An electronic device and its operating method that use artificial intelligence to provide predictive data on the physical properties of a composite material according to its composition ratio.

Abstract

According to one aspect of this disclosure, a method for providing predictive data on the physical properties of a composite according to its composition ratio using artificial intelligence may include the steps of: confirming data related to at least one composition constituting the composite; encoding the physical property data of the composite synthesized based on the data related to the at least one composition into first physical property data; calculating an embedding result corresponding to the first physical property data via a first model corresponding to an artificial intelligence learning model based on a neural network; and providing predictive data on the physical properties of the composite for second physical property data based on the calculated embedding result.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to an electronic device and an operating method thereof that provide physical property prediction data of a composite according to a composition ratio using artificial intelligence. More specifically, the present disclosure relates to providing data obtained by learning basic physical property experiment data of a composite and predicting the physical properties of the composite generated according to a composition ratio.

Background Art

[0002] A chemical composite has specific physical properties depending on the content of the compositions constituting it and the corresponding ratio. The formulation data of the compositions in the experimental unit of a chemical composite with specific chemical physical properties is limited. Considering the recent development of artificial intelligence technology, it is possible to develop a model learned by enhancing and synthesizing data using limited learning data.

[0003] By synthesizing data similar to the limitedly provided experimental data, increasing the similarity to the synthesized data, and having it learned by an artificial intelligence learning model, it can be used to update the artificial intelligence learning model for confirming the results that a user intends to obtain. As a result, problems of data learning according to the composition configuration of a chemical composite and generation of prediction data for the learned content have been raised.

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, such a conventional electronic device and an operating method thereof that provide physical property prediction data of a composite according to a composition ratio using artificial intelligence have a problem in that the physical properties of the composite cannot be accurately predicted based on the learned content.

[0005] The embodiments of this disclosure aim to solve various problems, including those described above, and provide an electronic device and its operating method that can provide data predicting the physical properties of a composite by sampling similar data synthesized based on limited training data and using it as input along with already given experimental data, utilizing a specific activation function. However, these problems are illustrative and do not limit the scope of this disclosure. [Means for solving the problem]

[0006] According to one aspect of this disclosure, a method for providing predictive data on the physical properties of a composite according to its composition ratio using artificial intelligence may include the steps of: confirming data related to at least one composition constituting the composite; encoding the physical property data of the composite synthesized based on the data related to the at least one composition into first physical property data; calculating an embedding result corresponding to the first physical property data via a first model corresponding to an artificial intelligence learning model based on a neural network; and providing predictive data on the physical properties of the composite for second physical property data based on the calculated embedding result.

[0007] According to this embodiment, the step of confirming data related to the at least one composition includes the steps of confirming experimental data and synthetic data for the complex, and sampling data related to the at least one composition based on the experimental data and synthetic data, wherein the data related to the at least one composition includes the type, content, and harmonization data of the at least one composition constituting the complex.

[0008] According to this embodiment, a method for providing composite property prediction data according to composition ratio using artificial intelligence further includes the step of confirming data related to the sampled at least one composition as an input dataset, wherein the input dataset has been preprocessed as a training input dataset for the first model, and the synthesized data is synthesized from the experimental data through data augmentation.

[0009] According to this embodiment, the step of encoding the first physical property data may include the steps of setting the physical property information of the composite to be predicted via the first model, and performing batch normalization by representing data related to the composition of the composite in individual features to confirm the set physical property information.

[0010] According to this embodiment, the step of calculating the embedding result corresponding to the first physical property data may include the step of generating masked features through masking of individual features of the first physical property data.

[0011] According to this embodiment, a method for providing composite property prediction data according to composition ratio using artificial intelligence may include the steps of: calculating a scale term based on the masked features to determine the importance of the individual features; confirming the decision output of the individual domains via an activation function layer based on the importance; and calculating the final decision output of the individual domains via a fully connected layer as an embedding result corresponding to the first property data.

[0012] According to this embodiment, the step of confirming the decision output of the individual domains includes the step of confirming the first decision output of the first domain among the individual domains, and the step of confirming the second decision output of the second domain among the individual domains based on the first decision output, wherein the individual domains, including the first domain and the second domain, are configured as the fully connected layer.

[0013] According to this embodiment, the step of providing predictive data for the properties of the composite may include the step of providing second property data that differs in at least part from first property data learned through the first model.

[0014] According to this embodiment, the step of providing predictive data for the physical properties of the composite material may include the step of confirming the ground truth for the second physical property data and the step of evaluating the accuracy of the second physical property data based on the ground truth.

[0015] According to this embodiment, an electronic device that provides predictive data on the physical properties of a composite according to its composition ratio using artificial intelligence includes a communication unit, a memory, and at least one processor electrically connected to the communication unit and the memory. The at least one processor can be configured to verify experimental data for compositional harmony of a composite having specific physical properties based on data related to at least one composition constituting the composite, encode the physical property data of the composite synthesized based on the data related to the at least one composition into first physical property data, calculate an embedding result corresponding to the first physical property data via a first model corresponding to an artificial intelligence learning model based on a neural network, and provide predictive data on the physical properties of the composite for second physical property data based on the calculated embedding result.

[0016] Other aspects, features, and advantages not mentioned above will become apparent from the specific details, claims, and drawings for carrying out the invention described below.

