Composition search method and cured epoxy product

Abstract

To provide: a composition search method that enables efficient search for the composition of cured epoxy products; and cured epoxy products obtained using the composition search method.SOLUTION: A composition search method according to an embodiment includes: a first acquisition step of acquiring a first image, which is an image showing an internal structure of a first cured epoxy product that includes a matrix containing an epoxy main agent and an acid anhydride curing agent, and silica as a filler; a first-generation step of subjecting the first image to a binarization treatment to generate a first binarization image; a first calculation step of subjecting the first binarization image to a phase data analysis to calculate a first feature amount; and a construction step of constructing a model that predicts strength characteristics of the cured epoxy product from feature amounts based on images representing the internal structure of the cured epoxy product, by using machine learning based on a dataset that includes the first feature amount and the measured value of strength characteristics of the first cured epoxy product.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] An embodiment of the present invention relates to a composition exploration method and a cured epoxy product. [Background technology]

[0002] Searching for the composition of epoxy cured materials requires a lot of time, cost, and human resources. For example, in the case of silica (SiO2)-filled epoxy resin, the composition, such as the compounding conditions of the silica to be filled, is considered, a prototype is made, and then physical properties such as strength characteristics are obtained, and the results are reflected in the next prototype. Specific compounding conditions include the silica filling rate, selection of particle shape (spherical or crushed), combination ratio of different particle shapes, particle size, particle size distribution, etc. In many cases, the composition of these compounding conditions is determined by the intuition and experience of a skilled engineer, and then the process of making numerous prototypes and obtaining physical properties is repeated to search for the best compounding conditions. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-38344 Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present invention is to provide a composition searching method capable of efficiently searching for the composition of an epoxy cured product, and an epoxy cured product obtained by the composition searching method. [Means for solving the problem]

[0005] The composition searching method according to the embodiment includes a first acquisition step of acquiring a first image that represents an internal structure of a first epoxy cured material that includes a matrix containing an epoxy base agent and an acid anhydride curing agent, and silica as a filler; a first generation step of binarizing the first image to generate a first binary image; a first calculation step of calculating a first feature amount by performing topological data analysis on the first binary image; and a construction step of constructing a model that predicts strength characteristics of the epoxy cured material from the feature amount based on the image that represents the internal structure of the epoxy cured material by machine learning based on a dataset that includes the first feature amount and actual measured values ​​of strength characteristics of the first epoxy cured material. [Brief description of the drawings]

[0006] [Figure 1] FIG. 1 shows the hardware configuration of a composition search device. [Diagram 2] A diagram showing the procedure of a method for searching for the composition of an epoxy cured material in the phase of constructing a predictive model. [Diagram 3] FIG. 3 is a diagram showing a schematic diagram of the processing procedure shown in FIG. 2; [Figure 4] FIG. 3 is a diagram showing a specific example of data generated in steps SA2 to SA5 of FIG. 2; [Diagram 5] FIG. 1 is a diagram showing a procedure of a composition exploration method for an epoxy cured material in a first composition exploration phase. [Figure 6] FIG. 6 is a diagram showing a schematic diagram of the processing procedure shown in FIG. 5; [Figure 7] FIG. 6 is a diagram showing a specific example of data generated in steps SB3 and SB5 of FIG. 5; [Figure 8] FIG. 1 is a diagram showing a procedure of a composition exploration method for an epoxy cured material in a second composition exploration phase. [Figure 9] FIG. 9 is a diagram showing a schematic diagram of the processing procedure shown in FIG. 8; [Figure 10] FIG. 9 is a diagram showing a specific example of data generated in steps SC2 and SC4 of FIG. 8. [Figure 11] FIG. 1 is a diagram showing an example of a persistence diagram of a cured material having certain strength characteristics. [Figure 12]An example of a SEM image (binarized) with the area corresponding to the special generator superimposed. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] Hereinafter, the composition exploration method and the epoxy cured product according to the present embodiment will be described with reference to the drawings.

[0008] (Composition search method) The composition searching method according to the present embodiment relates to a method for searching for the composition of an epoxy cured material using a composition searching device.

