WO2023008740A1 - Method of determining elastic constants for anisotropic material - Google Patents

Abstract

The present invention provides a method for determining an elastic constant for an anisotropic material, which can determine an anisotropic elastic constant only with a single core sample and thus can dramatically reduce time and cost.

Description

How to Calculate the Elasticity Constant for Anisotropic Materials

The present invention relates to a method for calculating the elastic constant for an anisotropic material.

Anisotropic rocks such as gneiss and shale exist in various strata, and industries in various resource fields target these strata. Shale, a representative anisotropic rock, contains a large amount of shale gas, and the global market for shale gas was valued at $68.9 billion in 2019 (Grand View Research, 2020). In addition, research is being conducted to build a high-level radioactive waste disposal site in the deep ground worldwide, including countries such as Finland, Sweden, and Korea, and geothermal energy and carbon dioxide underground storage are also growing industrially. In order to further advance these technologies, a high understanding of the ground layer is required.

In particular, in these technologies, it is very important to grasp information about the elastic constant of the rock constituting the ground in order to design drilling methods or conditions. However, since rocks often have anisotropy in which the value of the elastic constant varies depending on the direction, the physical behavior of the entire rock layer can be understood only by grasping the elastic constant of the rock in various directions. Conventionally, in order to determine the elastic constant for each direction of a rock, several cores must be sampled in a plurality of directions, and the elastic constants must be measured for each direction for each core. It was necessary to use a special loader. Accordingly, the application of these methods has a problem of considerable cost, but a level of technology capable of effectively reducing time and cost has not been developed so far.

(Patent Document 1) Patent Publication No. 10-2017-0092830

One aspect of the present invention is to provide a method for calculating the elastic constant for an anisotropic material, which can drastically reduce time and cost by calculating the anisotropic elastic constant using only a single core sample.

The object of the present invention is not limited to the foregoing. Anyone with ordinary knowledge in the technical field to which the present invention belongs will have no difficulty in understanding the additional objects of the present invention from the contents throughout the present specification.

One aspect of the present invention,

The learning strain obtained by applying a local load to the learning anisotropic core sample and the strain of a plurality of data sets consisting of the learning elastic constant of the learning anisotropic core sample as an input value and the elastic constant as an output value Prepare a computer system that performs machine learning doing;

Measuring an actual strain by applying a local load to the anisotropic core sample; and

obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

Including, it provides a method for calculating the elastic constant for an anisotropic material.

In addition, another aspect of the present invention,

Preparing a plurality of data sets consisting of a strain for learning obtained by applying a local load to an anisotropic core sample for learning and an elastic constant for learning of the anisotropic core sample for learning;

causing a computer system to perform machine learning using strains of the plurality of data sets as input values and elastic constants as output values;

Measuring an actual strain by applying a local load to the anisotropic core sample; and

obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

Including, it provides a method for calculating the elastic constant for an anisotropic material.

According to one aspect of the present invention, it is possible to calculate the anisotropic elastic constant using only a single core sample, and it is possible to provide a method for calculating the elastic constant for an anisotropic material, which can significantly reduce time and cost.

Various and beneficial advantages and effects of the present invention are not limited to the above description, and will be more easily understood in the process of describing specific embodiments of the present invention.

1 schematically shows the sequence of a method for calculating the elastic constant for an anisotropic material using artificial intelligence according to an implementation example of the present invention.

Figure 2 is a schematic diagram schematically showing a method for calculating the elastic constant through a concentrated load test, which is an embodiment of the present invention, Figure 2 (a) shows the structure of a flexible platen, Figure 2 (b) shows the core sample 2(c) schematically shows a structure in which a flexible platen is placed on one end of a core sample so as to contact only a part of the surface of one end of the core sample.

3 shows a schematic diagram of a core sample during a concentrated load test, which is an implementation example.

4 is a schematic diagram schematically showing a method for measuring an indirect tensile test, which is an embodiment of the present invention.

Figure 5 shows the attachment position of the strain measuring sensor during the concentrated load test, which is an embodiment of the present invention, and the arrow indicates the load position.

6 shows a conceptual diagram for the direction of the elastic constant and the isotropic plane, which is an embodiment of the present invention.

7 is a schematic diagram of an artificial neural network constructed from a machine learning computer system, which is an implementation example of the present invention.

8 is an average of the elastic constants of a specific combination in order to evaluate the accuracy of the elastic constants of E ₁ , E ₂ , and G ₂ for partial samples in an embodiment of the present invention in order to evaluate the accuracy of the machine learning computer system. After calculating the strain by performing a computer numerical simulation experiment on an inhomogeneous sample with It is a schematic diagram comparing (and its average value) and the elastic constant values calculated by performing a computer numerical simulation experiment.

9 is a computer numerical simulation experiment performed on a heterogeneous sample having an average elastic constant of a specific combination in order to evaluate the accuracy of the elastic constants of ν ₁ and ν ₂ for partial samples in an embodiment of the present invention, and strain When the elastic constant was calculated by substituting the obtained strain into the machine learning computer system, each elastic constant value (and its average value) obtained using the machine learning computer system and the computer numerical simulation experiment were performed to obtain It is a schematic diagram comparing the calculated elastic constant values.

10 shows the attachment position of the strain measuring sensor on the surface of the anisotropic core sample during the concentrated load test in the embodiment of the present invention, and the arrow indicates the load position.

Terms used herein are for describing specific embodiments and are not intended to limit the present invention. Also, the singular forms used herein include the plural forms unless the related definition clearly dictates the contrary.

The meaning of "comprising" as used in the specification specifies a component, and does not exclude the presence or addition of other components.

Unless otherwise defined, all terms including technical terms and scientific terms used in this specification have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. The terms defined in the dictionary are interpreted to have a meaning consistent with the related technical literature and the currently disclosed content.

Hereinafter, preferred embodiments of the present invention will be described. However, the embodiments of the present invention can be modified in many different forms, and the scope of the present invention is not limited to the embodiments described below. In addition, the embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art.

Conventional methods for obtaining the elastic constant of anisotropic rocks have engineering and economical limitations in practical applications. Specifically, in order to calculate the elastic constant of anisotropic rock, the existing methods have limitations such as requiring two or more core samples taken from various directions or requiring the use of a special loader, which is costly. There was a problem that was growing exponentially.

Accordingly, the inventors of the present invention, as a result of intensive examination to solve the above-described problem, have invented a method capable of sufficiently calculating the elastic constant of an anisotropic material even if only a single core specimen is used using a general load machine. In particular, in the case of the elastic constant, it was found that it can be easily obtained by learning the previously analyzed data set without using complicated numerical analysis or a plurality of specimens, and the present invention was completed. It is explained in detail below.

A method for calculating the elastic constant for an anisotropic material according to an aspect of the present invention,

The learning strain obtained by applying a local load to the learning anisotropic core sample and the strain of a plurality of data sets consisting of the learning elastic constant of the learning anisotropic core sample as an input value and the elastic constant as an output value Prepare a computer system that performs machine learning doing;

Measuring an actual strain by applying a local load to the anisotropic core sample; and

obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

Including, it provides a method for calculating the elastic constant for an anisotropic material.

On the other hand, another aspect of the present invention,

Preparing a plurality of data sets consisting of a strain for learning obtained by applying a local load to an anisotropic core sample for learning and an elastic constant for learning of the anisotropic core sample for learning;

causing a computer system to perform machine learning using strains of the plurality of data sets as input values and elastic constants as output values;

Measuring an actual strain by applying a local load to the anisotropic core sample; and

obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

Including, it provides a method for calculating the elastic constant for an anisotropic material.

Figure 1 of the present application schematically shows the procedure of the calculation method of the elastic constant for the anisotropic material, and the configuration of the present invention will be described in detail below.

