KR102423096B1 - Method For Virtual Data Generation of Crack Patterns - Google Patents

Abstract

The present invention provides a virtual defect data generation method for virtually generating defect image data of a ceramic surface generated in a ceramic press process, wherein at least one required input value regarding a two-dimensional model and input data regarding a contour profile are inputted. Step, a second step of generating a two-dimensional contour profile according to the data input from the first step, a two-dimensional defect model by applying a gradient descent method algorithm to analyze the two-dimensional contour profile generated from the second step A third step of extracting and digitalizing the two-dimensional defect model into digital data, a fourth step of extracting and classifying defect information in the two-dimensional model in which the two-dimensional defect model is digitally digitized, and the extracted and classified defect information into each two-dimensional model a fifth step of generating a star data set; It is characterized in that it includes.
Accordingly, a sufficient amount of data may be secured by virtually generating a pattern for a surface defect occurring in the ceramic press process. In addition, by learning through machine learning based on virtual defect data obtained through big data, information on defect patterns for various properties can be secured, thereby minimizing the defect rate for surface defects occurring in the press process.

Description

Method For Virtual Data Generation of Crack Patterns

The present invention relates to a method for generating virtual defect data.

In the current era of big data analysis and artificial intelligence technology (especially deep learning-related research among machine learning), securing data is an essential condition.

In particular, in the field of image information, for example, companies that have secured big data (eg, Google, Facebook, etc.) are researching and utilizing the latest artificial intelligence technologies such as deep learning.

Image big data is used as training data for artificial intelligence technologies to learn cognition and reasoning capabilities of computer systems. Currently, these artificial intelligence technologies are showing performance that approaches or exceeds human capabilities in areas such as speech recognition, image recognition, and text interpretation. Due to its high cognitive performance, the technology is currently recognized as having an innovative impact that will replace many service jobs.

However, there is a problem in that it is difficult to apply the latest technology because it is difficult to secure a sufficient amount of learning data in industries other than companies or fields where big data can be secured through social networks, especially manufacturing related industries.

On the other hand, lowering the defect rate in the manufacturing process in the ceramic material parts industry is one of the most important issues for reducing cost and period.

Specifically, in order to minimize surface defects (cracks) occurring in the ceramic press process, researches using machine learning are actively conducted on big data of defect generation patterns of the results in the press process. In this process, the most difficult and important task is to secure a large amount of data (eg, defect images), but there was a problem that actual data collection in the field had to be very limited.

An object of the present invention is to provide a method for generating virtual defect data in which a sufficient amount of data can be secured by virtually generating a pattern for a surface defect occurring in a ceramic press process, which was created to solve the above problem. is to do

Another object of the present invention is to provide a method for generating virtual defect data that can secure a defect image with a shape as similar as possible to a surface defect pattern that may occur during product manufacturing by generating data on surface defects using a gradient descent algorithm. is to do

Another object of the present invention is to minimize the defect rate for surface defects occurring in the press process by securing information on defect patterns for various properties by learning through machine learning based on virtual defect data obtained through big data. It is to provide a method for generating virtual defect data that can

Other detailed objects of the present invention will be clearly grasped and understood by experts or researchers in this technical field through the specific contents described below.

The present invention provides a virtual defect data generation method for virtually generating defect image data of a ceramic surface generated in a ceramic press process, wherein at least one required input value regarding a two-dimensional model and input data regarding a contour profile are inputted. Step, a second step of generating a two-dimensional contour profile according to the data input from the first step, a two-dimensional defect model by applying a gradient descent method algorithm to analyze the two-dimensional contour profile generated from the second step A third step of extracting and converting the two-dimensional defect model into digital data, a fourth step of extracting and classifying defect information from the digital dataized two-dimensional defect model, and converting the extracted and classified defect information into a data set for each two-dimensional model a fifth step of generating

The first step is a step 1-1 of determining the size (Nx, Ny) of the two-dimensional model, a step 1-2 of determining the number of defects (Nc) and the length of the defects (Lc) of the two-dimensional model, and steps 1-3 of determining the number Nd of the two-dimensional contour profile and steps 1-4 of determining the required number of two-dimensional models.

