KR102842452B1 - True-false defect detection system using deep learning algorithm-based ultrasound scan data and the method of thereof - Google Patents

Abstract

A system and method for detecting true-false defects using ultrasonic scan data based on a deep learning algorithm are disclosed. According to one embodiment of the present invention, a system for detecting true-false defects using ultrasonic scan data based on a deep learning algorithm comprises: a data collection unit that simultaneously collects B-Scan data and C-Scan data and configures the B-Scan data and the C-Scan data in a one-pair format; a data learning unit that trains an artificial intelligence learning model that detects at least one defect pattern based on the B-Scan learning data and the C-Scan learning data; a defect detection unit that inputs the C-Scan data collected by the data collection unit into the artificial intelligence learning model generated by the data learning unit to identify at least one defect area and generates coordinate data corresponding to the defect area; And a defect determination unit that sets a region of interest by mapping the X-axis coordinate of the coordinate data of the defect region generated by the defect detection unit to the B-Scan data, checks whether there is a defective region within the region of interest from the B-Scan data, and determines that there is no defective region as a false defect if there is, and determines that there is a defective region as a true defect if there is.

Description

TRUE-FALSE DEFECT DETECTION SYSTEM USING DEEP LEARNING ALGORITHM-BASED ULTRASOUND SCAN DATA AND THE METHOD OF THEREOF

The present invention relates to a true-false defect detection system and method using ultrasonic scan data based on a deep learning algorithm, and more particularly, to a true-false defect detection system and method using ultrasonic scan data based on a deep learning algorithm, which collects an ultrasound image including C-Scan data and B-Scan data from an analysis target, and, when a defective area is included in the C-Scan data, analyzes the B-Scan data to determine whether the defective area is true-false.

Ultrasonic testing is a nondestructive testing (NDT) technique widely used to identify internal defects in materials. Ultrasonic waves are emitted from the target and, based on the reflected and transmitted signals, can reveal internal structures. They are also ideal for quantitatively analyzing the location, size, and shape of defects. In particular, ultrasonic data is sensitive to depth information and density differences within a material, making it ideal for detecting various types of defects (e.g., cracks, voids, discontinuities).

In this regard, conventional defect determination technologies using ultrasonic inspection have been used, including simple signal-based analysis that detects defects by utilizing the intensity (amplitude) and time (Time-of-Flight, ToF) information of the reflected signal, and image processing-based analysis that converts signals into images to create 2D or 3D images and determines defects through these images. However, these methods have the problem of leading to unnecessary warnings (over-inspection) during the manufacturing process in distinguishing between genuine and false defects, which lowers the production yield.

Therefore, research is required on technologies that can analyze defects using deep learning algorithms based on ultrasonic scan data, efficiently distinguish between genuine and false defects, and provide accurate defect locations and characteristics.

Korean Patent Publication No. 10-082268 Korean Patent Publication No. 10-2017-0101834

The present invention provides a system and method for detecting genuine and false defects using ultrasonic scan data based on a deep learning algorithm, which can effectively distinguish between genuine and false defects and reduce the rate of misdiagnosis due to small pixel sizes in a technique for detecting defects using only existing C-Scan data by analyzing a defect area based on B-Scan data that provides vertical cross-sectional information based on the intensity and reflection time of an ultrasonic signal for an analysis target and C-Scan data that collects reflection data at a specific depth and expresses it as a two-dimensional image of a horizontal plane.

In addition, the purpose is to provide a true-false defect detection system and method using ultrasonic scan data based on a deep learning algorithm that enables more accurate detection of a defective area by primarily detecting a defective area from C-Scan data and analyzing B-Scan data corresponding to the C-Scan data in which the defective area is detected to determine whether the defect is true-false.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems to be solved by the present invention that are not mentioned herein will be clearly understood by a person having ordinary skill in the technical field to which the present invention pertains from the description below.

A true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention includes a data collection unit that simultaneously collects B-Scan data and C-Scan data and configures the B-Scan data and the C-Scan data in a one-pair format, a data learning unit that trains an artificial intelligence learning model that detects at least one defect pattern based on the B-Scan learning data and the C-Scan learning data, a defect detection unit that inputs the C-Scan data collected by the data collection unit into the artificial intelligence learning model generated by the data learning unit to identify at least one defect area and generates coordinate data corresponding to the defect area, and a defect determination unit that sets an area of interest by mapping an X-axis coordinate among the coordinate data of the defect area generated by the defect detection unit to the B-Scan data, and determines whether or not a defective area exists within the area of interest from the B-Scan data, and determines a false defect if a defective area does not exist, and determines a true defect if a defective area exists.

In addition, the data collection unit is characterized by including an ultrasonic generator that emits ultrasonic waves in the range of 1 to 50 MHz to a preset analysis target, an ultrasonic receiver that collects a reflection signal reflected from the analysis target and a transmission signal transmitted through the analysis target, a first scan data generation unit that generates the C-Scan data including depth information of the analysis target and internal structure information of a horizontal plane based on the reflection signal and the transmission signal received by the ultrasonic receiver, a second scan data generation unit that generates the B-Scan data including cross-sectional image information corresponding to the depth information of the analysis target based on the reflection signal and the transmission signal received by the ultrasonic receiver, and a group setting unit that groups the C-Scan data generated by the first scan data generation unit and the B-Scan data generated by the second scan data generation unit in a one-pair format.

