KR20210157224A - Apparatus and method for detecting fault - Google Patents

Abstract

An apparatus for detecting a fault according to an embodiment of the present invention comprises: a measurement unit configured to measure an eddy current for each position in a measurement section from a first position to a second position with respect to multiple objects under inspection; and a control unit configured to receive an eddy current measurement value for each position with respect to each of the multiple objects under inspection from the measurement unit, generate multiple eddy current profiles representing corresponding relations between eddy current measurement positions and eddy current measurement values based on the received eddy current measurement value for each position, extract multiple property values corresponding to multiple factors from the generated multiple eddy current profiles, generate multiple factor groups according to combination of the multiple factors, determine an estimation state of each of the multiple objects under inspection by using a corresponding property value and a preset classification model for each of the generated multiple factor groups, determine fault detection performance on the multiple objects under inspection for each of the multiple factor groups on the basis of determined multiple estimation states and preset reference states for each of the multiple objects under inspection, and set any one among the multiple factor groups as a reference factor group for the multiple objects under inspection on the basis of the fault detection performance determined for each of the factor groups. The present invention can enhance accuracy and reliability in detecting a fault of the object under inspection.

Description

Fault detection apparatus and method {APPARATUS AND METHOD FOR DETECTING FAULT}

The present invention relates to an apparatus and method for detecting a defect, and more particularly, to an apparatus and method for non-destructively detecting a defect in an object using an eddy current.

In general, an inspection apparatus measures a characteristic of interest from a product to determine the quality of the product, and selects a defective product by using the measured characteristic of interest as an index. In this case, the inspection performance of the inspection device depends on how accurately the characteristic of interest can be measured.

In addition, even if the inspection device accurately measures the characteristic of interest, there is a problem in that the quality of the product is not clearly distinguished only by the single characteristic. For example, when a measured single characteristic value is a value near the boundary of a reference value, a problem of classifying a product that is actually normal as a defective product or classifying a product that is actually defective as a normal product occurs.

In order to solve this problem, even if the measurement conditions set in the inspection device are adjusted, there is a limit in judging the quality of a product using only a single index. This is because the quality of a product can be evaluated based on multiple indicators, and when the quality of a product is judged using only a single indicator, the quality of the product is judged without considering the other indicators at all.

In order to solve a problem that may occur when inspecting the quality or performance of a product using only a single indicator, it is necessary to develop a technology capable of inspecting the quality or performance of a product using at least two indicators.

The present invention has been devised to solve the above problems, and an object of the present invention is to provide a defect detection apparatus and method for detecting a defect in a subject by using a plurality of factor groups.

Other objects and advantages of the present invention may be understood by the following description, and will become more clearly understood by the examples of the present invention. In addition, it will be readily apparent that the objects and advantages of the present invention can be realized by means and combinations thereof indicated in the claims.

A defect detection apparatus according to an aspect of the present invention includes: a measuring unit configured to measure an eddy current for each position in a measurement section from a first position to a second position with respect to a plurality of inspected objects; and receiving eddy current measurement values for each position of each of the plurality of inspected objects from the measurement unit, and generating a plurality of eddy current profiles indicating a corresponding relationship between eddy current measurement positions and eddy current measurement values based on the received eddy current measurement values for each position and extracting a plurality of feature values corresponding to a plurality of factors from the generated plurality of eddy current profiles, generating a plurality of factor groups according to a combination of the plurality of factors, and for each of the generated plurality of factor groups, corresponding The predicted state of each of the plurality of subjects is determined using a feature value and a preset classification model, and based on the determined plurality of predicted states and a preset reference state for each of the plurality of subjects, the plurality of factor groups Defect detection performance for the plurality of inspected objects is determined every time, and any one of the plurality of factor groups is set as a reference factor group for the plurality of inspected objects based on the determined defect detection performance for each of the plurality of factor groups It may include a control unit configured to do so.

The control unit determines whether the determined predicted state and the preset reference state are the same for each of the plurality of inspected objects, and determines the defect detection performance for the plurality of inspected objects for each of the plurality of factor groups It may be configured to calculate an outflow rate and an overdetection rate for

The control unit may be configured to set a factor group having the lowest calculated outflow rate among the plurality of factor groups as a reference factor group for the plurality of subjects.

When there are a plurality of factor groups having the lowest calculated outflow rate, the control unit sets the factor group having the lowest calculated overdetection rate among the plurality of factor groups having the lowest calculated outflow rate as a reference factor group for the plurality of subjects. can be configured to be set to .

The control unit, when the plurality of classification models are provided, determines the defect detection performance of the plurality of objects to be inspected with respect to the plurality of factor groups for each of the plurality of classification models, and based on the determined defect detection performance, the plurality of classification models By setting any one of the classification models of , as the reference classification model, the reference factor group and the reference classification model for the plurality of subjects may be set.

The control unit classifies the plurality of subjects into a learning subject and a detection subject, and classifies the plurality of feature values extracted from the learning subject and a preset reference state for the learning subject. A model is trained, and the detection target for each of the plurality of factor groups based on the learned classification model, a plurality of feature values extracted with respect to the detection target, and a preset reference state for the detection target. and may be configured to calculate a classification accuracy, a leak rate, and an overdetection rate for determining the defect detection performance for .

The control unit may be configured to calculate, with the classification accuracy, a ratio of an object having the same predicted state and the preset reference state determined according to the learned classification model among the detection objects.

The control unit classifies the plurality of subjects into a learning group to which the learning subject belongs and a detection group to which the detection subject belongs, and the classification accuracy for the classified detection group, the leakage rate, and the After calculating the inspection rate, the plurality of inspected objects are reclassified so that each of the plurality of inspected objects is included in the detection group at least once, and the classification accuracy for the reclassified detection group, the leakage rate, and It may be configured to calculate the overdetection rate.

The control unit calculates an average classification accuracy of the classification accuracy calculated for the detection group for each of the plurality of factor groups, selects a predetermined number of factor groups in the order of increasing the calculated average classification accuracy, and selects a predetermined number of factor groups from among the selected factor groups. It may be configured to set a factor group having the lowest calculated lowest leak rate as a reference factor group for the plurality of subjects.

The control unit, when the factor group having the lowest leakage rate is plural, the factor group having the lowest calculated lowest overdetection rate among the plurality of factor groups having the lowest leakage rate as a reference factor group for the plurality of subjects can be configured to set.

The control unit, when the factor group having the lowest overdetection rate is plural, selects the factor group having the highest calculated average classification accuracy from among the plurality of factor groups having the lowest lowest overcheck rate as a reference factor for the plurality of subjects It can be configured to set as a group.

The control unit sets some sections of the measurement section in the generated eddy current profile as a regression formula application section, sets a linear regression model for the eddy current measurement value for each position in the set regression formula application section, and the regression formula The average of the eddy current measured values of the application section, the standard deviation of the eddy current measured values of the regression formula application section, the length of the regression formula application section, the slope of the set linear regression model, and the eddy current measured value of the regression formula application section and the set and extract at least one of mean square errors between linear regression models as the feature value.

