Apparatus and method for designing a multilayer film

Abstract

A design apparatus and method for a multilayer film are disclosed. The design device for the multi-layer film can model a laminated structure of the multi-layer film to be designed; collecting dart falling impact strength data of the single-layer film forming the laminated structure as stress-strain data; reading a value pre-stored in a storage space accessible by the design device for the multilayer film, and acquiring a feature setting mode for specifying different feature setting methods according to the read value; setting a plurality of physical indexes selected from the falling dart impact strength data and the thickness of the single-layer film as features according to a feature setting mode; predicting the falling dart impact strength of the multilayer film using a machine learning model, the machine learning model being selected from a plurality of supervised learning models and being trained using the features as independent variables and using the falling dart impact strength of the multilayer film as a target variable; and generating design data of the multilayer film by combining the predicted values of the other characteristics with the predicted value of the falling dart impact strength.

Description

Technical Field

[0001] Cross-references to related applications

[0002] This application claims priority and benefit to Korean Patent Application Nos. 10-2023-0116487 and 10-2024-0117595, filed with the Korean Intellectual Property Office on September 1, 2023 and August 30, 2024, the entire contents of which are incorporated herein by reference.

[0003] This disclosure relates to design apparatus and methods for multilayer films. Background Technology

[0004] Polymer films are thin, flat structures made of polymer materials. Here, the polymer can have a long-chain molecular structure, with monomers linked as repeating units within the long chain. The thickness of polymer films can be varied, up to the millimeter (mm) level. Polymer films offer numerous advantages, including flexibility, lightweight, cost-effectiveness, and processability. Specifically, due to their excellent flexibility, polymer films can be made into various shapes and sizes, thus enabling their application in a wide range of product designs. Furthermore, their lightweight nature makes them suitable for products requiring portability, reducing transportation and handling costs. Additionally, polymer films can be manufactured using relatively inexpensive raw materials and mass-produced, making them cost-effective. Moreover, polymer films can be applied to various manufacturing and processing techniques, such as thermal processing, extrusion, and bonding, allowing for easy adjustment of product shape, size, and thickness. Finally, polymer films exhibit high durability and strength, and some polymer films possess excellent optical properties, excellent water resistance, and breathability as needed.

[0005] Multilayer membranes can be composed of two or more layers of polymers or other materials. This multilayer structure enhances the overall membrane performance by combining the unique properties of each layer. For example, one layer might be used to increase durability and strength, another to provide water resistance or moisture resistance, and yet another to provide resistance to specific chemicals. In this way, by combining layers with various properties, multilayer membranes can be suitable for a wider range of applications and specific requirements than single-layer membranes. Summary of the Invention

[0006] Technical issues

[0007] This disclosure seeks to provide an apparatus and method for designing multilayer membranes that can predict the properties of multilayer membranes using only information about the single-layer membranes forming the multilayer membrane when designing the multilayer membrane.

[0008] Technical solutions

[0009] According to an example embodiment, a design apparatus for multilayer membranes is provided. This apparatus designs a multilayer membrane by executing program code loaded into one or more memory devices via one or more processors, based on properties predicted using only information about a single-layer membrane. The program code can be executed to: model the laminated structure of the multilayer membrane to be designed; collect dart impact strength data of the single-layer membrane forming the laminated structure as stress-strain data; read values ​​pre-stored in a storage space accessible to the design apparatus for multilayer membranes, and obtain feature setting patterns for specifying different feature setting methods based on the read values; set multiple physical parameters selected from the dart impact strength data and the thickness of the single-layer membrane as features according to the feature setting patterns; select at least one of multiple supervised learning models capable of performing regression analysis as a machine learning model; predict the dart impact strength of the multilayer membrane using the machine learning model, which is trained using the features as independent variables and the dart impact strength of the multilayer membrane as the target variable; and generate design data for the multilayer membrane by combining the predicted values ​​of other properties with the predicted values ​​of the dart impact strength to meet the design requirements of the multilayer membrane.

[0010] In some example implementations, when the feature setting mode includes a first feature setting mode, setting a feature may include setting at least some of yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus and 2% modulus, and the thickness of the monolayer film as features.

[0011] In some example implementations, selection as a machine learning model may include: selecting one of a partial least squares (PLS) model, a LAST regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAST regression model, and random forest model may: receive a vector generated based on at least some of yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus, and the thickness of the monolayer film, and may be trained to predict the dart impact strength of the multilayer film.

[0012] In some example implementations, when the feature setting mode includes a second feature setting mode, setting a feature may include setting at least some of yield stress, yield strain, necking stress, necking strain, fracture stress and fracture strain, and the thickness of the monolayer film as a feature.

[0013] In some example implementations, selection as a machine learning model may include: selecting one of a PLS model, a LAS regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAS regression model, and random forest model may: receive a vector generated based on at least some of yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain, and the thickness of the monolayer membrane, and may be trained to predict the dart impact strength of the multilayer membrane.

[0014] In some example implementations, when the feature setting mode includes a third feature setting mode, setting a feature may include setting the yield stress, yield strain, and yield stress divided by yield strain (yield stress / yield strain) as indicators, as well as the thickness of the monolayer film as features.

[0015] In some example implementations, selection as a machine learning model may include: selecting one of a PLS model, a LAS regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAS regression model, and random forest model may: receive a vector generated based on the yield stress, yield strain, and an index of yield stress divided by yield strain, as well as the thickness of the monolayer membrane, and may be trained to predict the dart impact strength of the multilayer membrane.

[0016] In some example implementations, generating design data for a multilayer membrane may include: combining a predicted value of at least one of the multilayer membrane’s stiffness, strength, strain, tear strength, high-speed dart impact strength, seal initiation temperature (SIT), haze, moisture permeability, and air permeability with a predicted value of the dart impact strength; generating design data based on the combination when the multilayer membrane designed based on the combination of predicted values ​​meets the design requirements; and changing the combination to another combination and redesigning the multilayer membrane when the multilayer membrane designed based on the combination of predicted values ​​does not meet the design requirements.

[0017] In some example implementations, the multilayer membrane can be modeled as a multilayer membrane model, which may include a first monolayer membrane and a second monolayer membrane laminated on top of each other. The multilayer membrane model may include n (n is a multiple of 6) model layers laminated on top of each other, and the first monolayer membrane corresponds to two or more model layers among the model layers, and the second monolayer membrane corresponds to two or more other model layers among the model layers.

[0018] In some example implementations, program code may be executed to additionally display the model layers on a display device and to provide a user interface that allows the user to set the model layer corresponding to the first monolayer membrane and the model layer corresponding to the second monolayer membrane.