In addition, the defect detection unit inputs C-Scan data into an artificial intelligence learning model generated by a data learning unit to generate object detection result data including at least one defect area information and transmits the data to the defect judgment unit, and in the case where the object detection result data is not generated, generates a defect judgment termination signal and transmits the signal to the defect judgment unit so that the process of determining a false defect and a true defect through the defect judgment unit is not performed.

In addition, the defect detection unit is characterized in that, when object detection result data is generated, C-Scan data is stored as defect judgment data.

In addition, the defect detection unit is characterized in that, when a defect area is identified from C-Scan data, it creates a bounding box in the defect area, extracts x _min , y _min , x _max , and y _max coordinates of the defect area, creates an origin marker in the C-Scan data, and applies an XY two-dimensional coordinate system based on the origin marker to create coordinate data corresponding to the x _min , y _min , x _max , and y _max coordinates.

In addition, the defect judgment unit is characterized by including an analysis range setting unit that sets an X-axis analysis range of B-Scan data based on coordinate data generated by the defect detection unit, an interest area setting unit that sets at least one interest area within the X-axis analysis range set by the analysis range setting unit, and a signal size analysis unit that analyzes a signal size within the interest area to determine a genuine defect and a false defect.

In addition, the defect determination unit is characterized in that, if the B-Scan data is determined to be a genuine defect, the B-Scan data is stored as genuine defect result data, and if the B-Scan data is determined to be a false defect, the B-Scan data is stored as false defect result data.

In addition, the data collection unit includes a fault diagnosis unit that monitors whether a fault has occurred in the ultrasonic generator, and an alarm generation unit that generates a user alarm signal in response to the diagnosis result of the fault diagnosis unit, and the fault diagnosis unit analyzes the size of the ultrasonic signal, which is an output value of the ultrasonic generator, and if the upper and lower difference value (N _{dif_num} ) calculated by [Mathematical Formula 1] is greater than a preset number standard value (S _standard ), and at the same time, the upper and lower variance difference value (N _{dif_disp} ) calculated by [Mathematical Formula 2] is greater than a preset variance standard value (D _standard ), it is characterized in that it diagnoses that a fault has occurred in the ultrasonic generator.

[Mathematical Formula 1]

(Here, N _{up_num} is the number of ultrasonic signal values exceeding the average value within a preset interval, N _{down_num} is the number of ultrasonic signal values below the average value within a preset interval, and N _{dif_num} is the absolute value of the difference between N _{up_num} and N _{down_num} , respectively.)

[Equation 2]

(Here, is the sum of the variances of each ultrasonic signal value exceeding the average value within a preset interval. is the sum of the variances of each ultrasonic signal value below the average within a preset interval, and N _{dif_disp} is and ) means the absolute value of the difference between

In addition, the fault diagnosis unit is characterized in that it diagnoses that a fault has occurred in the ultrasonic generator when the condition of clause 7 is satisfied and the skewness (γ) calculated by [Mathematical Formula 5] is greater than a preset skewness standard value (SK _standard ).

[Equation 5]

(Here, γ represents the skewness of the ultrasonic signal value, N represents the number of ultrasonic signal values within the interval, O(t _n ) represents the ultrasonic signal value at time t _n , av represents the average value within the interval, and σ represents the standard deviation within the interval.)

In addition, the fault diagnosis unit is characterized in that it divides the preset section at preset time intervals, and if P _new satisfying [Mathematical Expression 3] and [Mathematical Expression 4] is 1, it resets the output value of the n+3rd ultrasonic generator to the starting value of the new section.

[Equation 3]

(Here, I _n is the slope connecting the consecutive ultrasonic signal values, t _n is the time when the nth ultrasonic signal value was output, t _n-1 is the time when the n-1th ultrasonic signal value was output, O(t _n ) is the ultrasonic signal value at time t _n , and O(t _n-1 ) is the ultrasonic signal value at time t _n-1 .)

[Equation 4]

(Here, I _standard means the standard slope)

According to the present invention, by analyzing a defect area based on B-Scan data that provides vertical cross-sectional information based on the intensity and reflection time of an ultrasonic signal for an analysis target and C-Scan data that collects reflection data at a specific depth and expresses it as a two-dimensional image of a horizontal plane, the present invention has the effect of reducing the occurrence rate of misdiagnosis due to small pixel size in a technology that detects defects using only C-Scan data and effectively distinguishing between genuine defects and false defects.

In addition, by primarily detecting a defective area from C-Scan data and analyzing B-Scan data corresponding to the C-Scan data in which the defective area is detected to determine whether the defect is genuine or false, it has the effect of enabling more accurate detection of the defective area.

FIG. 1 is a configuration diagram of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention.
FIG. 2 and FIG. 3 are diagrams for explaining a data collection unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention.
FIG. 4 and FIG. 5 are diagrams for explaining a defect judgment unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention.
FIGS. 6 and 7 are diagrams for explaining a fault diagnosis unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention.

Specific details, including the problems to be solved, means of solving them, and effects of the invention, are included in the embodiments and drawings described below. The advantages and features of the present invention, and methods for achieving them, will become clearer with reference to the embodiments described below in detail, along with the accompanying drawings.

The scope of the present invention is not limited to the embodiments described below, and various modifications may be made by a person having ordinary knowledge in the relevant technical field without departing from the technical spirit of the present invention.

Hereinafter, the title of the invention of the present invention is described in detail with reference to the attached drawing 1.