The control unit determines a plurality of peaks in each of the plurality of eddy current profiles, and as the measurement position moves from the first position to the second position among the determined plurality of peaks, the upper end of the section in which the eddy current measurement value increases The upper peak located at the lower end and the lower end peak located at the lower end are determined as one peak pair, and the difference between the eddy current measurement value between the lower end peak and the upper peak among the determined plurality of peak pairs is large in the order of the first pair of peaks , a second peak pair, and a third peak pair, and set the third peak pair as the regression expression application interval.

The control unit, the eddy current measured value of the upper peak of the first pair of peaks, the difference between the eddy current measurement position between the upper and lower peaks of the first pair of peaks, the eddy current measured value of the lower peak of the second pair of peaks, the first To extract at least one of the difference in eddy current measurement positions between the upper peak and the lower peak of the two peak pairs, and the difference in eddy current measurement positions between the upper peak of the first pair of peaks and the lower peak of the second pair of peaks, as the feature value. can be configured.

A defect detection method according to another aspect of the present invention includes: an eddy current measuring step of measuring an eddy current for each position in a measurement section from a first position to a second position with respect to a plurality of inspected objects; an eddy current profile generation step of generating a plurality of eddy current profiles indicating a correspondence relationship between an eddy current measurement position and an eddy current measurement value based on the eddy current measurement value for each position measured in the eddy current measurement step; a feature value extraction step of extracting a plurality of feature values corresponding to a plurality of factors from the plurality of generated eddy current profiles; A prediction for generating a plurality of factor groups according to a combination of the plurality of factors, and determining the predicted state of each of the plurality of subjects by using a corresponding feature value and a preset classification model for each of the generated plurality of factor groups state determination step; Defect detection performance for determining defect detection performance for the plurality of inspected objects for each of the plurality of factor groups based on the plurality of predicted states determined in the predictive state determining step and preset reference states for each of the plurality of inspected objects judgment step; and a reference factor group setting step of setting any one of the plurality of factor groups as a reference factor group for the plurality of objects to be inspected based on the defect detection performance determined for each of the plurality of factor groups.

According to one aspect of the present invention, there is an advantage that an optimal reference factor group for detecting a defect in an object can be set. In addition, according to an aspect of the present invention, since the defect of the subject is detected according to the reference factor group, the accuracy and reliability of the defect detection of the subject may be improved.

Effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

Since the following drawings attached to the present specification serve to further understand the technical spirit of the present invention together with the detailed description of the present invention to be described later, the present invention should not be construed as limited only to the matters described in such drawings.
1 is a diagram schematically illustrating a defect detection apparatus according to an embodiment of the present invention.
2 is a diagram schematically illustrating an eddy current profile generated by a defect detection apparatus according to an embodiment of the present invention.
3 is a diagram illustrating a first table in which a plurality of factors extracted by a defect detection apparatus according to an embodiment of the present invention are defined.
4 is a diagram schematically illustrating a second table for the defect detection performance of a plurality of factor groups determined by the defect detection apparatus according to an embodiment of the present invention.
5 is a diagram schematically illustrating a defect detection method according to another embodiment of the present invention.

The terms or words used in the present specification and claims should not be construed as being limited to their ordinary or dictionary meanings, and the inventor may properly define the concept of the term in order to best describe his invention. Based on the principle that there is, it should be interpreted as meaning and concept consistent with the technical idea of the present invention.

Accordingly, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiment of the present invention and do not represent all the technical spirit of the present invention, so at the time of the present application, various It should be understood that there may be equivalents and variations.

In addition, in the description of the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

Terms including an ordinal number such as 1st, 2nd, etc. are used for the purpose of distinguishing any one of various components from the others, and are not used to limit the components by such terms.

Throughout the specification, when a part "includes" a certain element, it means that other elements may be further included, rather than excluding other elements, unless specifically stated otherwise.

In addition, a term such as a control unit described in the specification means a unit for processing at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software.

In addition, throughout the specification, when a part is "connected" with another part, it is not only "directly connected" but also "indirectly connected" with another element interposed therebetween. include

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically illustrating a defect detection apparatus 100 according to an embodiment of the present invention.

Referring to FIG. 1 , a defect detection apparatus 100 according to an embodiment of the present invention may include a measurement unit 110 and a control unit 120 .

The measuring unit 110 may be configured to measure an eddy current for each position in a measurement section from a first position to a second position with respect to a plurality of subjects.

Here, the subject may be a metallic material. Preferably, the subject may be a metallic conductive film. More preferably, the object to be inspected may be a metal thin film having a thickness of several μm.

Preferably, the measurement unit 110 may measure the eddy currents for each position of the plurality of objects as a value for the phase.

For example, in the process of measuring the eddy current for each position of the subject by the measuring unit 110 , the position of the measuring unit 110 may be moved and the position of the subject may be fixed. In this case, the measuring unit 110 may move along the longitudinal direction or the width direction of the subject and measure the eddy current for each position from one end (first position) to the other end (second position) of the subject.

As another example, in the process of measuring the eddy current for each position of the subject by the measuring unit 110 , the position of the measuring unit 110 may be fixed and the position of the subject may be moved.

The control unit 120 may be configured to receive the eddy current measurement values for each position of the plurality of subjects from the measurement unit 110 .

Specifically, the control unit 120 may be communicatively connected to the measurement unit 110 . For example, the control unit 120 may be connected to the measurement unit 110 by wire or wirelessly, and may receive an eddy current measurement value for each position of the plurality of subjects from the measurement unit 110 . Here, the eddy current measured value for each location may include a measured location where the eddy current is measured and a measured eddy current value.

The control unit 120 may be configured to generate a plurality of eddy current profiles indicating a corresponding relationship between the eddy current measurement position and the eddy current measurement value based on the received eddy current measurement value for each position.

Specifically, the eddy current profile may be a profile indicating a correspondence between an eddy current measurement position and an eddy current measurement value at the corresponding position.

2 is a diagram schematically illustrating an eddy current profile generated by the defect detection apparatus 100 according to an embodiment of the present invention.

Referring to FIG. 2 , the controller 120 may generate an eddy current measurement position and an eddy current profile for the eddy current measurement value at the corresponding position based on the eddy current measurement value for each position received from the measurement unit 110 .

For example, in the embodiment of FIG. 2 , the eddy current measurement position of the object may be a first position to a second position. A first position of the subject may correspond to one end of the subject, and a second position of the subject may correspond to the other end of the subject. That is, the eddy current may be measured with respect to the entire subject.

As another example, the measuring unit may measure the eddy current only in a part of the subject.

The controller 120 may be configured to extract a plurality of feature values corresponding to a plurality of factors from a plurality of generated eddy current profiles.