[0017] Furthermore, these general and specific aspects can be implemented using systems, computer programs, or any combination of systems and computer programs. [Effects of the Invention]

[0018] According to the exemplary embodiments of this disclosure configured as described above, an artificial intelligence learning model can be trained via multiple activation functions from a limited training dataset to predict the physical properties of a composite more accurately, and predictive data for specific properties desired by the user can be generated based on performance evaluation results. Of course, such effects do not limit the scope of this disclosure. [Brief explanation of the drawing]

[0019] [Figure 1] This is a schematic block diagram showing the internal components of an electronic device that uses artificial intelligence according to an exemplary embodiment of the present disclosure to provide predictive data on the physical properties of a composite according to its composition ratio. [Figure 2] This is a schematic block diagram illustrating a system that provides predictive data on the physical properties of a composite according to the composition ratios of exemplary embodiments of this disclosure. [Figure 3] This is a schematic diagram illustrating a method for providing predictive data on the physical properties of a composite according to the composition ratios of exemplary embodiments of the present disclosure. [Figure 4] This flowchart schematically illustrates a method for providing predictive data on the physical properties of a composite according to the composition ratios of exemplary embodiments of this disclosure. [Figure 5] This flowchart schematically illustrates the process for verifying data related to the composite composition according to the exemplary embodiments of this disclosure. [Figure 6]A block diagram schematically showing a physical property prediction data encoding and embedding process of a composite according to a composition ratio according to an exemplary embodiment of the present disclosure. [Figure 7] An exemplary diagram schematically showing a physical property prediction data encoding process of a composite according to a composition ratio according to an exemplary embodiment of the present disclosure. [Figure 8] A flowchart schematically showing a physical property prediction data embedding process of a composite according to a composition ratio according to an exemplary embodiment of the present disclosure. [Figures 9a-9b] An exemplary diagram of physical property prediction data of a composite according to a composition ratio corresponding to an activation function according to an exemplary embodiment of the present disclosure. [Figure 10] An exemplary diagram schematically showing a performance evaluation result screen of physical property prediction data of a composite according to a composition ratio according to an exemplary embodiment of the present disclosure.

Mode for Carrying Out the Invention

[0020] The present disclosure can be subjected to various conversions and can have various embodiments. Therefore, specific embodiments are shown in the drawings and will be described in detail in the detailed description. The effects and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described in detail later together with the drawings. However, the present disclosure is not limited to the embodiments disclosed below and can be implemented in various forms.

[0021] In the following embodiments, terms such as first, second, etc. are not used in a limiting sense and are used for the purpose of distinguishing one component from another component.

[0022] In the following embodiments, singular expressions include plural expressions unless the context clearly indicates otherwise.

[0023] In the following embodiments, terms such as “includes” or “has” mean that the features or components described in the specification are present, and do not preclude the possibility that one or more other features or components may be added.

[0024] In the following embodiments, when a part such as a layer, region, or component is located above or above another part, this includes not only cases where it is directly above another part, but also cases where another region, component, etc. is interposed between them.

[0025] In the drawings, the size of components may be exaggerated or reduced for illustrative purposes. For example, the size and thickness of each component shown in the drawings are arbitrarily shown for illustrative purposes and are not necessarily limited to those shown in this disclosure.

[0026] Where other embodiments are possible, a particular sequence of operations may be performed in a different order than that described. For example, two steps described consecutively may be performed substantially simultaneously, or in the reverse order of the description.

[0027] In this specification, "A and / or B" means that it is either A, B, or A and B. And "at least one of A and B" means that it is either A, B, or A and B.

[0028] In the following embodiments, when layers, regions, components, etc. are connected, this includes cases where layers, regions, components are directly connected, and / or indirectly connected with other layers, regions, components interposed between them. For example, in this specification, when layers, regions, components, etc. are said to be electrically connected, this refers to cases where layers, regions, components, etc. are directly electrically connected, and / or indirectly electrically connected with other layers, regions, components, etc. interposed between them.

[0029] The x, y, and z axes are not limited to the three axes on a Cartesian coordinate system, but can be interpreted in a broader sense that includes them. For example, the x, y, and z axes may be orthogonal to each other, or they may point in different directions that are not orthogonal to each other.

[0030] The advantages and features of this disclosure, as well as the methods for achieving them, will become apparent by referring to the embodiments described below in detail with the accompanying drawings. However, this disclosure is not limited to the embodiments disclosed below and can be implemented in various forms, and these embodiments are provided merely to complete the disclosure and to fully inform a person of the ordinary skill in the art to which this disclosure belongs of its scope, which is defined only by the claims.

[0031] The terms used in this disclosure are for illustrative purposes only and are not intended to limit the disclosure. In this disclosure, the singular form may also include the plural form unless otherwise specified in the context. The terms “comprises” and / or “comprising” used in this disclosure do not exclude the presence or addition of one or more other components in addition to the components mentioned. The same drawing reference numerals throughout the disclosure refer to the same components, and “and / or” may include each of the components mentioned and all combinations of one or more of them. Although terms such as “first,” “second,” etc., are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are used simply to distinguish one component from another. Thus, it goes without saying that the first component mentioned below may be the second component within the technical concept of this disclosure.

