KR20220160401A - Deterioration degree evaluation apparatus and method for evaluating deterioration degree - Google Patents

Abstract

A degradation evaluation method according to an embodiment of the present invention includes steps of: creating a plurality of image data by photographing a plurality of sections of a certain facility; generating a plurality of binary images by binarizing data of a plurality of images; extracting information of a plurality of pipes included in the plurality of binary images; inputting the information of the plurality of pipes into a pre-learned algorithm, grouping the plurality of binary images, and ranking the deterioration degrees of each group; and determining a final number of groups and a degree of deterioration of each group based on at least one of operation history and maintenance information of the facility. One objective of the present invention is to provide the deterioration evaluation device and the deterioration evaluation method capable of evaluating the degree of deterioration in an optimized method.

Description

Deterioration evaluation device and degradation evaluation method

The present invention relates to a deterioration evaluation device and a deterioration evaluation method, and more particularly, to an apparatus and method capable of evaluating the deterioration degree by utilizing image preprocessing and unsupervised learning classification techniques for defect detection of equipment. .

During preventive maintenance of major facilities such as boilers and turbines of power plants, in order to non-destructively evaluate the degree of deterioration of target parts, the metal surface is copied as shown in FIG. 1 through polishing and etching processes, and then the reference image and A technique for evaluating damage types and deterioration grades is used through a comparative method.

1 is a view showing a film after surface replication and replication for boiler damage evaluation.

2 shows reference images for each deterioration grade based on VTT of steel X20, which is a representative steel type applied to standard coal-fired power plants.

Depending on the heat treatment temperature and time, the microstructure is divided into six stages from A to F, and the life consumption rate is finally obtained by linking this with the creep porosity and microcrack formation criteria.

The conventional deterioration evaluation method has the following problems.

First, the images used for the classification of deterioration in the existing VTT report were made by exposing test pieces of each steel type to a constant temperature for a certain period of time, and there is a big difference from the actual operating environment of power plant facilities.

During facility operation, the temperature of tubes and pipes is not kept constant, and temperature deviations occur depending on the location of the same parts.

In addition, it is exposed to various multiaxial stress states according to the pressure and temperature change due to the internal fluid, thermal stress due to contraction and expansion, residual stress generated during the manufacture of pressure vessels, and part shape.

Therefore, rather than the existing method of classifying the degree of deterioration based on this after manufacturing the test piece considering only the temperature, it is a method that can increase reliability by relatively comparing the images of actual parts to set the order and classify them.

Second, comparison and classification of existing surface reproduction images are performed by experts in the field, but when there is a large amount of data acquired in the field, human error may occur because it takes a long time to read.

In addition, since power generation facilities are judged by other experts rather than the same person over a long period of time, a problem in that standards for ratings vary from person to person may occur.

3 is a comparison of some images among existing analysis results.

The two figures are the results of analyzing different facilities in the same year. The picture on the right is an image of Unit 6, which was determined to be the DE stage, which is in the middle of the D and E stages.

Comparing both sides, it was determined that the E stage on the left was relatively more degraded than the DE stage on the right. However, when compared again with the reference image for each grade in FIG. 2 , which is the criterion for judgment, it can be seen that the right picture is more deteriorated than the left picture.

Therefore, it can be seen that errors may occur frequently during the process of judging an image by an expert.

Lastly, since the images in the existing VTT report provide only one or two pictures for each grade, it is difficult to select an accurate grade.

Accurate classification is possible only when various images are provided according to magnification according to grain size, and it is difficult to use as a reference point when classifying images due to low resolution and contrast ratio.

That is, in the conventional deterioration evaluation method, errors may occur in the process of repeatedly judging a large amount of data when determining the image deterioration level by the judgment of existing experts. As a result, errors may occur due to differences in standards.

It is necessary to develop an algorithm that prevents human errors and enables rapid diagnosis by excluding judgment by human intuition as much as possible and classifying grades by applying artificial intelligence techniques when analyzing such metal tissue replication images.

One object of the present invention is to provide a deterioration evaluation device and a deterioration evaluation method capable of evaluating the degree of deterioration in an optimized method.

Another object of the present invention is to provide a deterioration evaluation device and a deterioration evaluation method capable of evaluating the deterioration degree of equipment using an unsupervised learning classification technique.

