understand and verbalize / quantify, as described above, but also advanced information that humans cannot currently understand or express, from such video information. Thus, according to the third embodiment, by using ultrasound as a modality and irradiating the intermediate with ultrasound, it is possible to obtain image information containing even more advanced information as an intermediate descriptor, in addition to quantitative and / or qualitative information including particle size, density, dispersion state, flow state such as flow velocity, and viscoelasticity of the contents of the intermediate.

[0067] (Characteristic prediction processing, and process parameter adjustment processing) Next, referring to Figures 7 to 9, the characteristic prediction processing and process parameter adjustment processing performed by the information processing device 10 (Figure 1) or 10c (Figure 5) will be described. Figure 7 is a flowchart showing the characteristic prediction processing and parameter adjustment processing performed by the information processing device 10 or 10c.

[0068] (Step S11) The acquisition unit 100 acquires intermediate descriptors related to the intermediate object. For example, the acquisition unit 100 irradiates the intermediate object with ultrasound and acquires the received ultrasound signal. The processing unit 110 generates an ultrasound image (intermediate descriptor) from the received signal through predetermined processing. The ultrasound image may be one or more time-series still images, or a moving image over a predetermined period of time.

[0069] (Step S12) The prediction unit 120 uses a first trained model to predict the characteristics of the final product from the intermediate descriptor. For example, the ultrasonic image acquired in step S11 is input to the first trained model, and the characteristics of the final product are predicted as its output. These characteristics may be, for example, numerical data or information expressed as a classification result of the characteristics. In the case of a classification result, the likelihood is calculated for each classification candidate, and the classification with the highest likelihood may be obtained as the classification result.

[0070] (Step S13) The information processing device 10 or 10c evaluates the prediction results (characteristics of the final product, etc.) obtained in step S12. Specifically, it compares the prediction results with the predetermined characteristics of the final product to be realized. If the two match or fall within a predetermined difference (target achieved), the information processing device 10 or 10c skips step S14 and terminates the process (end). On the other hand, if the prediction results do not fall within a predetermined difference (target not achieved), the information processing device 10 or 10c proceeds to step S14.

[0071] (Step S14) The first process parameter adjustment unit 131 adjusts the process parameters using the second trained model. Specifically, it inputs the intermediate descriptor obtained in step S11 to the second trained model and obtains process parameters as its output. These process parameters may be displayed as recommended adjustment values ​​on a display unit, or they may be reflected in the raw material composition in the input unit 51 and the processing conditions of the first process unit 52, as shown in Figures 1 and 5. Furthermore, if the process parameters include the processing conditions of the second process unit 53, they may be reflected in the process conditions of the second process unit 53.

[0072] (Training the first pre-trained model) Figure 8 is a flowchart illustrating the processing of the machine learning method for the first trained model. The processing shown in Figure 8 uses the training sample data stored in the first memory unit 141. The first memory unit 141 stores multiple sets of data as training sample data, which are combinations of intermediate descriptors acquired by the acquisition unit 100 and characteristic data (ground truth data) of the final manufactured product acquired by the characteristic acquisition unit 102. In the following, it is assumed that the information processing device 10 or 10c functions as a learning machine to perform machine learning. However, it is not limited to this, and other cloud computers may also perform machine learning as learning machines (the same applies to the processing in Figure 9 described later). Furthermore, in the following, a learning method using a neural network constructed by combining perceptrons in a learning machine (not shown) will be described, but it is not limited to this, and various methods can be adopted as long as it is supervised learning. For example, random forests, support vector machines (SVMs), boosting, Bayesian network linear discriminant analysis, nonlinear discriminant analysis, etc. can be applied (the same applies to the processing in Figure 9 described later).

[0073] (Step S21) The information processing device 10 or 10c, which functions as a learning machine, reads training sample data, which is training data, from the first storage unit 141. Initially, the learning machine reads the intermediate descriptor of the first set of intermediate products (hereinafter also referred to as input data) and the corresponding characteristic data of the final product (hereinafter also referred to as output data). Here, for example, the intermediate descriptor is an ultrasound image.

[0074] (Step S22) The learning machine inputs the input data (intermediate descriptors) from the loaded training sample data into the neural network. Initially, the neural network is configured with weights specified by the user.