[0009] Fig. 1 is a diagram showing the hardware configuration of a composition search device 100. As shown in Fig. 1, the composition search device 100 is a computer having a processor 1, a ROM (Read Only Memory) 2, a RAM (Random Access Memory) 3, an auxiliary storage device 4, an input device 5, a display device 6, and a communication device 7. Transmission and reception of data and various signals between the processor 1, ROM 2, RAM 3, the auxiliary storage device 4, the input device 5, the display device 6, and the communication device 7 is performed via a bus.

[0010] The processor 1 is an integrated circuit that controls the overall operation of the composition search device 100. For example, the processor 1 has a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor) and / or an FPU (Floating-Point Unit). The processor 1 may also have an internal memory and an I / O interface. The processor 1 executes various processes by interpreting and calculating programs stored in advance in the ROM 2 or the auxiliary storage device 4. The processor 1 may be realized in part or in whole by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0011] The processor 1 executes a composition exploration method for exploring the composition of an epoxy cured material. The epoxy cured material according to this embodiment includes a matrix (base material) containing an epoxy base agent and an acid anhydride curing agent, and silica (SiO2) as a filler. The epoxy cured material can be used in a wide variety of products and / or applications ranging from daily necessities to factory products. For example, the epoxy cured material can be used as a resin or adhesive for semiconductors, switchgear, etc. The composition exploration method is divided into a phase of constructing a prediction model (hereinafter, construction phase) and a phase of exploring the composition of the epoxy cured material using the prediction model (hereinafter, composition exploration phase).

[0012] In the construction phase, the processor 1 executes a first acquisition step, a first generation step, a first calculation step, and a construction step. In the first acquisition step, the processor 1 acquires a first image that is an image representing the internal structure of a first epoxy cured product. Here, the first epoxy cured product means an epoxy cured product to be subjected to machine learning. In the first generation step, the processor 1 performs binarization on the first image to generate a first binarized image. In the first calculation step, the processor 1 performs topological data analysis on the first binarized image to calculate a first feature amount. In the construction step, the processor 1 constructs a prediction model that predicts the strength characteristics of the epoxy cured product from the feature amount based on the image representing the internal structure of the epoxy cured product by machine learning based on a data set including the first feature amount and an actual measured value of the strength characteristics of the first epoxy cured product. In the first calculation step, the processor 1 performs topological data analysis on the first binarized image to calculate a first analysis value, and vectorizes the first analysis value to calculate the first feature amount. More specifically, in the first calculation step, the processor 1 performs persistent homology as a topological data analysis on the first binarized image to calculate a persistence map, and vectorizes the persistence map to calculate a first feature amount. The first persistence map is an example of a first analysis value.

[0013] In the composition search phase, the processor 1 executes a second acquisition step, a second generation step, a second calculation step, and a prediction step. In the second acquisition step, the processor 1 acquires a second image representing the internal structure of the second epoxy cured material. The second epoxy cured material means the epoxy cured material whose composition is to be searched for. In the second generation step, the processor 1 performs binarization on the second image to generate a second binarized image. In the second calculation step, the processor 1 performs topological data analysis on the second binarized image to calculate a second feature amount. In the prediction step, the processor 1 predicts the strength characteristics of the second epoxy cured material from the second feature amount using a prediction model. Here, in the second calculation step, the processor 1 performs topological data analysis on the second binarized image to calculate a second analysis value, and vectorizes the second analysis value to calculate the second feature amount. More specifically, in the second calculation step, the processor 1 performs persistent homology as a topological data analysis on the second binarized image to calculate a second persistence map, and vectorizes the second persistence map to calculate a second feature amount. The second persistence map is an example of a second analysis value.

[0014] The ROM 2 is a non-volatile memory that stores various data. For example, the ROM 2 stores data and setting values ​​used when the processor 1 executes various processes. The ROM 2 may include a non-transitory computer-readable storage medium that stores a program executed by the processor 1.

[0015] The RAM 3 is a volatile memory used for reading and writing data. The RAM 3 temporarily stores data used when the processor 1 executes various processes. The RAM 3 provides a work area for the processor 1.