First, according to one embodiment of the present invention, the strain of the data set consisting of the learning strain obtained by applying a local load to the learning anisotropic core sample and the learning elastic constant of the learning anisotropic core sample as an input value, and the elastic constant as an output value. A step of preparing a computer system performing machine learning may be performed (see FIG. 1A ).

Alternatively, according to one embodiment of the present invention, a step of preparing a plurality of data sets consisting of a learning strain obtained by applying a local load to an anisotropic core sample for learning and an elastic constant for learning of the anisotropic core sample for learning is performed, and then the plurality A step of causing the computer system to perform machine learning using the strain of the data set as an input value and the elastic constant as an output value may be followed (see FIG. 1B).

According to one embodiment of the present invention, the strain for learning and the elastic constant for learning included in the data set may be 100 or more sets in terms of securing accuracy of output values. For greater accuracy, there may be more than 1,000 sets, and in one implementation, more than 1,800 sets. The larger the number of data sets, the larger the size of the secured database, thereby improving the accuracy of calculating the elastic constant for anisotropic materials. Therefore, the upper limit of the number of data sets may not be particularly limited, and as a non-limiting example, 5,000 can be a dog

Also, in the present invention, the strain input to the data set may be referred to as 'learning strain' in order to distinguish it from the 'actual strain' actually measured to calculate the elastic constant of the anisotropic core material. Similarly, elastic constants are also referred to as 'learning elastic constants' that are input to the data set. The elastic constant calculated by the actual strain can be called 'actual elastic constant' apart from this. Of course, the actual strain and the actual elastic constant may also be included in the data set to increase the learning accuracy, and in this case, they may serve as the learning strain and the elastic constant for learning.

According to one embodiment of the present invention, the learning strain included in the data set may be measured under the same load condition or substantially the same load condition. This is because even if the elastic constant is the same, the strain pattern can be different when the loading conditions are different.

There is no particular limitation on how to obtain the strain for learning and the elastic constant for learning. However, as one method, under the condition of applying a local load to at least one of both ends of the anisotropic core sample, while changing the value of the input elastic constant, performing a computer numerical simulation experiment to obtain a strain repeatedly to obtain a plurality of data There are ways to secure a set. At this time, the method described below may be applied in the same way to the method of collecting the anisotropic core sample, the method of attaching the strain measuring sensor, and the method of applying a local load.

At this time, during the computer numerical simulation experiment, equipment commonly used in the art, such as a program for computer numerical simulation of Comsol Multiphysics, may be used. Therefore, according to one embodiment of the present invention described above, even if numerical analysis using a complicated formula used in the existing method for calculating the elastic constant is not applied, it is relatively easy and accurate using a commonly used computer numerical simulation experiment. A high elastic constant can be calculated.

The computer system refers to an artificial intelligence program or a system equipped with the same. As the computer system, machine learning may be performed in advance by using the strain for learning included in the data set as an input value and the elastic constant for learning as an output value. Therefore, the computer system can calculate the actual elastic constant with high accuracy when the actual strain obtained under the same load condition is input.

The aforementioned machine learning is a term generally widely used in artificial intelligence, and the method is not particularly limited in the present invention. However, one method of machine learning applicable to the present invention will be described as follows.

According to an implementation example of the present invention, an artificial neural network that calculates an elastic constant as an output value when a strain is input as an input value can be constructed by performing machine learning by training the data set with artificial intelligence. The principle on which such an artificial neural network is built is similar to a regression problem for a general linear function. That is, to create a function that tells the relationship between input values and output values of given data sets. The artificial intelligence learning method is a widely known method such as machine learning, and any method among previously known methods may be used in the present invention. Machine learning is a very well-known method, but in brief, it has the following concept. For example, assuming that there is data between study hours and test scores, we want to model the relationship between study hours and test scores using these data. At this time, it can be considered that the relationship between the study time and the test score has a linear function (test score = a × study time + b) relationship, followed by finding the values of a and b. The values of a and b are obtained at the point where the difference between the linear function describing the relationship between the actual data and the actual data is the minimum, and the process of finding the values of a and b at this time uses an optimization method called gradient descent. done through After first guessing the values of a and b, the difference between the linear function obtained from them and the actual data is obtained, and the values of a and b are corrected through gradient descent. From this, by repeating the process of obtaining the difference between the linear function and the actual data, the values of a and b are finally found. Although a simple example has been described in the above description, the artificial neural network also has a principle similar to the above-described example, and a schematic diagram of the artificial neural network used for artificial intelligence learning, which is an implementation example of the present invention, is shown in FIG. 7 .

In the input layer of the first column among the five columns vertically in FIG. 7 herein, the number of nodes is determined according to the number of strain measuring sensors attached to the surface of the core sample. 7 shows an example in which 8 strain measuring sensors are attached to the surface of an anisotropic core sample, and accordingly, 8 strain values are input to the input layer of the first column of FIG. 7 . As each of these columns is connected to the next column, an artificial neural network is created, and a modeling that predicts one elastic integer from the final column obtained from this is built. Therefore, when calculating a total of 5 elastic constants, since the artificial neural networks shown in FIG. 7 are independently performed for each elastic constant, 5 artificial neural networks are independently formed.

As such, eight strain values are input to the input layer of the first column in FIG. 7, and then each node in the input layer of the first column is connected to each node of the second column by a line, and through each line Node values in the second column are determined using the strain values. The first node present at the top of the second column follows the same calculation as the following relational expression A.

[relational expression A]

value1 = 1/[1 + exp(-(a11×ε1 + a12×ε2 + … + a18×ε8))]

(In the relational expression A, value 1 represents the value of the first node in the second column, a11 to a18 represent the weight of each strain for determining the value of the first node in the second column, respectively, and ε1 to ε8 are 8, respectively. represents two strain values.)

Subsequently, the second node of the second column also follows the same calculation, and specifically follows the same calculation as the following relational expression B.

[relational expression B]

value2 = 1/[1 + exp(-(a21×ε1 + a22×ε2 + … + a28×ε8))]

(In relational expression B, value 2 represents the value of the second column, second node, and the definitions of a21 to a28 and ε1 to ε8 are respectively for the second node of the second column, except for the relational expression A Same as definition.)

This same process can be repeated from the third column to the last column. Subsequently, a11, a12, . . . , a18, a21, a22, … By changing the values of unknown factors such as a28, a function called artificial neural network finds the relationship between input and output of real data.

To this end, initially arbitrarily a11, a12, ... To find the difference between the calculated output value and the actual data output by giving values to a28, etc., and to reduce the difference, a11, a12, ... Modify values such as a28. In this way, the process of correcting the values of the unknowns using the difference between the output values is called back-propagation. While performing such backpropagation, the artificial intelligence learns the process of the artificial neural network finding a relationship between real data.

[relational expression C]

(In the relational expression C, E represents the sum of squares of the difference between the actual value and the predicted value, c represents the c-th data set, j represents the j-th elasticity constant, and y _j,c represents the j-th elasticity of the c-th data set represents the predicted value of an integer, and d _j,c represents the actual value of the j-th elastic constant of the c-th data set. Since each artificial neural network predicts only one elastic integer, j is set from 1 to 1.)

Through the artificial intelligence learning process as described above, when calculating the elastic constant for an anisotropic material in the conventional method, as opposed to requiring a separate operation by a real user, according to the present invention, machine learning without a separate operation by an actual user By simply inputting the measured strain into the computer system, the predicted value of the elastic constant can be immediately obtained. Therefore, since it can be easily used in combination with existing laboratory equipment, it is possible to drastically reduce time and cost.