A deterioration evaluation method according to an embodiment of the present invention includes generating a plurality of image data by photographing a plurality of sections of a certain facility; generating a plurality of binary images by binarizing the plurality of image data; extracting a plurality of pipe information included in the plurality of binary images; inputting the plurality of pipe information into a pre-learned algorithm, grouping the plurality of binary images, and ranking deterioration degrees of each group; and determining the final number of groups and the deterioration degree of each group based on at least one of operation history and maintenance information of the facility.

In an embodiment, the generating of the plurality of binary images may include generating the plurality of binary images from the plurality of image data by using an adaptive threshold technique.

In an embodiment, the generating of the plurality of binary images may include dividing the plurality of image data into several regions, setting a threshold value, and setting a contrast ratio of the threshold value as a boundary to divide the image into two regions. It is characterized in that the plurality of binary images are generated by dividing.

In an embodiment, the plurality of pipe information includes at least one of the size, number, and circumferential length of particles.

In an embodiment, the extracting of the plurality of pipe information may include extracting the plurality of pipe information by using a contour technique indicating a particle boundary.

In an embodiment, the contour technique is characterized in that it identifies the boundary of a particle by connecting the edge boundary of a part having the same color or the same color intensity.

In an embodiment, the pre-learned algorithm includes a clustering method, which is one of the unsupervised learning methods, and the step of ranking the deterioration degrees of each group uses the clustering method, which is one of the unsupervised learning methods. It is characterized by grouping a plurality of binary images.

In an embodiment, the clustering method is characterized in that it is a merged clustering method.

In an embodiment, the step of ranking the degree of deterioration of each group may include grouping a plurality of binary images into an arbitrary number of groups using a dendrogram, and classifying the degree of deterioration of each group.

In an embodiment, the method further includes generating a damage map by connecting the images classified for each group.

An apparatus for evaluating deterioration degree according to another embodiment of the present invention includes an image acquisition module for generating a plurality of image data by photographing a plurality of sections of a certain facility; and generating a plurality of binary images by binarizing the plurality of image data, extracting a plurality of pipe information included in the plurality of binary images, and inputting the plurality of pipe information to a pre-learned algorithm, and a control unit that groups binary images of , ranks the degree of deterioration of each group, and determines the final number of groups and the degree of deterioration of each group based on at least one of operation history and maintenance information of the facility.

In an embodiment, the control unit may generate the plurality of binary images from the plurality of image data by utilizing an adaptive threshold technique.

In an embodiment, with respect to the plurality of image data, the control unit divides the image into several regions, sets a threshold value, sets a contrast ratio of the threshold value as a boundary, divides the image data into two regions, and divides the plurality of binary images into two regions. characterized by the creation of

In an embodiment, the plurality of pipe information includes at least one of the size, number, and circumferential length of particles.

In an embodiment, the control unit is characterized in that the plurality of pipe information is extracted by using a contour technique representing the boundary of the particle.

In an embodiment, the contour technique is characterized in that it identifies the boundary of a particle by connecting the edge boundary of a part having the same color or the same color intensity.

In an embodiment, the pre-learned algorithm includes a clustering method, which is one of the unsupervised learning methods, and the control unit groups the plurality of binary images using the clustering method, which is one of the unsupervised learning methods. to be characterized

In an embodiment, the clustering method is characterized in that it is a merged clustering method.

In an embodiment, the controller may group a plurality of binary images into an arbitrary number of groups using a dendrogram, and classify the deterioration degree of each group.

In an embodiment, the controller may generate a damage map by connecting images classified for each group.

A computer program according to an embodiment of the present invention is combined with a computer that is hardware and stored in a computer-readable recording medium to perform a degradation evaluation method.

Other specific details of the invention are included in the detailed description and drawings.

According to the present invention, as one of the digital transformations, the present invention develops a degradation evaluation model, which is an image preprocessing and unsupervised learning classification method for boiler damage evaluation, and improves accuracy, thereby reducing investment costs during power plant operation.

In addition, the present invention is capable of predicting and improving reliability of facility operation linked to real-time operation information, is used for technology consulting in the power generation field using digital power plant implementation technology, and is used for virtual power plant performance and deterioration state simulation algorithms of the boiler digital twin system. can be utilized

The 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 below.

1, 2 and 3 are conceptual diagrams for explaining a conventional deterioration evaluation method.
4 is a conceptual diagram showing the deterioration degree evaluation device of the present invention.
5 is a flowchart for explaining a representative deterioration evaluation method of the present invention.
6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20 show the deterioration observed in FIG. 5 It is a conceptual diagram for explaining the evaluation method.