[0016] The auxiliary storage device 4 is a non-volatile memory that stores various data. For example, the auxiliary storage device 4 stores data and setting values ​​used when the processor 1 executes various processes, data generated by various processes in the processor 1, etc. The auxiliary storage device 4 is configured with an HDD (Hard Disk Drive), an SSD (Solid State Drive), an integrated circuit storage device, etc. The auxiliary storage device 4 may have a non-transitory computer-readable storage medium that stores a program executed by the processor 1.

[0017] The input device 5 accepts various operation inputs from an operator. Examples of the input device 5 that can be used include a keyboard, a mouse, various switches, a touch pad, and a touch panel display. An electrical signal corresponding to the accepted operation input (hereinafter, an operation signal) is supplied to the processor 1.

[0018] The display device 6 displays various data under the control of the processor 1. A CRT (Cathode-Ray Tube) display, a liquid crystal display, an organic EL (Electro Luminescence) display, an LED (Light-Emitting Diode) display, a plasma display, or any other display may be appropriately used as the display device 6. The display device 6 may be a projector.

[0019] The communication device 7 includes a communication interface such as a network interface card (NIC) for performing data communication with various devices connected to the composition search device 100 via a network. An operation signal may be supplied from a computer connected via the communication device 7 or an input device provided in the computer, and various data may be displayed on a display device or the like provided in the computer connected via the communication device 7. However, for the sake of simplicity of the following description, unless otherwise specified, it is assumed that the source of the operation signal is the input device 5 and the display destination of the various data is the display device 6. The input device 5 can be replaced by a computer connected via the communication device 7 or an input device provided in the computer, and the display device 6 can be replaced by a display device or the like provided in the computer connected via the communication device 7.

[0020] The composition search device 100 does not need to include all of the processor 1, ROM 2, RAM 3, auxiliary storage device 4, input device 5, display device 6, and communication device 7. If necessary, some of the processor 1, ROM 2, RAM 3, auxiliary storage device 4, input device 5, display device 6, and communication device 7 may not be provided. The composition search device 100 may be provided with any additional hardware device useful for executing the process according to this embodiment. The composition search device 100 does not need to be physically composed of one computer, and may be composed of a computer system having multiple computers connected to each other so as to be able to communicate via wires or a network line. The allocation of the series of processes according to this embodiment to the multiple processors 1 implemented in each of the multiple computers can be arbitrarily set. All the processors 1 may execute all the processes in parallel, or a specific process may be assigned to one or some of the processors 1, and the series of processes according to this embodiment may be executed by the entire computer system.

[0021] Next, a method for searching for a composition of an epoxy cured material using the composition searching device 100 will be described in several examples. In the following description, the epoxy cured material may be simply referred to as a cured material.

[0022] <Example 1> As Example 1, a processing example of the construction phase of a predictive model will be described with reference to Figs. 2 to 4. Fig. 2 is a diagram showing a processing procedure of a composition searching method for an epoxy cured material in the construction phase of a predictive model. Fig. 3 is a diagram showing a schematic diagram of the processing procedure shown in Fig. 2. Fig. 4 is a diagram showing a specific example of data generated in steps SA2 to SA5 in Fig. 2.

[0023] First, a plurality of first cured products are prepared (step SA1). The plurality of first cured products are prepared so as to have different compositions and / or strength properties. The composition includes compounding conditions such as silica filling rate, selection of particle shape (spherical or crushed), combination compounding ratio of different particle shapes, particle size, particle size distribution, etc. The strength property is assumed to be bending strength. As shown in FIG. 4, two types of first cured products, that is, cured product A and cured product B, are taken as examples. The cured product A has a bending strength of 168 MPa, and the cured product B has a bending strength of 131 MPa. In this embodiment, for convenience of explanation, two types of first cured products are taken as examples, but it is desirable to prepare 10 or more types, preferably 100 or more types. The first cured products may be prepared by any device and any method.