Then, in order to calculate the actual elastic constant of the new anisotropic core sample whose elastic constant is not known using the computer system, a step of measuring an actual strain by applying a local load to the anisotropic core sample is performed.

According to one embodiment of the present invention, the above-described actual strain may be measured under the same load condition as the load condition used when obtaining the strain for learning and the elastic constant for learning. At this time, the step of measuring the actual strain may be performed prior to obtaining the plurality of data sets consisting of the above-described strain for learning and the elastic constant for learning, or after obtaining the plurality of data sets.

According to one embodiment of the present invention, when the actual strain is measured, the shape and size of the anisotropic core sample, the strain measurement position (ie, the strain measuring sensor attachment position), the strain measurement direction, etc. It may be the same as when obtaining the elastic constant for learning. Alternatively, the shape and size of the anisotropic core sample when the actual strain is measured, the strain measuring position (ie, the strain measuring sensor attachment position), the strain measuring direction, etc. may be slightly different from when obtaining the strain for learning and the elastic constant for learning. The different cases will be described in more detail later. At this time, the method described below may be applied in the same way to the method of collecting the anisotropic core sample, the method of attaching the strain measuring sensor, and the method of applying a local load.

Subsequently, a step of obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system is performed.

That is, the actual strain of the anisotropic core sample measured by applying the aforementioned local load can be input to the computer system that performed the aforementioned machine learning, and the computer system outputs an estimate of the actual elastic constant based on the already learned artificial neural network. can do. In one embodiment of the present invention, the predicted value output may be calculated as an actual elastic constant.

According to one embodiment of the present invention, elastic constants that can be calculated using the computer system include Young's modulus (E), shear modulus (G), Poisson's ratio (ν), and the like, , A conceptual diagram of the direction of the elastic constant and the isotropic plane is shown in FIG. As an example, the number of elastic constants that can be calculated in the present invention may be five, and examples thereof include E ₁ , E ₂ , G ₂ , ν ₁ and ν ₂ . As shown in FIG. 6, E ₁ , E ₂ , and G ₂ are the Young's modulus on the xz plane, the Young's modulus on the y-axis, and the shear modulus on the xy plane or yz plane, respectively, when the isotropic plane is placed parallel to the xz plane. it means. In addition, ν ₁ is the first Poisson's ratio as a negative value of the ratio of strain in the z-axis direction to strain in the x-axis when a uniaxial compression test parallel to the x-axis is performed, and ν ₂ is a uniaxial compression test parallel to the y-axis. , the second Poisson's ratio is a negative value of the ratio of the strain in the z-axis or x-axis direction to the strain in the y-axis.

As described above, according to the present invention, by calculating the elastic constant for an anisotropic material using artificial intelligence, it is economical and simple compared to the conventional elastic constant calculation method, and the elastic constant with relatively low error rate and excellent accuracy is effectively calculated. can do.

On the other hand, although not particularly limited, optionally, the step of evaluating the accuracy of the computer system may be further included.

As a method for evaluating the accuracy of the above-described machine learning computer system, as an example, a strain for evaluation obtained by applying a local load to an anisotropic core sample for evaluation and a plurality of evaluations consisting of an elastic constant for evaluation of the anisotropic core sample for evaluation Prepare the data set. In this case, the number of data sets for evaluation may be smaller than the number of data sets described above, and as an example, may be 6 or more, 50 or more, or 100 or more. However, like the upper limit of the number of data sets described above, the upper limit of the number of data sets for evaluation is also preferable for accuracy evaluation as the value is higher, so this is not particularly limited. However, as an example, the upper limit of the number of data sets for evaluation may be smaller than the upper limit of the number of the plurality of data sets composed of the strain for learning and the elastic constant for learning.

In addition, although not particularly limited, the evaluation data set may be obtained by repeatedly performing a computer numerical simulation experiment to obtain a corresponding strain for evaluation while changing the value of the elastic constant for evaluation. Here, the above description can be equally applied to the method of conducting the computer numerical simulation experiment.

After securing the above-described data set for evaluation, the value of the elastic constant obtained and predicted from the above-described machine learning computer system is compared with the value of each elastic constant obtained from the above-described data set for evaluation, and from the following relational expression 1 A case where the calculated error rate is 10% or less is evaluated as pass. At this time, the elastic constants for confirming the error rate include Young's modulus (E), shear modulus (G), and the like.

[Relationship 1]

[|E _true - E _predicted |/ E _true ] × 100 (%)

(E _true represents the value of each elastic constant obtained from the evaluation data set for each elastic constant, and E _predicted represents the value of the predicted elastic constant obtained from the machine learning computer system for each elastic constant.)

After securing the above-described data set for evaluation, the value of the elastic constant obtained and predicted from the above-described machine learning computer system is compared with the value of each elastic constant obtained from the above-described data set for evaluation, and from the following relational expression 2 A case where the calculated error rate is 0.03 or less is evaluated as pass. At this time, the elastic constant for confirming the error rate below includes Poisson's ratio (ν) and the like.

[Relationship 2]

|E _true - E _predicted |

(In the above relational expression 2, E _true represents the value of each elastic constant obtained from the evaluation data set for each elastic constant, and E _predicted represents the value of the predicted elastic constant obtained from the machine learning computer system for each elastic constant indicate.)

Meanwhile, hereinafter, a method for measuring the strain for learning and the actual strain will be described in detail. That is, according to one embodiment of the present invention, since the strain for learning and the actual strain can be measured under the same load condition as described above, the following description can be applied equally to the measurement of the strain for learning and the actual strain. In addition, it is not particularly limited thereto, and as long as the desired effect can be achieved in the present invention, even if the learning strain and the actual strain measurement conditions are slightly different, it may be included in the scope of the present invention. It will be described in more detail below.

First, an anisotropic core sample (or a core sample for learning) is prepared. The anisotropic core sample may be taken from an anisotropic material, or a commercially available anisotropic core sample may be used. Such anisotropic materials include rocks having anisotropy such as gneiss and shale. In addition, the anisotropic core sample may be collected using a coring method well known in the art and may have a columnar shape. At this time, the cross section of the pillar shape is not particularly limited, but may have a circular shape due to the nature of the coring method. However, it is not necessarily limited to a circular shape, and may have various cross sections by changing the sampling method or post-processing. In addition, the columnar shape referred to in the present invention generally means a shape in which the length in the axial direction is longer than the width of the cross section, and is not particularly limited as long as it can be recognized as a columnar shape in the art.

In addition, the shape and size of the anisotropic core sample described above in the present invention are not limited, and can be applied to various shapes and sizes of samples. That is, in one embodiment of the present invention, if the anisotropic core sample is applied to the computer system that performed the above-described machine learning and can be accurately reproduced to obtain the actual elastic constant, the shape and size of the anisotropic core sample are not particularly limited. , the elastic constant calculation method according to the present invention can also be applied to anisotropic core samples of various shapes and sizes.

In addition, according to one embodiment of the present invention, the anisotropic core sample may be the same as the anisotropic core sample for learning. Alternatively, if it is applied to the computer system that performed the above-described machine learning and is accurately reproducible in obtaining the actual elastic constant, even if the anisotropic core sample is different from the learning anisotropic core sample, it may be included in the scope of the present invention. As an example, if an indirect tensile test described later is applied when measuring the strain for learning, the anisotropic core sample for learning may be a partial sample taken from the anisotropic core sample. On the other hand, if the concentration load test is applied to both the measurement of the strain for learning and the actual strain, the diameter of the anisotropic core sample for learning and the anisotropic core sample may be the same, and the length in the axial direction is about 9 cm or more. As long as they are satisfied, they may be slightly different within a range not impairing the object of the present invention.