[0032] The term “exemplary” is used in this disclosure to mean “exemplary or used exemplary.” Any embodiment described as “exemplary” in this disclosure should not be construed as necessarily preferable or superior to other embodiments.

[0033] Embodiments of the present disclosure can be described in terms of functions or blocks that perform functions. Blocks referred to as “parts” or “modules” in the present disclosure may be physically composed of analog or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memories, passive electronic components, active electronic components, optical components, and hardwired circuits, and may be selectively driven by firmware and software. Furthermore, the term “part” as used in the present disclosure means software, FPGAs, or hardware elements such as ASICs, and a “part” can play any role. However, “part” is not limited to software or hardware. A “part” may be configured to reside in an addressable storage medium and may be configured to regenerate one or more processors. Thus, as an example, a “part” may include elements such as software elements, object-oriented software elements, class elements, and task elements, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within elements and "parts" can be combined with fewer elements and "parts," or further separated into additional elements and "parts."

[0034] Embodiments of the present disclosure can be configured using at least one software program executed by at least one hardware device that can perform network management functions to control elements.

[0035] Spatially relative terms such as "below," "beneath," "lower," "above," and "upper" can be used to easily describe the correlation between one component and another, as shown in the diagram. Spatially relative terms should be understood as terms that include different orientations of the components in use or operation, in addition to the orientation shown in the diagram. For example, if the components shown in the diagram are turned over, a component described as "below" or "beneath" of another component may be positioned "above" of the other component. Thus, the exemplary term "below" can include both downward and upward directions. Components can also be oriented in other directions, and therefore spatially relative terms can be interpreted according to their orientation.

[0036] Unless otherwise defined, all terms used in this disclosure (including technical and scientific terms) should be used in a way that is commonly understood by those skilled in the art in which this disclosure pertains. Furthermore, terms defined in commonly used dictionaries should not be interpreted ideally or excessively unless specifically defined otherwise.

[0037] Embodiments of the present disclosure will be described in detail below with reference to the attached drawings. When referring to the drawings, identical or corresponding components will be given the same reference numerals, and redundant descriptions thereof will be omitted.

[0038] Figure 1 is a schematic block diagram showing the internal components of an electronic device that provides predictive data on the physical properties of a composite according to its composition ratio using artificial intelligence according to an exemplary embodiment of the present disclosure.

[0039] The electronic device 100 (for example, an electronic device that provides predictive data on the physical properties of a composite) may include a processor 110, a communication unit 120, a memory 130, and the like. The internal components that the electronic device 100 may include are not limited to these. The electronic device 100 of this disclosure may perform the functions of the processor 110 via a separate processing server or cloud server instead of the processor 110.

[0040] An electronic device 100 according to an embodiment may be a type of server, a central processing unit, an application program server, etc. The server may include a device that provides data to other devices connected to a network via an application or the web. For example, other devices may include devices such as desktops, laptops, tablets, and mobile terminals. In another example, the electronic device 100 may be a device that incorporates an electronic device with specifications capable of performing the operations of the Disclosure.

[0041] Referring to Figure 1, the processor 110 may be configured to store data for an algorithm or a program that reproduces an algorithm for controlling the operation of components within the electronic device 100, and to perform data provision operations that predict the physical properties of the composite according to the composition ratio using the data stored in memory 130. In this case, the processor 110 and memory 130 may each be configured on separate chips, or the processor 110 and memory 130 may be configured on a single chip.

[0042] The processor 110 can control one or more of the components discussed above in order to realize the various embodiments of this disclosure shown in Figures 2 to 10 below in the electronic device 100.

[0043] The communication unit 120 according to this embodiment may include one or more components that enable communication with an external device, for example, at least one of a broadcast receiving module, a wired communication module, a wireless communication module, a short-range communication module, and a location information module.

[0044] An input / output interface (not shown) according to an embodiment serves as a passage to various types of external devices connected to the electronic device 100 of the present disclosure. Such an input / output interface may include at least one of the following: a wired / wireless headset port, an external charger port, a wired / wireless data port, a memory card port, a port for connecting a device with an identification module (SIM), an audio I / O (input / output) port, a video I / O (input / output) port, or an earphone port. The electronic device 100 of the present disclosure can perform appropriate control related to the external device connected to the input / output interface.

[0045] The memory 130 in this embodiment can store data that supports various functions of the electronic device 100, programs for the operation of the processor 110, and input / output data (e.g., images, videos, etc.). The memory 130 can store multiple application programs (applications) driven by the electronic device 100, data for the operation of the electronic device 100, and instructions. At least some of these application programs can be downloaded from an external server via wireless communication.

[0046] Such memory 130 may include at least one type of storage medium from among flash memory type, hard disk type, SSD type (Solid State Disk type), SDD type (Silicon Disk Drive type), multimedia card micro type, card type memory (e.g., SD or XD memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. Furthermore, although memory 130 is separate from the electronic device 100, it may be a database connected by wire or wireless.

[0047] At least one component can be added or removed in accordance with the performance of the components shown in Figure 1. Furthermore, it will be readily apparent to those with ordinary skill in the art that the relative positions of the components can be changed in accordance with the performance or structure of the device.