[0075] (Step S23) The learning machine compares and evaluates the prediction results of the neural network, i.e., the estimated characteristics of the final product, with the output data (characteristics data), which is the ground truth data.

[0076] (Step S24) The learning machine adjusts the parameters (weights) of the neural network based on the comparison results. For example, by performing backpropagation, it adjusts and updates the parameters (weights) to reduce the error in the comparison results.

[0077] (Step S25) If the learning machine has finished processing all of the training sample data (YES), it proceeds to step S26; otherwise, it returns to step S21, reads the next training sample data, and repeats the process from step S21 onward.

[0078] (Step S26) The learning machine stores the first trained model constructed through the previous processing into a memory area (for example, the memory in the prediction unit 120) and terminates (end). From here on, the processing shown in Figure 7 (step S12) is performed using this first trained model.

[0079] (Training the second pre-trained model) Figure 9 is a flowchart illustrating the processing of the machine learning method for the second trained model. The processing shown in Figure 9 uses the training sample data stored in the second memory unit 142. The second memory unit 142 stores multiple sets of data, which are combinations of intermediate descriptors and process parameters (ground truth data), as training sample data.

[0080] (Step S31) The information processing device 10 or 10c, which functions as a learning device, reads training sample data, which is training data, from the second storage unit 142. Initially, the learning device reads the intermediate descriptors of the first set of intermediate objects (hereinafter also referred to as input data) and the process parameter data of the first manufacturing process corresponding to them (hereinafter also referred to as output data). Here, for example, the intermediate descriptors are ultrasound images.

[0081] (Step S32) The learning machine inputs the input data (intermediate descriptors) from the loaded training sample data into the neural network. Initially, the neural network is configured with weights specified by the user.

[0082] (Step S33) The learning machine compares and evaluates the prediction results of the neural network, i.e., the estimated process parameter data, with the ground truth data, which is the output data (process parameters).

[0083] (Step S34) The learning machine adjusts the parameters (weights) of the neural network based on the comparison results. For example, by performing backpropagation, it adjusts and updates the parameters (weights) to reduce the error in the comparison results.

[0084] (Step S35) If the learning machine has finished processing all of the training sample data (YES), it proceeds to step S36; otherwise, it returns to step S31, reads the next training sample data, and repeats the process from step S31 onwards.

[0085] (Step S36) The learning machine stores the second trained model constructed through the previous processing into a memory area (for example, the memory in the first process parameter adjustment unit 131) and terminates (end). From this point onward, the processing shown in Figure 7 (step S14) is performed using this second trained model.

[0086] In addition, in Figure 9, the learning machine may be configured to reverse the relationship between the input data and the output data, using process parameters as input data and intermediate descriptors as output data, predicting intermediate descriptors from the process parameters, and comparing and evaluating the results with the correct data (intermediate descriptors).

[0087] (Characteristic prediction processing, and process parameter adjustment processing) Next, with reference to Figures 10 and 11, the characteristic prediction processing and process parameter adjustment processing performed by the information processing device 10b (Figure 4) will be described. Figure 10 is a flowchart of the characteristic prediction processing and parameter adjustment processing performed by the information processing device 10b. This figure is a modified version of some of the steps in Figure 7. Therefore, steps that have the same content as in Figure 7 are given the same number and their explanation is omitted, while only the differences from Figure 7 are given new numbers and their content is explained below.

[0088] (Step S44) The second process parameter adjustment unit 132 adjusts the process parameters using the third trained model. Specifically, the predicted results of the characteristics of the final product obtained in step S12 are input to the third trained model, and process parameters are obtained as the output. These process parameters may be displayed as recommended adjustment values ​​on the display unit, etc., or they may be reflected in the raw material composition etc. in the input unit 51 and the processing conditions of the first process unit 52, etc., as shown in Figure 4, etc. Furthermore, if the processing parameters include the processing conditions of the second process unit 53, they may be reflected in the process conditions of the second process unit 53.

[0089] (Training the third pre-trained model) Figure 11 is a flowchart showing the processing of the machine learning method for the third trained model. The processing shown in Figure 11 uses the training sample data stored in the third memory unit 143. The third memory unit 143 stores multiple sets of data as training sample data, which combine the characteristics of the final product with process parameters (ground truth data). This figure is a modified version of some steps in Figure 9. Therefore, steps with the same content are given the same number and their explanations are omitted, and only the differences from Figure 9 are given new numbers and their contents are explained below.