According to the research results of the present inventors, the anisotropic material having a unique elastic constant exhibits a unique strain behavior by an applied load. By using this, the elastic constant of the anisotropic material can be obtained without using two or more core samples taken from various directions like conventional methods, and thus time and cost can be greatly reduced. To this end, in the present invention, a local load is applied to the anisotropic core sample (or the anisotropic core sample for learning). Hereinafter, a method of applying a local load in the present invention will be described in detail.

According to one embodiment of the present invention, a load is applied to both ends of the prepared anisotropic core sample (or an anisotropic core sample for learning, hereinafter, the display of 'anisotropic core sample for learning' is omitted), and the load is an anisotropic core sample can be applied at both ends of At this time, at least one of the loads applied from both ends of the anisotropic core sample may be a local load applied only to a part of the end surface instead of the entire end surface. As described above, since there is a close relationship between the strain behavior obtained by the local load and the elastic constant of the anisotropic core sample, when obtaining the above-described data set, the strain for learning and the elastic constant for learning can be stored as one data set. . A plurality of such data sets are acquired and stored in a database, so that artificial intelligence can be used for learning (eg, machine learning).

There is no particular limitation on how to obtain the strain and elastic constant data sets. However, if one non-limiting method is described, the following method may be used. That is, in one embodiment of the present invention, the strain is measured while applying a uniform load to the entire cross section of the sample. However, when a uniform load is applied to the entire cross-section of the sample, the same stress state is applied to all locations of the material, so in order to calculate the elastic constant for each direction of the material, two or more core samples are prepared and several experiments are performed. it was necessary to do However, as in the present invention, by applying a local load to at least one of both ends of the sample, a different stress state is formed for each position of the material based on one anisotropic core sample, so even in the case of an anisotropic material, a single The elastic constant in each direction can be easily calculated with only the core sample of

According to one embodiment of the present invention, for example, as shown in FIGS. 2 (a) to (c) herein, the flexible pressure plate 20 may be provided so as to contact only a portion of the end surface to which the load is applied ( see Figure 2(c)). Therefore, since the flexible platen 20 partially occupies the end surface 101 of the core sample 100 (ie, the surface in the axial direction X with respect to the core sample 100), the core sample 100 The end surface 101 of ) has an area 102 in contact with the soft platen 20 and an area 103 not in contact with the soft platen. At this time, the end surface 101 means the surface of the sample viewed from the direction in which a load is applied to the core sample. In addition, the axial direction (X) may be the same as the direction in which the load is applied to the core sample (corresponding to 'Load' in FIG. 2) during the concentrated load test (except for the indirect tensile test) .

According to one embodiment of the present invention, the flexible platen 20 is not particularly limited, but has a lower Young's modulus than that of the sample, but a yield stress equal to or greater than that of the sample (ie, a yield stress greater than or equal to that of an anisotropic material). use the material If the above conditions are satisfied, the material of the flexible platen is not particularly limited, but since the flexible platen can be easily manufactured in a desired shape by a 3D printing method, as an example, the flexible platen can be made of a material used as a material for a 3D printer. there is. Veroclear can be selected and used as one example that can be advantageously used as a flexible platen. By using a flexible pressure plate that satisfies these conditions, a load applied from a loader described later can be uniformly and concentratedly transmitted to a specific portion of the core sample. Due to this, the flexible platen can effectively calculate the elastic constant of the sample by controlling the sample not to be destroyed under high stress despite the concentrated load. In addition, since the above-described flexible platen is relatively inexpensive, it can be easily used to calculate the elastic constant without a large increase in manufacturing cost.

According to one embodiment of the present invention, the Young's modulus (Y ₁ ) of the flexible platen is less than 1/10 of the Young's modulus (Y ₀ ) of the core sample (ie, Y ₁ /Y ₀ ≤ 1/10) . For example, considering the Young's modulus of a general rock, 3D printer materials such as the aforementioned Veroclear, or materials such as PC, ULTEM ^TM 9085 Resin, and ULTEM ^TM 1010 Resin may be used. When the Young's modulus of the flexible platen exceeds 1/10 of the Young's modulus (Y ₀ ) of the core sample (ie, when the Y ₁ /Y ₀ exceeds 1/10), the load transmitted to the contact surface becomes heterogeneous. However, there may be a problem of increasing inaccuracy in reproducing the experiment by computer numerical simulation. The smaller the value of Y ₁ /Y ₀ is 1/10 or less, the more uniform the load transmitted in the contact area becomes, which is preferable in terms of numerical simulation. However, there is a possibility that the contact area increases due to the Poisson effect (a phenomenon in which the pressure plate expands laterally during a compression test). However, in the case of experiments using the above-mentioned materials such as Veroclear, ULTEM ^TM 9085 Resin, and ULTEM ^TM 1010 Resin, the Poisson phenomenon was negligible, and in reality, the Young's modulus was much smaller than the above-mentioned materials such as Veroclear, and the yield stress was still high. Given that it is difficult to find materials of this level, it may be meaningless to calculate the lower limit of Y ₁ /Y ₀ . Therefore, in the present invention, the lower limit of the value of Y ₁ /Y ₀ is not separately limited. However, as a non-limiting example, when the allowable limit of the amount of change in the width of the pressurized area (corresponding to 'Width' in FIG. 3) is about 1 mm, the lower limit of Y ₁ is approximately 0.5 GPa, and Y ₁ /Y ₀ The lower limit of may be approximately 1/100.

According to one embodiment of the present invention, the flexible pressure plate 20 is provided to contact only a portion of the end surface 101, but the flexible pressure plate 20 may be positioned differently according to the coring direction of the core sample. For example, the area where the flexible platen contacts the core sample is the farthest from the point closest to one end surface of the core sample among the points of the curve 600 where the isotropic surface 500 of the core sample meets the circumferential surface. A straight line drawn from each point toward the end surface may include a line segment connecting the two points where the end surface meets the end surface.

Specifically, as shown in FIGS. 5 (b) to (d), when the coring direction φ of the x-axis exceeds 0° and is 45° or less, as shown in FIG. 3 of the present application, the isotropic surface 500 of the core sample and The flexible platen 20 is placed side by side on the area 1000 where the curve 600 where the circumferential surfaces of the sample (ie, surfaces other than both end surfaces of the core sample) meet is parallel to the end surface 101 of the sample. can be located. In addition, as shown in FIGS. 5 (e) to (f), when the coring direction φ of the x-axis exceeds 45 ° and is 90 ° or less, the flexible platen ( 20) can be placed side by side.

In addition, as shown in FIG. 5(a), when the coring direction φ is 0°, since the entire curve 600 is parallel to the end surface 101, in this case, the end surface of the core sample The flexible platen 20 can be positioned on the 101 in any direction.

On the other hand, as shown in FIG. 5(g), when the coring direction φ is 90°, the area where the flexible platen contacts the core sample is a curve 600 where the isotropic surface 500 of the core sample meets the surface in the circumferential direction It may include a line segment connecting two points that meet the surface of one end of the core sample. That is, the end surface 101 of the core sample in a direction parallel to the line segment connecting the two points where the isotropic surface 500 of the core sample meets the circumferential surface and the line segment connecting the two points where the curve 600 meets one end surface of the core sample A flexible platen 20 may be placed on the top. Therefore, when the flexible platen 20 is placed on the end surface 101 of the sample, the center of the area 102 where the end surface 101 of the core sample 100 contacts the soft platen 20 and the end of the sample The centers of the surfaces 101 can be made coincident.

According to one embodiment of the present invention, as one method of measuring the above-described strain, a plurality of (ie, two or more) different attachment positions and attachment directions (one or more) are attached to the anisotropic core sample. After attaching the strain measuring sensor, there is a method to measure the strain value. The strain measuring sensor is also called a strain gauge, and refers to a device attached to the surface of a core sample and measuring strain at that point. In the present invention, in order to explain the attachment position of the above-described strain measuring sensor, a form in which the strain measuring sensor 10 is attached to the core sample 100 is schematically shown in FIG. 2 (b).