FIG. 1 is a block diagram of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention, FIGS. 2 and 3 are diagrams for explaining a data collection unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention, FIGS. 4 and 5 are diagrams for explaining a defect judgment unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention, and FIGS. 6 and 7 are diagrams for explaining a fault diagnosis unit of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention.

Example 1

Referring to FIG. 1, a true-false defect detection system (100) utilizing ultrasonic scan data based on a deep learning algorithm according to an embodiment of the present invention may include a data collection unit (110), a data learning unit (120), a defect detection unit (130), and a defect judgment unit (140).

More specifically, the data collection unit (110) collects B-Scan data and C-Scan data simultaneously, configures the B-Scan data and the C-Scan data in a one-pair format, the data learning unit (120) trains an artificial intelligence learning model that detects at least one defect pattern based on the B-Scan learning data and the C-Scan learning data, the defect detection unit (130) inputs the C-Scan data collected by the data collection unit (110) into the artificial intelligence learning model generated by the data learning unit (120) to identify at least one defect area, and generates coordinate data corresponding to the defect area, and the defect determination unit (140) sets a region of interest by mapping the X-axis coordinate of the coordinate data of the defect area generated by the defect detection unit (130) to the B-Scan data, and checks whether there is a defective area in the region of interest from the B-Scan data, and if the defective area does not exist, determines it as a false defect, and if the defective area exists, determines it as a genuine defect. It was.

At this time, as shown in FIGS. 2 and 3, the data collection unit (110) includes an ultrasonic generator (111) that emits ultrasonic waves in the range of 1 to 50 MHz to a preset analysis target, an ultrasonic receiver (112) that collects a reflection signal reflected from the analysis target and a transmission signal transmitted through the analysis target, a first scan data generation unit (113) that generates the C-Scan data including depth information of the analysis target and internal structure information of a horizontal plane based on the reflection signal and the transmission signal received by the ultrasonic receiver (112), a second scan data generation unit (114) that generates the B-Scan data including cross-sectional image information corresponding to the depth information of the analysis target based on the reflection signal and the transmission signal received by the ultrasonic receiver (112), and the C-Scan data generated by the first scan data generation unit (113) and the B-Scan data generated by the second scan data generation unit (114) based on gate (layer) information. It may include a group setting unit (115) that groups in a one-pair format.

At this time, the gates can be distinguished based on the size of the reflection signal and the transmission signal according to the analysis target, and the C-Scan data (310) and the B-Scan data (320) can be generated.

That is, the C-Scan data (320) may be generated by separating the surface scan image of the analysis target by gate, and the B-Scan data (320) may be generated by separating the cross-sectional scan image of the analysis target by gate.

Meanwhile, the data learning unit (120) may further include a labeling generation unit that generates a label in the form of a bounding box for a defective area in the C-Scan data based on the one-pair format data generated by the group setting unit (115), crops the defective area in the B-Scan data, refines and processes the image data, and then generates a label for classification learning, a first scan data learning unit that learns a supervised learning model for object recognition for object detection while simultaneously generating a label corresponding to the C-Scan data based on the labeling generation unit, and a second scan data learning unit that learns a classification model for the labeling of the B-Scan data generated by the labeling generation unit.

At this time, the results learned for the C-Scan data and the B-Scan data can be stored in a pre-designated learning result DB.

For example, an object detection model for the C-Scan data can be learned based on the following [Mathematical Formula 6], and a classification model for the B-Scan data can be learned based on the following [Mathematical Formula 7].

[Equation 6]

Here, L _total is the total loss, L _loc is the bounding box location (localization) loss, L _cls is the class classification loss, and L _reg is the bounding box size and ratio (regression) loss.

[Equation 7]

Here, L _cls is the classification loss, yisms is the true class label, means the predicted probability.

For example, the object detection model may perform a learning process to minimize the overall loss, and the classification model may perform a learning process to minimize the class classification loss.

Meanwhile, the defect detection unit (130) can input the C-Scan data into the artificial intelligence learning model generated by the data learning unit (120) to generate object detection result data including at least one defect area information.

At this time, the object detection result data may include x _min , y _min , x _max , and y _max coordinate data corresponding to the defective area.

For example, when the C-Scan data is generated by dividing into the five gates, the defect detection unit (130) inputs the C-Scan data for each gate into the artificial intelligence learning model, and can generate the object detection result data by collecting the gate information and the defect area information of the C-Scan data in which the defect pattern is detected among the C-Scan data.

Meanwhile, if the object detection result data is not generated, the defect detection unit (130) can generate a defect determination end signal and transmit it to the defect determination unit (140) so that the determination process of the false defect and the true defect through the defect determination unit (140) is not performed.

That is, if at least one defect pattern is not detected in the C-Scan data, it can be determined that no defect area is detected in the B-Scan data either.

At this time, the defect detection unit (130) can store the C-Scan data as defect determination data when the object detection result data is generated.

That is, the ultrasound scan data analyzed through the defect detection unit (130) is stored in the learning DB and can be used as learning data for the artificial intelligence learning model of the data learning unit (120).

At this time, the data learning unit (120) can collect scan data added to the learning DB and learning data previously stored in the learning DB at preset intervals (e.g., every 3 months) and re-perform the learning process of the artificial intelligence learning model, and the learning result data generated through re-performing the learning can be updated in the learning result DB.