[0024] When step SA1 is performed, the processor 1 acquires an image (hereinafter, internal image) showing the internal structure of each of the first cured products created in step SA1 (step SA2). The internal image is preferably an image (hereinafter, SEM image) generated by electronic scanning with a scanning electron microscope (SEM) for the sake of convenience of binarization performed in a later step. The observation magnification is preferably a magnification that makes it easy to observe the shape and particle size of silica in the epoxy cured product, and is preferably about 100 to 500 times as an example. The internal image may be a grayscale image that expresses black and white shades in a predetermined gradation, or a color image that expresses RGB elements of color in a predetermined bit, but as an example, it is assumed to be a grayscale image. The internal image is taken into the composition searching device 100 by any means from the scanning electron microscope. Hereinafter, the internal image related to the first cured product will be referred to as a first internal image.

[0025] After step SA2, processor 1 binarizes each of the first internal images acquired in step SA2 (step SA3). Specifically, processor 1 binarizes the pixel values ​​of the first internal images using a predetermined threshold value. The threshold value may be set to a pixel value that allows the silica portion to be distinguished from other portions. In FIG. 4, the silica portion is drawn in black, and the other portions are drawn in white. A first binary image is generated from the first internal images by binarization.

[0026] When step SA3 is performed, the processor 1 performs topological data analysis on the first binarized image generated in step S3 (step S4). A first analysis value is obtained by the topological data analysis. For the topological data analysis, a 0th-order or 1st-order persistent homology is used. In the following, an explanation will be given using an example of 1st-order persistent homology. Persistent homology is a method for calculating a feature quantity by tracing the generation and disappearance of holes (called genus in geometric terms) that appear in the process of time-series changes in a sequence of graphs (simplicial complexes). The time series of simplicial complexes is called filtration. For example, if the radius of each point in a given set of points is gradually increased, and when the circles overlap, the points are connected with edges, and when multiple circles overlap, a surface is formed, then circles appear over time in the graph that initially contained only points, and as time passes, a surface appears. The change in the graph obtained in this way is an example of filtration. In persistent homology, the birth and death times of each genus that appears in the filtration are calculated. In persistent homology, each genus is associated with multiple combinations of birth and death times (birth and death pairs). A birth and death pair is also called a generation.

[0027] As shown in Figure 4, a diagram based on plots of birth and annihilation pairs of each genus on a two-dimensional coordinate plane (hereinafter referred to as the coordinate plane) with the vertical axis defined as birth time and the horizontal axis defined as death time is called a persistence diagram. The value of each plot in the persistence diagram represents the number (frequency) of combinations corresponding to the plot. The values ​​may be calculated discretely for each bin of birth and annihilation times. Bins can be set to any width, including zero.

[0028] A method for generating a persistence diagram will be briefly described. First, the processor 1 performs persistent homology on the first binarized image to calculate pairs of generation and annihilation times (generation and annihilation pairs) for each genus, and plots the generation and annihilation pairs on a coordinate plane to generate an original persistence diagram. The processor 1 then sets bins of a predetermined width for the original persistence diagram, and calculates, for each bin, a count value such as the number of generation and annihilation pairs or an arbitrary weighting value for the number. The processor 1 then assigns the count value corresponding to the bin on the coordinate plane. This generates a first persistence diagram. For example, a persistent image method may be used as a method for generating a first persistence diagram from an original persistence diagram. The first persistence diagram is an example of a first analysis value.

[0029] When persistent homology is applied to a binary image of an epoxy cured material, the persistence diagram is considered to represent a feature that aggregates geometric information such as the size (particle size) and quantity (filling rate) of silica, the relationship between shapes, and the positional relationship of the distribution of silica in the cured material. In other words, the geometric information represents a part of the composition. For example, in the case of silica with a small particle size, genus disappears immediately just by increasing the radius. In other words, genus with a short time from generation to disappearance (survival time) can be interpreted as corresponding to silica with a small particle size. In addition, for spherical silica, since a given point is arranged at an equal distance from the center of the silica, genus is generated immediately when the radius of each point is increased, but it takes a long time to disappear. Conversely, for silica with a crushed shape, the distance from the center is different, so it takes a long time to generate, but it takes a short time to disappear. In this way, the persistence diagram represents the relationship between the generation and disappearance of genus and the geometric information of silica.