According to one embodiment of the present invention, as can be seen in FIG. 2 (b), the strain measuring sensor 10 is attached to and attached to each other on a surface other than both end surfaces of the anisotropic core sample 100 Two or more are attached such that at least one of the directions is different, so that strain at the point of attachment can be measured.

Alternatively, when the anisotropic core sample is cylindrical, on the surface in the circumferential direction (Y) excluding both end surfaces of the anisotropic core sample 100 (ie, the circumferential surface), (attachment position to each other and Two or more strain measuring sensors 10 (different in at least one of attachment directions) may be attached to measure the strain at the attachment point. That is, the strain measurement can be applied to any surface of the core sample within a range that does not impair the object of the present invention, as long as it is not the surface of both ends to which the load is applied.

At this time, two or more strain measuring sensors 10 attached to the two or more points may be attached to the same part on the surface of the anisotropic core sample 100 in different measurement directions, or two or more of the core sample 100 Each may be attached to two or more different points on the surface.

According to one embodiment of the present invention, in order to calculate the elastic constant with relatively high accuracy for an anisotropic material using only a single anisotropic core sample, the attachment position and number of the strain measuring sensors can be controlled. That is, the number of strain measuring sensors attached to the surface of the anisotropic core sample may be plural (ie, 2 or 3 or more), and the plurality of strain measuring sensors may differ in at least one of the attachment position and direction of attachment. can

Meanwhile, since accuracy improves as the value of the number of strain measuring sensors increases, an upper limit on the number of strain measuring sensors may not be separately limited. However, as an example, the number of strain measuring sensors is 200 on the assumption that a strain gauge from Kyowa with a length of 16 mm and a width of 5.2 mm is tightly attached to the sample surface based on the case of using a sample with a radius of 54 mm and a height of 108 mm. Can be used up to dogs. However, since the number of channels corresponding to the sensors of the data acquisition system used in actual experiments is usually several to dozens, it can be set to less than 100.

In addition, according to one embodiment of the present invention, in order to improve the accuracy of the elastic constant, the number of positions for attaching strain measuring sensors on the surface of the core sample may be two or more, more preferably three or more. .

According to one embodiment of the present invention, when the strain measuring sensor 10 is attached, the axial direction from the end surface 101 of the anisotropic core sample (or the surface of the side where the soft platen with respect to the core sample contacts) (One or more) strain measuring sensors 11 or 12 are attached to the first position 1 separated by a predetermined distance in (X). In one embodiment of the present invention, the strain measuring sensor 11 or 12 is 0.5 times to the width of the flexible platen 20 (corresponding to 'Width' in FIG. 3) from the end surface 101 of the core sample to the core sample. It may be attached at a first location less than 1.5 times the diameter of . At this time, the width of the flexible platen 20 (corresponding to 'Width' in FIG. 3) is the shortest distance of the narrow side based on the area where the flexible platen 20 contacts the end surface 101 of the core sample. can mean

On the other hand, when the first position is less than 0.5 times the width of the flexible platen 20 from the end surface of the anisotropic core sample, the unevenness of the load transmitted by the flexible platen due to experimental errors may cause a problem that greatly affects the strain can In addition, when the first position is greater than 1.5 times the diameter of the core sample from the end surface of the anisotropic core sample, the effect of the concentrated load is dispersed, resulting in a decrease in the accuracy of calculating the elastic constant.

Then, (one or more) additional strain measuring sensors 21 or 22 can be attached at a second position 2 further away in the axial direction X from the end surface 101 than the first position. . In this way, the strain value for each point measured from the strain measuring sensor attached to two or more different points can be used to predict the elastic constant by injecting it as an input value of a machine learning computer system described later.

One implementation example of the attachment form of the above-described strain measuring sensor is shown in FIG. 3 . As can be seen in FIG. 3 , according to one embodiment of the present invention, two or more strain measuring sensors 11 and 12 having different strain measuring directions may be attached to the first position 1 . Similarly, two or more strain measuring sensors 21 and 22 having different strain measuring directions may be attached to the second position 2 as well.

Here, the different directions of measuring the strain may mean that the directions of the strain designed to be measured by the strain measuring sensor attached to the surface of the anisotropic core sample are different. For example, when a strain gauge is used as a strain measuring sensor, it may mean that the direction in which the grids of the strain gauge are aligned is the direction of the strain the strain gauge is designed to measure, and the directions of the grids are different from each other. For example, at least one of two or more strain measuring sensors attached to any one location (eg, location (1)) may have the same strain measurement direction as the axial direction (X) of the core sample, , Another one may be the same as the direction in which the strain measurement direction is perpendicular to the axial direction (X).

That is, as can be seen in FIG. 3, the strain measuring sensor 11 and the strain measuring sensor 12 are attached to the first position 1 at the same distance from the end surface 101 of the anisotropic core sample, but each strain is measured. Since the strain directions designed to be measured by the sensors are different from each other, this corresponds to an example in which the strain measurement directions are different. For example, in FIG. 3 , the strain measuring sensor 11 corresponds to an example in which the strain measuring direction is perpendicular to the axial direction X, and the strain measuring sensor 12 corresponds to the strain measuring direction in the axial direction. Corresponds to the same (or parallel) example as (X). Therefore, in FIG. 3, the strain measuring sensors 21 and 22 are also attached to each other at the same position as the second position 2, but the strain measuring directions are different from each other, and the strain measuring sensors 31 and 32 and (33) also correspond to examples in which the attachment positions are the same, but the strain measurement directions are different.

On the other hand, in the present specification, in the form of attaching the strain measuring sensors so that the strain measuring directions are different from each other at any one of the above positions, the strain measuring directions need not be the same, and the desired effect in the present invention can be achieved. If this can be achieved, various changes can be made in the measurement direction of the strain, and this is not particularly limited.

In addition, according to one embodiment of the present invention, when the strain measuring sensor 10 is attached, the axis relative to the portion 102 in contact with the flexible pressure plate 20 on the end surface 101 of the core sample. (One or more) strain measurements at locations A (on the circumferential surface) included in parallel region 1000 in direction X (e.g., corresponding to locations (1) and (2) in FIG. 3). A sensor 10 may be attached.

Then, on the end surface 101 of the core sample, not included in the parallel area 1000 in the axial direction X to the portion 102 with which the flexible platen 20 is in contact (on the circumferential surface) ) (one or more) additional strain measuring sensors may be attached to position B (for example, corresponding to position (3) in FIG. 3). That is, the B position is the circumferential direction (Y ) can be located on the surface. At this time, as shown in FIG. 3, the position A may be the same as the first position described above, or may be the same as the first position and the second position.

Alternatively, according to one embodiment of the present invention, two or more strain measuring sensors (different from each other at least one of the attachment position and attachment direction) may be attached to the A position. Alternatively, three or more strain measuring sensors (where one or more of the attachment positions and attachment directions are different from each other) may be attached to the B position.

One form in which three or more strain measuring sensors are attached to the position B is shown in FIG. 3, and three strain measuring sensors 31, 32, and 33 having different strain measuring directions are attached to position (3). You can check. At this time, the above description may be equally applied to the measurement direction of the strain.

Alternatively, as an example in which the number of the above-described attached strain measuring sensors is three or more, as shown in FIG. 3, in the axial direction with respect to the portion 102 where the flexible platen 20 contacts the end surface 101 of the core sample. Two or more strain measuring sensors are attached to a position A included in the parallel area 1000 to (X), and included in the area 2000 other than the parallel area 1000 (ie, the parallel area (1000) may include a form in which one or more strain measuring sensors are attached to position B). At this time, at least one of the attachment position and attachment direction of the strain measuring sensors attached to the A position may be different from each other.