[0090] (Step S51) The information processing device 10b, which functions as a learning machine, reads training sample data, which is training data, from the third storage unit 143. Initially, the learning machine reads the first set of data on the characteristics of the final manufactured product (hereinafter also referred to as input data) and the corresponding data on the process parameters of the first manufacturing process (hereinafter also referred to as output data).

[0091] (Step S52) The learning machine inputs the input data (characteristic data, etc.) from the loaded training sample data into the neural network. Initially, the neural network is configured with weights specified by the user.

[0092] (Step S53) The learning machine compares the prediction results of the neural network, i.e., the estimated process parameter data, with the ground truth data, which is the output data (process parameters).

[0093] (Step S56) The learning machine stores the third trained model constructed through the previous processing into a memory area (for example, the memory in the second process parameter adjustment unit 132) and terminates (end). From here on, the processing shown in Figure 10 (step S44) is performed using this third trained model.

[0094] In addition, in Figure 11, the learning machine may be configured to reverse the relationship between the input data and the output data, using process parameters as input data and data such as the characteristics of the final product as output data, predicting the data such as the characteristics of the final product from the process parameters, and comparing and evaluating the result with the correct data (characteristics data).

[0095] Thus, in a design and manufacturing process to obtain a final product having predetermined characteristics from raw materials through intermediates, the information processing apparatus 10, 10b, or 10c according to this embodiment can acquire intermediate descriptors, which are data indicating the structure and / or state of the intermediates, as ultrasonic images or the like, predict and evaluate the characteristics of the final product using the intermediate descriptors, and appropriately adjust process parameters based on the results.

[0096] <Examples> The following describes examples of applications of the information processing device 10, etc., according to this embodiment, with reference to Figures 12 to 16.

[0097] (Example 1) Figure 12 shows a schematic configuration of the design and manufacturing process in Example 1. Example 1 is an example in which the acquisition unit 100, the characteristic acquisition unit 102, and the information processing device 10 are applied to application industry 1 (bio-manufacturing using Euglena). Note that the information processing device 10 may be replaced with the information processing device 10b or 10c (the same applies to Examples 2 to 5 below). The acquisition unit 100 acquires various intermediate descriptors from the intermediate product (culture medium) in the hydrothermal treatment and hydrogenation treatment in the first process unit 52. In the example shown in Figure 12, the acquisition unit 100 can acquire intermediate descriptors such as the presence and density of microorganisms (foreign matter), the size, shape, and movement state of Euglena, etc., from the intermediate product obtained by the hydrothermal treatment in the preceding stage. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. Furthermore, the acquisition unit 100 can acquire intermediate descriptors such as sulfur content and nitrogen content from the intermediate product obtained by the hydrogenation treatment in the subsequent stage. The prediction unit 120 uses some or all of these intermediate descriptors to predict the characteristics of the final product. The first process parameter adjustment unit 131 outputs appropriate values ​​for the process parameters of the input unit 51 and the first process unit 52 based on the intermediate descriptors, and adjusts the process parameters based on these values.

[0098] (Example 2) Figure 13 is a diagram illustrating the schematic configuration of the design and manufacturing process in Example 2. Example 2 is an example in which the acquisition unit 100, the property acquisition unit 102, and the information processing device 10 are applied to application industry 2 (composite material development). The acquisition unit 00 acquires various intermediate descriptors from the intermediate material (molten resin) in the injection injection, rolling, casting, and cooling solidification processes in the first process unit 52. In the example shown in Figure 13, the acquisition unit 100 can acquire the flow velocity distribution of the molten resin, fiber orientation, etc., as intermediate descriptors from the intermediate material obtained by the preceding injection injection, rolling, and casting processes. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. Furthermore, the acquisition unit 100 can acquire the solidification status, temperature distribution, etc., as intermediate descriptors from the intermediate material obtained by the subsequent cooling solidification process. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. The prediction unit 120 uses some or all of these acquired intermediate descriptors to predict the properties of the final product. Furthermore, the first process parameter adjustment unit 131 outputs appropriate values ​​for the process parameters of the input unit 51 and the first process unit 52 based on the intermediate descriptor, and adjusts the process parameters based on these values.