3 is a diagram illustrating a first table T1 in which a plurality of factors extracted by the defect detection apparatus 100 are defined according to an embodiment of the present invention.

For example, in the embodiment of FIG. 3 , the number of factors may be ten. That is, the controller 120 may extract 10 feature values corresponding to a total of 10 factors from each eddy current profile.

A process in which the controller 120 extracts a plurality of feature values corresponding to a plurality of factors will be described later. However, a plurality of factors that can be extracted from the eddy current profile are not limited by the embodiment of FIG. 3 . That is, it is noted that the controller 120 may extract more various factors for detecting the defect of the inspected object from the eddy current profile.

The controller 120 may be configured to generate a plurality of factor groups according to a combination of the plurality of factors.

Specifically, the plurality of factor groups may be preset to include one or more factors. Also, each of the plurality of factor groups may be set so as not to overlap each other.

For example, in the embodiment of FIG. 3 , the number of the plurality of factors is ten. Since at least one factor must be included in the factor group, the number of the plurality of factor groups generated by the controller 120 ^{is calculated as “2 10} -1”, and may be a total of 1,023.

The controller 120 may be configured to determine the predicted state of each of the plurality of subjects by using a corresponding feature value and a preset classification model for each of the plurality of generated factor groups.

Here, the classification model may be a model preset to determine whether the object to be inspected is defective based on the input feature value and to output the determination result. For example, the classification model may be a logistic regression model, a random forest classifier, or a support vector machine. That is, various classification models in addition to the logistic regression classification model, the random forest classifier, and the support vector machine may be applied to the preset classification model.

For example, the controller 120 may input a feature value corresponding to the first factor group to the classification model. In addition, the controller 120 may determine the predicted state of the subject with respect to the first factor group according to a result output from the classification model. That is, as in the previous embodiment, when 1,023 factor groups are generated by the controller 120 , the predicted state of the subject may be determined for each of the 1,023 factor groups.

Hereinafter, for convenience of explanation, the state of the inspected object will be described as a normal state or a defective state. However, the state of the subject may be more subdivided and set.

The control unit 120 may be configured to determine defect detection performance for the plurality of inspected objects for each of the plurality of factor groups based on the determined plurality of predicted states and preset reference states for each of the plurality of inspected objects. .

Specifically, the control unit 120 determines whether the predicted state determined for each of the plurality of inspected objects and the preset reference state are the same, and the control unit 120 determines the leak rate for determining the defect detection performance for the plurality of inspected objects for each of the plurality of factor groups. and calculating an overdetection rate.

Here, the leak rate may be a ratio of the subject whose predicted state is determined to be a normal state among subjects whose reference state is a defective state. That is, the leak rate may be a ratio in which an inspected object in a defective state is erroneously determined to be in a normal state. For example, it is assumed that the reference state of the M inspected objects among the N inspected objects is a defective state. When the predicted state of the K subjects among the M subjects is determined to be a normal state, the outflow rate may be calculated according to a formula of “K÷M×100”.

In addition, the overcheck rate may be a ratio of the subject whose predicted state is determined to be a defective state among the subjects whose reference state is a normal state. That is, the overdetection rate may be a ratio in which an inspected object in a normal state is erroneously determined as a defective state. For example, it is assumed that the reference state of L test objects among the N test objects is a normal state. When the predicted state of the J objects among the L objects is determined to be a defective state, the overcheck rate may be calculated according to a formula of “J÷L×100”.

The controller 120 may be configured to set any one of the plurality of factor groups as a reference factor group for the plurality of inspected objects based on the defect detection performance determined for each of the plurality of factor groups.

Preferably, the controller 120 may be configured to set a factor group having the lowest calculated outflow rate among the plurality of factor groups as a reference factor group for the plurality of subjects. If there are a plurality of factor groups having the lowest calculated leakage rate, the control unit 120 sets the factor group having the lowest calculated overdetection rate among the plurality of factor groups having the lowest calculated leakage rate as a reference factor group for the plurality of subjects. can be configured to be set to .

Since the leak rate is a rate at which the predicted state of the inspected object, which is in the defective state, is incorrectly determined as the normal state, the controller 120 may first consider the leak rates for a plurality of factor groups when setting the reference factor group, and then consider the overdetection rate. For example, if the overdetection rate is prioritized over the outflow rate, the ratio of erroneously judging an inspected object in a normal state as a defective state may decrease, but may increase the ratio of erroneously determining an inspected object in a defective state as a normal state. In this case, since there is a problem that the inspected object in a defective state can be delivered, the defect detection apparatus 100 according to an embodiment of the present invention may preferentially consider the leak rate rather than the overdetection rate.

4 is a diagram schematically illustrating a second table T2 for defect detection performance of a plurality of factor groups determined by the defect detection apparatus 100 according to an embodiment of the present invention.

Specifically, the number of factor groups included in the second table T2 is 15 in total. That is, the second table T2 is a table including the top 15 among 1,023 factor groups generated from factors of the first table T1 .

Referring to FIG. 4 , when the first classification model M1 is used, the leakage rate for the 14th factor group including R_MSE, R_Mean, R_Stdev, R_Slope, R_Distance, Distance, Max, and Min is the lowest. Accordingly, the control unit 120 may set the 14th factor group as a reference factor group for detecting defects of a plurality of objects by using the first classification model M1.

For example, the control unit 120 generates an eddy current profile for the second target in the process of detecting a defect in the second target of the same type as the target, and a reference factor group (fourteenth factor group) from the generated eddy current profile Only a plurality of feature values corresponding to can be extracted. In addition, the controller 120 may detect a defect in the second subject by using the plurality of extracted feature values and the first classification model M1 .

The defect detection apparatus 100 according to an embodiment of the present invention has the advantage of being able to set an optimal factor group for detecting a defect in an object to be inspected. Accordingly, the defect detection apparatus 100 has the advantage of accurately and quickly detecting a defect in the object to be inspected.

On the other hand, the control unit 120 provided in the defect detection apparatus 100 according to an embodiment of the present invention is a processor, ASIC (application-specific integrated circuit) known in the art to execute various control logics performed in the present invention. , other chipsets, logic circuits, registers, communication modems, data processing devices, and the like. Also, when the control logic is implemented in software, the controller 120 may be implemented as a set of program modules. In this case, the program module may be stored in the memory and executed by the controller 120 . The memory may be inside or outside the control unit 120 , and may be connected to the control unit 120 by various well-known means.

Also, the defect detection apparatus 100 according to an embodiment of the present invention may further include a storage unit 130 . The storage unit 130 may store programs and data used in the control unit 120 . That is, the storage unit 130 may store data necessary for each component of the defect detection apparatus 100 to perform an operation and function, a program or data generated while the operation and function are performed. The storage unit 130 is not particularly limited in its type as long as it is a known information storage means capable of writing, erasing, updating and reading data. As an example, the information storage means may include RAM, flash memory, ROM, EEPROM, registers, and the like. Also, the storage unit 130 may store program codes in which processes executable by the control unit 120 are defined. For example, the storage unit 130 may store the eddy current data for each position measured by the measurement unit 110 . Then, the control unit 120 may access the storage unit 130 to obtain the stored eddy current data for each location.