On the other hand, according to one embodiment of the present invention, in terms of improving the accuracy of the elastic constant secured, the number of strain measuring sensors attached to the surface of the core sample may be 5 or more, in which case Therefore, it is possible to make it 7 or more.

As one implementation example of controlling the number of the attached strain measuring sensors to 5 or more, the core sample is at a portion 102 in contact with the flexible pressure plate 20 on the end surface 101 of the core sample. A position (1 and 2 in FIG. 3) included in the parallel area 1000 in the axial direction (X) for the axis and a B position not included in the parallel area 1000, and the above-described A position And three or more strain measuring sensors (different from each other at least one of the attaching position and attaching direction) at least one position selected from the position B, and one or more strain measuring sensors may be attached to the remaining positions. For example, after attaching three strain measuring sensors (at least one of the attachment positions and attaching directions different from each other) to at least one position selected from positions A and B, and attaching one strain measuring sensor to the remaining positions, One or more strain measuring sensors may be additionally attached to at least one of the positions A and B so that at least one of the attachment position and direction is different from the strain measuring sensor already attached to at least one position.

Alternatively, as another embodiment, it is included in the area 1000 parallel to the axial direction (X) of the portion 102 in contact with the end surface 101 of the core sample. Three or more strain measuring sensors may be attached to position A (on the core sample surface (circumferential direction (Y) surface)) as much as possible. In addition, the end surface 101 of the core sample is not included in the area 1000 parallel to the area 1000 in the axial direction X with respect to the portion 102 in contact with the flexible platen 20 (the surface of the core sample (circumference) One or more strain measuring sensors (31, 32) having different strain measurement directions at position B (or, position B included in the region 2000 other than the parallel region 1000) on the surface in the direction (Y)) , 33) can be attached.

Alternatively, as another embodiment in which the number of attached strain measuring sensors is 5 or more, in the A position, a certain distance away from the end surface 101 of the core sample in the axial direction (X) (the surface of the core sample ( One or more strain measuring sensors (at least one of 11 or 12) may be attached at a first location (1) on the circumferential (Y) surface). At this time, when two or more strain measuring sensors are attached to the first position, strain measuring directions may be different from each other. In addition, during the A position, one or more strain measuring sensors may be attached to a second position (2) further away from the end surface (101) of the core sample in the axial direction (X) than the first position, At this time, at least one of the strain measuring sensors attached to the second location may have the same strain measuring direction as a direction perpendicular to the axial direction X (ie, corresponding to 21 in FIG. 3 ). On the other hand, although not particularly limited, the measurement sensor attached to the B position and the strain measurement sensor attached to the second position have the same shortest distance from the end surface 101 of the core sample in the axial direction (X). (At this time, the shortest distance described above is measured based on the center in the measurement direction of strain of each measuring sensor). Therefore, as one implementation example in which the number of the above-described attached strain measuring sensors is 5 or more, in FIG. 3, the strain measuring sensors are at least {one or more of (11) and (12)}, (21), (31) , (32), (33).

As one implementation example of controlling the number of the attached strain measuring sensors to be 7 or more, the axial direction (X ), but included in the parallel region 1000 (corresponding to position A), at a first position 1 spaced a certain distance from the end surface 101 of the core sample in the axial direction (X), two or more strains Measurement sensors 11 and 12 may be attached. In addition, it is included in the parallel area 1000 in the axial direction (X) with respect to the portion 102 in contact with the flexible platen 20 to the end surface 101 of the core sample (corresponding to position A), Two or more strain measuring sensors 21 and 22 may be attached to a second position 2 farther from the end surface 101 in the axial direction X than the first position. In addition, a third position (2000) not included in the parallel area 1000 in the axial direction (X) with respect to the portion 102 where the flexible platen 20 contacts the end surface 101 of the core sample. Three or more strain measuring sensors 31, 32 and 33 may be attached to (3).

At this time, when a plurality of the strain measuring sensors are attached, one or more of the attachment position and attachment direction may be different from each other. In addition, although not particularly limited, the measuring sensor attached to the third position (3) and the strain measuring sensor attached to the second position (2), from the end surface 101 of the core sample in the axial direction (X ) may be the same (at this time, the shortest distance described above is measured based on the center in the measurement direction of the strain of each measuring sensor). Therefore, as one implementation example in which the number of the above-described attached strain measuring sensors is 7 or more, the strain measuring sensors are at least in FIG. 3, (11), (12), (21), (22), (31), ( 32) and (33).

On the other hand, as described above, by controlling the number of strain measuring sensors and the number of attachment positions, it is possible to calculate the elastic constant for an anisotropic material with only a single core sample, thereby dramatically reducing time and cost, as well as , it is possible to increase the accuracy of the elastic constant calculated from the anisotropic material.

However, in the present invention, the number and position of the strain measuring sensor are not limited to the above-described form, and can be changed in various forms. As an example, various examples of changing the attachment position and number of strain measuring sensors during a concentrated load test are shown in FIG. 5 . In FIG. 5, the attachment position on the y-axis of the strain measuring sensor in the concentrated load test along the coring direction of the x-axis is shown.

That is, after the flexible platen 20 is brought into contact with only a portion of the end surface 101 of the core sample 100 described above, on the soft platen 20, the axial direction (X) of the core sample 100 After applying a concentrated load, which is a local load, by , the strain value is measured by the strain measuring sensor 10 (see FIG. 2(b)). The direction of applying the aforementioned concentrated load (corresponding to 'Load' in FIG. 2) may be the same as the axial direction (X) of the core sample. At this time, a concentrated load test method commonly known in the art can be equally applied to the present invention, except that a local load is applied using the above-described flexible platen 20.

As such, by applying the concentrated load test method of applying a local load to the core sample 100 using the above-described flexible platen 20, without using additional experimental equipment such as a loader with specially controlled conditions, Even using a commonly known compression loader, the elastic constant of an anisotropic material can be calculated relatively accurately. That is, the present invention has the advantage that it can be realized without adding a large cost because the compressive loader and the strain measuring sensor used in general experiments using compressive loads can be used as they are.

Meanwhile, according to one embodiment of the present invention, when obtaining the plurality of data sets, an indirect tensile test may be additionally performed on the partial sample 200 additionally taken from the anisotropic core sample for learning.

Specifically, according to one embodiment of the present invention, during the indirect tensile test, applying a load from both ends in the radial direction 50 of the partial sample 200; and measuring a strain value on the surface of the partial sample 200 to which the load is applied. The measurement method of this indirect tensile test is schematically shown in FIG. 4 .

At this time, according to one embodiment of the present invention, when measuring the strain value on the surface of the partial sample 200, the surface on the surface 500 in the axial direction 60 of the columnar partial sample 200 After attaching two or more strain measuring sensors 71 and/or 72 to, strain values at each attachment point may be measured.

In the indirect tensile test, the number of strain measuring sensors 71 and 72 attached to the axial surface of the partial sample may be two or more, and in this case, the two or more strain measuring sensors are attached to each other in the position and direction of attachment. At least one (more than one) may be different. That is, the strain measuring sensors may be attached to the same attachment location so that strain measurement directions are different from each other, or may be attached to different attachment locations.

The description of the measurement direction of the strain can be applied in the same manner as described above, except for the fact that it is an indirect tensile test. In addition, the axial direction 60 of the partial sample 200 in the indirect tensile test may coincide with the axial direction X of the core sample in the above-described concentrated load test.

Meanwhile, in the present specification, except for the above description, conventional methods in the art may be equally applied to the indirect tensile test.