Therefore, the above artificial intelligence learning model can accurately and quickly detect defect patterns included in the C-Scan data by repeating the learning process in a three-month cycle and utilizing more learning data.

Meanwhile, the defect detection unit (130), when the defect area is identified from the C-Scan data, creates a bounding box for the defect area, extracts x _min , y _min , x _max , and y _max coordinates of the defect area, creates an origin marker in the C-Scan data, and applies an XY two-dimensional coordinate system based on the origin marker to create the coordinate data corresponding to the x _min , y _min , x _max , and y _max coordinates.

For example, if a defect pattern is detected at gate 3 of the C-Scan data, the defect detection unit (130) can extract the outline of the defect area from the C-Scan data of gate 3 and create a bounding box to include all of the outlines.

Additionally, when the above bounding box is created, an origin marker can be created on one side of the C-Scan data, and the origin marker can be defined as (0, 0).

Accordingly, the defect detection unit (130) generates an x-axis coordinate based on the column pixel value of the C-Scan data, and generates a y-axis coordinate based on the row pixel value of the C-Scan data, and can generate the coordinate data based on the x _min , y _min , x _max , and y _max coordinates of the defect area.

Meanwhile, referring to FIGS. 4 and 5, the defect determination unit (140) may include an analysis range setting unit (141) that sets an X-axis analysis range of the B-Scan data based on the coordinate data generated by the defect detection unit (130), an interest area setting unit (142) that sets at least one interest area within the X-axis analysis range set by the analysis range setting unit (141), and a signal size analysis unit (143) that analyzes the signal size within the interest area to determine the true defect and the false defect.

The above defect judgment unit (140) maps the X-axis coordinates of x _min and x _max among the coordinate data of the defect area generated by the defect detection unit (130) to the B-Scan data, crops the area of interest mapped from the B-Scan data into an image, and extracts a feature vector by calling the classification model learning result from the learning result DB stored in the data learning unit (120).

In addition, the defect judgment unit (140) performs similarity distance measurement and final judgment, and enables measurement through cosine similarity by comparing the pseudo-true feature vector.

For example, if the distance value between the learned feature vector (f _ref ) and the embedding vector to be tested (f _new ) is proportional, the similarity is close to 1 regardless of the size, so a true defect can be determined, and if the similarity is close to 0, a false defect can be determined. This maximizes the performance of defect detection.

At this time, the feature vector can be extracted based on the following [Mathematical Formula 8], the cosine similarity can be calculated based on the following [Mathematical Formula 9], and the cosine distance can be generated based on the following [Mathematical Formula 10].

[Equation 8]

Here, f is a feature vector, R represents a real number space, and d represents the dimension of the feature vector.

[Equation 9]

[Equation 10]

Here, f is a feature vector, f _ref is a learned feature vector (f _ref ), and f _new is an embedding vector to be tested.

For example, the analysis range setting unit (141) can set the X-axis analysis range based on the distance from the x _min , y _min , x _max , and y _max coordinates to the outer line of the bounding box.

In addition, when the X-axis analysis range is set through the analysis range setting unit (141), the interest region setting unit (142) can call the B-Scan data of the same gate as the C-Scan data analyzed as having a defective area in the defect detection unit (130), and set an interest region in the B-Scan data based on the X-axis analysis range set by the analysis range setting unit (141).

For example, if there are three bounding boxes (51, 52, 53) generated in the C-Scan data (311), three regions of interest (54, 55, 56) can also be generated in the B-Scan data (321).

Accordingly, the signal size analysis unit (143) analyzes the signal size within the region of interest generated by the region of interest setting unit (142) to determine the true defect and the false defect. If the signal size within the region of interest (54, 55) exceeds a preset threshold, it can be determined as a true defect in which a defective region exists, and if the signal size within the region of interest (56) is less than the threshold, it can be determined as a false defect in which the defective region does not exist.

Therefore, even if three defective areas are detected in the C-Scan data, it can be determined that there are actually two defective areas in the analysis target based on the two genuine defects detected from the B-Scan data.

Meanwhile, the defect determination unit (140) stores the B-Scan data as the B-Scan learning data in the learning DB. If the B-Scan data is determined to be a genuine defect, the B-Scan data is stored as defect learning data. If the B-Scan data is determined to be a false defect, the B-Scan data is stored as non-defect learning data. This can be used as learning data to improve the learning ability of the artificial intelligence learning model of the data learning unit (120).

Meanwhile, the data collection unit (110) may further include a fault diagnosis unit that monitors whether a fault has occurred in the ultrasonic generator (111) and an alarm generation unit that generates a user alarm signal in response to the diagnosis result of the fault diagnosis unit.

More specifically, the fault diagnosis unit can determine whether the ultrasonic generator (111) is faulty by analyzing the size (output value) of the ultrasonic signal generated from the ultrasonic generator (111) in real time and determining whether abnormal data is output. Here, the fault of the ultrasonic generator (111) can be determined not only when the operation of the ultrasonic generator (111) is stopped, but also when the output error of the ultrasonic signal generated from the ultrasonic generator (111) becomes larger than the allowable limit.

To this end, if the upper and lower difference value (N _{dif_num} ) calculated by [Mathematical Formula 1] below is greater than the preset number standard value (S _standard ), and at the same time, the upper and lower dispersion difference value (N _{dif_disp} ) calculated by [Mathematical Formula 2] below is greater than the preset dispersion standard value (D _standard ), it can be determined that a failure has occurred in the ultrasonic generator (111) or that the output error has exceeded the allowable limit.