[0019] A design method for multilayer membranes is disclosed, which designs multilayer membranes based on properties predicted using only information about monolayer membranes. The design method is executed by a computing device including one or more processors and one or more memory devices. The method may include: modeling the laminated structure of the multilayer membrane to be designed; collecting dart impact strength data of the monolayer membranes forming the laminated structure as stress-strain data; reading values ​​pre-stored in storage accessible by the computing device, and obtaining feature setting patterns for specifying different feature setting methods based on the read values; setting multiple physical parameters selected from the dart impact strength data and the thickness of the monolayer membrane as features according to the feature setting patterns; selecting at least one of multiple supervised learning models capable of performing regression analysis as a machine learning model; predicting the dart impact strength of the multilayer membrane using the machine learning model, which is trained using the features as independent variables and the dart impact strength of the multilayer membrane as the target variable; and generating design data for the multilayer membrane by combining the predicted values ​​of other properties with the predicted values ​​of the dart impact strength to meet the design requirements of the multilayer membrane.

[0020] In some example implementations, when the feature setting mode may include a first feature setting mode, setting a feature may include setting at least some of yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus and 2% modulus, and the thickness of the monolayer film as features.

[0021] In some example implementations, selection as a machine learning model may include: selecting one of a PLS model, a LAS regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAS regression model, and random forest model may: receive a vector generated based on at least some of yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus, and the thickness of the monolayer film, and may be trained to predict the dart impact strength of the multilayer film.

[0022] In some example implementations, when the feature setting mode includes a second feature setting mode, setting a feature may include setting at least some of yield stress, yield strain, necking stress, necking strain, fracture stress and fracture strain, and the thickness of the monolayer film as a feature.

[0023] In some example implementations, selection as a machine learning model may include: selecting one of a PLS model, a LAS regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAS regression model, and random forest model may: receive a vector generated based on at least some of yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain, and the thickness of the monolayer membrane, and may be trained to predict the dart impact strength of the multilayer membrane.

[0024] In some example implementations, when the feature setting mode includes a third feature setting mode, setting a feature may include setting the yield stress, yield strain, yield stress divided by yield strain, and the thickness of the monolayer as features.

[0025] In some example implementations, selection as a machine learning model may include: selecting one of a PLS model, a LAS regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, LAS regression model, and random forest model may: receive a vector generated based on the yield stress, yield strain, and an index of yield stress divided by yield strain, as well as the thickness of the monolayer membrane, and may be trained to predict the dart impact strength of the multilayer membrane.

[0026] In some example implementations, generating design data for a multilayer membrane may include: combining a predicted value of at least one of the multilayer membrane’s stiffness, strength, strain, tear strength, high-speed dart impact strength, seal initiation temperature (SIT), haze, moisture permeability, and air permeability with a predicted value of the dart impact strength; generating design data based on the combination when the multilayer membrane designed based on the combination of predicted values ​​meets the design requirements; and changing the combination to another combination and redesigning the multilayer membrane when the multilayer membrane designed based on the combination of predicted values ​​does not meet the design requirements.

[0027] In some example implementations, the multilayer membrane can be modeled as a multilayer membrane model, which may include a first monolayer membrane and a second monolayer membrane laminated on top of each other. The multilayer membrane model may include n (n is a multiple of 6) model layers laminated on top of each other, and the first monolayer membrane corresponds to two or more model layers among the model layers, and the second monolayer membrane corresponds to two or more other model layers among the model layers.

[0028] In some example implementations, the design method may further include: displaying the model layers on a display device, and providing a user interface that allows a user to set the model layer corresponding to the first monolayer film and the model layer corresponding to the second monolayer film.

[0029] Beneficial effects

[0030] According to the implementation method, when designing multilayer membranes, the properties of the multilayer membrane can be predicted using only information about the single-layer membranes that constitute the multilayer membrane, and the predictions about other properties can be considered together to design a multilayer membrane that can meet a given design requirement. Attached Figure Description

[0031] Figure 1 This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment.

[0032] Figure 2 This is a graph showing stress-strain data for a monolayer membrane according to an example embodiment.

[0033] Figure 3 This is a flowchart illustrating a design method for multilayer films according to an example embodiment.

[0034] Figure 4 This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment.

[0035] Figure 5 This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment.

[0036] Figure 6 and Figure 7 This is a diagram illustrating a user interface according to an example implementation.

[0037] Figure 8 This is a diagram used to describe a computing device according to an example implementation. Detailed Implementation

[0038] In the following description, exemplary embodiments of the present disclosure will be described more fully with reference to the accompanying drawings to facilitate practice by those skilled in the art. As will be appreciated by those skilled in the art, modifications can be made to the described embodiments in various ways without departing from the spirit or scope of the present disclosure. Therefore, the drawings and description are to be considered illustrative rather than restrictive in nature. Throughout the specification, similar reference numerals denote similar elements.

[0039] Throughout the specification and claims, unless explicitly stated otherwise, the word "comprising" and variations such as "including" or "containing" shall be construed as implying inclusion of the stated elements but not excluding any other elements. Terms including ordinal numbers such as first, second, etc., may be used to describe various components, but components are not limited to these terms. The terms above are used only for the purpose of distinguishing one component from another.

[0040] Terms such as “…unit,” “…device,” and “module” used in this specification may refer to a unit capable of performing at least one function or operation described in the specification, which may be implemented as hardware or circuitry, software, or a combination of hardware or circuitry and software. Furthermore, at least some components or functions of the design apparatus and method for multilayer films according to the exemplary embodiments described below may be implemented as programs or software, and such programs or software may be stored on a computer-readable medium.

[0041] Figure 1 This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment. Figure 2 This is a graph showing stress-strain data for a monolayer membrane according to an example embodiment.

[0042] Reference Figure 1 The design apparatus 10 for multilayer films according to the example embodiment can execute program code or instructions loaded into one or more memory devices via one or more processors. For example, the design apparatus 10 for multilayer films can be implemented as follows regarding... Figure 8 The computing device 50 is described. In this case, one or more processors may correspond to processor 510 of computing device 50, and one or more memory devices may correspond to memory 520 of computing device 50. Program code or instructions executed by one or more processors may use only information about a single-layer film to predict the properties of a multilayer material (i.e., a multilayer film), and design the multilayer film based on the predicted properties. In this specification, the term "module" is used to logically separate these functions executed by program code or instructions.

[0043] The design apparatus 10 for multilayer membranes may include all or at least a portion of a multilayer membrane modeling module 110, a data collection module 120, a feature setting module 130, a model selection module 140, a learning module 150, a prediction module 160, and a design data generation module 162. For example, the learning module 150 may be implemented within the design apparatus 10. Alternatively, the learning module 150 may be implemented externally to the design apparatus 10, and the design apparatus 10 may be equipped with a machine learning model trained by the learning module 150.