As such, by additionally conducting an indirect tensile test following the intensive load test, the accuracy of the elastic constant calculated from the anisotropic material can be further improved.

According to the above-described method for calculating the elastic constant of the present invention, even a single core specimen obtained from an anisotropic rock through one-time coring is sufficiently applicable, and only a small cost is required to manufacture a specially designed ductile pressure plate. And cost can be greatly reduced, so it can be applied very economically. Thus, the present invention has significant economic advantages compared to existing methods.

Therefore, according to the present invention, since the average elastic constant can be calculated economically and simply for anisotropic materials, it is highly applicable to rocks handled in fields such as rock engineering, petroleum engineering, and resource engineering. In addition, since it can be applied to various types of samples, it can be widely applied in fields such as construction environment engineering and materials engineering.

Hereinafter, the present invention will be described in more detail through examples. However, it should be noted that the following examples are only for explaining the present invention through examples, and are not intended to limit the scope of the present invention. This is because the scope of the present invention is determined by the matters described in the claims and the matters reasonably inferred therefrom.

(Example 1)

After preparing Asan gneiss as an anisotropic material, the rock was cored to obtain a cylindrical anisotropic core sample having a length of 10.8 cm in the axial direction and a diameter of 5.4 cm. Subsequently, a flexible pressure plate having a width of 5.4 cm, a length of 2.2 cm, and a height of 2 cm was prepared using Veroclear manufactured by a 3D printer. The flexible pressure plate was brought into contact with only a part of the surface of either end of both ends of the collected anisotropic core sample. The isotropic plane of the rock was cored so that 1) φ = 70 degrees and 2) φ = 90 degrees with the surface on which the rock sample is placed.

Subsequently, a strain measuring sensor was attached on the surface (ie, the circumferential surface) of the core sample except for the end surface, and the specific attachment position of the strain measuring sensor is shown in FIG. 10a (corresponding to '1) 70 degrees' and It is shown in Figure 10b (corresponding to '2) 90 degrees' above. Thereafter, a local load was applied to the flexible platen in the axial direction of the core sample using a compressive strength tester from MTS. Then, the strain value was measured with the strain measuring sensor.

Next, using Comsol's Multiphysics program, the same diameter (i.e., the diameter of the end surface: 5.4 cm), height (i.e., axial length: 10.8 cm), load condition, and strain A numerical model with measurement locations was created.

In the numerical model, the strain was obtained by changing the combination of elastic constants. A total of 1800 data sets are obtained, and one data set contains 5 elastic constants and 8 strains (10 when indirect tensile tests are added). The elastic constant combinations of the 1800 data sets are shown in Tables 1 and 2 below, respectively.

Tables 1 and 2 show what values (E, E', υ, υ', G') were put into each elastic constant and data were created. Then, 5 × 5 × 3 × 3 × 4 = 900 combinations of elastic constants that can be created for each table, and 1800 data sets are formed in total. (E=E1, E'=E2, υ = υ1, υ'= υ2, G'=G2)

E(GPa)	40	44	48	52	56
E'(GPa)	16	20	24	28	32
υ	0.1	0.2	0.3	-	-
υ'	0.1	0.2	0.3	-	-
G'(GPa)	10	12	14	16	-

E(GPa)	64	67	70	73	76
E'(GPa)	48	51	54	57	60
υ	0.1	0.2	0.3	-	-
υ'	0.1	0.2	0.3	-	-
G'(GPa)	14	16	18	20	-

After obtaining a data set for training in which the number of the aforementioned data sets is 1800, the data set was used to train a computer program by machine learning. Through this, an artificial neural network is formed, and 8 strains (10 when indirect tension is included) are input to the input layer of the artificial neural network thus formed, and 1 elastic constant is output to the output layer. Five artificial neural networks were created according to the five elastic constants. The structure of this artificial neural network is shown in FIG. 7 .

In order to evaluate whether learning was completed, the accuracy was evaluated using [Relationship 1] and [Relationship 2] described above. After preparing a total of 6 unlearned evaluation data sets as shown in Table 3, the elastic constant values obtained by putting them into the artificial neural network described in Table 4 below were compared with the elastic constant values in the evaluation data set.

for evaluation data	set 1	set 2	set 3	set 4	set 5	set 6
E1 (GPa)	45.8	38.1	68.4	74.2	69.9	62.3
E2 (GPa)	16.5	20.1	19.2	20.6	52.6	55.1
ν1	0.13	0.23	0.19	0.18	0.23	0.27
v2	0.23	0.14	0.13	0.19	0.26	0.21
G2 (GPa)	12	6.2	12.1	17.5	16.8	17

artificial neural network output value	set 1	set 2	set 3	set 4	set 5	set 6
E1 (GPa)	45.800	38.100	68.397	74.183	69.900	62.300
E2 (GPa)	16.498	20.112	19.202	20.612	52.600	55.100
ν1	0.127	0.224	0.190	0.179	0.232	0.271
v2	0.238	0.116	0.115	0.185	0.270	0.213
G2 (GPa)	12.001	6.171	12.099	17.489	16.800	17

All the elastic constant values of the six sets produced by the artificial neural network are compared with the elastic constant values in the actual evaluation data to satisfy [Relational Expression 1] and [Relational Expression 2]. Through this, it was judged that the learning was completed.

After learning is completed, the elastic constant is calculated by inputting the strain values obtained from the actual experiment into the artificial neural network. The results obtained by following the above procedure are shown in Tables 5 to 8 below.

At this time, the definitions of E ₁ , E ₂ , and G ₂ are Young's modulus on the xz plane, Young's modulus on the y-axis, and the xy plane or yz plane, respectively, when the isotropic plane is placed parallel to the xz plane as shown in FIG. is the shear modulus of elasticity at ν ₁ is the first Poisson's ratio as a negative value of the ratio of strain in the z-axis direction to strain in the x-axis when a uniaxial compression test parallel to the x-axis is performed, and ν ₂ is a uniaxial compression test parallel to the y-axis. The second Poisson's ratio is the negative value of the ratio of the strain in the z-axis or x-axis direction to the strain in the y-axis.

8 and 9 show the accuracy of the elastic constants of E ₁ , E ₂ , and G ₂ and the elastic constants of ν ₁ and ν ₂ from Example 1 after preparing 100 additional non-learned heterogeneous evaluation data sets. The results of each evaluation of accuracy are shown. Specifically, FIG. 8 is a computer numerical simulation experiment performed on a heterogeneous sample having an average elastic constant of a specific combination to obtain a strain, and then substituting the obtained strain into a machine learning computer system to calculate the elastic constant. The elastic constant values of E ₁ , E ₂ , and G ₂ calculated using the learned computer system (squares in FIG. 8) and their average values (thick lines in FIG. 8) and elasticity obtained by performing computer numerical simulation experiments Integer values (dotted lines) are compared. 9 is a machine learning computer Each elastic constant value of ν ₁ and ν ₂ calculated using the system (square in FIG. 9) and their average value (bold line in FIG. 9), and elastic constant values obtained by performing a computer numerical simulation experiment (dotted line) is compared to It was confirmed from FIGS. 8 and 9 that the average value of each elastic constant was similar to the elastic constants obtained from the actual machine learning computer program.

(Comparative Example 1)

Using a conventional method, core samples having the same shape were taken in the same manner as in Example 1, but two core samples were taken so that the respective coring angles were 0 ° and 45 °. For a sample with a coring angle of 0 °, two strain measuring sensors in different directions were attached to the center of the sample, and for a sample with a coring angle of 45 °, two sensors with different strain measurement directions were placed at position (3) in FIG. A strain measuring sensor was attached, and two strain measuring sensors having different strain measuring directions were attached to position (2) in FIG. 3 .