[Mathematical Formula 1]

Here, N _{up_num} represents the number of ultrasonic signal values exceeding the average value within a preset interval, N _{down_num} represents the number of ultrasonic signal values below the average value within a preset interval, and N _{dif_num} (upper and lower difference value) represents the absolute value of the difference between N _{up_num} and N _{down_num} , respectively.

[Equation 2]

Here, is the sum of the variances of each ultrasonic signal value exceeding the average value within a preset interval. is the sum of the variances of each ultrasonic signal value below the average within a preset interval, and N _{dif_disp} (upper and lower variance difference value) is and Each represents the absolute value of the difference.

in other words, is the sum of the variances for each ultrasonic signal value exceeding the average value within a preset interval. For example, if there are three ultrasonic signal values exceeding the average value within a preset interval, up_num is 3, U ₁ is the variance of the first ultrasonic signal value exceeding the average value, U ₂ is the variance of the second ultrasonic signal value exceeding the average value, and U ₃ is the variance of the third ultrasonic signal value exceeding the average value.

likewise, is the sum of the variances for each ultrasonic signal value that is less than the average within a preset interval. For example, if there are two ultrasonic signal values that are less than the average within a preset interval, down_num is 2, D ₁ is the variance of the first ultrasonic signal value that is less than the average, and D ₂ is the variance of the second ultrasonic signal value that is less than the average.

Accordingly _{, the fault diagnosis unit can determine that a fault has occurred in the ultrasonic generator (111) or that the output error has exceeded the allowable limit if N dif_num calculated} by [Mathematical Formula 1] is greater than a preset number standard value (S _standard ) and N dif_disp calculated by [Mathematical Formula 2] is greater than a preset dispersion standard value (D _standard ).

Meanwhile, the principle of setting the preset section above is to divide the section at preset time intervals, but in the case where P _new calculated according to [Mathematical Formula 3] and [Mathematical Formula 4] below is 1, the section can be adjusted by resetting the n+3rd ultrasonic signal value as the starting value of a new section.

[Equation 3]

Here, I _n represents the slope connecting the consecutive ultrasonic signal values, t _n represents the time when the nth ultrasonic signal value is output, t _n-1 represents the time when the n-1th ultrasonic signal value is output, O(t _n ) represents the ultrasonic signal value at time t _n , and O(t _n-1 ) represents the ultrasonic signal value at time t _n-1 .

[Equation 4]

Here, I _standard means the standard slope.

Hereinafter, the specific contents of the methods for determining whether the ultrasonic generator (111) is faulty will be described in detail with reference to FIGS. 6 to 8.

Figure 6 is a graph showing the analysis results of an ultrasonic signal according to the present invention in terms of time (X-axis) and output value (Y-axis).

Figure 6 displays time (t) on the X-axis and output values on the Y-axis, and data is displayed divided into sections (P1, P2, P3, P4), and each section can be divided into preset time intervals according to the time flow of the ultrasonic signal value.

Here, 30 on the Y-axis is the average of the ultrasonic signal values within each section, and whether or not the ultrasonic generator (111) is broken is determined based on how much each ultrasonic signal value exceeds or falls below the average.

For example, the six ultrasonic signal values in the P1 section are 32, 29, 27, 29, 31, and 32, respectively, so the deviations are 2, -1, -3, -1, 1, 2, and the variances are 4, 1, 9, 1, 1, and 4.

Accordingly, the number of ultrasonic signal values exceeding the average of ultrasonic signal values, 30, is 3, and the number of ultrasonic signal values below the average of ultrasonic signal values, 30, is 3, so N _{dif_num} calculated according to [Mathematical Formula 1] is 0.

In addition, the variances of the ultrasonic signal values exceeding the average of the ultrasonic signal values, 30, are 4, 1, 4, and the sum is 9, and the variances of the ultrasonic signal values below the average of the ultrasonic signal values, 30, are 1, 9, 1, and the sum is 11, so the absolute value of the difference between the sum values calculated according to the above [Mathematical Formula 2], N _{dif_disp} , becomes 2.

At this time, if the preset number standard value (S _standard ) is 3 and the preset dispersion standard value (D _standard ) is initially set to 6, both N _{dif_num} and N _{dif_disp} are not greater than the number standard value (S _standard ) and the dispersion standard value (D _standard ), so it can be determined that no failure has occurred in the ultrasonic generator (111) in the P1 section.

If P2 is calculated in the same manner as P1 above, the six ultrasonic signal values in the P2 section are 31, 31, 31, 24, 31, 32, respectively, so the deviation is 1, 1, 1, -6, 1, 2, and the variance is 1, 1, 1, 36, 1, 4.

Accordingly, the number of ultrasonic signal values exceeding the average of ultrasonic signal values, 30, is 5, and the number of ultrasonic signal values below the average of ultrasonic signal values, 30, is 1, so N _{dif_num} calculated according to [Mathematical Formula 1] is 4.

In addition, the variance of each ultrasonic signal value exceeding the average of 30, which is the ultrasonic signal value, is 1, 1, 1, 1, 4, and the sum is 8, and the variance of the ultrasonic signal value below the average of 30, which is the ultrasonic signal value, is 36, and the sum is also 36. Therefore, N _{dif_disp} , which is the absolute value of the difference between the sum values calculated according to the above [Mathematical Formula 2], becomes 28.