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

In addition, since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

Terminology used herein is for describing the embodiments and is not intended to limit the present invention. In this specification, singular forms also include plural forms unless specifically stated otherwise in a phrase. As used herein, "comprises" and/or "comprising" does not exclude the presence or addition of one or more other elements other than the recited elements.

Terms including ordinal numbers such as first and second described herein may be used to describe various components, but the components are not limited by the terms. That is, the above terms are used only for the purpose of distinguishing one component from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the present invention. The term and/or includes any combination of a plurality of related recited items or any of a plurality of related recited items.

In addition, when a component is referred to as being “connected” or “connected” to another component, it may be directly connected or connected to the other component, but other components may exist in the middle. have. On the other hand, when an element is referred to as “directly connected” or “directly connected” to another element, it should be understood that no other element exists in the middle.

In addition, terms used in this application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, the terms "include" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

In addition, unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this application, it should not be interpreted in an ideal or excessively formal meaning. don't

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

4 is a conceptual diagram showing the deterioration degree evaluation device of the present invention.

Referring to FIG. 4 , the deterioration evaluation apparatus according to an embodiment of the present invention may acquire a surface image of a facility (eg, a boiler or a turbine) 200 in the field.

The deterioration evaluation apparatus 100 may include an image acquisition module 110, a mechanism and driving module 120, a lighting module 130, and a control unit 180 (or an operation module).

The image acquisition module 110 may be, for example, an image sensor (or camera).

The image acquisition module 110 may generate a plurality of image data by photographing a plurality of sections of a facility (eg, a boiler facility, a turbine, etc.).

The image acquisition module 110 is a part that captures a surface image of a facility, and acquires an image of hundreds of magnifications (eg, 500 times) from the surface through light and digital zoom.

The mechanism and the driving module 120 bring the image acquisition module into close contact with the curved equipment surface, and the image sensor moves to acquire an image and drive and control it in the z-axis direction with a precision of several tens of microns to focus.

The lighting module 130 may be connected to the mechanism and driving module 120 and the image acquisition module 110 to provide light to the surface so as to obtain a clear image.

The control unit 180 (or operating module) includes a function of storing, correcting, and dividing the acquired image, and classifying the degradation grade.

5 is a flowchart for explaining a representative deterioration evaluation method of the present invention.

Referring to FIG. 5 , the controller 180 of the deterioration evaluation device may control the image acquisition module to generate a plurality of image data by photographing a plurality of sections of a certain facility (S210).

The control unit 180 may binarize the plurality of image data generated by the image acquisition module 110 to generate a plurality of binary images (S220), and extract a plurality of pipe information included in the plurality of binary images. Yes (S230).

The controller 180 may generate the plurality of binary images from the plurality of image data by utilizing an adaptive threshold technique.

The controller 180 may use average and Gaussian methods among adaptive threshold techniques, and may set the block size to 3 to 15 and the differential coefficient C value to 2 to 10.

Also, with respect to the plurality of image data, the controller 180 divides the image into several regions, sets a threshold value, and divides the image into two regions by setting the contrast ratio of the threshold value as a boundary to generate the plurality of binary images. can do.

The plurality of pipe information may include at least one of the size, number, and circumferential length of particles.

The controller 180 may extract the plurality of pipe information by using a contour technique representing the boundary of the particle.

The contour technique may refer to a technique of identifying the boundary of a particle by connecting the edge boundary of a part having the same color or the same color intensity.

In addition, the control unit 180 inputs the plurality of pipe information to the previously learned algorithm, groups the plurality of binary images, ranks the deterioration degree of each group (S240), and records operation history and maintenance of the facility. Based on at least one of the pieces of information, the final number of groups and the degree of deterioration of each group may be determined (S250).

The pre-learned algorithm may include a clustering method, which is one of unsupervised learning methods. The clustering technique may be an agglomerative clustering technique.

The controller 180 may group the plurality of binary images using a clustering method (merging grouping method), which is one of the unsupervised learning methods.

The controller 180 may calculate the number, area, and circumference of the particles in the image through the contour technique, then obtain the average area and average circumference for each image, and analyze them using the merged clustering technique.

In addition, the control unit 180 may determine the number of groups when the reduction ratio of the deformation value starts to fall below a threshold value as the number of groups increases by utilizing the elbow technique.