Hereinafter, a process in which the controller 120 extracts a plurality of feature values corresponding to a plurality of factors will be described.

The controller 120 may be configured to determine a plurality of peaks in each of the plurality of eddy current profiles.

For example, in the embodiment of FIG. 2 , the controller 120 controls the first peak P1, the second peak P2, the third peak P3, the fourth peak P4, the fifth peak P5, and A sixth peak P6 may be determined. Here, the peak may be a point at which the slope changes from positive to negative or from negative to positive in the eddy current profile. That is, in the eddy current profile, the control unit 120 is a point at which the instantaneous rate of change of the eddy current measured value for the measuring position is 0, and the instantaneous rate of change of the eddy current measured value for the measuring position based on the point is positive to negative or negative to A positively changing point can be determined as a peak.

In the embodiment of FIG. 2 , the first peak P1 , the third peak P3 , and the fifth peak P5 are points at which the instantaneous rate of change of the eddy current measurement value with respect to the measurement position is 0, and based on the corresponding points It may be a point at which the instantaneous rate of change of the eddy current measurement value for the measurement location changes from negative to positive.

In addition, in the embodiment of FIG. 2 , the second peak P2 , the fourth peak P4 , and the sixth peak P6 are points at which the instantaneous rate of change of the eddy current measurement value for the measurement position is 0, and the corresponding points As a reference, it may be a point at which the instantaneous rate of change of the eddy current measurement value for the measurement position changes from positive to negative.

The control unit 120 includes an upper peak located at the upper end and a lower peak located at the lower end in a section in which the eddy current measurement value increases as the measurement position moves from the first position to the second position among the determined plurality of peaks. It can be configured to determine one peak pair.

For example, in the embodiment of FIG. 2 , as the measurement position is moved from the first position to the second position, the section in which the eddy current measurement value increases is the first section between the first peak P1 and the second peak P2, the second It may be a second section between the third peak P3 and the fourth peak P4 and a third section between the fifth peak P5 and the sixth peak P6 . Accordingly, the controller 120 may determine the first peak P1 as the lower peak of the first section and the second peak P2 as the upper peak as one pair of peaks. Also, the controller 120 may determine a third peak P3 that is a lower peak of the second section and a fourth peak P4 that is an upper peak of the second section as a single peak pair. Also, the controller 120 may determine the fifth peak P5 as the lower peak and the sixth peak P6 as the upper peak of the third section as one peak pair.

In addition, the control unit 120 determines a first peak pair, a second peak pair, and a third peak pair according to an order in which a difference between the eddy current measurement value between the lower peak and the upper peak among the plurality of determined peak pairs is large. can be configured to

Specifically, the controller 120 may calculate the difference between the eddy current measurement value between the upper peak and the lower peak included in one pair of peaks. Preferably, the controller 120 may be configured to determine the first peak pair, the second peak pair, and the third peak pair according to the order in which the difference in the eddy current measurement values is large.

For example, in the embodiment of Figure 2, the difference in the eddy current measurement value between the upper peak and the lower peak included in one peak pair is the largest in the peak pair including the first peak (P1) and the second peak (P2), A peak pair including the fifth peak P5 and the sixth peak P6 may be the second largest, and the peak pair including the third peak P3 and the fourth peak P4 may be the smallest. Therefore, the control unit 120 determines the first peak (P1) and the second peak (P2) as the first peak pair, and determines the fifth peak (P5) and the sixth peak (P6) as the second peak pair, , the third peak P3 and the fourth peak P4 may be determined as the third peak pair.

Also, the control unit 120 may be configured to set the third peak pair as a regression expression application period.

Specifically, the control unit 120 sets some sections of the measurement section in the generated eddy current profile as a regression formula application section, and sets a linear regression model for the eddy current measurement value for each position in the set regression formula application section can be configured.

Here, the regression equation application section may be a section in which a linear regression model is set based on regression analysis. For example, in the embodiment of FIG. 2 , the controller 120 may set a linear regression model with respect to the eddy current measurement values from the L3 position to the L4 position.

On the other hand, when the control unit 120 determines a plurality of peaks based on the instantaneous rate of change of the eddy current measurement value for the measurement position in the eddy current profile, the third peak (P3) and the fourth peak (P4) according to the eddy current measurement value A peak corresponding to may not be accurately determined.

In this case, the controller 120 determines a measurement position symmetrical to the measurement position L1 of the first peak P1 based on the measurement position L2 of the second peak P2, and the measurement determined from the eddy current profile A point corresponding to the position may be determined as the third peak P3.

For example, based on the measurement position L2 of the second peak P2 , the control unit 120 may have a measurement position symmetrical with the measurement position L1 of the first peak P1 is L3 . That is, the difference (L2-L1) between the measurement position L2 of the second peak P2 and the measurement position L1 of the first peak P1 is between L3 and the measurement position L2 of the second peak P2. It may be equal to the difference (L3-L2).

More specifically, the control unit 120 calculates the difference in the measurement position between the measurement position L2 of the second peak P2 and the measurement position L1 of the first peak P1 according to the formula of “L2-L1”. can In addition, the controller 120 may calculate the sum of the measurement position L2 of the second peak P2 and the calculated measurement position difference L2-L1 according to the equation “L2+(L2-L1)”. In addition, the controller 120 may determine the calculated “(2×L2)-L1” as the measurement position L3 of the third peak P3. In this case, the eddy current measurement value A3 of the third peak P3 may be an eddy current measurement value corresponding to the measurement position L3 of the third peak P3 in the eddy current profile. In a similar manner, the control unit 650 Based on the measurement position (L5) of the fifth peak (P5), determine the measurement position symmetrical to the measurement position (L6) of the sixth peak (P6), and determine the point corresponding to the measurement position determined in the eddy current profile It can be determined as 4 peaks (P4).

For example, based on the measurement position L5 of the fifth peak P5, the controller 650 may have a measurement position symmetrical with the measurement position L6 of the sixth peak P6 is L4. That is, the difference (L6-L5) between the measurement position (L5) of the fifth peak (P5) and the measurement position (L6) of the sixth peak (P6) is between L4 and the measurement position (L5) of the fifth peak (P5) It may be equal to the difference (L5-L4).

More specifically, the control unit 650 calculates the difference in the measurement position between the measurement position L5 of the fifth peak P5 and the measurement position L6 of the sixth peak P6 according to the formula of “L6-L5”. can In addition, the control unit 650 may calculate the difference between the measurement position L5 of the fifth peak P5 and the calculated measurement position difference L6-L5 according to the formula “L5-(L6-L5)” . Then, the controller 650 may determine the calculated “L6-(2×L5)” as the measurement position L4 of the fourth peak P4 . In this case, the eddy current measurement value A4 of the fourth peak P4 may be an eddy current measurement value corresponding to the measurement position L4 of the fourth peak P4 in the eddy current profile.