Similarly, if the preset number standard value (S _standard ) is 3 and the preset dispersion standard value (D _standard ) is 6, both N _{dif_num} and N _{dif_disp} are greater than the number standard value (S _standard ) and the dispersion standard value (D _standard ), so it can be determined that a failure has occurred in the ultrasonic generator (111) in the P2 section.

In this case, since N _{dif_num} and N _{dif_disp} exceeding the count standard (S _standard ) and the variance standard (D _standard ) are calculated in the P2 section, the failure of the ultrasonic generator (111) can be immediately confirmed, but in the second embodiment of the present invention, considering a situation where a temporary error value may have been output in a specific section, if N _{dif_num} and N _{dif_disp} exceeding the count standard (S _standard ) and the variance standard (D _standard ) are continuously calculated in multiple sections (e.g., 3 sections), the failure of the ultrasonic generator (111) can be definitively diagnosed.

To this end, in the P3 section, if the upper and lower difference values (N _{dif_num} ) and the upper and lower variance difference values (N _{dif_disp} ) are calculated according to the above [Mathematical Formula 1] and the above [Mathematical Formula 2], in the case of the P3 section, the six ultrasonic signal values are 26, 32, 32, 25, 33, and 32, respectively, so the deviations are -4, 2, 2, -5, 3, and 2, and the variances are 16, 4, 4, 25, 9, and 4.

Accordingly, the number of ultrasonic signal values exceeding the average of ultrasonic signal values, 30, is 4, and the number of ultrasonic signal values below the average of ultrasonic signal values, 30, is 2, so N _{dif_num} calculated according to [Mathematical Formula 1] is 2.

In addition, the variances of the ultrasonic signal values exceeding the average of the ultrasonic signal values, 30, are 4, 4, 9, and 4, respectively, and the sum is 21, and the variances of the ultrasonic signal values below the average of the ultrasonic signal values, 30, are 16, 25, and the sum is 41, so the absolute value of the difference between the sum values calculated according to the above [Mathematical Formula 2], N _{dif_disp} , is 20.

Similarly, if the preset number standard value (S _standard ) is 3 and the preset dispersion standard value (D _standard ) is 6, the N _{dif_disp} is greater than the dispersion standard value (D _standard ), but N _{dif_num} is not greater than the number standard value (S _standard ), so it can be determined that no failure occurred in the ultrasonic generator (111) in the P3 section.

That is, in the P2 section, N _{dif_num} and N _{dif_disp} exceeding the count standard value (S _standard ) and the variance standard value (D _standard ) were calculated, but in _the P3 section, N dif_num and N dif_disp exceeding the count standard value (S _standard ) and the variance standard value (D _standard ) were not calculated, so in a number of consecutive sections, _{N dif_num} and N _{dif_disp} exceeding the count standard value (S standard) and the _variance standard value (D _standard ) were not calculated, and therefore, it can be determined that the ultrasonic generator (111) did not malfunction.

In addition, in the third embodiment of the present invention, if N _{dif_num} and N _{dif_disp} exceeding the count standard value (S _standard ) and the dispersion standard value (D _standard ) are calculated more than twice within three sections, it can be determined that a failure has occurred in the ultrasonic generator (111) even if they did not occur continuously, and if N _{dif_num} and N _{dif_disp} exceeding the count standard value (S _standard ) and the dispersion standard value (D _standard ) are calculated in the P4 section, it can be determined that a failure has occurred in the ultrasonic generator (111) because N _{dif_num} and N _{dif_disp} exceeding the count standard value (S _standard ) and the dispersion standard value (D _standard ) are calculated in two sections among the P2 to P4 sections.

In addition, in the fourth embodiment of the present invention, N _{dif_num} and N _{dif_disp} exceeding the number standard value (S _standard ) and the dispersion standard value (D _standard ) according to the above [Mathematical Formula 1] and the above [Mathematical Formula 2] are calculated, and at the same time, whether the asymmetry of the data distribution is high is determined, and only when all of these are satisfied, it is determined that a failure has occurred in the ultrasonic generator (111), thereby preventing the failure from being determined indiscriminately.

In order to determine the asymmetry of data distribution, the skewness γ of the ultrasonic signal value according to [Mathematical Formula 5] below can be calculated, and it can be determined whether the skewness (absolute value) is greater than a preset skewness standard value (SK _standard ).

[Equation 5]

Here, γ represents the skewness of the ultrasonic signal value, N represents the number of ultrasonic signal values within the interval, O(t _n ) represents the ultrasonic signal value at time t _n , av represents the mean within the interval, and σ represents the standard deviation within the interval.

When calculating the skewness γ of the P1 section according to the above [Mathematical Formula 5], the number of ultrasonic signal values N is 6, the average av is 30, and the standard deviation σ is 1.82, so the skewness γ becomes -0.33. At this time, when the skewness standard value SK _standard is 1.5, the absolute value of the skewness γ of the P1 section, 0.33, is smaller than the skewness standard value SK _standard of 1.5, so it can be determined that the asymmetry is not large in the P1 section.

Similarly, when calculating the skewness γ of the P2 section, the number of ultrasonic signal values N is 6, the average av is 30, and the standard deviation σ is 2.71, so the skewness γ becomes -1.71. Therefore, the absolute value of the skewness γ of the P2 section, 1.71, is greater than the skewness standard SK _standard of 1.5, so it can be determined that the P2 section has a large asymmetry.