The second step is a 2-1 step of generating coordinates of a plurality of point charges located in the set two-dimensional model, setting an arbitrary charge amount, and deriving a potential distribution by the coordinates of the plurality of point charges and the amount of charge Step 2-2 and Step 2-3 of generating a two-dimensional contour profile based on the derived dislocation distribution.

The second step is the second step of generating coordinates of a plurality of point charges located within the set size of the two-dimensional model, the second step of setting an arbitrary amount of charge, the coordinates of the plurality of point charges using the finite element analysis method and It is characterized in that it comprises a step 2-B of deriving a potential distribution by performing electrostatic analysis with the amount of electric charge as an input value and a step 2-C of generating a two-dimensional contour profile based on the derived potential distribution.

The second step is characterized in that a two-dimensional contour profile is generated based on a fractal technique, Random Midpoint Displacement Fractal.

The third step is a 3-1 step of selecting an arbitrary n-th coordinate for generating an n-th defect with respect to a given contour profile, and a 3-2 step of searching for a dislocation value at the n-th coordinate , step 3-3 of searching for potential values at the n+q coordinate adjacent to the n-th coordinate, and determining the n-th defect by comparing the potential value of the n-th coordinate with the potential value of the n+q coordinate It is characterized in that it includes steps 3-4.

The two-dimensional defect model is characterized in that the positions of the n-th defect to the n+q-th defect extracted from the contour profile are displayed on the two-dimensional model.

According to the present invention, a sufficient amount of data can be secured by virtually generating a pattern for a surface defect occurring in a ceramic press process.

In addition, by generating data on surface defects using a random step algorithm, it is possible to secure a defect image in the form of a surface defect pattern that may occur during product manufacturing as much as possible.

In addition, by learning through machine learning based on virtual defect data obtained through big data, information on defect patterns for various properties can be secured, thereby minimizing the defect rate for surface defects occurring in the press process.

1 is a flowchart of an embodiment of generating a two-dimensional contour profile using an equation of Coulomb's law among virtual defect data generating methods according to an embodiment of the present invention.
2 is a flowchart of an embodiment of generating a two-dimensional contour profile using a finite element analysis method among virtual defect data generation methods according to an embodiment of the present invention.
3 is a diagram illustrating a two-dimensional contour profile generated by a finite element analysis method.
4 is a flowchart of an embodiment of generating a two-dimensional contour profile based on a random midpoint displacement fractal, which is a fractal technique, among methods for generating virtual defect data according to an embodiment of the present invention.
5 is a diagram illustrating a two-dimensional contour profile generated by a random midpoint displacement method.
6 is a conceptual diagram that can assist in the description of the third step according to an embodiment of the present invention.
7 is a diagram illustrating a two-dimensional defect model extracted from a two-dimensional contour profile using a gradient descent algorithm.
8 is a view showing the extracted two-dimensional defect model on the two-dimensional model.
9 is a table in which defect-related information secured by a method of generating virtual defect data is classified.

Since the present invention can have various changes and can have various forms, specific embodiments are illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to the specific disclosed form, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as "comprises" or "comprising" are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification is present, but one or more other features It is to be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, elements, parts, or combinations thereof.

In addition, the defect data generation method described in the present application may be implemented as a computing device having a control unit, a calculation unit, etc., but is not limited thereto, and the subject performing each specific step is each step for rapid simulation and calculation. It may be a separate configuration that performs the

Meanwhile, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention belongs.

Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

With reference to the drawings below, preferred embodiments of the present invention will be described in more detail.

The method for generating virtual defect data according to an embodiment of the present invention is largely divided into five steps. Representatively, it includes an input step (S100), a contour profile generation step (S200), a two-dimensional defect model extraction step (S300), a defect information extraction and classification step (S400), and a data set generation step (S500). Hereinafter, each step will be described in order.