Referring to FIG. 3 , the controller 120 controls the mean (R_Mean) of the eddy current measured values of the regression formula application section, the standard deviation (R_Stdev) of the eddy current measured values of the regression formula application section, and the set slope (R_Slope) of the linear regression model. ), at least one of the mean square error (R_MSE) between the eddy current measurement value of the regression formula application section and the set linear regression model, and the length (R_Distance) of the regression formula application section can be configured to extract as the feature value. .

For example, in the embodiment of FIG. 2 , the regression equation application section may be a section from the third peak P3 to the fourth peak P4 . The controller 120 may extract the average of the eddy current measurement values measured by the measurement unit 110 from the L3 position to the L4 position as a feature value corresponding to the R_Mean factor. Also, the controller 120 may calculate the standard deviation of the eddy current measurement value measured by the measurement unit 110 from the L3 position to the L4 position as a feature value corresponding to the R_Stdev factor. Also, the controller 120 may extract the length from the L3 position to the L4 position as a feature value corresponding to the R_Distance factor. Also, the controller 120 may extract the slope of the linear regression model applied from the L3 position to the L4 position as a feature value corresponding to the R_Slope factor. In addition, the control unit 120 may extract the mean square error between the eddy current measurement value measured by the measurement unit 110 from the L3 position to the L4 position and the function value of the linear regression model as a feature value corresponding to the R_MSE factor.

In addition, referring to FIG. 3 , the control unit 120 includes the eddy current measurement value Max of the upper peak of the first pair of peaks, and the difference (Peak_Diff_1) of the eddy current measurement positions between the upper and lower peaks of the first peak pair. ), the eddy current measurement value of the lower peak of the second pair of peaks (Min), the difference in the eddy current measurement positions between the upper and lower peaks of the second pair of peaks (Peak_Diff_2), and the upper peak of the first pair of peaks and the It may be configured to extract at least one of a difference (Distance) of an eddy current measurement position between the lower peaks of the second peak pair as the feature value.

For example, in the embodiment of FIG. 2 , the bottom peak of the first pair of peaks is the first peak (P1), and the top peak is the second peak (P2). Accordingly, the controller 120 may extract the eddy current measurement value A2 of the second peak P2 as a feature value corresponding to the Max factor. Also, the controller 120 may extract the difference between the eddy current measurement position L2 of the second peak P2 and the eddy current measurement position L1 of the first peak P1 as a feature value corresponding to the Peak_Diff_1 factor. .

Also, in the embodiment of FIG. 2 , the lower peak of the second pair of peaks is the fifth peak (P5), and the upper peak is the sixth peak (P6). Accordingly, the controller 120 may extract the eddy current measurement value A5 of the fifth peak P5 as a feature value corresponding to the Min factor. Also, the controller 120 may extract the difference between the eddy current measurement position L6 of the sixth peak P6 and the eddy current measurement position L5 of the fifth peak P5 as a feature value corresponding to the Peak_Diff_2 factor.

In addition, in the embodiment of Figure 2, the control unit 120, the eddy current measurement position (L5) of the fifth peak (P5), which is the lower peak of the second peak pair, and the second peak (P2), which is the upper peak of the first peak pair ) of the eddy current measurement position (L2) can be extracted as a feature value corresponding to the Distance factor.

In the above, the feature values that can be extracted by the controller 120 have been described with reference to FIGS. 2 and 3 . However, it should be noted that the types of factors that can be extracted from the eddy current profile in order to detect a defect in the object are not limited by the above-described example.

When a plurality of classification models are provided, the controller 120 may be configured to determine the defect detection performance of the plurality of inspected objects for the plurality of factor groups with respect to each of the plurality of classification models.

Referring to FIG. 4 , a first classification model M1 and a second classification model M2 may be provided. The controller 120 may determine the defect detection performance for a plurality of factor groups by using each of the first classification model M1 and the second classification model M2 . Specifically, the controller 120 may calculate the outflow rate and the overdetection rate for a plurality of factor groups using each of the first classification model M1 and the second classification model M2 . In addition, the controller 120 may determine the defect detection performance for the plurality of factor groups based on the leak rate and/or the overdetection rate.

The control unit 120 is configured to set the reference factor group and the reference classification model for the plurality of subjects by setting any one of the plurality of classification models as a reference classification model based on the determined defect detection performance can be

Referring to the second table T2 of FIG. 4 , the lowest outflow rate for the first classification model M1 is 5%, and the lowest outflow rate for the second classification model M2 is 0%. That is, when the defect of the subject is detected using the first classification model M1 and the 14th factor group (a factor group including R_MSE, R_Mean, R_Stdev, R_Slope, R_Distance, Distance, Max, and Min), the defect The probability of erroneously determining the subject in a normal state as a normal state may be 5%. On the other hand, the second classification model (M2) and the twelfth factor group (a factor group including R_MSE, R_Mean, R_Stdev, R_Slope, R_Distance, and Distance) or the thirteenth factor group (R_MSE, R_Mean, R_Stdev, R_Slope, R_Distance, Max) , and a factor group including Min), when a defect of the subject is detected, the probability of erroneously determining the subject in a defective state as a normal state may be 0%. Accordingly, the controller 120 may set the second classification model M2 as a reference classification model for a plurality of subjects.

The defect detection apparatus 100 according to an embodiment of the present invention may further set a classification model as well as a combination of factors optimized for detecting defects of a plurality of inspected objects. Accordingly, based on the reference factor group and the reference classification model, defect detection can be performed quickly and accurately for a plurality of objects and objects of the same type.

The control unit 120 may be configured to classify the plurality of subjects as a learning subject and a detection subject.

For example, 100 subjects may be classified into one subject group of 10 each. That is, 100 subjects may be classified into 10 subject groups, and each subject group may include 10 subjects. The control unit 120 may classify 10 subjects included in the first subject group as detection subjects, and classify 90 subjects included in the second to tenth subject groups as learning subjects. .

The control unit 120 may be configured to learn the classification model based on a plurality of feature values extracted with respect to the subject for learning and a preset reference state for the subject for learning.

For example, the control unit 120 may generate an eddy current profile for each of the 90 subjects classified as subjects for learning. That is, a total of 90 eddy current profiles may be generated for the learning target. The controller 120 may extract a plurality of feature values from each eddy current profile, and train the classification model based on the extracted feature values and the reference state of the subject for learning.

The control unit 120 is configured to control the detection target for each of the plurality of factor groups based on a learned classification model, a plurality of feature values extracted with respect to the detection target, and a preset reference state for the detection target. and may be configured to calculate a classification accuracy, a leak rate, and an overdetection rate for determining the defect detection performance for .