[0048] On the other hand, the disclosed embodiments can be implemented in the form of a recording medium that stores computer-executable instructions. The instructions may be stored in the form of program code, which, when executed by a processor, can generate a program module to perform the operations of the disclosed embodiments. The recording medium can be implemented as a computer-readable recording medium.

[0049] Computer-readable storage media include all types of storage media that store instructions that can be deciphered by a computer. Examples include ROM (Read Only Memory), RAM (Random Access Memory), magnetic tape, magnetic disks, flash memory, and optical data storage devices.

[0050] Figure 2 is a schematic block diagram illustrating a system that provides predictive data for the physical properties of a composite according to the composition ratios of an exemplary embodiment of the present disclosure.

[0051] Referring to Figure 2, the system that provides data predicting the physical properties of a composite according to its composition ratio using artificial intelligence (hereinafter referred to as "the system") may include an electronic device 100 and an external device 200. For example, the electronic device 100 and the external device 200 may be connected to a network via a communication unit (for example, the communication unit 120 in Figure 1).

[0052] The system according to the embodiment may provide predicted data about the physical properties of a composite composed according to the ratio of its components in the electronic device 100. For example, the electronic device 100 can augment its data by data synthesis via physical property data of composites composed according to compositional harmony, which is pre-stored in memory (e.g., memory 130 in Figure 1) or received from an external device 200. Here, even the same composite can have various physical properties depending on the ratio of the components that make up the composite. That is, the physical properties of a composite may differ depending on the compositional harmony. Specifically, the electronic device 100 can pre-process the physical property data of the composite and perform learning using artificial intelligence. More specifically, the electronic device 100 can augment input data, such as experimental data (e.g., physical property data), through the trained model. This makes it possible for the system to provide predicted data for the physical properties of composites according to compositional harmony that correspond to the results that the user wishes to analyze, as well as data on the accuracy of the predicted data.

[0053] Figure 3 is a schematic diagram illustrating a method for providing predictive data on the physical properties of a composite according to the composition ratios of exemplary embodiments of this disclosure.

[0054] Referring to Figure 3, an electronic device (e.g., electronic device 100 in Figure 1) can perform the processes of learning a model and augmenting data to provide predictive data on the physical properties of a composite composed of elements with their respective content and ratios according to compositional harmony. Specifically, the electronic device can infer changes in the physical properties of a composite due to compositional harmony. More specifically, by learning the physical property data of composites with known compositional harmony, the electronic device can predict the physical properties of composites with unlearned compositional harmony.

[0055] According to the embodiment, a processor (e.g., processor 110 in Figure 1) can receive data related to at least one composition constituting the complex from an external device (e.g., external device 200 in Figure 2) via a communication unit (e.g., communication unit 120 in Figure 1). This can correspond to the electronic device collecting (Data Collect) (310). The data related to at least one composition may be experimental data and / or synthetic data. As an example, the processor can receive and verify experimental data, which is measured experimentally to obtain material properties of the complex through compositional harmony. In this case, the experimental data may be included in the data related to at least one composition. As another example, the processor can verify synthetic data, which is synthesized by augmenting experimental data. In this case, the synthetic data may be included in the data related to at least one composition. Furthermore, the synthetic data may be data synthesized by the processor through learning based on experimental data, similar to the experimental data. That is, the processor can augment synthetic data by synthesizing experimental data that is restrictively given for more accurate prediction of the complex's material properties, and can learn the experimental and synthetic data together and utilize them for predicting the complex's material properties.

[0056] According to one embodiment, the processor can select data in which the properties of the composite to be predicted and the types of compositions constituting the composite are similar. The processor can then perform a preprocessing process on the selected data. The preprocessing process may be a process of processing the experimental data to facilitate training of an artificial intelligence learning model (e.g., a property prediction model), thereby selecting data in which the type of composition and the properties are the same from a form in which the harmonic information and property information of the compositions are contained in separate tables.

[0057] Experimental data in an external device or memory may have a unique ID (ID) for each complex, set to distinguish each piece of harmonic information. The processor can verify discrete data for the composition and its content in the experimental data and obtain pre-processed data in the first column and first row. The processor can convert each composition into the first column and its content into the first row, based on the unique ID of the complex. For example, the first column may consist of PPO, HIPS, CNT, Clay, CF, GF, Antioxidant, and Coupling agent. The first row may be the corresponding values ​​in the first column that correspond to the unique ID.

[0058] The processor can examine the discrete data for the physical properties and property values ​​of the experimental data and obtain pre-processed data in the second column and second row. For the data in the property table, the processor can convert the property values ​​to the second row, with the physical properties of the composition in the second column, based on the unique ID. For example, the second column may consist of bending strength, bending modulus, tensile strength, impact strength, and thermal distortion temperature. The second row may be the corresponding values ​​in the second column that correspond to the unique ID.

[0059] Subsequently, the processor can combine the first and second columns into a third column based on the unique ID of the complex, and combine the first and second rows into a third row. The data thus preprocessed can then be sorted by the unique ID of the complex that the user is looking for. That is, the processor can transform the form of the data frame by merging the first and second columns and the first and second rows based on the unique ID of the complex. The processor can perform training on the data related to the compositions obtained by performing the preprocessing process, or it can synthesize and augment the data via a Tabular Data Generative Model (TDGM). Here, the TDGM corresponds to a model that samples according to the distribution map of the experimental data corresponding to the original data and generates synthetic data with a distribution map similar to the original data, and can be trained using Wasserstein Distance. The TDGM can be configured to use a Tabular GAN or a Variational Autoencoder-based model.