The input step (the first step) ( S100 ) is a step in which at least one required input value for the two-dimensional model and input data for the contour profile are input. In the data input in this step (S100), the size (Nx, Ny) of the two-dimensional model, the number of defects (Nc) in the two-dimensional model, and the length (Lc) of the defects are input. The conditions for the size of the 2D model, the number of defects, and the length of the defects are as follows.

(1) The number of defects (Nc) is a natural number. For example, it may have a range of Nc = 1 to 5.

(2) The length of the defect (L _c ) < the horizontal size of the two-dimensional model (L _x ), and L _c < the vertical size of the two-dimensional model (L _y ) are satisfied. That is, it must be smaller than the maximum size of the two-dimensional model. For example, if L _x = 500, L _y = 500, then L _c =50, 100, 150, 200, 250

Then, the number of contour profiles to be created (N _d ) is input. The number of contour profiles (N _d ) in the embodiment of the present invention is 10. When all variables are determined, the total number of generated two-dimensional models N _total satisfies Equation 1 below.

[Equation 1]

N _total = (N _d ) * (# of N _c ) * (# of L _c * N _r )

(Where # of Nc is the number of defects (N _c ) (in the embodiment of the present invention, it is 5), # of L _c is the number of lengths (in the embodiment of the present invention, it is 5), N _r denotes an arbitrary number (in the embodiment of the present invention, it is 4).

When substituting the variables illustrated above, N _total is (N _d ) * (# of N _c ) * (# of L _c * N _r ) = 10 * 5 * 4 * 5 = 1000.

Contour profile generating step (second step) (S200) is a step of generating a two-dimensional contour profile according to input data. According to an embodiment of the present invention, there are three methods for generating a two-dimensional contour profile.

(1) A method of establishing an equation for Coulomb's law and generating a contour profile for the equation

1 is a flowchart of an embodiment of generating a two-dimensional contour profile using an equation of Coulomb's law among virtual defect data generating methods according to an embodiment of the present invention.

The method for generating a contour profile according to an embodiment of the present invention generates arbitrary point charges having an arbitrary amount of charge at a random location, calculates a potential value due to Coulomb force by each of the point charges, and calculates the calculated It means a method of setting the potential value according to the position as a contour profile.

In step 2-1 ( S210 ), coordinates of a plurality of point charges located in the set two-dimensional model are generated, and an arbitrary amount of charge is set. For example, (Example of equipotential lines), a plurality of point charges located in a region satisfying 1 < x _n < N _x , 1 < y _n < N _y are generated. It is assumed that each point charge has respective charge quantities Q ₁ , Q ₂ ,... Q _n . Assuming three point charges, the three point charges have values (x ₁ , y ₁ , Q ₁ ), (x ₂ , y ₂ , Q ₂ ), and (x ₃ , y ₃ , Q ₃ ). (Where x, y, n are real numbers.)

Step 2-2 ( S220 ) is a step of deriving a potential distribution based on the coordinates of a plurality of point charges and the amount of charge. That is, it means calculating the potential profile due to the Coulomb interaction of each point charge.

Step 2-3 ( S230 ) is a step of generating a two-dimensional contour profile based on the derived dislocation distribution. The potential value V _{(i, j)} at a specific arbitrary point (x, y) satisfies the expression of Coulomb's law. For all pixels with coordinates (i, j) (1 <= i <= N _x , 1 <= j <= N _y ), calculate the potential value V _{(i, j)} for the data (i, Create j, V _{(I, j)} ).

(2) A method of generating a two-dimensional contour profile using the finite element analysis method

2 is a flowchart of an embodiment of generating a two-dimensional contour profile using a finite element analysis method among virtual defect data generating methods according to an embodiment of the present invention.

The method for generating a contour profile according to an embodiment of the present invention generates arbitrary point charges having an arbitrary amount of charge at a random location, and applies finite element analysis (FEA) to each of the point charges using finite element analysis (FEA). Conduct electrostatic analysis by This refers to a method of setting the potential value calculated through simulation as a contour profile. Step 2-A 210 ′ and step 2-C 230 ′ according to an embodiment of the present invention are similar to steps 2-1 , 210 and 2-3, but are calculated through equations There is a difference in that the electrostatic finite element analysis is performed at a given location rather than that.