[0099] (Example 3) Figure 14 is a diagram illustrating the schematic configuration of the design and manufacturing process in Example 3. Example 3 is an example of applying the acquisition unit 100, the property acquisition unit 102, and the information processing device 10 to the resin molding field, as an application industry 3 (chemical compound manufacturing). The acquisition unit 100 acquires various intermediate descriptors from the intermediate material (flow resin) during the injection process and solidification process (including holding pressure and cooling) in the first process unit 52. In the example shown in Figure 14, the acquisition unit 100 can acquire the flow velocity distribution of the resin, the resin orientation, etc., as intermediate descriptors from the intermediate material obtained by the preceding injection process. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. Furthermore, the acquisition unit 100 can acquire the solidification status of the resin, the temperature distribution, etc., as intermediate descriptors from the intermediate material obtained by the subsequent solidification process. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. The prediction unit 120 uses some or all of these acquired intermediate descriptors to predict the properties of the final product. Furthermore, the first process parameter adjustment unit 131 outputs appropriate values ​​for the process parameters of the input unit 51 and the first process unit 52 based on the intermediate descriptor, and adjusts the process parameters based on these values.

[0100] (Example 4) Figure 15 is a diagram illustrating the schematic configuration of the design and manufacturing process in Example 4. Example 4 is an example of applying the acquisition unit 100, the characteristic acquisition unit 102, and the information processing device 10 to either application industry 4 (food manufacturing) or application industry 6 (biopharmaceuticals). The acquisition unit 00 acquires various intermediate descriptors from the intermediate product (culture medium, etc.) in the cell proliferation process and the fermentation / purification process in the first process unit 52. In the example shown in Figure 15, the acquisition unit 100 can acquire cell size, density, motility, etc. as intermediate descriptors from the intermediate product obtained by the preceding cell proliferation process. These intermediate descriptors may also be acquired as ultrasound images using ultrasound. The acquisition unit 100 can also acquire fermentation state, pH, etc. as intermediate descriptors from the intermediate product obtained by the subsequent fermentation / purification process. The prediction unit 120 uses some or all of these acquired intermediate descriptors to predict the characteristics of the final product. Furthermore, the first process parameter adjustment unit 131 outputs appropriate values ​​for the process parameters of the input unit 51 and the first process unit 52 based on the intermediate descriptor, and adjusts the process parameters based on these values.

[0101] (Example 5) Figure 16 is a diagram illustrating the schematic configuration of the design and manufacturing process in Example 5. Example 5 is an example of applying the acquisition unit 100, the characteristic acquisition unit 102, and the information processing device 10 to applicable industry 5 (cosmetics manufacturing). The acquisition unit 00 acquires various intermediate descriptors from the intermediate material (fluid) in the bulk manufacturing (mixing) process and the emulsification process in the first process unit 52. In the example shown in Figure 16, the acquisition unit 100 can acquire flow velocity distribution, particle size distribution, etc., as intermediate descriptors from the intermediate material obtained by the preceding bulk manufacturing process. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. The acquisition unit 100 can also acquire emulsification status, density distribution, etc., as intermediate descriptors from the intermediate material obtained by the subsequent emulsification process. These intermediate descriptors may also be acquired as ultrasonic images using ultrasound. The prediction unit 120 uses some or all of these acquired intermediate descriptors to predict the characteristics of the final product. Furthermore, the first process parameter adjustment unit 131 outputs appropriate values ​​for the process parameters of the input unit 51 and the first process unit 52 based on the intermediate descriptor, and adjusts the process parameters based on these values.

[0102] Thus, the information processing device, etc., according to this embodiment can be applied to various industries, and by using the acquired intermediate descriptors, the characteristics of the final product in each applicable industry can be predicted, enabling efficient manufacturing.

[0103] The configuration of the information processing device, etc., described above is intended to illustrate the main configuration in order to explain the features of the above embodiment, and is not limited to the above configuration; various modifications can be made within the scope of the claims. Furthermore, it does not preclude configurations that are generally found in information processing devices, etc.