[0030] After step SA4, the processor 1 vectorizes each of the first persistence maps generated in step SA4 (step SA5). A first vector is obtained by vectorization. As a vectorization method, for example, a method of converting the persistence map into a one-dimensional numerical vector may be used. The vertical axis of the vector shown in FIG. 4 represents the value of the persistent image, and the horizontal axis represents the bin number.

[0031] When step SA5 is performed, the processor 1 creates a dataset by linking the first vector obtained in step SA5 with the strength characteristic of the first cured product created in step SA1 (step SA6). That is, the dataset is a combination of a numerical value representing the first strength characteristic of the first cured product and a second vector. A dataset is created for each first cured product. The dataset is used as learning data for machine learning for the prediction model. The numerical value of the first strength characteristic is used as ground truth data. The strength characteristic of the first cured product may be obtained by any method, such as a strength test on the first cured product.

[0032] When step SA6 is performed, the processor 1 executes machine learning based on the multiple data sets created in step SA6, inputs a vector based on an internal image of the cured product, and generates a prediction model that outputs the strength characteristics of the cured product (step SA7). Specifically, the processor 1 inputs a first vector from the data set into an unlearned prediction model to obtain a model output, calculates an error between the model output and the first strength characteristic, and updates the parameters of the prediction model so as to minimize the error. As the prediction model, any neural network having at least network layers such as an input layer, a hidden layer, and an output layer may be used. The network structure of the prediction model may be arbitrarily designed from a fully connected layer, a CNN, a pooling layer, a normalization layer, or other network layers. The parameter update is repeated until any update end condition is satisfied. The update end condition can be set arbitrarily, but may be set to, for example, that a predetermined number of data sets have been processed, that the accuracy has reached a predetermined value, or the like. When the update end condition is satisfied, the prediction model is completed.

[0033] After step SA7 is performed, the processor 1 outputs the prediction model generated in step SA7 (step SA8). The prediction model or parameters of the prediction model are stored in the auxiliary storage device 4 or the like.

[0034] This completes the process of the construction phase according to the first embodiment.

[0035] The above-mentioned Patent Document 1 provides a search method for predicting the properties of a cured product by vectorizing feature quantities related to the composition of an amine-cured epoxy resin, such as the molecular weight of each epoxy or amine, the compounding ratio, and the curing temperature, and creating a prediction model using the actual measured values ​​of the adhesive properties corresponding to each as training data. However, according to this method, the target of machine learning is limited to those with a clear material composition, such as a data set in which developers themselves determine the composition and produce prototypes and obtain the physical properties. Therefore, there is a risk of a decrease in prediction accuracy due to reasons such as a lack of data sets for machine learning and overlearning of the prediction model.

[0036] According to the first embodiment, a prediction model for predicting the strength properties of an epoxy cured material can be constructed based on a dataset including the feature amount obtained by topological data analysis of an internal image of the epoxy cured material and the strength properties of the epoxy cured material. As described above, the feature amount is obtained by topological data analysis, specifically, persistent homology, and therefore reflects the composition of the epoxy cured material. Since the strength properties of the epoxy cured material are strongly correlated with the internal structure, it is possible to construct a prediction model with relatively high accuracy. In addition, it is only necessary to know the strength properties of the epoxy cured material used in the dataset, and the composition may be unclear or unknown. Therefore, according to the present embodiment, it is possible to easily collect a dataset, and as a result, it is possible to improve the prediction accuracy of the prediction model.

[0037] <Example 2> As Example 2, a processing example of the first composition search phase will be described with reference to Figs. 5 to 7. Fig. 5 is a diagram showing a processing procedure of a composition search method for an epoxy cured material in the first composition search phase. Fig. 6 is a diagram showing a schematic processing procedure shown in Fig. 5. Fig. 7 is a diagram showing a specific example of data generated in steps SB3 and SB5 in Fig. 5. The persistence diagram shown in Fig. 7 represents a first-order persistence diagram.