[0044] The multilayer membrane modeling module 110 can model the laminated structure of the multilayer membrane to be designed. A multilayer membrane can include multiple monolayer membranes. Here, the multiple monolayer membranes can include monolayer membranes of a specific type and different types, and the unique properties of the monolayer membranes can vary depending on their type. Furthermore, the lamination order of the monolayer membranes within the multilayer membrane can also vary. The multilayer membrane modeling module 110 can model the multilayer membrane by setting the type of monolayer membrane, the lamination order of the monolayer membranes, etc., for the multilayer membrane whose properties are to be predicted.

[0045] In some example implementations, the multilayer membrane modeling module 110 can model a multilayer membrane comprising k monolayer membranes (where k is an integer greater than or equal to 2). For example, the multilayer membrane modeling module 110 can model a multilayer membrane having a first membrane layer, a second membrane layer, ..., a kth membrane layer sequentially laminated from the bottom. For each of the k monolayer membranes forming the multilayer membrane, a feature set by the feature setting module 130 described below can be applied.

[0046] In some example implementations, the multilayer membrane modeling module 110 can model a multilayer membrane comprising six single-layer membranes. For example, the multilayer membrane modeling module 110 can model a multilayer membrane having a first membrane layer, a second membrane layer, a third membrane layer, a fourth membrane layer, a fifth membrane layer, and a sixth membrane layer sequentially laminated from the bottom.

[0047] The data collection module 120 can collect dart impact strength data of the monolayer membranes forming the laminated structure modeled by the multilayer membrane modeling module 110. Here, dart impact strength can represent the strength that a material (here, the monolayer membrane) can withstand when subjected to a physical impact. Dart impact strength can be used as an indicator to measure, for example, the degree of damage to the monolayer membrane when a heavy object is dropped from a predetermined height. The design device 10 for multilayer membranes can predict the dart impact strength of the multilayer membrane using machine learning based on the dart impact strength data of each of the multiple monolayer membranes forming the laminated structure.

[0048] In some example implementations, the data collection module 120 can collect dart impact strength data of the monolayer membrane as stress-strain data. See also... Figure 2 Stress-strain data can represent the mechanical properties of a monolayer membrane and can be expressed in the form of stress-strain curves. Stress-strain curves can show the stress that occurs when a force is applied to a monolayer membrane, and how the membrane deforms as a result.

[0049] In some example implementations, stress-strain data may include at least one of yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus. Here, the yield point is the point at which the material moves from an elastically deformable region to an inelastically deformable region, and after the yield point, even if the stress applied to the material is removed, the material may not fully recover its original shape. The yield stress is the stress value at the yield point, and a higher value may indicate that the material can withstand greater stress before deformation. Resilience is the area under the stress-strain curve up to the yield point and can be a measure of how well the material can recover its original shape after deformation at a specific stress when the stress is removed. Strain hardening refers to the phenomenon that the material strengthens to withstand additional stress with increasing strain after the yield point, and failure can be the point at which the material may no longer withstand stress and is destroyed. The strain hardening modulus is expressed as the slope of the stress-strain curve in the strain hardening region and can be used as an indicator of the rate of material strengthening. The flow energy can be the area under the stress-strain curve from the yield point to strain hardening, and the strain hardening energy can be the area under the stress-strain curve from strain hardening to failure. Toughness is the total area under the stress-strain curve, indicating how much energy a material can absorb, and the 1% modulus and 2% modulus indicate the stress at 1% and 2% strain, respectively. The values ​​of the 1% modulus and 2% modulus can be used as measurements indicating strength within the material's elastic range.

[0050] In some example implementations, the data collection module 120 can collect dart impact strength data of the monolayer membrane by performing a test according to "Method A" of ASTM D1709. Here, the test may be to use a free-falling dart to evaluate the impact strength, toughness, etc. of the monolayer membrane. For more detailed information relating to ASTM D1709, please refer to the ASTM standard document, and therefore its description will be omitted in this specification.

[0051] In some example embodiments, the data collection module 120 may acquire stress-strain curves as data based on the results of uniaxial tensile tests performed in the machine direction (MD) and transverse direction (TD). Here, MD can be the direction in which the membrane moves between the rollers of the manufacturing machine during membrane manufacturing, and TD can be perpendicular to MD, i.e., orthogonal to the rollers of the manufacturing machine. Meanwhile, the uniaxial tensile test may be performed by measuring the resistance of a single-layer membrane when stretched in a uniaxial direction. In this case, the data collection module 120 can collect data on the stress-strain curves by means of a test according to ASTM D882. More detailed information relating to ASTM D882 can be found in the ASTM standard document, and therefore its description will be omitted in this specification.

[0052] The feature setting module 130 can perform feature setting according to multiple feature setting modes. The multiple feature setting modes are stored and managed as predetermined values ​​in a storage space accessible by the processor of the multilayer film design device 10, and the multilayer film design device 10 can read the values ​​pre-stored in the storage space and determine the feature setting mode for specifying different feature setting methods based on the read values.

[0053] In some example implementations, when the feature setting mode is the first feature setting mode, the feature setting module 130 can set multiple physical indicators selected from the dart impact intensity data of the monolayer membrane and the thickness of the monolayer membrane as features. In machine learning, features are input variables used to predict a target (i.e., the properties of the multilayer membrane) and can represent individual independent variables of the data. Features can provide information related to the dependent or target variable to be predicted in the dataset and have a significant impact on the performance of the machine learning model 20. The learning module 150 can use features to perform training of the machine learning model 20. In this case, features constitute part of the data input to the machine learning model 20, and the machine learning model 20 can train a mode based on the data and perform predictions. In particular, feature engineering, such as selecting or transforming features, can be an important process for improving the performance of the machine learning model 20. In other words, selecting appropriate features or appropriately transforming features can have a significant impact on the high performance of the machine learning model 20.

[0054] In some example implementations, the feature setting module 130 can set yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, some of the 1% modulus and 2% modulus, and the thickness of the monolayer membrane as features. That is, the feature setting module 130 can set only some of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply features including only some of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane to a first membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features including only selected some of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane to a second membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features, including only some of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane, to the k-th membrane layer to train the machine learning model 20. In this case, training can be performed by applying the same type of features to all k monolayer membranes, or conversely, by applying features of a different type to one of the k monolayer membranes than to the other membrane layers. This example implementation can be employed in environments with limited computational resources for training and predicting the machine learning model 20.

[0055] In some example implementations, the feature setting module 130 can set yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, all of the 1% modulus and 2% modulus, and the thickness of the monolayer membrane as features. That is, the feature setting module 130 can set all of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply the features including all of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane to the first membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply the features including all of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane to the second membrane layer among the k monolayer membranes. Furthermore, the learning module 150 can apply the features including all of the stress-strain data collected by the data collection module 120 and the thickness of the monolayer membrane to the kth membrane layer to perform training of the machine learning model 20. This example implementation can be used in an environment with sufficient computing resources for training and prediction of the machine learning model 20.