[0104] Each embodiment and each example may be applied in combination. For example, the first to third embodiments may be applied in combination with each other, and these combinations may be applied to the first to fifth embodiments. In addition, the information processing device 10 (or 10b, 10c) may output the prediction results (characteristics of the target object) obtained in step S12 to the user. For example, the information processing device 10 may display the prediction results on a display unit or transmit the prediction results to a pre-configured terminal device (PC). Upon receiving this, the user can recognize in advance how much the predicted characteristics of the final product deviate from the pre-configured characteristics of the final product to be achieved, or how close it is to the target.

[0105] Furthermore, the means and methods for performing various processing in the information processing devices 10, 10b, or 10c according to the above-described embodiment can be implemented by either a dedicated hardware circuit or a programmed computer. The program may be provided, for example, on a computer-readable recording medium such as a USB memory stick or a DVD (Digital Versatile Disc)-ROM, or it may be provided online via a network such as the Internet. In this case, the program recorded on the computer-readable recording medium is usually transferred to and stored in a storage unit such as a hard disk. The program may also be provided as a standalone application software, or it may be incorporated into the software of the device as a function of the device. [Explanation of Symbols]

[0106] 10 Information Processing Devices 110 Processing Unit 120 Prediction Section 131 First process parameter adjustment unit 132 Second process parameter adjustment unit 141 First Memory Unit 142 Second Memory Unit 143 Third Memory Unit 100 Acquisition Department 102 Characterization Unit 101 Ultrasonic Transceiver Unit 111 Image Data Generation Unit

Claims

1. In the design and manufacturing process of an object, which is a final product having predetermined characteristics, starting from raw materials and progressing through intermediates, An information processing device comprising a prediction unit that predicts the characteristics of an object based on an intermediate descriptor, which is data indicating the state and / or structure of the intermediate object obtained.

2. The information processing apparatus according to claim 1, wherein the intermediate is a fluid that can move and / or change over time.

3. The information processing apparatus according to claim 1, wherein the intermediate has a phase different from that of the object.

4. The information processing apparatus according to claim 1, wherein the intermediate descriptor includes image information and / or feature quantities extracted from the image information.

5. The information processing apparatus according to claim 4, wherein the image information includes a tomographic image.

6. The information processing apparatus according to claim 4, wherein the intermediate descriptor is data obtained by irradiating the intermediate object with a wave of a predetermined frequency and detecting at least one of the transmitted wave, reflected wave, and scattered wave from the intermediate object.

7. The information processing apparatus according to claim 1, wherein the prediction unit predicts the characteristics of the object from the intermediate descriptor using a first trained model that outputs the characteristics of the object based on the intermediate descriptor.

8. The information processing apparatus according to claim 1, wherein the intermediate descriptors are multiple for a single object.

9. The information processing apparatus according to claim 8, wherein the intermediate descriptor is selected from a plurality of intermediate descriptors according to the characteristics.

10. The information processing apparatus according to claim 8, wherein the intermediate descriptor is selected from a plurality of intermediate descriptors according to the constituent size of the object whose characteristics are determined.

11. The aforementioned intermediate descriptors are multiple for a single object, The information processing apparatus according to claim 6, wherein a selection is made from a plurality of intermediate descriptors by setting the frequency of the wave or the signal length.

12. The information processing apparatus according to claim 1, further comprising a process parameter adjustment unit that adjusts the process parameters based on the intermediate descriptor using a second trained model that outputs process parameters relating to the intermediate based on the intermediate descriptor.

13. In the design and manufacturing process of an object, which is a final product having predetermined characteristics, starting from raw materials and progressing through intermediates, (a) A step of obtaining an intermediate descriptor which is data indicating the state and / or structure of the intermediate, A method for predicting characteristics, comprising the step (b) predicting the characteristics of the object based on the intermediate descriptor obtained in step (a).

14. The characteristic prediction method according to claim 13, wherein in step (b), the characteristics of the object are predicted from the intermediate descriptor using a first trained model that outputs the characteristics of the object based on the intermediate descriptor.

15. In the design and manufacturing process of an object, which is a final product having predetermined characteristics, starting from raw materials and progressing through intermediates, (a) A step of obtaining an intermediate descriptor which is data indicating the state and / or structure of the intermediate, A control program for causing a computer to perform a process including the step (b) of predicting the characteristics of the object based on the intermediate descriptor obtained in step (a).

Images