[0038] As shown in Figs. 5 and 7, first, the composition of the second cured product is specified (step SB1). The second cured product means the cured product whose composition is to be searched. The composition may be specified by specifying the blending conditions such as diameter, particle size distribution, and packing rate. In the first step SB1, the composition does not need to be optimal. After step SB1 is performed, a very small test piece of the second cured product is created according to the composition specified in step SB1 (step SB2). The second cured product may be created by any device and any method.

[0039] Steps SB3 to SB6 may be performed in the same manner as SA2 to SA5, respectively. When step SB2 is performed, the processor 1 acquires a second internal image representing the internal structure of the second cured material created in step SB2 (step SB3). When step SB3 is performed, the processor 1 binarizes the internal image acquired in step SB3 (step SB4). A second binarized image is generated by the binarization. When step SB4 is performed, the processor 1 performs topological data analysis on the second binarized image generated in step SB4 (step SB5). As in step SA4, persistent homology is used for the topological data analysis. A second persistent diagram is generated by performing persistent homology. When step SB5 is performed, the processor 1 vectorizes the second persistent diagram generated in step SB5 (step SB6). A second vector is obtained by the vectorization. A one-dimensional numerical vector based on the persistent diagram is generated by the vectorization.

[0040] When step SB6 is performed, the processor 1 predicts the strength characteristic (second strength characteristic) of the second cured product created in step SB2 from the second vector obtained in step SB6 using the prediction model output in step SA8 (step SB7). Specifically, the processor 1 applies the second vector to the prediction model and outputs a predicted value of the second strength characteristic corresponding to the second vector.

[0041] When step SB7 is performed, the processor 1 judges whether the second strength characteristic obtained in step SB7 is the desired strength characteristic (step SB8). The judgment method is not particularly limited, but as an example, the processor 1 may judge whether the error between a desired value of the strength characteristic input in advance and a predicted value of the second strength characteristic is within an allowable range. As another example, the judgment may be made by an operator or the like. Specifically, the processor 1 first displays the predicted value of the second strength characteristic on the display device 6. The operator judges whether the predicted value is the desired strength characteristic, and inputs the judgment result, such as whether it is within the allowable range or not, via the input device 5. The processor 1 may judge whether the strength characteristic obtained in step SB7 is the desired strength characteristic according to the input judgment result.

[0042] If it is determined in step SB8 that the second strength characteristic is not the desired strength characteristic (step SB8: NO), steps SB1 to SB8 are repeated again for another second cured product. In this manner, steps SB1 to SB8 are repeated until the desired strength characteristic is obtained. Then, if it is determined in step SB8 that the second strength characteristic is the desired strength characteristic (step SB8: YES), the processor 1 determines the corresponding composition as the composition of the desired second cured product (step SB9).

[0043] According to the second embodiment, by using a prediction model, it is possible to obtain strength characteristics from a second feature amount such as a vector based on an internal image of a small amount of epoxy cured material created experimentally. This makes it possible to obtain the strength characteristics of a new cured material without creating a large-scale test piece. Also, according to the second embodiment, if a feature amount is obtained, the corresponding strength characteristics are known, so it is not necessary to determine the composition in detail for all test pieces. For example, it is possible to simply specify the composition in the early stage of FIG. 5 (a stage where the number of updates is relatively small), and then determine the composition in detail when the predicted value of the strength characteristics approaches the desired value. In this way, it is possible to efficiently search for the composition of an epoxy cured material.

[0044] <Example 3> As Example 3, a processing example of the second composition search phase will be described with reference to Figs. 8 to 10. Fig. 8 is a diagram showing a processing procedure of a composition search method for an epoxy cured material in the second composition search phase. Fig. 9 is a diagram showing a schematic processing procedure shown in Fig. 8. Fig. 10 is a diagram showing a specific example of data generated in steps SC2 and SC4 in Fig. 8. The persistence diagram shown in Fig. 10 represents a first-order persistence diagram.

[0045] As shown in Figs. 8 and 9, first, the composition of the second cured product is specified (step SC1). The second cured product means the cured product whose composition is to be searched for. The composition may be specified by specifying blending conditions such as diameter, particle size distribution, and filling rate. The composition may be specified by any method. For example, the user may manually input the composition via the input device 5, or the processor 1 may calculate the composition based on a predetermined algorithm. At least in the first step SC1, the composition does not need to be optimal. Also, in the second composition search phase, unlike the first composition search phase, it is not necessary to actually create the second cured product.