[0060] The processor according to the embodiment can, after selecting the physical properties of the composite to be predicted after learning via an artificial intelligence learning model, execute a data representation (320) process that defines a data representation method. For example, when the processor predicts a physical property of a composite such as bending strength, it can classify data on the composition and the content of that composition based on the compositional harmony of the composite into input data features, which are the input dataset for the artificial intelligence learning model. In this case, the features corresponding to the input dataset of the artificial intelligence learning model may be encoded through a feature transformer to perform a regression or classification process of the physical property to be predicted. The encoded results are generated as a decision output via an activation function layer, and the decision outputs of individual domains can be derived through an FC layer (Fully Connected layer) to finally predict the physical property.

[0061] The processor according to the embodiment can perform a Knowledge Transfer (330) process by assigning weights to individual features via an artificial intelligence learning model. The Knowledge Transfer process may be a process of configuring individual domains as fully connected layers via specific activation function layers to extract prediction results for the properties of the complex. The processor can then evaluate the accuracy of the artificial intelligence learning model's performance in predicting the properties of the complex by comparing the results of the predictions with Ground Truth. This can be called the Evaluation (340) process.

[0062] Figure 4 is a schematic flowchart illustrating a method for providing predictive data on the physical properties of a composite according to the composition ratios of exemplary embodiments of this disclosure.

[0063] Referring to Figure 4, the processor (for example, processor 110 in Figure 1) can provide specific physical properties of a composite that the user wishes to understand based on the results learned using an artificial intelligence learning model. That is, the processor can provide the user with predicted physical properties corresponding to the compositional ratio of a particular composite via a physical property prediction model.

[0064] In step S100, the processor can examine data related to at least one composition constituting the composite. The data related to at least one composition may be experimental data and / or synthetic data. The synthetic data may be data synthesized by the processor by learning based on the experimental data, similar to the experimental data. That is, the processor can synthesize synthetic data by combining experimental data that is constrained for more accurate prediction of the composite's properties, and can learn the experimental and synthetic data together to utilize in predicting the composite's properties.

[0065] According to the embodiment, data extraction for constructing experimental data may be performed through a method of extracting composition harmony-related paragraphs from text within a document and a method of extracting composition harmony-related data from a table. In this case, the text extraction model can, in the case of a PDF document, convert the text within the PDF document into an mmd file using an OCR-based model (e.g., Nougat) so that Python can recognize it, and then extract the text. The extracted text can be classified into classes related to composition harmony within the input document using an object name recognition model (e.g., MaterialsBERT). In this case, if there is text containing composition content or physical property information of composition harmony, the relevant paragraph can be stored. In addition, experimental data can be extracted by adaptively adjusting the size of the bounding box when recognizing the table structure and by sharing attention modules to cooperate related tasks. The processor can then check the experimental data thus extracted from a database in memory (e.g., memory 130 in Figure 1) or an external device (e.g., external device 200 in Figure 2).

[0066] In step S200, the processor can encode the physical property data of the composite, synthesized based on data related to at least one composition, into first physical property data. At this time, the processor can set the physical property information of the composite that it intends to predict via the artificial intelligence learning model corresponding to the first model. That is, the process of setting the physical property information can correspond to the process of encoding it into first physical property data corresponding to individual features. The processor can then perform batch normalization by representing the data related to the composition of the composite as individual features to verify the set physical property information.

[0067] In step S300, the processor can calculate the embedding results corresponding to the first physical property data via the first model. The processor can perform masking on individual features of the first physical property data and generate masked features corresponding to the masked results.

[0068] In step S400, the processor can provide composite property prediction data for the second property data based on the calculated embedding results. The processor can provide results predicting the composite property according to composition ratios that have not been previously learned by learning from previously known data (e.g., experimental data) and / or data similar to previously known data (e.g., synthetic data) via an artificial intelligence learning model. The processor can also additionally perform an accuracy evaluation of the artificial intelligence learning model used for the property prediction results. This allows the processor to update the model in order to improve its accuracy. In other words, the processor can retrain and update the artificial intelligence learning model with data that is closer to the correct answer in order to improve its accuracy.

[0069] Figure 5 is a flowchart illustrating the process of verifying data related to the composite composition according to exemplary embodiments of this disclosure.

[0070] In step S110, the processor (for example, processor 110 in Figure 1) can verify data related to the compositions constituting the complex. The data related to the compositions constituting the complex may include experimental data and synthetic data. The data related to the compositions constituting the complex may also include data on the type, content, and harmony of at least one composition.

[0071] In step S120, the processor can perform a sampling process based on experimental and synthetic data. Here, sampling may be a process in which the processor constructs an input dataset for an artificial intelligence learning model based on experimental and synthetic data. That is, the process of extracting and preprocessing tabular data to construct an input dataset for an artificial intelligence learning model can be called the sampling process.

[0072] In step S130, the processor can generate an input dataset for the artificial intelligence learning model. The artificial intelligence learning model can utilize individual features of the composite's physical properties or data encoded from these individual features as its input dataset after a sampling process. The processor can use individual features of the composite's physical properties as the input dataset and perform learning using the predicted results of the composite's physical properties according to the composition ratio of the composite as the output dataset.