Accordingly, in the fourth embodiment, it can be determined that a failure has occurred in the ultrasonic generator (111) only when N _{dif_num} and N _{dif_disp} exceeding the number standard value (S _standard ) and the dispersion standard value (D _standard ) according to [Mathematical Formula 1] and [Mathematical Formula 2] are calculated in the same manner as described above, and the skewness calculated by [Mathematical Formula 3] is greater than the skewness standard value, so that the asymmetry of the data distribution is high.

Meanwhile, in Fig. 6, an example in which sections (P1, P2, P3, P4) are divided into preset time intervals is described, but in the case where P _new calculated according to [Mathematical Formula 3] and [Mathematical Formula 4] is 1, the section can be adjusted by resetting the nth ultrasonic signal value to the starting value of the new section.

Referring to FIGS. 7 and 8, the slope of consecutive ultrasonic signal values is calculated, and if the slope is greater than the reference slope (I _standard ) three times in a row, it is appropriate to judge that there is a normal change in the ultrasonic signal value (a rise or fall in the continuous ultrasonic signal value) and reset the section, so the section can be adjusted by resetting the n+3rd ultrasonic signal value as the starting value of a new section.

That is, as illustrated in FIG. 7, since the slope I ₈ connecting the ultrasonic signal values of t ₈ and t ₇ , the slope I ₉ connecting the ultrasonic signal values of t ₉ and t ₈ , and the slope I ₁₀ connecting the ultrasonic signal values of t ₁₀ and t ₉ are all greater than the reference slope (I _standard ) of 0.3, P _new becomes 1, and the P3 section and subsequent sections can be reset so that the n+3rd ultrasonic signal value t ₁₁ becomes the starting value of the new section.

Through this, it is possible to prevent the occurrence of inaccurate fault judgment according to [Mathematical Formula 1] and [Mathematical Formula 2] in sections where the ultrasonic signal value naturally rises or falls, and thus, in the present invention, fault judgment with higher reliability can be achieved.

For example, the fault diagnosis unit may monitor whether a fault occurs in the ultrasonic receiver (112) by setting the size of the reflected signal and the transmitted signal received by the ultrasonic receiver (112) as output values in the same manner as the method of diagnosing a fault in the ultrasonic generator (111).

Example 2

Meanwhile, a true-hypothetical defect detection method of a true-hypothetical defect detection system (100) using deep learning algorithm-based ultrasonic scan data according to an embodiment of the present invention comprises: a first step in which the data collection unit (110) collects B-Scan data and C-Scan data simultaneously and configures the B-Scan data and the C-Scan data in a one-pair format; a second step in which the data learning unit (120) generates an artificial intelligence learning model that detects at least one defect pattern based on the B-Scan learning data and the C-Scan learning data stored in a pre-designated learning DB; a third step in which the defect detection unit (130) inputs the C-Scan data collected by the data collection unit (110) into the artificial intelligence learning model generated by the data learning unit (120) to identify at least one defect area and generate coordinate data corresponding to the defect area; and a third step in which the defect determination unit (140) determines the X-axis coordinate of the coordinate data of the defect area generated by the defect detection unit (130) as the B-Scan data. It may include a fourth step of mapping data, checking whether there is a defective area from the B-Scan data, determining a false defect if there is no defective area, and determining a true defect if there is a defective area.

Therefore, according to the present invention as described above, by analyzing a defect area based on B-Scan data that provides vertical cross-sectional information based on the intensity and reflection time of an ultrasonic signal for an analysis target and C-Scan data that collects reflection data at a specific depth and is expressed as a two-dimensional image of a horizontal plane, a true-false defect detection system and method using ultrasonic scan data based on a deep learning algorithm that can reduce the occurrence of misdiagnosis due to a small pixel size in a technology that detects defects using only existing C-Scan data and can effectively distinguish between true defects and false defects can be provided.

In addition, a true-false defect detection system and method utilizing ultrasonic scan data based on a deep learning algorithm capable of more accurate detection of a defective area can be provided by primarily detecting a defective area from C-Scan data and analyzing B-Scan data corresponding to the C-Scan data in which the defective area is detected to determine whether the defect is true-false.

In addition, according to an embodiment of the present invention, a control method of a true-false defect detection system using ultrasonic scan data based on a deep learning algorithm may be recorded on a computer-readable medium including program commands for performing various computer-implemented operations. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination. The program commands may be specially designed and configured for the present invention or may be known and usable by those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

Although the embodiments of the present invention have been described with limited examples and drawings, the embodiments of the present invention are not limited to the embodiments described above, and various modifications and variations are possible based on this description by those skilled in the art to which the present invention pertains. Therefore, the embodiments of the present invention should be understood solely by the scope of the claims set forth below, and all equivalent or equivalent modifications thereof are deemed to fall within the scope of the present invention.