[0056] In some example implementations, the stress-strain data may include first stress-strain data and second stress-strain data. The first stress-strain data may represent the mechanical properties of the monolayer membrane in a first direction, and the second stress-strain data may represent the mechanical properties of the monolayer membrane in a second direction perpendicular to the first direction. Specifically, the first direction may be MD, and the second direction may be TD. The feature setting module 130 may set a plurality of physical indices selected from at least one of the first stress-strain data and the second stress-strain data as features.

[0057] In some example implementations, the feature setting module 130 can set some of the following in a first direction: yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus; and some of the following in a second direction: yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus; and the thickness of the monolayer membrane. That is, the feature setting module 130 can set only a selected portion of the first stress-strain data and only a selected portion of the second stress-strain data collected by the data collection module 120, and the thickness of the monolayer membrane, as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply only a selected portion of the first stress-strain data and only a selected portion of the second stress-strain data collected by the data collection module 120, and the thickness of the monolayer membrane, as features to the first membrane layer among the k monolayer membranes. Furthermore, the learning module 150 can apply features, including only a selected portion of the first stress-strain data and only a selected portion of the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, to the second membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features, including only a selected portion of the first stress-strain data and only a selected portion of the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, to the kth membrane layer to perform training of the machine learning model 20.

[0058] In some example implementations, the feature setting module 130 can set all of the yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus in the first direction, and all of the yield point, yield stress, resilience, strain hardening, failure, strain hardening modulus, flow energy, strain hardening energy, toughness, 1% modulus, and 2% modulus in the second direction, as well as the thickness of the monolayer membrane, as features. That is, the feature setting module 130 can set all of the first stress-strain data and all of the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply all of the first stress-strain data and all of the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, as features to the first membrane layer among the k monolayer membranes. Furthermore, the learning module 150 can apply features including all the first stress-strain data and all the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, to the second membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features including all the first stress-strain data and all the second stress-strain data collected by the data collection module 120, as well as the thickness of the monolayer membrane, to the k-th membrane layer to perform training of the machine learning model 20.

[0059] In some other example implementations, when the feature setting mode is the second feature setting mode, the feature setting module 130 can set yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain as features. Here, yield stress, yield strain, necking strain, and fracture stress can be features corresponding to yield stress, yield point, strain hardening, and failure of the stress-strain data described above. Meanwhile, necking stress and fracture strain can be new specified features derived from stress-strain data analysis. Here, yield stress can represent the stress at the yield point, which is the point where the material exceeds its elastic limit and begins permanent deformation, i.e., the point where the material moves from an elastic deformation region to an inelastic deformation region, and yield strain can represent the strain corresponding to the yield stress. Necking stress represents the stress at the point where the cross-section of the material begins to locally narrow (necking point), and typically the load-bearing capacity of the material can decrease sharply after the necking point. Necking strain can represent the strain at the necking point. Fracture stress can represent the stress at the point where the material finally fractures, and at the fracture point, the material may no longer support the load. Fracture strain can represent the strain corresponding to the fracture stress.

[0060] In some example embodiments, the feature setting module 130 can set at least some of the yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in each of the first and second directions, as well as the thickness of the monolayer membrane, as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply features including at least some of the yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the first direction, at least some of the yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the second direction, and the thickness of the monolayer membrane to the first membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features including at least some of the yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the first direction, at least some of the yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the second direction, and the thickness of the monolayer membrane to the second membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features, including at least some of yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the first direction, and at least some of yield stress, yield strain, necking stress, necking strain, fracture stress, and fracture strain in the second direction, as well as the thickness of the monolayer membrane, to the k-th membrane layer to train the machine learning model 20. In this case, training can be performed by applying the same type of features to all k monolayer membranes, or conversely, by applying features of a different type to one of the k monolayer membranes than to the other membrane layers.

[0061] In some other example implementations, when the feature setting mode is the third feature setting mode, the feature setting module 130 can set the yield stress, yield strain, and yield stress divided by yield strain as features. Specifically, the feature setting module 130 can set the yield stress, yield strain, and all indices of yield stress divided by yield strain in each of the first and second directions, as well as the thickness of the monolayer membrane, as features. When the multilayer membrane modeling module 110 models a multilayer membrane comprising k monolayer membranes, the learning module 150 can apply the features including the yield stress, yield strain, and all indices of yield stress divided by yield strain in the first direction, the yield stress, yield strain, and all indices of yield stress divided by yield strain in the second direction, and the thickness of the monolayer membrane to the first membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply the features including the yield stress, yield strain, and all indices of yield stress divided by yield strain in the first direction, the yield stress, yield strain, and all indices of yield stress divided by yield strain in the second direction, and the thickness of the monolayer membrane to the second membrane layer among the k monolayer membranes. Additionally, the learning module 150 can apply features including yield stress, yield strain, and all indices of yield stress divided by yield strain in the first direction, yield stress, yield strain, and all indices of yield stress divided by yield strain in the second direction, and the thickness of the monolayer membrane to the k-th membrane layer to perform training of the machine learning model 20.

[0062] For example, when the multilayer membrane modeling module 110 models a multilayer membrane comprising six monolayer membranes, it can apply seven features to a membrane layer, including yield stress, yield strain, and all indices of yield stress divided by yield strain in the first direction, yield stress, yield strain, and all indices of yield stress divided by yield strain in the second direction, and the thickness of a monolayer membrane. Since there are a total of six layers, a total of 42 features can be applied.

[0063] The model selection module 140 can select a machine learning model to be used to predict the physical properties (i.e., dart impact strength) of the multilayer membrane. Specifically, the model selection module 140 can select at least one machine learning model from a number of supervised learning models capable of performing regression analysis to predict the dart impact strength of the multilayer membrane. Supervised learning is a method of training a model using labeled training data, and in supervised learning, the model is trained to establish a relationship between input variables and a target variable, and can use this relationship to predict the value of the target variable for new input data. In some example implementations, the model selection module 140 can select a supervised learning model capable of performing regression analysis from partial least squares (PLS), linear regression, multinomial regression, ridge regression, lassore regression, logistic regression, decision trees, random forests, gradient boosting, and neural networks. PLS models can be used to model the relationship between multiple independent variables and multiple dependent variables (or target variables), and can identify independent variables that are determined to be important when predicting the dependent variable. In linear regression, the model is trained to establish a linear relationship between input variables and the target variable, and in multinomial regression, the model can train a complex relationship by extending linear regression to include higher-order terms of the input variables. Ridge regression and Lasso regression can add regularization to linear regression to prevent overfitting and adjust model complexity. Logistic regression, a classification algorithm used to solve binary classification problems, can also be used for regression analysis because it predicts outcomes as probabilities. Decision trees, random forests, and gradient boosting train complex nonlinear relationships, and they can be used for both regression and classification problems. Neural networks can solve various types of regression problems, and deep learning, in particular, is very effective for training complex nonlinear relationships. Of course, the supervised learning model selected by the model selection module 140 is not limited to those listed above, and other types of models not listed can be selected considering specific implementation objectives, implementation environments, required performance, etc.