In more detail, the controller 120 may be configured to calculate, with classification accuracy, a ratio of an inspected object having the same predicted state and a preset reference state determined according to a learned classification model among detection objects.

For example, as a result of the control unit 120 detecting the defects of 10 inspection objects for detection using the first classification model M1 and the first factor group, it is determined that the number of inspected objects having different predicted states and reference states is one. Assume The controller 120 may calculate the classification accuracy for the first classification model M1 and the first factor group according to a formula of “9÷10×100”.

More specifically, the control unit 120 classifies the plurality of inspected objects into a learning group to which the learning object belongs and a detection group to which the detection object belongs, and the classification accuracy for the classified detection group , the outflow rate, and may be configured to calculate the over-gum rate.

Thereafter, the controller 120 may be configured to reclassify the plurality of inspected objects so that each of the plurality of inspected objects is included in the detection group at least once.

That is, the control unit 120 may classify each of the plurality of inspected objects as an inspection object for detection at least once. Preferably, the control unit 120 may classify the plurality of inspected objects as the detection objects once.

For example, it is assumed that the first subject group is selected as the detection group, the second to tenth subject groups are selected as the learning group, and the classification accuracy, leakage rate, and overdetection rate for the first subject group are calculated. . The control unit 120 may select the second subject group as the detection group, and select the first subject group and the third to tenth subject groups as the learning group. In addition, the controller 120 may learn the classification model based on the first subject group and the third to tenth subject groups. In this case, the controller 120 may train the initial classification model based on the second to tenth subject groups. Also, the controller 120 may train the initial classification model based on the first subject group and the third to tenth subject groups. That is, the classification model learned by the control unit 120 may be different from each other according to the subject group.

In addition, the control unit 120 may be configured to calculate the classification accuracy, the leak rate, and the overdetection rate for the reclassified detection group.

For example, in the previous embodiment, the control unit 120 may calculate classification accuracy, leakage rate, and overdetection rate for the first and second subject groups selected as the detection group. In this way, for each of the third to tenth subject groups, the control unit 120 may calculate classification accuracy, leakage rate, and overdetection rate for each of a plurality of factor groups and classification models.

The control unit 120 may be configured to calculate an average classification accuracy of the classification accuracy calculated for the detection group for each of the plurality of factor groups.

For example, it is assumed that a total of N subject groups are classified into M subject groups, and that J factor groups and one classification model are provided. For each of the J factor groups, the controller 120 may calculate classification accuracy, leakage rate, and overdetection rate for each of the M subject groups. That is, M classification accuracies, M leak rates, and M overcheck rates may be calculated for each of the J factor groups. The controller 120 may calculate the average of the M classification accuracies calculated for each factor group, and calculate the average accuracy of the corresponding factor group.

The controller 120 may be configured to select a predetermined number of factor groups in an order of increasing the calculated average classification accuracy.

For example, the 15 factor groups described in the second table T2 may be the top 15 factor groups having the highest average classification accuracy. Specifically, since the factor group must include at least one factor, the number of factor groups that can be generated with a total of 10 factors described in the first table T1 is 1,023. Among them, 15 factor groups having the highest average classification accuracy calculated for each factor group may be the factor groups described in the second table T2.

The control unit 120 may be configured to set a factor group having the lowest calculated lowest leakage rate among the selected factor groups as a reference factor group for the plurality of subjects.

For example, referring to the second table T2, for the first classification model M1, the lowest outflow rate of the 14th factor group is the lowest at 5%. Accordingly, when the first classification model M1 is used, the controller 120 may set the reference factor group for the plurality of subjects as the 14th factor group.

On the other hand, when there are a plurality of factor groups having the lowest leakage rate, the control unit 120 applies the factor group having the lowest calculated lowest overdetection rate among the plurality of factor groups having the lowest leakage rate to the plurality of subjects. may be configured to be set as a reference factor group for If there are a plurality of factor groups having the lowest overdetection rate, the controller 120 selects the factor group having the highest calculated average classification accuracy from among the plurality of factor groups having the lowest lowest overcheck rate. It can be configured to set as a reference factor group for the cadaver.

For example, referring to the second table T2 , with respect to the second classification model M2 , the lowest outflow rates of the twelfth factor group and the thirteenth factor group are both equal to 0%. Also, with respect to the second classification model M2, the lowest overdetection rate of the twelfth factor group and the thirteenth factor group is also the same at 0%. Accordingly, the controller 120 compares the classification accuracy (specifically, average classification accuracy) of the twelfth factor group and the thirteenth factor group, and selects the thirteenth factor group with higher classification accuracy as the second classification model M2. It can be set as a reference factor group when detecting defects of a plurality of inspected objects using

That is, the defect detection apparatus 100 according to an embodiment of the present invention classifies each of the plurality of inspected objects into a detection group once, so that the classification accuracy, the leakage rate, and cross-validation of the overcheck rate. In addition, the defect detection apparatus 100 has an advantage in that it can set the optimal factor group for the plurality of objects by setting the reference factor groups for the plurality of objects according to the result of the cross-verification.

Meanwhile, the control unit 120 selects the plurality of subjects as a learning group and a detection group so that each of the plurality of subjects is included in the detection group at least once, and a classification model based on the selected learning group and detection group can be independently trained multiple times.

In the process of learning the classification model, parameters of the classification model may be set according to an input learning group. That is, the parameters of the classification model may be set differently according to the input learning group. Therefore, the control unit 120 independently trains the classification model a plurality of times, compares the outflow rate and the overcheck rate calculated from each of the plurality of learned classification models, and selects any one of the plurality of learned classification models as the optimal classification model. You can choose.

For example, for the K-th factor group, it is assumed that N groups of subjects and one classification model are provided. The control unit 120 selects each of the N test subject groups as a detection group once, selects the remaining N-1 subject groups as a learning group, and independently separates the classification model based on the selected N-1 learning groups. can be trained N times. That is, through this process, N learned classification models may be generated. Here, since the training groups input to the N learned classification models are different from each other, parameters set in each of the N learned classification models may be different from each other. And, the control unit 120 can select any one of the N learned classification models as the optimal classification model for the K-th factor group based on the N leak rates and N over-test rates calculated from the N learned classification models. have.

Specifically, the controller 120 may select a learned classification model having the lowest calculated leakage rate among the N learned classification models as the optimal classification model. If the leakage rate calculated from the M learned classification models among the N learned classification models is the lowest, the learned classification model with the lowest calculated overcheck rate among the M learned classification models can be selected as the optimal classification model. have.

Thereafter, the control unit 120 selects a detection group so that each of the plurality of subjects is included in the detection group at least once, and inputs the selected detection group into the optimal classification model to obtain classification accuracy, leakage rate, and overdetection rate. can be calculated.