[0073] Figure 6 is a schematic block diagram illustrating the data coding and embedding process for predicting the physical properties of a composite according to composition ratios, according to an exemplary embodiment of the present disclosure.

[0074] The artificial intelligence learning model according to the embodiment may be the first model. For example, the artificial intelligence learning model may be a model for predicting the physical properties of a complex and may be a deep learning-based network designed for learning tabular data (e.g., tabular data). In this case, the first model may have an autoencoder structure consisting of an encoder and a decoder that reflect the variable selection features of a tree-based model. The first model can learn, perform classification and regression, and be input with tabular data containing composition harmony data and physical property data without any separate preprocessing.

[0075] The first model according to the embodiment may be configured to use a method that assigns weights to individual features rather than selecting only specific features. That is, the first model can learn by masking features in the input tabular data and going through steps for various domains. In this case, the importance of features can be determined for each step in each domain, masking can be performed, and only the masked features can be selected for learning, thereby improving prediction performance. The input features are encoded via a feature transformer. The processor (for example, processor 110 in Figure 1) can check the decision output of each domain via an activation function layer based on importance and calculate the final decision output as an embedded result.

[0076] The processor according to this embodiment can perform a batch normalization process 610. The processor can perform batch normalization on the features of the physical property data input to the artificial intelligence learning model, and then perform encoding on individual features.

[0077] The processor according to the embodiment can generate decision outputs at steps corresponding to individual domains by passing the encoded results through an activation function layer. This can be represented by 620 and 630 in Figure 6. For example, the processor can verify a first decision output in a first domain of the individual domains. In another example, the processor can verify a second decision output in a second domain of the individual domains based on the first decision output. In yet another example, the processor can extract second physical property data from a fully connected layer that connects the output terminals of the individual domains, including the first and second domains (640). This can be represented by a process in which the processor sums the results of the decision outputs at individual steps of the individual domains to generate an overall decision embedding, and then passes this embedding through a fully connected layer to finally calculate the classification or regression result.

[0078] The processor according to this embodiment can utilize an attention-based transformer layer as a mechanism for calculating the weight of the output sequence for each position in the input sequence. The attentive transformer layer can help the artificial intelligence learning model grasp the context and focus on relevant information by assigning importance to various positions in the input sequence. Such an attention mechanism can help model the relationship between the input sequence and the output sequence.

[0079] Subsequently, the processor can generate a trainable mask via a mask layer. A trainable mask can refer to a parameterized mask containing weights that are updated during the training of an artificial intelligence learning model. Such a mask can be adjusted during the training of the AI ​​learning model and can be used to adjust the importance of the model to specific inputs or outputs according to a particular purpose. For example, a mask used in an attention mechanism allows the AI ​​learning model to decide how much attention to a particular input. The mask is adjusted during the model's training and can help the model become more adaptable to the training data. In another example, a trainable mask can be used to dynamically adjust the weights to specific layers or features. The mask contains information about which features to use. Feature selection can be performed by multiplying the mask and features. Feature selection can be performed using masked features generated by multiplying the features from the previous domain step by the mask. In this process, the processor can calculate the importance of features through a learnable mask, generate masked features by applying them to the features of the previous domain, and then calculate a prior scale term to apply to the mask of the next domain. The prior scale term can indicate how much individual features were used in the previous domain and can be designed to reflect the variables selected in the previous domain with less weight.

[0080] For example, a processor can calculate the importance of individual features by masking the input features of an artificial intelligence learning model. This allows the processor to select important features to use when proceeding with regression or classification. This can correspond to a feature selection process. Also, if an artificial intelligence learning model has three input features, the processor can select important features by applying a mask such as [0.1,0.2,0.7] to the three input features using an attentive transformer and proceeding with learning. Representing data with masked features by masking the features makes it easier to select important features and transfer knowledge.

[0081] Figure 7 is a schematic diagram illustrating the data coding process for predicting the physical properties of a composite according to the composition ratios, according to an exemplary embodiment of the present disclosure.

[0082] In step S210, the processor (for example, processor 110 in Figure 1) can set up physical property information of the composite to be predicted via the first model. The processor sets up physical property information to generate an input dataset for the first model based on data related to the composition.

[0083] In step S220, the processor can perform batch normalization by representing the data related to the composition for verifying the set physical property information as individual features. The result of this batch normalization can be verified by the processor by generating an input dataset of data related to the sampled composition, corresponding to the encoded result. The input dataset can be characterized by being preprocessed into the input dataset for training the first model.

[0084] Figure 8 is a schematic flowchart illustrating the process of embedding composite property prediction data according to composition ratios in an exemplary embodiment of this disclosure.

[0085] In step S310, the processor (for example, processor 110 in Figure 1) can generate masked features. The processor generates masked features by masking individual features of the first physical property data. The processor can perform training by applying a mask to each individual feature that sums to 1.

[0086] In step S320, the processor can determine the importance of individual features. The processor determines importance by calculating a scale term based on the masked features. That is, by calculating a plier scale term from the masked features, the processor can determine the importance of individual features.