[0064] The learning module 150 can train the machine learning model 20 using features set by the feature setting module 130 as independent variables and the dart impact intensity of the multilayer membrane as the target variable. Therefore, by training the relationship between the features and the target variable, the machine learning model 20 can be trained to predict the properties of multilayer membranes for new data (i.e., property data of single-layer membranes). Specifically, the learning module 150 can use different learning methods depending on the model selected by the model selection module 140.

[0065] The prediction module 160 can use the machine learning model 20 trained by the learning module 150 to predict the dart impact intensity of the multilayer membrane. That is, the prediction module 160 can feed new data (i.e., property data of the single-layer membrane) to the machine learning model 20 and obtain the prediction result of the dart impact intensity of the multilayer membrane.

[0066] Specifically, the model selection module 140 can perform model selection according to multiple model selection modes. The multiple model selection modes are stored and managed as predetermined values ​​in a storage space accessible by the processor of the multilayer film design device 10, and the multilayer film design device 10 can read the values ​​pre-stored in the storage space and determine the model selection mode for specifying different model selection methods based on the read values.

[0067] In some example implementations, when the model selection mode is the first model selection mode, the model selection module 140 can select the PLS model as the machine learning model to be used for predicting the dart impact intensity of the multilayer membrane. As described above, the features set by the feature setting module 130 according to the feature setting mode can be combined in the form of a single vector and can be used as input to the PLS model. The dart impact intensity of the multilayer membrane is set as the target variable, and the learning module 150 can train the PLS model in a direction that maximizes the correlation between the input features and the target variable. In this case, the learning module 150 optimizes the "n_components" parameter so that the PLS model can properly describe the data, and the optimal "n_components" value can be determined through cross-validation. With the PLS model trained in this way, the dart impact intensity of the multilayer membrane can be predicted by inputting the features of a new monolayer membrane.

[0068] In some other example implementations, when the model selection mode is the second model selection mode, the model selection module 140 can select a LCA regression model as the machine learning model to be used to predict the dart impact intensity of the multilayer membrane. As described above, the features set by the feature setting module 130 according to the feature setting mode can be combined in the form of a single vector and can be used as input to the LCA regression model. The dart impact intensity of the multilayer membrane is set as the target variable, and the learning module 150 can train the LCA regression model in a direction that guides the model to select only important features through L1 regularization (a regularization that adds the sum of the absolute values ​​of the regression coefficients as a penalty to the target function). In this case, the learning module 150 optimizes the "alpha" parameter used to adjust the regularization strength during model training, and the optimal "alpha" value can be determined through cross-validation. With the LCA regression model trained in this way, the dart impact intensity of the multilayer membrane can be predicted by inputting features of a new monolayer membrane.

[0069] In some other example implementations, when the model selection mode is the third model selection mode, the model selection module 140 can select a random forest model as the machine learning model to be used to predict the dart impact intensity of the multilayer membrane. As described above, the features set by the feature setting module 130 according to the feature setting mode can be combined in the form of a single vector and used as input to the random forest model. The dart impact intensity of the multilayer membrane is set as the target variable, and the learning module 150 can be trained by setting "n_estimators", "max_depth", "min_samples_split", etc., as hyperparameters of the random forest model. The random forest model trained in this way can predict the dart impact intensity of the multilayer membrane by inputting features of a new monolayer membrane.

[0070] The design data generation module 162 can generate design data for multilayer membranes by combining predicted values ​​of other properties with predicted values ​​of dart impact strength obtained by the prediction module 160, in order to meet the design requirements of multilayer membranes. In some example embodiments, other properties may include at least one of the following: stiffness, strength, strain, tear strength, high-speed dart impact strength, seal initiation temperature (SIT), haze, moisture permeability, and air permeability of the multilayer membrane.

[0071] The design data generation module 162 can first combine the predicted values ​​of at least one of the following: stiffness, strength, strain, tear strength, high-speed dart impact strength, seal initiation temperature, haze, moisture permeability, and air permeability of the multilayer membrane with the predicted value of the dart impact strength. The design requirements for the multilayer membrane to be designed may include ensuring that mechanical, thermal, optical, barrier, and chemical properties meet the desired performance. When the multilayer membrane designed based on the first combination of predicted values ​​meets the design requirements, the design data generation module 140 can generate design data based on the first combination. In this case, the design data may include specific predicted values.

[0072] In contrast, when the multilayer membrane designed based on the first combination of predicted values ​​fails to meet design requirements, the design data generation module 140 can redesign the multilayer membrane by changing the first combination to a different combination. That is, the design data generation module 140 can perform a secondary combination of the predicted values ​​of at least one of the multilayer membrane's stiffness, strength, strain, tear strength, high-speed dart impact strength, sealing initiation temperature, haze, moisture permeability, and air permeability with the predicted value of the dart impact strength, making this combination different from the first combination. In this case, the predicted value of the dart impact strength in the first combination and the predicted value of the dart impact strength in the second combination can be different values.

[0073] When the multilayer membrane designed based on the second combination of predicted values ​​meets the design requirements, the design data generation module 140 can generate design data based on the second combination. Conversely, when the multilayer membrane designed based on the second combination of predicted values ​​also fails to meet the design requirements, the design data generation module 140 can change the second combination to another combination and repeat the process of redesigning the multilayer membrane.

[0074] According to this example embodiment, when designing multilayer membranes, the properties of the multilayer membrane can be predicted using only information about the single-layer membranes forming the multilayer membrane. Specifically, by performing learning and prediction using various indices of the single-layer membrane's dart impact strength stress-strain data and the single-layer membrane's thickness as features, the dart impact strength of the multilayer membrane can be predicted with high accuracy. Furthermore, by varying the number of indices selected as features when performance is prioritized and when computational resources are saved, flexible design is possible depending on the implementation purpose and environment of the design apparatus for the multilayer membrane. Moreover, by considering predictions of other properties of the multilayer membrane as well as predictions of the dart impact strength, a multilayer membrane that meets given design requirements can be designed.

[0075] Figure 3 This is a flowchart illustrating a design method for multilayer films according to an example embodiment.