For example, following the previous embodiment, it is assumed that N groups of subjects and one optimal classification model are provided for the K-th factor group. The controller 120 may input each of the N test subject groups into the optimal classification model to calculate N classification accuracies, N leak rates, and N overdetection rates. Thereafter, the controller 120 may calculate the average of the N classification accuracies to set the average classification accuracy for the K-th factor group and the optimal classification model. In addition, the controller 120 may set the lowest outflow rate among the calculated N outflow rates as the outflow rate for the K-th factor group and the optimal classification model. Similarly, the controller 120 may set the lowest overcheck rate among the calculated N overcheck rates as the overcheck rate for the K-th factor group and the optimal classification model.

The controller 120 may set the average classification accuracy, leakage rate, and overdetection rate for each factor group and each optimal classification model by performing the above-described process for each factor group and each classification model. In addition, the controller 120 may compare the defect detection performance of each of the plurality of factor groups to set a reference factor group and a reference classification model for the plurality of subjects.

That is, the defect detection apparatus 100 according to an embodiment of the present invention cross-verifies the classification accuracy, leakage rate, and overdetection rate for a plurality of inspected objects for each of the plurality of factor groups. The most optimized classification model can be created. Therefore, since even the same classification model can be learned differently to correspond to each factor group, the defect detection apparatus 100 has the advantage of being able to set the most suitable reference factor group and reference classification model for a plurality of subjects. .

A battery inspection apparatus according to another embodiment of the present invention may include the defect detection apparatus 100 according to an embodiment of the present invention.

More specifically, the object to be inspected may be composed of an electrode lead provided in a battery.

Here, the battery includes a negative terminal and a positive terminal, and refers to one physically separable independent cell. For example, one pouch-type lithium polymer cell may be regarded as a battery. The battery may include a positive electrode lead and a negative electrode lead as electrode leads. In addition, the defect detection apparatus 100 may detect a defect in the electrode lead provided in the battery.

For example, the measuring unit 110 may measure the eddy current for each position from one end to the other end of the electrode lead in the longitudinal direction or the width direction of the electrode lead.

The controller 120 may be configured to detect a defect in the electrode lead using the classification model and the reference factor group set for the electrode lead of the battery, and to diagnose whether the battery is defective based on the detection result. can

The controller 120 may preset a reference factor group and a reference classification model for the electrode lead. Then, the control unit 120 extracts a plurality of feature values corresponding to the factors included in the reference factor group from the eddy current profile for the electrode leads, and inputs the extracted feature values into the reference classification model to diagnose defects in the electrode leads can do.

For example, in the embodiment of FIG. 4 , it is assumed that the subject is an electrode lead. The controller 120 may set the second classification model M2 as a reference classification model. And, since the classification accuracy of the thirteenth factor group is higher than that of the twelfth factor group, the controller 120 may set the thirteenth factor group as the reference factor group.

Since the reference factor group and reference classification model optimized for the electrode lead are preset by the controller 120, the accuracy and reliability of defect detection for the electrode lead is very high, and whether the electrode lead is defective can be quickly diagnosed. .

5 is a diagram schematically illustrating a defect detection method according to another embodiment of the present invention.

Each step of the defect detection method may be performed by the defect detection apparatus 100 according to an embodiment of the present invention. Hereinafter, content overlapping with the previously described content will be briefly described.

5 , the defect detection method according to another embodiment of the present invention includes an eddy current measurement step (S100), an eddy current profile generation step (S200), a feature value extraction step (S300), a predicted state determination step (S400), and a defect It may include a detection performance determination step (S500) and a reference factor group setting step (S600).

The eddy current measuring step ( S100 ) is a step of measuring the eddy currents for each position in the measurement section from the first position to the second position with respect to the plurality of objects, and may be performed by the measuring unit 110 .

The eddy current profile generation step (S200) is a step of generating a plurality of eddy current profiles indicating the correspondence between the eddy current measurement position and the eddy current measurement value based on the eddy current measurement value for each position measured in the eddy current measurement step (S100). 120) can be carried out.

For example, in the embodiment of FIG. 2 , the controller 120 may generate an eddy current profile with respect to an eddy current measurement position and an eddy current measurement value for each of a plurality of inspected objects.

The feature value extraction step S300 is a step of extracting a plurality of feature values corresponding to a plurality of factors from a plurality of generated eddy current profiles, and may be performed by the controller 120 .

2 and 3 , the controller 120 may extract a plurality of feature values corresponding to a plurality of factors described in the first table T1 from each of the plurality of generated eddy current profiles.

In the predicting state determining step ( S400 ), a plurality of factor groups are generated according to a combination of the plurality of factors, and for each of the generated plurality of factor groups, a corresponding feature value and a preset classification model are used to generate the plurality of test objects. The step of determining each prediction state may be performed by the controller 120 .

The controller 120 may generate a plurality of factor groups by combining a plurality of factors. Preferably, at least one factor may be included in the factor group. For example, if the number of the plurality of factors is N, the number of the plurality of factor groups generated by the controller 120 is “2 ^N −1”.

In addition, the controller 120 may determine the predicted state of each of the plurality of subject groups for each of the plurality of factor groups.

The defect detection performance determination step S500 is performed on the basis of the plurality of predicted states determined in the predicted state determination step S400 and preset reference states for each of the plurality of inspected objects, the plurality of inspected states for each of the plurality of factor groups. As a step of determining the defect detection performance for the cadaver, it may be performed by the controller 120 .

Specifically, according to a result of comparing the predicted state and the reference state of each of the plurality of inspected objects, the controller 120 may calculate the leak rate and the overdetection rate to determine the defect detection performance for the plurality of factor groups.

The reference factor group setting step ( S600 ) is a step of setting any one of the plurality of factor groups as a reference factor group for the plurality of inspected objects based on the defect detection performance determined for each of the plurality of factor groups, and includes a control unit ( 120) can be carried out.

Preferably, the controller 120 may set a factor group having the lowest calculated outflow rate among the plurality of factor groups as the reference factor group for the plurality of subjects. If there are a plurality of factor groups having the lowest calculated leakage rate, the controller 120 may set a factor group having the lowest calculated overdetection rate among the plurality of factor groups as a reference factor group for the plurality of subjects.

The embodiment of the present invention described above is not implemented only through the apparatus and method, and may be implemented through a program for realizing a function corresponding to the configuration of the embodiment of the present invention or a recording medium in which the program is recorded. The implementation can be easily implemented by those skilled in the art to which the present invention pertains from the description of the above-described embodiments.

In the above, although the present invention has been described with reference to limited embodiments and drawings, the present invention is not limited thereto and will be described below with the technical idea of the present invention by those of ordinary skill in the art to which the present invention pertains. Of course, various modifications and variations are possible within the scope of equivalents of the claims.

In addition, since the present invention described above can be various substitutions, modifications and changes within the scope that does not depart from the technical spirit of the present invention for those of ordinary skill in the art to which the present invention pertains, the above-described embodiments and attachments It is not limited by the illustrated drawings, and all or part of each embodiment may be selectively combined and configured so that various modifications may be made.