[0046] When step SC1 is performed, the processor 1 acquires a virtual internal image of the second cured product having the composition specified in step SC1 (step SC2). Specifically, the processor 1 generates an internal image (virtual internal image) of the cured product (second cured product) having the composition specified in step SC1 by predictive calculation based on the composition.

[0047] Steps SC3 to SC5 may be performed in the same manner as steps SB4 to SB6. After step SC2, the processor 1 binarizes the virtual internal image acquired in step SC2 (step SC3), performs topological data analysis on the second binarized image generated in step SC3 (step SC4), and vectorizes the second persistence diagram generated in step SC4 (step SC5). A second vector is obtained by the vectorization.

[0048] When step SC5 is performed, the processor 1 predicts the second strength property of the second cured product having the composition specified in step SC1 from the second vector obtained in step SC5 using the prediction model output in step SA8 (step SC6). Specifically, the processor 1 applies the second vector to the prediction model and outputs a predicted value of the second strength property corresponding to the second vector.

[0049] After step SC6, the processor 1 determines whether or not the second strength characteristic obtained in step SC6 is a desired strength characteristic (step SC7). The determination in step SC7 may be performed in the same manner as in step SB8.

[0050] If it is determined in step SC7 that the second strength characteristic is not the desired strength characteristic (step SC7: NO), steps SC1 to SC7 are repeated again for another composition of the second cured product. In this manner, steps SC1 to SC7 are repeated until the desired strength characteristic is obtained. Then, if it is determined in step SC7 that the second strength characteristic is the desired strength characteristic (step SC7: YES), the processor 1 determines the corresponding composition as the desired composition of the second cured product (step SC8).

[0051] According to the third embodiment, it is possible to predict the strength characteristics of an epoxy cured product from the second feature amount such as a vector based on a virtual internal image corresponding to a specified composition. This makes it possible to obtain the strength characteristics of a new cured product without actually creating an epoxy cured product. According to the third embodiment, if the feature amount is obtained, the corresponding strength characteristics can be known, so for example, in the initial stage (stage with a relatively small number of updates) of FIG. 8, it is possible to simply specify the composition, and at the stage where the predicted value of the strength characteristics approaches the desired value, it is also possible to determine the composition in detail. It is also possible to combine the second embodiment and the third embodiment. That is, it is also possible to mix an update time in which a test piece is created and an update time in which a test piece is not created in the composition search phase. For example, it is also possible to predict the strength characteristics from the second vector based on a virtual internal image without creating a test piece at the initial stage, and at the stage where the predicted value of the strength characteristics approaches the desired value, a test piece is created, and the strength characteristics are predicted from the second vector based on an internal image obtained from the test piece.

[0052] The internal image used in Example 3 is not limited to a virtual internal image, and may be an internal image obtained by photographing an epoxy cured product with an electron scanning microscope, etc. In other words, an internal image with an unknown composition may be applied to a prediction model to predict strength characteristics.

[0053] The processing according to the above-described embodiments is not limited to the above-described examples, and various configurations and / or processing can be deleted, added, and / or modified without departing from the spirit of the invention.

[0054] As an example, the prediction model according to the above embodiment is assumed to input a vector based on a persistence diagram. However, the prediction model according to the present embodiment may input a persistence diagram. In this case, the prediction model may input the persistence diagram as image data, or may input the persistence diagram as a sequence of multiple pixel values ​​corresponding to multiple pixels constituting the persistence diagram. In other words, the persistence diagram is also an example of a feature to be input to the prediction model.

[0055] (Epoxy cured product) The relationship between persistence diagrams and strength properties of various epoxy cured materials produced in the construction phase and / or composition exploration phase was analyzed.

[0056] Fig. 11 is a diagram showing an example of a persistence diagram of an epoxy cured material having certain strength characteristics. The persistence diagram shown in Fig. 11 shows a first-order persistence diagram. Region 11 shown in Fig. 11 includes the ranges of (generation time, disappearance time) of approximately (1.2 to 3.5, 2.0 to 5.0). It was found that region 11 is a region where a generator (hereinafter, special generator) characteristic of an epoxy cured material with a bending strength of 160 MPa or more occurs.