[0087] In step S330, the processor can check the decision output of individual domains. The processor checks the decision output of individual domains via an activation function layer based on importance. An activation function can refer to a function that determines the output of a neuron in an artificial intelligence network. Each neuron can generate an output value by multiplying its input value by weights and applying a bias, then transmitting this result to the activation function. This output value can be passed to neurons in the next layer or to the final output, enabling model prediction and learning. Activation functions primarily introduce nonlinearity, which can help artificial intelligence learning models learn complex functions.

[0088] In step S340, the processor can calculate the embedding results via the fully connected layer. The final decision output of the individual domains can be derived as the embedding results corresponding to the first physical property data. That is, the processor can provide second physical property data from the fully connected layer to which the individual domains are connected, which differs from the first physical property data in at least part.

[0089] Figures 9a and 9b are illustrative diagrams of predicted physical properties of composites according to the composition ratio corresponding to the activation function in the exemplary embodiments of this disclosure.

[0090] The processor (for example, processor 110 in Figure 1) can utilize the Swish function 920 and / or the Mish function 930 as activation functions instead of the ReLU (Rectified Linear Unit) function 910. The ReLU function is a nonlinear function that outputs the input value as is when the input value is positive, and outputs zero when the input value is negative. While it can be computationally efficient and allow for faster learning, it has the drawback that the gradient becomes zero when the input value is negative, potentially leading to the loss of neurons.

[0091] The Swish function according to this embodiment can be expressed as shown in Equation 1 below.

[0092] (Math 1) Swish(x) = x·σ(βx)

[0093] In the above equation 1, σ is the sigmoid function, and β is a new hyperparameter of the Swish function. The Swish function has a form similar to the ReLU function, but it has the characteristic of changing into a smoother form.

[0094] The Mish function according to this embodiment can be expressed as shown in Equation 2 below.

[0095] (Math 2) Mish(x) = x·tanh(softplus(x))

[0096] In the above equation 2, softplus(x) is log(1+e x The Mish function represents this, and it can have a nonlinear and smooth form.

[0097] Referring to Figure 9a, the processor can extract more accurate output values ​​from negative input values ​​by using the Swish function and / or Mish function as activation functions instead of the ReLU function in the first model. The processor can use all of the first models that utilize each of the Swish function and Mish function layers to advance the prediction of the composite's physical properties and perform knowledge transfer with the model that shows higher accuracy. In other words, after setting the activation function for the first model to a two-track layer and training it, the processor can select the most accurate trained model from among the physical properties predicted as the second model trained through the Swish function or the third model trained through the Mish function and perform knowledge transfer.

[0098] The processor according to this embodiment can improve the accuracy of predicting the properties of composites by replacing the ReLU function with the Swish function and / or the Mish function. Referring to the example figure 940 in Figure 9b, we can see the property prediction data (e.g., second property data) that utilizes the harmonic data of mPPO. The example figure 940 in Figure 9b corresponds to the results of training a property prediction model (e.g., the first model) using a total of 1057 data points using 57 harmonic data points and 1000 synthetic harmonic data points, and making predictions on 25 test sets. Depending on the activation function, mean absolute error (MAE) and mean absolute percentage error (MAPE) can be used as indicators for evaluating the prediction performance of the model, and the f1-score can also be used. For the three properties of bending strength, bending modulus, and thermal distortion temperature, the results using the Swish function in example figure 940 showed the lowest error rate. For the two physical properties, tensile strength and impact strength, Figure 940 shows that the results using the Mish function exhibited the lowest error rate.

[0099] Figure 10 is an illustrative diagram (1000) schematically showing the performance evaluation result screen of the composite property prediction data according to the composition ratio according to an exemplary embodiment of the present disclosure.

[0100] Referring to Figure 10, the processor (for example, processor 110 in Figure 1) can derive performance evaluation results for the final characteristic prediction model. The processor can verify the ground truth for the second physical property data. Based on the ground truth, the processor can evaluate the accuracy of the second physical property data. The second physical property data differs from the first physical property data in at least part and may be data for the properties of a composite in which ground truth exists. For example, if the bending strength value for a specific composition harmony of mPPO is known as the ground truth, and the processor has not learned about it, the processor can verify how much the second physical property data corresponding to the bending strength of mPPO predicted through learning differs from the known ground truth. This verification of the difference can be considered a verification by deriving performance evaluation results for the final characteristic prediction model.

[0101] Referring to Figure 10, the x-axis 1010 could be strain (%), and the y-axis 1020 could be stress (MPa). That is, the processor can compare the ground truth value 1030 and the predicted value 1040 of the composite material's physical properties corresponding to the composition harmony, while keeping the physical properties to be predicted on the x-axis 1010 and y-axis 1020. In other words, by checking the error rate between the ground truth value 1030 and the predicted value 1040, the processor can determine which artificial intelligence learning model is superior. The artificial intelligence learning model here could be the second model, which is the first model trained with the Swish function, or the third model, which is the first model trained with the Mish function, and so on. The processor can provide the user with a screen showing the prediction performance results of the artificial intelligence learning model, as shown in Figure 10.