100: defect detection device
110: measurement unit
120: control unit
130: storage

Claims (15)

a measuring unit configured to measure an eddy current for each position in a measurement section from a first position to a second position with respect to a plurality of subjects; and
Receives eddy current measurement values for each position for each of the plurality of inspected objects from the measurement unit, and generates a plurality of eddy current profiles indicating a corresponding relationship between the eddy current measurement positions and the eddy current measurement values based on the received eddy current measurement values for each position, and , extracts a plurality of feature values corresponding to a plurality of factors from the generated plurality of eddy current profiles, generates a plurality of factor groups according to a combination of the plurality of factors, and for each of the generated plurality of factor groups, corresponding features The predicted state of each of the plurality of subjects is determined using a value and a preset classification model, and based on the determined plurality of predicted states and a preset reference state for each of the plurality of subjects, for each of the plurality of factor groups determine defect detection performance for the plurality of inspected objects, and set any one of the plurality of factor groups as a reference factor group for the plurality of inspected objects based on the determined defect detection performance for each of the plurality of factor groups A defect detection device comprising a configured control unit.
According to claim 1,
The control unit is
For each of the plurality of inspected objects, it is determined whether the determined predicted state and the preset reference state are the same, and for each of the plurality of factor groups, an outflow rate and A defect detection device, characterized in that configured to calculate an inspection rate.
3. The method of claim 2,
The control unit is
and a factor group having the lowest calculated leakage rate among the plurality of factor groups is configured to be set as a reference factor group for the plurality of inspected objects.
4. The method of claim 3,
The control unit is
When there are a plurality of factor groups having the lowest calculated outflow rate, the factor group having the lowest calculated overdetection rate among the plurality of factor groups having the lowest calculated outflow ratio is configured to set as a reference factor group for the plurality of subjects A defect detection device, characterized in that.
3. The method of claim 2,
The control unit is
When a plurality of classification models are provided, the defect detection performance of the plurality of inspected objects with respect to the plurality of factor groups is determined for each of the plurality of classification models, and among the plurality of classification models based on the determined defect detection performance. The defect detection apparatus according to claim 1, wherein the reference factor group and the reference classification model for the plurality of inspected objects are set by setting any one as a reference classification model.
According to claim 1,
The control unit is
Classifying the plurality of subjects into a learning subject and a detection subject, and learning the classification model based on a plurality of feature values extracted from the learning subject and a preset reference state for the learning subject, , a learned classification model, a plurality of feature values extracted for the detection object, and the defect on the detection object for each of the plurality of factor groups based on a preset reference state for the detection object and calculating a classification accuracy, a leak rate, and an overdetection rate for judging the detection performance.
7. The method of claim 6,
The control unit is
and calculating, with the classification accuracy, a ratio of an inspected object having the same predicted state and the preset reference state determined according to the learned classification model among the detection inspected objects.
7. The method of claim 6,
The control unit is
Classifying the plurality of subjects into a learning group to which the learning subject belongs and a detection group to which the detection subject belongs, and calculating the classification accuracy, the leakage rate, and the overdetection rate for the classified detection group Then, the plurality of subjects is reclassified so that each of the plurality of subjects is included in the detection group at least once, and the classification accuracy, the leak rate, and the overdetection rate for the reclassified detection group are calculated A defect detection device, characterized in that configured to calculate.
9. The method of claim 8,
The control unit is
The average classification accuracy of the classification accuracy calculated for the detection group is calculated for each of the plurality of factor groups, a predetermined number of factor groups are selected in the order of the calculated average classification accuracy, and the calculated lowest leakage rate among the selected factor groups The defect detection apparatus according to claim 1, wherein the lowest factor group is configured to be set as a reference factor group for the plurality of inspected objects.
10. The method of claim 9,
The control unit is
When there are a plurality of factor groups having the lowest leakage rate, the factor group having the lowest calculated lowest overdetection rate among the plurality of factor groups having the lowest leakage rate is configured to be set as a reference factor group for the plurality of subjects A defect detection device characterized in that.
11. The method of claim 10,
The control unit is
If there are a plurality of factor groups with the lowest overdetection rate, set a factor group with the highest calculated average classification accuracy among the plurality of factor groups with the lowest overdetection rate as a reference factor group for the plurality of subjects A defect detection device, characterized in that configured.
According to claim 1,
The control unit is
In the generated eddy current profile, some section of the measurement section is set as a regression formula application section, a linear regression model for the eddy current measurement value for each position is set in the set regression formula application section, and the eddy current of the regression formula application section The mean of the measured values, the standard deviation of the eddy current measured value of the regression formula application section, the length of the regression formula application section, the slope of the set linear regression model, and between the eddy current measured value of the regression formula application section and the set linear regression model and extracting at least one of the mean square errors as the feature value.
13. The method of claim 12,
The control unit is
A plurality of peaks are determined in each of the plurality of eddy current profiles, and as the measurement position is moved from the first position to the second position among the determined plurality of peaks, the upper end located at the upper end in the section where the eddy current measurement value increases A peak and a lower peak located at the lower end are determined as one peak pair, and a first pair of peaks and a second peak are determined according to the order in which the difference in the eddy current measurement value between the lower peak and the upper peak among the determined plurality of peak pairs is large. and determine a pair and a third peak pair, and set the third peak pair as the regression expression application interval.
14. The method of claim 13,
The control unit is
The eddy current measurement value of the upper peak of the first pair of peaks, the difference in the eddy current measurement positions between the upper and lower peaks of the first peak pair, the eddy current measured value of the lower peak of the second pair of peaks, the second peak pair At least one of a difference in an eddy current measurement position between an upper peak and a lower peak and a difference in an eddy current measurement position between an upper peak of the first pair of peaks and a lower peak of the second pair of peaks is configured to extract as the feature value fault detection device.
An eddy current measuring step of measuring an eddy current for each position in a measurement section from a first position to a second position with respect to a plurality of subjects;
an eddy current profile generation step of generating a plurality of eddy current profiles indicating a correspondence relationship between an eddy current measurement position and an eddy current measurement value based on the eddy current measurement value for each position measured in the eddy current measurement step;
a feature value extraction step of extracting a plurality of feature values corresponding to a plurality of factors from the plurality of generated eddy current profiles;
A prediction for generating a plurality of factor groups according to a combination of the plurality of factors, and determining a prediction state of each of the plurality of subjects by using a corresponding feature value and a preset classification model for each of the generated plurality of factor groups state determination step;
Defect detection performance for determining defect detection performance for the plurality of inspected objects for each of the plurality of factor groups based on the plurality of predicted states determined in the predictive state determining step and preset reference states for each of the plurality of inspected objects judgment step; and
and a reference factor group setting step of setting any one of the plurality of factor groups as a reference factor group for the plurality of inspected objects based on the defect detection performance determined for each of the plurality of factor groups; Way.

Images