3 is a diagram illustrating a two-dimensional contour profile generated by a finite element analysis method. Later, a gradient descent method based on a change in the slope of the potential value may be applied.

(3) This is a method of generating a two-dimensional contour profile based on the fractal technique, Random Midpoint Displacement Fractal.

The height of the grid (2 x 2, 4 x 4, 8 x 8, 16 x 16, ...2 ⁿ x 2 ⁿ ) of which the 2D surface increases to the power of 2 n when the random midpoint method is applied to the fractal surface. can create From this, a contour profile expressed in x, y, and z values can be generated.

4 is a flowchart of an embodiment of generating a two-dimensional contour profile based on a random midpoint displacement fractal, which is a fractal technique, among methods for generating virtual defect data according to an embodiment of the present invention.

5 is a diagram illustrating a two-dimensional contour profile generated by a random midpoint displacement method.

The two-dimensional defect model extraction step (third step) (S300) is a step of extracting a two-dimensional defect model by analyzing a two-dimensional contour profile by applying a gradient descent method algorithm, and converting the two-dimensional defect model into digital data. Hereinafter, the operation in this step will be described with reference to FIGS. 6 to 8 .

The step of selecting an arbitrary n-th defect coordinate (step 3-1) (S310) is an arbitrary n-th coordinate ( Choose P _n = (i ₁ , j ₁ )).

The step of searching for the dislocation value at the nth defect coordinate (step 3-2) (S320) is the dislocation value (V _n(in ) at the nth coordinate (P _n = (i ₁ , j ₁ )) in the contour profile. _{, jn)} ).

The step of searching for dislocation values at the n+q-th defect coordinates (step 3-3) (S330) is a step of searching for dislocation values at the n+q-th coordinates adjacent to the n-th coordinates. In this case, the n+q-th coordinate adjacent to the n-th coordinate may be expressed as P _n+q = (i _1+q , j _1+q )). Referring to FIG. 6 , the n+q-th coordinate means coordinates adjacent to the n-th coordinate in the vertical, left, and right directions. For example, if it is assumed that n is 1 in the first execution step, the adjacent second coordinates of the first coordinate are P _a = (i ₁ -1, j ₁ ), P _b = (i ₁ +1, j, respectively) ₁ ), P _c = (i ₁ , j ₁ -1), and P _d = (i ₁ , j ₁ +1) may be provided. The third coordinate means a coordinate of a position adjacent to the second coordinate, and is selected so that there is no overlapping coordinate.

Comparing the potential values and determining the nth defect (steps 3-4) (S340) is the n-th coordinate (P _n ) of the potential value (V _n ) and the n+q coordinate (P _n+q ) This is a step of determining the nth defect by comparing the potential values (V _n+q ).

First, potential values V _a , V _b , V _c , and V _d at adjacent coordinates are searched for. Find the smallest potential value (V _min ) and the corresponding point coordinate P _min =(i _min , j _min ) among potential values V _a , V _b , V _c , and V _d , and find the smallest potential value V _min and the potential value V ₁ as a reference is compared.

At this time, when V _min >= V ₁ , it is determined that the first coordinate is a defect. On the other hand, if V _min < V ₁ , P ₂ =(i ₂ , j ₂ ) = P _min is set. In other words. Determining that the first coordinate is not a defect means proceeding to determine a defect at the second coordinates. By repeating this process, coordinates P ₁ , P ₂ ... P _n+q corresponding to the nth defect can be found. Since the above coordinates are all adjacent coordinates, it can be seen that the junction of all points is the length of the nth defect. However, the length of the nth defect must be less than or equal to the length L _c of the largest defect initially set. (Distance P _n P _n+q <= L _c )

Steps 3-2 (S320) to (S340) are repeated until the initially set number of defects (N _c ). That is, repeating this process determines the presence or absence of a defect by comparing the dislocation values in all regions of the full (N _x , N _y ) size two-dimensional model with adjacent dislocation values, and derives the length of the defect.