[0076] Reference Figure 3 The design method for multilayer membranes according to the example embodiment includes: modeling the laminated structure of the multilayer membrane to be designed (S301); collecting dart impact strength data of the single-layer membrane forming the laminated structure (S302); reading values ​​pre-stored in a storage space, and obtaining a feature setting mode for specifying different feature setting methods based on the read values ​​(S303); setting multiple physical indicators selected from the dart impact strength data and the thickness of the single-layer membrane as features according to the feature setting mode (S304); selecting at least one of multiple supervised learning models capable of performing regression analysis as a machine learning model (S305); training the machine learning model using the feature as an independent variable and using the dart impact strength as a target variable (S306); predicting the dart impact strength of the multilayer membrane using the trained machine learning model (S307); and generating design data for the multilayer membrane by combining the predicted values ​​of other characteristics and the predicted values ​​of the dart impact strength to meet the design requirements of the multilayer membrane (S308).

[0077] For more specific details regarding the design methods for multilayer films, please refer to and apply the description of the exemplary embodiments described in this specification, and therefore, redundant descriptions thereof will be omitted herein.

[0078] Figure 4This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment.

[0079] Reference Figure 4 The multilayer membrane design apparatus 10 according to the example embodiment may include all or at least a portion of a multilayer membrane modeling module 110, a data collection module 120, a feature setting module 130, a model selection module 140, a learning module 150, a prediction module 160, a design data generation module 162, a correlation coefficient analysis module 170, and a PLS analysis module 180. For example, the learning module 150 may be implemented within the multilayer membrane design apparatus 10. Alternatively, for example, the learning module 150 may be implemented outside the multilayer membrane design apparatus 10, and the multilayer membrane design apparatus 10 may be equipped with a machine learning model trained by the learning module 150. For the multilayer membrane modeling module 110, data collection module 120, feature setting module 130, model selection module 140, learning module 150, prediction module 160, and design data generation module 162, unless they contradict or are inconsistent with the description in this example embodiment, reference can be made to the above reference. Figure 1 and Figure 2 The descriptions of the multilayer membrane modeling module 110, data collection module 120, feature setting module 130, model selection module 140, learning module 150, prediction module 160, and design data generation module 162 are described herein, and therefore, their redundant descriptions will be omitted.

[0080] The correlation coefficient analysis module 170 can analyze the correlation coefficients between mechanical properties in stress-strain data, and the PLS analysis module 180 can use PLS analysis based on the correlation coefficients analyzed by the correlation coefficient analysis module 170 to predict the dart impact strength value of the multilayer membrane. PLS analysis can be used to model the relationship between multiple independent variables and multiple dependent variables (or target variables), and can identify independent variables that are determined to be important when predicting the dependent variable. Specifically, the correlation coefficient analysis module 170 can construct independent and dependent variables based on various indices of the dart impact strength stress-strain data of a single-layer membrane, and find the component with the largest covariance. This generates a new subspace in which the relationship between the independent and dependent variables is most fully expressed. The PLS analysis module 180 can repeatedly adjust the variables proportionally to each covariance of the new subspace to find new components until the desired number of components are extracted. In this way, the properties of the multilayer membrane can be predicted in conjunction with or in place of the prediction module 160.

[0081] In some example implementations, the correlation coefficient analysis module 170 may be implemented logically separate from the feature setting module 130, or it may be implemented logically included in the feature setting module 130. When the correlation coefficient analysis module 170 is implemented logically included in the feature setting module 130, the correlation coefficient between the mechanical properties of the stress-strain data can be analyzed as a preprocessing step for data dimensionality reduction when multiple physical parameters selected from the dart impact strength data of the monolayer membrane and the thickness of the monolayer membrane are set as features.

[0082] Meanwhile, in some example implementations, the PLS analysis module 180 can be implemented either logically separate from the model selection module 140, or logically included within the model selection module 140. When the PLS analysis module 180 is implemented as logically included within the model selection module 140, the model for performing the PLS analysis can be selected when choosing a machine learning model to be used to predict the impact intensity of a falling dart.

[0083] According to this exemplary embodiment, by performing separate analysis on various indices of the dart impact strength stress-strain data of monolayer films that exhibit high collinearity, the accuracy of predicting the dart impact strength of multilayer films can be further improved.

[0084] Figure 5 This is a block diagram illustrating a design apparatus for multilayer films according to an example embodiment.

[0085] Reference Figure 5 The multilayer membrane design apparatus 10 may include all or at least a portion of a multilayer membrane modeling module 110, a data collection module 120, a feature setting module 130, a model selection module 140, a learning module 150, a prediction module 160, a design data generation module 162, and a model layer setting interface providing module 190. For example, the learning module 150 may be implemented within the multilayer membrane design apparatus 10. Alternatively, for example, the learning module 150 may be implemented externally to the multilayer membrane design apparatus 10, and the multilayer membrane design apparatus 10 may be equipped with a machine learning model trained by the learning module 150. For the multilayer membrane modeling module 110, data collection module 120, feature setting module 130, model selection module 140, learning module 150, prediction module 160, and design data generation module 162, unless they contradict or are inconsistent with the description in this example embodiment, reference can be made to the above references. Figure 1 and Figure 2 The descriptions of the multilayer membrane modeling module 110, data collection module 120, feature setting module 130, model selection module 140, learning module 150, prediction module 160, and design data generation module 162 are described herein, and therefore, their redundant descriptions will be omitted.

[0086] The multilayer membrane modeled in the multilayer membrane modeling module 110 can be modeled as a multilayer membrane model. Here, the multilayer membrane model can be in the form of multiple empty slots arranged sequentially in the vertical direction. Specifically, the multilayer membrane model can include n slots (n is a multiple of 6) stacked on top of each other, i.e., model layers. In addition, when the multilayer membrane includes a first monolayer membrane and a second monolayer membrane laminated on top of each other, the first monolayer membrane can correspond to two or more model layers in the model layers, and the second monolayer membrane can correspond to two or more other model layers in the model layers. The data collection module 120 can collect dart impact intensity data of the first monolayer membrane and the second monolayer membrane based on the multilayer membrane model.

[0087] The model layer setting interface module 190 enables users to set model layers. To this end, the model layer setting interface module 190 provides a user interface that displays the model layers on a display device and allows users to set the model layer corresponding to the first monolayer membrane and the model layer corresponding to the second monolayer membrane.

[0088] Figure 6 and Figure 7 This is a diagram illustrating a user interface according to an example implementation.