[0057] FIG. 12 is a diagram showing an example of an SEM image (binarized) on which an area 12 corresponding to a special generator is superimposed. The area 12 is drawn in gray in FIG. 12. The SEM image shown in FIG. 12 is the original SEM image from which the persistence diagram shown in FIG. 11 was generated, and is derived from the same cured product. The processor 1 estimated the area 12 in the SEM image from the persistence diagram shown in FIG. 11 using an inverse analysis technique. The special generator is seen in a structure in which small-sized silica is arranged near large-sized silica. It was confirmed that the blending and arrangement of large-sized silica and small-sized silica relieves stress, resulting in high bending strength.

[0058] That is, the values ​​in the range of (creation time, extinction time) = (1.2~3.5, 2.0~5.0) of the first-order persistence diagram are 10 -2 As described above, an epoxy cured product having a composition in which the bending strength is less than 1.2 has high bending strength. An epoxy cured product having such a composition can be obtained, for example, by the composition search method according to the present embodiment. The ranges of (generation time, disappearance time) do not need to be strictly (generation time, disappearance time) = (1.2 to 3.5, 2.0 to 5.0), and may be appropriately changed depending on the composition of the matrix and filler, etc.

[0059] Thus, according to the above embodiment, it is possible to provide a composition searching method capable of efficiently searching for the composition of an epoxy cured material, and an epoxy cured material obtained by the composition searching method.

[0060] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0061] 1...processor, 2...ROM, 3...RAM, 4...auxiliary storage device, 5...input device, 6...display device, 7...communication device, 100...composition search device.

Claims

1. A first acquisition step of acquiring a first image which is an image showing an internal structure of a first epoxy cured product including a matrix including an epoxy base material and an acid anhydride curing agent, and silica as a filler; a first generating step of generating a first binary image by binarizing the first image; a first calculation step of calculating a first feature amount by performing a topological data analysis on the first binarized image; a construction step of constructing a model for predicting strength characteristics of the epoxy cured material from features based on an image representing an internal structure of the epoxy cured material by machine learning based on a data set including the first features and actual measured values ​​of strength characteristics of the first epoxy cured material; A composition exploration method comprising:

2. a second acquisition step of acquiring a second image representing an internal structure of the second epoxy cured material; a second generating step of generating a second binarized image by binarizing the second image; a second calculation step of calculating a second feature amount by performing a topological data analysis on the second binarized image; and a prediction step of predicting a strength characteristic of the second epoxy cured material from the second feature amount by using the model. The composition searching method according to claim 1.

3. The composition exploration method according to claim 2 , wherein the second acquisition step acquires, as the second image, an electron scanning microscope image generated by electronically scanning the second epoxy cured product with an electron scanning microscope.

4. The composition exploration method according to claim 2 , wherein the second acquisition step acquires, as the second image, a virtual scanning electron microscope image generated by a predictive calculation based on a composition of the second epoxy cured material.

5. The first calculation step includes: performing the topological data analysis on the first binarized image to calculate an analysis value; vectorizing the analytic value to calculate the first feature amount; The composition searching method according to claim 1.

6. The first calculation step includes: applying persistent homology to the first binarized image as the topological data analysis to calculate a persistence diagram; vectorizing the persistence map to calculate the first feature amount; The composition searching method according to claim 5.

7. The composition exploration method according to claim 1 or 2, wherein the image is generated by electronically scanning the epoxy cured product with a scanning electron microscope.

8. 3. The composition exploration method according to claim 1, wherein the strength characteristic is bending strength.

9. An epoxy cured material comprising a matrix containing an epoxy base material and an acid anhydride curing agent, and silica as a filler, The primary persistence diagram based on the electron scanning microscope image of the epoxy cured product has a value of 10 in the range of (1.2 to 3.5, 2.0 to 5.0) (time of formation, time of disappearance). －２ having a composition less than Epoxy hardened material.

Images