It means to continuously search for defects until the number of derived defects becomes equal to the number of defects (N _c ) set in the first step ( S100 ). A two-dimensional defect model is derived by finding all defects, and the derived two-dimensional defect model is shown in FIG. 7 . This two-dimensional defect model is displayed on the two-dimensional model. This is shown in FIG. 8 .

In the case of an embodiment of the present invention, if it corresponds to a defect, it is expressed as 1, and if it is not a defect, it is expressed as 0. The data property of the position corresponding to the coordinates of the defects on the bitmap corresponding to the two-dimensional model size (L _x , L _y ) and the overall coordinate data property is set to 0 is changed to 1. That is, according to the presence or absence of a defect, data is expressed in binary. However, there may be other embodiments in which defects have arbitrary color values (RGB values) in addition to being expressed in binary numbers, which may be digitized in various ways.

Defect information extraction and classification step (fourth step) (S400) extracts defect information, such as the number of defects, the length of the defect, position coordinates, etc., and classifies it according to each condition. Such an embodiment is shown in FIG. 9 .

The data set creation step (S500) is a step of generating the extracted and classified defect information as a data set for each two-dimensional model.

As described above, the present invention has been described with specific matters such as specific components, limited embodiments, and drawings, but these are only provided to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments No, various modifications and variations are possible from these descriptions by those of ordinary skill in the art to which the present invention pertains.

Therefore, the spirit of the present invention should not be limited to the described embodiments, and not only the claims described below, but also all of the claims and equivalents or equivalent modifications will be said to belong to the scope of the spirit of the present invention.

Claims (7)

A virtual defect data generation method for virtually generating defect image data of a ceramic surface generated in a ceramic press process, comprising:
A first step of inputting at least one or more required input values for the two-dimensional model and input data for the contour profile;
a second step of generating a two-dimensional contour profile according to the data input from the first step;
a third step of extracting a two-dimensional defect model by analyzing the two-dimensional contour profile generated from the second step by applying a gradient descent method algorithm, and converting the two-dimensional defect model into digital data;
a fourth step of extracting and classifying defect information in the digital dataized two-dimensional defect model; and
A fifth step of generating the extracted and classified defect information as a dataset for each two-dimensional model,
The first step is
Step 1-1 of determining the size (Nx, Ny) of the two-dimensional model;
Step 1-2 of determining the number of defects (Nc) and the length (Lc) of the defects of the two-dimensional model;
Steps 1-3 of determining the number (Nd) of the two-dimensional contour profile; and
Steps 1-4 of determining the number of required two-dimensional models; delete The method of claim 1,
The second step is
a 2-1 step of generating coordinates of a plurality of point charges located in the set two-dimensional model and setting an arbitrary amount of charge;
a 2-2 step of deriving a potential distribution based on the coordinates and the amount of charge of the plurality of point charges; and
Step 2-3 of generating a two-dimensional contour profile based on the derived dislocation distribution; The method of claim 1,
The second step is
a step 2-A of generating coordinates of a plurality of point charges located within the set size of the two-dimensional model and setting an arbitrary amount of charge;
Step 2-B to derive a potential distribution by performing electrostatic analysis using the coordinates and charge amounts of the plurality of point charges as input values using the finite element analysis method; and
A method of generating virtual defect data comprising: a step 2-C of generating a two-dimensional contour profile based on the derived dislocation distribution. The method of claim 1,
The second step is
A method for generating virtual defect data, characterized in that a two-dimensional contour profile is generated based on a fractal technique, Random Midpoint Displacement Fractal. The method of claim 1,
The third step is
a 3-1 step of selecting an arbitrary n-th coordinate for generating an n-th defect with respect to a given contour profile;
a 3-2 step of searching for a potential value at the n-th coordinate;
a 3-3 step of searching for potential values at an n+q-th coordinate adjacent to the n-th coordinate; and
and a third-fourth step of determining an n-th defect by comparing the potential value of the n-th coordinate with the potential value of the n+q-coordinate. 7. The method of claim 6,
The two-dimensional defect model is a method for generating virtual defect data, characterized in that the positions of the n-th defect to the n+q-th defect extracted from the contour profile are displayed on the two-dimensional model.

Images