[0089] Reference Figure 6 The model layer setting interface module 190 can present the first to sixth model layers to the user via a display device. These layers are represented as n (n is a multiple of 6) slots stacked on top of each other. The user can input information about how many model layers a single-layer membrane occupies, the stacking order of the single-layer membranes, and the type and unique properties of the single-layer membrane assigned to which model layer. For example, as shown in the figure, the user can input the following information: by assigning the first single-layer membrane to the first to third model layers, and the second single-layer membrane to the fourth to sixth model layers, the first and second single-layer membranes each occupy three model layers, and the second single-layer membrane is laminated on top of the first single-layer membrane. Although not shown, the user can input the type and other special characteristics of each of the first and second single-layer membranes.

[0090] Next, refer to Figure 7 The user can input the following information: by assigning a first monolayer membrane to the first to second model layers, a second monolayer membrane to the third to fourth model layers, and a third monolayer membrane to the fifth to sixth model layers, each of the first to third monolayer membranes occupying two model layers, the second monolayer membrane being laminated on the first monolayer membrane, and the third monolayer membrane being laminated on the second monolayer membrane. Although not shown, the user can input the type and other special characteristics of each of the first to third monolayer membranes.

[0091] According to this example implementation, for example, when there are two or three types of monolayer membranes forming the multilayer membrane, the feature setting module 130 can set features based on the same six-layer model structure. Therefore, the learning module 150 can train the machine learning model 20 using features from the same number of model layers as independent variables and the dart impact intensity of the multilayer membrane as the target variable, regardless of the number of types of monolayer membranes forming the multilayer membrane. Therefore, since the machine learning model 20 can be applied regardless of the number of types of monolayer membranes forming the multilayer membrane, compared to the case where different layer structures need to be trained for each type of monolayer membrane forming the multilayer membrane, high prediction accuracy can be maintained while reducing implementation complexity and computational cost.

[0092] Figure 8 This is a diagram used to describe a computing device according to an example implementation.

[0093] Reference Figure 8 The design apparatus and method for multilayer films according to the example embodiment can be implemented using a computing device 50. The computing device 50 can be implemented as various types of electronic devices, servers, or similar devices, and its functions can be implemented through a combination of software and hardware.

[0094] The computing device 50 may include at least one of a processor 510, a memory 530, a user interface input device 540, a user interface output device 550, and a storage device 560, which communicate via a bus 520. The computing device 50 may also include a network interface 570 electrically connected to the network 40. The network interface 570 may send signals to or receive signals from other entities via the network 40.

[0095] Processor 510 can be implemented as various types of computing units, such as microcontroller units (MCUs), application processors (APs), central processing units (CPUs), graphics processing units (GPUs), neural processing units (NPUs), and quantum processing units (QPUs). Processor 510 is a semiconductor device that executes instructions stored in memory 530 or storage device 560 and can play a crucial role in the system. The program code and data stored in memory 530 or storage device 560 instruct processor 510 to perform specific tasks, thereby enabling the overall operation of the system. Processor 510 can be configured to achieve the above-mentioned... Figures 1 to 7 The various functions and methods described.

[0096] Memory 530 and storage device 560 may include various forms of volatile or non-volatile storage media for storing and accessing data in the system. For example, memory 530 may include read-only memory (ROM) 531 and random access memory (RAM) 532. In some example embodiments, memory 530 may be embedded within processor 510, in which case data transfer between memory 530 and processor 510 can be very fast. In some other example embodiments, memory 530 may be located external to processor 510, in which case memory 530 may be connected to processor 510 via various data buses or interfaces. Such connections may be implemented via various means known in the art, such as high-speed peripheral component interconnect (PCIe) interfaces or memory controllers for high-speed data transfer.

[0097] In some exemplary embodiments, at least some of the components or functions of the apparatus and method for designing multilayer films according to the example embodiments may be implemented as programs or software running on computing device 50, and the programs or software may be stored on a computer-readable medium. Specifically, the computer-readable medium according to the example embodiments may record a program for causing a computer including processor 510 to execute programs or instructions stored in memory 530 or storage device 560 to perform the steps included in the method for designing multilayer films according to the example embodiments.

[0098] In some example implementations, at least some of the configurations or functions of the design apparatus and method for multilayer films according to the example implementations may be implemented using the hardware or circuitry of the computing device 50, or may be implemented as a separate hardware or circuitry that may be electrically connected to the computing device 50.

[0099] According to the example implementation, when designing multilayer membranes, the properties of the multilayer membrane can be predicted using only information about the single-layer membranes forming the multilayer membrane. Specifically, by performing learning and prediction using various indices of the single-layer membrane's dart impact strength stress-strain data and the single-layer membrane's thickness as features, the dart impact strength of the multilayer membrane can be predicted with high accuracy. Furthermore, by varying the number of indices selected as features when performance is prioritized and when computational resources are saved, flexible design is possible depending on the implementation purpose and environment of the design apparatus for the multilayer membrane. Moreover, by considering predictions of other properties of the multilayer membrane as well as predictions of the dart impact strength, a multilayer membrane that meets given design requirements can be designed.

[0100] Although the embodiments of this disclosure have been described in detail above, the scope of this disclosure is not limited thereto, but may include various modifications and alterations made by those skilled in the art using the basic concept of this disclosure as defined in the claims.

Claims

1. A design device for a multilayer film that designs a multilayer film based on properties of the multilayer film predicted using only information about a single layer film by executing program code loaded into one or more memory devices via one or more processors, wherein the program code is executed to: model a laminate structure of the multilayer film to be designed; collect dart impact strength data of the single layer film forming the laminate structure as stress-strain data; read values pre-stored in a storage space accessible to the design device for a multilayer film, and acquire a feature setting mode for specifying a different feature setting method according to the read values; set a plurality of physical indices selected from the dart impact strength data and a thickness of the single layer film as features according to the feature setting mode; select at least one of a plurality of supervised learning models capable of regression analysis as a machine learning model; predict a dart impact strength of the multilayer film using the machine learning model trained using the features as independent variables and the dart impact strength of the multilayer film as a target variable; and generate design data of the multilayer film by combining predicted values of other properties with the predicted value of the dart impact strength to satisfy design requirements of the multilayer film.

2. The design device according to claim 1, wherein: when the feature setting mode includes a first feature setting mode, the setting as features includes: setting at least some of a yield point, a yield stress, a resilience, a strain hardening, a failure, a strain hardening modulus, a flow energy, a strain hardening energy, a toughness, a 1% modulus, and a 2% modulus, and a thickness of the single layer film as the features.

3. The design device according to claim 2, wherein: the selection as a machine learning model includes: selecting one of a partial least squares (PLS) model, a lasso regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, the lasso regression model, and the random forest model: receives a vector generated from at least some of the yield point, the yield stress, the resilience, the strain hardening, the failure, the strain hardening modulus, the flow energy, the strain hardening energy, the toughness, the 1% modulus, and the 2% modulus, and the thickness of the single layer film, and is trained to predict a dart impact strength of the multilayer film.