After performing a uniaxial compression test in which a uniform load is applied to both ends of each sample, the strain value was measured with the strain sensor, and the elastic constant value determined by applying the relational expression between each strain and elastic constant to the least square method It was determined by the elastic constant value of the core sample. The results obtained by following the above procedure are shown in Tables 5 to 8 below.

(Comparative Example 2)

After taking an anisotropic core sample in the same manner as in Example 1, the same local load was applied using the same flexible platen, and a strain measuring sensor was attached to the same position.

Thereafter, the strain value is measured with the strain measuring sensor, and the input elastic constant value is changed using the Comsol Multiphysics program, and a computer numerical simulation experiment is performed under the same load condition as that applied to the core sample. The strain at was calculated. The elastic constant value when the sum of the squared residuals between the measured strains and the calculated strains was the smallest was determined as the elastic constant value of the core sample. On the other hand, the Gauss-Newton method was used to optimize the elastic constant, and the substitution method was used to optimize the angle. The results obtained by following the above procedure are shown in Tables 5 to 8 below.

(When φ = 70 degree coring is performed and indirect tensile test is not included)

	E1 (GPa)	E2 (Gpa)	ν1	v2	G2 (Gpa)
Example 1	52.3	35.2	0.34	0.08	14.5
Comparative Example 1	57.57	32.99	0.1376	0.1177	16.2
Comparative Example 2	47.9	34.4	0.21	0.07	13.0

(When performing φ = 70 degree coring and including indirect tensile test)

	E1 (GPa)	E2 (GPa)	ν1	v2	G2 (GPa)
Example 1	46.4	21.8	0.13	0.12	14.7
Comparative Example 1	57.57	32.99	0.1376	0.1177	16.2
Comparative Example 2	45.0	26.4	0.16	0.12	12.6

(If φ =90 degree coring is performed and indirect tensile test is not included)

	E1 (GPa)	E2 (GPa)	ν1	v2	G2 (GPa)
Example 1	51.8	18.0	0.16	0.05	23.0
Comparative Example 1	57.57	32.99	0.1376	0.1177	16.2
Comparative Example 2	52.1	14.9	0.15	0.05	29.0

(When performing φ =90 degree coring and including indirect tensile test)

	E1 (GPa)	E2 (GPa)	ν1	v2	G2 (GPa)
Example 1	46.0	23.6	0.15	0.10	24.7
Comparative Example 1	57.57	32.99	0.1376	0.1177	16.2
Comparative Example 2	51.7	23.8	0.13	0.13	19.8

As can be seen from the experimental results of Tables 5 to 8, in the case of Example 1 using an artificial neural network, it was confirmed that very similar results were obtained compared to Comparative Example 1 using an existing method. Through this, it can be seen that the artificial neural network properly calculated the elastic constant.

In addition, Example 1 using an artificial neural network has an advantage over Comparative Example 1. That is, in the experimental method, the existing method requires excessive time and cost because samples with different coring angles must be taken, but the method using artificial neural networks uses only a single core sample, so time and cost can be reduced. .

In addition, compared to the method of Comparative Example 2, there is an advantage when the elastic constant has to be calculated for a large amount of samples. In the case of Comparative Example 2, since optimization is performed for each case when the strain is input from each sample, there is a disadvantage in that calculation takes a long time when a large amount of samples are to be analyzed. In contrast, in the case of the method using the artificial neural network as in Example 1 of the present application, since the artificial neural network has already completed learning using the learning data (1800 data), even if a large number of cases are entered, the result can be confirmed immediately after entering the value. could For this reason, according to the present invention, the analysis time could be drastically shortened.

[Description of code]

100: core sample

X: axial direction of the core sample

Y: circumferential direction of the core sample

10: strain measurement sensor

20: flexible pressure plate

101: end surface (or axial direction (X) surface) of core sample

102: at the end surface 101 of the core sample, a region in contact with the soft platen 20

103: on the end surface 101 of the core sample, a region not in contact with the soft platen 20

1, 2, 3: Position where the strain measurement sensor is attached

11, 12, 21, 22, 31, 32, 33: strain measurement sensor

71, 72: strain measurement sensor

200: partial sample

50: In the case of an indirect tensile test, the radial direction of the partial sample 200

60: axial direction of the partial sample 200 during indirect tensile test

300: the axial direction (60) surface of the part sample (200)

500: isotropic surface of the core sample

600: curve formed by the isotropic surface 500 of the core sample on the surface in the circumferential direction (ie, the surface excluding both end surfaces of the core sample)

71, 72: strain measurement sensor

1000: area parallel to the core sample in the axial direction (X) with respect to the portion 102 in contact with the flexible platen 20

2000: Area 2000 that is not a parallel area 1000 in the axial direction X with respect to the portion 102 where the flexible platen 20 contacts the core sample

Claims (13)

1. The learning strain obtained by applying a local load to the learning anisotropic core sample and the strain of a plurality of data sets consisting of the learning elastic constant of the learning anisotropic core sample as an input value and the elastic constant as an output value Prepare a computer system that performs machine learning doing;

   Measuring an actual strain by applying a local load to the anisotropic core sample; and

   obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

   Comprising, a method for calculating the elastic constant for an anisotropic material.
2. Preparing a plurality of data sets consisting of a strain for learning obtained by applying a local load to an anisotropic core sample for learning and an elastic constant for learning of the anisotropic core sample for learning;

   causing a computer system to perform machine learning using strains of the plurality of data sets as input values and elastic constants as output values;

   Measuring an actual strain by applying a local load to the anisotropic core sample; and

   obtaining an actual elastic constant by inputting the actual strain into the machine-learned computer system;

   Comprising, how to calculate the elastic constant for an anisotropic material.
3. According to claim 1 or 2,

   The method of calculating the elastic constant for an anisotropic material, wherein the number of the plurality of data sets is 1,000 or more.
4. According to claim 1 or 2,

   The actual strain is measured under the same load condition as the load condition used when measuring the strain for learning, a method for calculating the elastic constant for an anisotropic material.
5. According to claim 1 or 2,

   The plurality of data sets are obtained by repeatedly performing a computer numerical simulation experiment to obtain a corresponding learning strain while changing the value of the learning elastic constant, Method for calculating the elastic constant for an anisotropic material.
6. According to claim 1 or 2,

   The anisotropic core sample for learning and the method for calculating the elastic constant for the anisotropic material, using only one core sample.
7. According to claim 1 or 2,

   The local load is that at least one of the loads applied at both ends of the anisotropic core sample is applied only to a part of the end surface rather than the entire end surface, Calculation method of the elastic constant for the anisotropic material.
8. According to claim 1 or 2,

   The method of calculating the elastic constant for an anisotropic material, wherein the local load is provided with a soft pressure plate such that it contacts only a part of the end surface to which the local load is applied to the anisotropic core sample.
9. According to claim 8,

   The Young's modulus of the flexible platen is less than 1/10 of the Young's modulus of the core sample.
10. According to claim 1 or 2,

    A method for calculating the elastic constant for an anisotropic material by attaching two or more strain sensor to the surface of the anisotropic core sample and measuring the strain value at each attachment point.
11. According to claim 10,

    The method of calculating the elastic constant for the anisotropic material, wherein the number of strain measuring sensors attached to the surface of the anisotropic core sample is 5 or more.
12. According to claim 1 or 2,

    When obtaining the plurality of data sets, additionally performing an indirect tensile test on a partial sample additionally taken from the anisotropic core sample for learning, a method for calculating the elastic constant for the anisotropic material.
13. According to claim 1 or 2,

    A method for calculating the elastic constant for an anisotropic material, further comprising evaluating the accuracy of the machine-learned computer system.

Images