110: Data collection unit 111: Ultrasonic generator
112: Ultrasonic receiver
113: First scan data generation unit
114: Second scan data generation unit
115: Group Settings
120: Data Learning Department
130: Defect detection unit
140: Defect determination section 141: Analysis scope setting section
142: Area of Interest Setting Section
143: Signal size analysis unit

Claims (10)

A data collection unit that simultaneously collects B-Scan data and C-Scan data and configures the B-Scan data and the C-Scan data in a one-pair format;
A data learning unit that trains an artificial intelligence learning model that detects at least one defect pattern based on B-Scan learning data and C-Scan learning data;
A defect detection unit that inputs the C-Scan data collected by the data collection unit into the artificial intelligence learning model generated by the data learning unit to identify at least one defect area and generate coordinate data corresponding to the defect area; and
A defect determination unit that sets a region of interest by mapping the X-axis coordinates of the coordinate data of the defect region generated by the defect detection unit to the B-Scan data, checks whether there is a defective region within the region of interest from the B-Scan data, and determines that there is no defective region as a false defect if there is, and determines that there is a defective region as a true defect if there is;

The above data collection unit,
An ultrasonic generator that emits ultrasonic waves in the range of 1 to 50 MHz to a preset analysis target;
An ultrasonic receiver that collects a reflection signal reflected from the analysis target and a transmission signal passing through the analysis target;
A first scan data generation unit that generates the C-Scan data including depth information of the analysis target and internal structure information of the horizontal plane based on the reflection signal and the transmission signal received by the ultrasonic receiver;
A second scan data generation unit that generates the B-Scan data including cross-sectional image information corresponding to the depth information of the analysis target based on the reflection signal and the transmission signal received by the ultrasonic receiver; and
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that it includes a group setting unit that groups the C-Scan data generated by the first scan data generation unit and the B-Scan data generated by the second scan data generation unit into the one-pair format. delete In the first paragraph,
The above defect detection unit,
By inputting the C-Scan data into the artificial intelligence learning model generated in the data learning unit, object detection result data including at least one defect area information is generated and transmitted to the defect determination unit,

A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that when the object detection result data is not generated, a defect determination end signal is generated and transmitted to the defect determination unit so that the determination process of the false defect and the true defect through the defect determination unit is not performed. In the third paragraph,
The above defect detection unit,
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that when the above object detection result data is generated, the C-Scan data is stored as defect judgment data. In the first paragraph,
The above defect detection unit,
When the defective area is identified from the C-Scan data, a bounding box is created in the defective area, and the x _min , y _min , x _max and y _max coordinates of the defective area are extracted.
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that an origin marker is generated in the C-Scan data, and coordinate data corresponding to the x _min , y _min , x _max and y _max coordinates is generated by applying an XY two-dimensional coordinate system based on the origin marker. In paragraph 5,
The above defect determination unit,
An analysis range setting unit that sets the X-axis analysis range of the B-Scan data based on the coordinate data generated by the defect detection unit;
An area of interest setting unit for setting at least one area of interest within the X-axis analysis range set in the above analysis range setting unit; and
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that it includes a signal size analysis unit that analyzes the signal size within the region of interest to determine the true defect and the false defect. In the first paragraph,
The above defect determination unit,
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that when the B-Scan data is determined to be the true defect, the B-Scan data is stored as true defect result data, and when the B-Scan data is determined to be the false defect, the B-Scan data is stored as false defect result data. In the first paragraph,
The above data collection unit,
A fault diagnosis unit that monitors whether a fault occurs in the above ultrasonic generator; and
Further comprising an alarm generating unit that generates a user alarm signal in response to the diagnosis result of the above fault diagnosis unit;

The above fault diagnosis unit,
A true-false failure detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that the ultrasonic generator is diagnosed to have a failure if the upper and lower difference value (N _{dif_num} ) calculated by [Mathematical _Formula 1] below by analyzing the size of the ultrasonic signal, which is the output value of the ultrasonic generator, is greater than a preset number standard value (S standard), and at the same time, the upper and lower dispersion difference value (N _{dif_disp} ) calculated by [Mathematical Formula 2] below is greater than a preset dispersion standard value (D _standard ).

[Mathematical Formula 1]

(Here, N _{up_num} is the number of ultrasonic signal values exceeding the average value within a preset interval, N _{down_num} is the number of ultrasonic signal values below the average value within a preset interval, and N _{dif_num} is the absolute value of the difference between N _{up_num} and N _{down_num} , respectively.)

[Equation 2]

(Here, is the sum of the variances of each ultrasonic signal value exceeding the average value within a preset interval. is the sum of the variances of each ultrasonic signal value below the average within a preset interval, and N _{dif_disp} is and ) means the absolute value of the difference between In paragraph 8,
The above fault diagnosis unit,
A true-false failure detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that it diagnoses that a failure has occurred in the ultrasonic generator when the condition of the above clause 7 is satisfied and the skewness (γ) calculated by [Mathematical Formula 5] below is greater than a preset skewness standard value (SK _standard ).

[Equation 5]


(Here, γ represents the skewness of the ultrasonic signal value, N represents the number of ultrasonic signal values within the interval, O(t _n ) represents the ultrasonic signal value at time t _n , av represents the average value within the interval, and σ represents the standard deviation within the interval.) In paragraph 9,
The above fault diagnosis unit,
A true-false defect detection system utilizing ultrasonic scan data based on a deep learning algorithm, characterized in that a preset section is divided into preset time intervals, and when P _new satisfying [Mathematical Formula 3] and [Mathematical Formula 4] below is 1, the output value of the n+3rd ultrasonic generator is reset to the starting value of a new section.

[Equation 3]

(Here, I _n is the slope connecting the consecutive ultrasonic signal values, t _n is the time when the nth ultrasonic signal value was output, t _n-1 is the time when the n-1th ultrasonic signal value was output, O(t _n ) is the ultrasonic signal value at time t _n , and O(t _n-1 ) is the ultrasonic signal value at time t _n-1 .)

[Equation 4]

(Here, I _standard means the standard slope)