4. The design device according to claim 1, wherein: when the feature setting mode includes a second feature setting mode, the setting as features includes: setting at least some of a yield stress, a yield strain, a necking stress, a necking strain, a breaking stress, and a breaking strain, and a thickness of the single layer film as the features.

5. The design device according to claim 4, wherein: the selection as a machine learning model includes: selecting one of a partial least squares (PLS) model, a lasso regression model, and a random forest model as the machine learning model, and ​ the machine learning model selected from the PLS model, the Lasso regression model, and the random forest model: receives a vector generated from at least some of the yield stress, yield strain, necking stress, necking strain, breaking stress, and breaking strain, and the thickness of the single layer film, and is trained to predict the dart impact strength of the multi-layer film.

6. The design apparatus of claim 1, wherein: when the feature setting mode includes a third feature setting mode, the setting as a feature includes: setting a yield stress, a yield strain, and an index of yield stress divided by yield strain, and a thickness of the single layer film as the feature.

7. The design apparatus of claim 6, wherein: the selection as a machine learning model includes: selecting one of a partial least squares (PLS) model, a Lasso regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, the Lasso regression model, and the random forest model: receives a vector generated from at least some of the yield stress, yield strain, and an index of yield stress divided by yield strain, and a thickness of the single layer film, and is trained to predict the dart impact strength of the multi-layer film.

8. The design apparatus of claim 1, wherein: the generating design data of the multi-layer film includes: combining a predicted value of at least one of a stiffness, a strength, a strain, a tear strength, a high speed dart impact strength, a seal initiation temperature (SIT), a haze, a moisture permeability, and an air permeability of the multi-layer film with the predicted value of the dart impact strength; when the multi-layer film designed based on the combination of the predicted values satisfies the design requirement, generating the design data based on the combination; and when the multi-layer film designed based on the combination of the predicted values does not satisfy the design requirement, changing the combination to another combination, and redesigning the multi-layer film.

9. The design apparatus of claim 1, wherein: the multi-layer film is modeled as a multi-layer film model, the multi-layer film includes a first single layer film and a second single layer film laminated to each other, the multi-layer film model includes n (n is a multiple of 6) model layers laminated to each other, and the first single layer film corresponds to two or more model layers among the model layers, and the second single layer film corresponds to two or more other model layers among the model layers.

10. The design apparatus of claim 9, wherein: the program code is executed to further: display the model layers on a display device, and provide a user interface that allows a user to set the model layers corresponding to the first single layer film and the model layers corresponding to the second single layer film.

11. A design method for a multi-layer film that designs the multi-layer film based on properties of the multi-layer film predicted using only information about a single layer film, the design method being executed by a computing device including one or more processors and one or more memory devices, the design method including: modeling a laminated structure of the multi-layer film to be designed; collecting dart impact strength data of the single layer film forming the laminate structure as stress-strain data; reading values pre-stored in a storage space accessible by the computing device, and acquiring a feature setting pattern for specifying a different feature setting method according to the read values; setting, as features, a plurality of physical indices selected from the dart impact strength data and a thickness of the single layer film according to the feature setting pattern; selecting at least one of a plurality of supervised learning models capable of regression analysis as a machine learning model; predicting a dart impact strength of the multi-layer film using the machine learning model trained using the features as independent variables and the dart impact strength of the multi-layer film as a target variable; and generating design data of the multi-layer film by combining predicted values of other characteristics with the predicted value of the dart impact strength to satisfy design requirements of the multi-layer film.

12. The design method of claim 11, wherein: when the feature setting pattern includes a first feature setting pattern, the setting as features includes: setting at least some of a yield point, a yield stress, a resilience, a strain hardening, a failure, a strain hardening modulus, a flow energy, a strain hardening energy, a toughness, a 1% modulus, and a 2% modulus, and a thickness of the single layer film as the features.

13. The design method of claim 12, wherein: the selecting as a machine learning model includes: selecting one of a partial least squares (PLS) model, a lasso regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, the lasso regression model, and the random forest model: receives a vector generated from at least some of the yield point, the yield stress, the resilience, the strain hardening, the failure, the strain hardening modulus, the flow energy, the strain hardening energy, the toughness, the 1% modulus, and the 2% modulus, and the thickness of the single layer film, and is trained to predict a dart impact strength of the multi-layer film.

14. The design method of claim 11, wherein: when the feature setting pattern includes a second feature setting pattern, the setting as features includes: setting at least some of a yield stress, a yield strain, a necking stress, a necking strain, a breaking stress, and a breaking strain, and a thickness of the single layer film as the features.

15. The design method of claim 14, wherein: the selecting as a machine learning model includes: selecting one of a partial least squares (PLS) model, a lasso regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, the lasso regression model, and the random forest model: receives a vector generated from at least some of the yield stress, the yield strain, the necking stress, the necking strain, the breaking stress, and the breaking strain, and the thickness of the single layer film, and is trained to predict a dart impact strength of the multi-layer film.

16. The design method of claim 11, wherein: when the feature setting mode includes a third feature setting mode, the setting as a feature includes: setting a yield stress, a yield strain, and an index of the yield stress divided by the yield strain, and a thickness of the single layer film as the feature.

17. The design method of claim 16, wherein: the selecting as a machine learning model includes: selecting one of a partial least squares (PLS) model, a lasso regression model, and a random forest model as the machine learning model, and the machine learning model selected from the PLS model, the lasso regression model, and the random forest model: receives a vector generated from the yield stress, the yield strain, and the index of the yield stress divided by the yield strain, and the thickness of the single layer film, and is trained to predict the dart impact strength of the multi-layer film.

18. The design method of claim 11, wherein: the generating the design data of the multi-layer film includes: combining a predicted value of at least one of a stiffness, a strength, a strain, a tear strength, a high speed dart impact strength, a seal initiation temperature (SIT), a haze, a moisture permeability, and an air permeability of the multi-layer film with the predicted value of the dart impact strength; when the multi-layer film designed based on the combination of the predicted values satisfies the design requirement, generating the design data based on the combination; and when the multi-layer film designed based on the combination of the predicted values does not satisfy the design requirement, changing the combination to another combination, and redesigning the multi-layer film.

19. The design method of claim 11, wherein: the multi-layer film is modeled as a multi-layer film model, the multi-layer film includes a first single layer film and a second single layer film laminated to each other, the multi-layer film model includes n (n is a multiple of 6) model layers laminated to each other, and the first single layer film corresponds to two or more model layers among the model layers, and the second single layer film corresponds to two or more other model layers among the model layers.

20. The design method of claim 19, further comprising: displaying the model layers on a display device, and providing a user interface that allows a user to set the model layers corresponding to the first single layer film and the model layers corresponding to the second single layer film.

Images