[0102] According to the embodiment, the processor can determine which activation function layer is advantageous for learning and select or update the artificial intelligence learning model. For example, the processor can adopt an artificial intelligence learning model that utilizes the Swish function based on performance evaluation results and provide the user with material property prediction data. In another example, the processor can adopt an artificial intelligence learning model that utilizes the Mish function based on performance evaluation results and provide the user with material property prediction data. In yet another example, the processor can adopt an artificial intelligence learning model that utilizes the Swish function for some material property predictions and an artificial intelligence learning model that utilizes the Mish function for some material property predictions and provide the user with material property prediction data. The reason the processor can selectively proceed with learning as described above is that it learns about the individual material properties of the complex individually during the learning process.

[0103] According to the embodiment, the processor can generate synthetic data similar to the published experimental data in order to overcome the difficulty of obtaining minimal data for training a model that predicts the physical properties of a composite, which is limited by the number of published experimental data. This can correspond to a process in which experimental data, which is structured in a tabular format via the processor, is trained and preprocessed, and then data similar to the experimental data is generated to augment the data, as described above.

[0104] According to the embodiment, the processor can perform reverse engineering to accelerate harmonic optimization by target physical properties for harmonic optimization of the composite. The processor can leverage an actor-critic reinforcement learning model to generate novel composition harmonics suitable for the target physical properties in the composition harmonics of existing composites.

[0105] Up to this point, the description has mainly focused on electronic devices, but this disclosure is not limited to them. For example, a method for manufacturing such electronic devices is also said to fall within the scope of this disclosure.

[0106] Although this disclosure has been described with reference to embodiments shown in the drawings, these are merely illustrative, and it will be understood by those ordinary skill in the art that various modifications and equivalent other embodiments are possible therefrom. Accordingly, the true scope of technical protection of this disclosure should be determined by the technical idea of ​​the appended claims.

Claims

1. In a method for providing predictive data on the physical properties of a composite according to its composition ratio using artificial intelligence, A step of verifying data related to at least one composition constituting the complex via a processor of an electronic device, The steps include encoding the physical property data of the composite synthesized via the processor based on data related to the at least one composition into first physical property data, The steps include using the processor to calculate an embedding result corresponding to the first physical property data via a first model that corresponds to an artificial intelligence learning model based on a neural network, The step of providing prediction data for the properties of the composite relative to the second property data based on the embedding results calculated via the processor, The step of calculating the embedding result corresponding to the first physical property data is: The steps include: using the processor to generate masked features through masking of individual features of the first physical property data; The steps include: calculating a scale term based on the masked features via the processor to determine the importance of the individual features; The process involves the processor confirming the decision output of individual domains via an activation function layer based on their importance, The process includes the step of using the processor to calculate the final decision output of the individual domains via the fully connected layer as an embedding result corresponding to the first physical property data, Method for providing composite material property prediction data.

2. The step of verifying data related to the at least one composition is: The steps include: confirming experimental data and synthetic data for the composite via the processor; The step of sampling data related to the at least one composition based on the experimental data and the synthesis data via the processor, The data relating to the at least one composition is, The type, content, and harmonization data of at least one composition constituting the composite, The method for providing composite material property prediction data according to feature 1.

3. The process further includes confirming the data related to the sampled at least one composition via the processor as an input dataset, The input dataset is characterized in that it has been preprocessed as a training input dataset for the first model. The aforementioned synthesized data is synthesized from the aforementioned experimental data through data augmentation. The method for providing composite material property prediction data according to feature 2.

4. The step of encoding the first physical property data is: The steps include using the processor to set the physical property information of the composite to be predicted via the first model, The process includes the step of representing data related to the composition of the composite in individual features for verifying the set physical property information via the processor, and performing batch normalization. A method for providing composite material property prediction data according to claim 1.

5. The step of checking the decision output of the individual domain is: The steps include: confirming the first decision output in the first domain among the individual domains via the processor; The process includes the step of confirming the second decision output in the second domain of the individual domains based on the first decision output via the processor, Each of the individual domains, including the first and second domains, is comprised of the full connectivity layer. The method for providing composite material property prediction data according to feature 1.

6. The step of providing predictive data on the physical properties of the composite is: The step of using the processor to provide second physical property data that is at least partially different from the first physical property data learned through the first model, A method for providing composite material property prediction data according to claim 1.

7. The step of providing predictive data on the physical properties of the composite is: The steps include: confirming the ground truth for the second physical property data via the processor; The process includes the step of evaluating the accuracy of the second physical property data based on the ground truth via the processor, A method for providing composite material property prediction data according to claim 6.

8. An electronic device that uses artificial intelligence to provide predictive data on the physical properties of a composite according to its composition ratio, Communications Department and, Memory and The system comprises the communication unit and at least one processor electrically connected to the memory, The aforementioned at least one processor is We examine the data related to at least one of the compositions that make up the complex. The physical property data of the composite synthesized based on the data related to the at least one composition is encoded into first physical property data. Masked features are generated for individual features of the first physical property data via a first model corresponding to an artificial intelligence learning model based on a neural network. Based on the masked features, a scale term is calculated to determine the importance of each individual feature. Based on the aforementioned importance, the decision output of individual domains is checked via the activation function layer. The final decision output of the individual domains is converted via a fully connected layer to calculate the embedding result corresponding to the first physical property data. Based on the calculated embedding results, the system is configured to provide predictive data for the properties of the composite with respect to the second property data. Electronic device for providing material property prediction data.

Images