The controller 180 may group a plurality of binary images into an arbitrary number of groups using a dendrogram, and classify (or rank) the degree of deterioration of each group.

The controller 180 may generate a damage map by connecting images classified for each group.

The control unit 180 may determine the final number of groups and the degree of deterioration based on a facility maintenance and operation history for a certain facility (eg, power plant, boiler, turbine, etc.).

At this time, the controller 180 may connect the images classified for each group to neighboring areas by n×n (n is a natural number), draw a damage map for a wide area, and convert the histogram to compare.

Hereinafter, a deterioration evaluation method will be described in more detail with reference to the accompanying drawings.

6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20 show the deterioration observed in FIG. 5 It is a conceptual diagram for explaining the evaluation method.

Fig. 6 shows a conceptual diagram of an evaluation method for evaluating a degree of deterioration of a facility from an image acquired from a certain facility.

Referring to FIG. 6 , the controller 180 may perform steps of acquiring a large number of surface replication images from vulnerable parts for each facility and converting the acquired images into binary images using an adaptive threshold technique. .

In addition, the control unit 180 analyzes the area and circumference (interface length) of each particle in the image based on the contour, and unsupervised learning from the characteristics of the number, area, and circumference of the particles for each analyzed image. A step of classifying all images into a random number of groups by performing image classification using the merged clustering technique and ranking the entire deterioration degree may be performed.

In addition, the control unit 180 may finally perform a step of determining the final number of groups and the degree of deterioration based on the power plant facility history.

Conventionally, after images are acquired, an expert reads each image and groups them by degree of deterioration.

However, as the operating period of the facility gradually increases, the degree of deterioration increases, and in order to increase the reliability of the evaluation method, it is necessary to increase the location and area of image replication, and the research institute still has tens of thousands of images.

In addition, in order to reflect the facility operation history at the time of judgment, it is essential to develop a classification algorithm and carry out collectively and quickly from the acquired image to the final deterioration degree judgment rather than making a direct judgment by an existing expert.

The controller 180 may analyze the acquired image data using image analysis software.

The image analysis software is capable of image optimization functions such as auto focus, brightness, and contrast control, as well as basic image analysis functions such as grain size, phase fraction, thickness measurement, and comparison of published materials, and supervised learning-based image analysis functions. can be done

- Image Optimization: Auto Focus, Brightness, Contrast Control, Panorama, Depth Enhancement, etc.

- Image analysis: grain size, phase fraction, thickness measurement, comparison of published materials, etc.

- Phase analysis using machine learning (ZEN Intelesis): supervised learning-based phase classification

Although it is possible to analyze the number, area, and circumferential length of particles through the existing image analysis function, the problem is how clearly the particles can be distinguished from the acquired image (image data).

Conventionally, there was a disadvantage in that the area of dark particles in each corner area where overall brightness was dark was greatly analyzed because it was classified only by absolute contrast difference.

On the other hand, the supervised learning-based phase classification technique used in the deterioration evaluation apparatus of the present invention reinforces these disadvantages, and as shown in FIG. 7, it is possible to distinguish particles very clearly and clearly.

The two pictures shown above in FIG. 7 show that particles are classified in the conventional method, and the two pictures shown below in FIG. 7 show pictures in which particles are separated from images using supervised learning-based image analysis software.

As shown in FIG. 7 , an image expressed in only two colors is referred to as a binary image.

This can be divided into two areas by setting the contrast ratio of the threshold value between black and white as a boundary.

When creating such a binary image, it is possible to set a threshold value by dividing the image into several areas, not based on the entire image, and this is called an adaptive threshold technique.

By optimizing the existing code provided as an open source, it is possible to clearly distinguish the area and boundary of the particle regardless of the location of the acquired image as shown in FIG. 7 .

In addition, the number, size, and circumference of each particle in one image can be obtained by optimizing the contour technique for the size, number, and circumference of particles that could be implemented in the existing image analysis function.

An image contour refers to a line connecting edge boundaries of parts having the same color or color intensity, and the controller 180 can quantitatively analyze the characteristics of particles in the image using the contour technique.

Existing deterioration classification criteria classify images into six grades from A to F.

As for the actual equipment grade, as the operating period increases, the initial state A and B grades are extremely rare, and mainly three grades from D to F and intermediate grades are made and used.

Therefore, rather than classifying grades by making absolute standards, it is necessary to variably classify grades according to criteria that can be effectively classified according to the condition of each facility.

In order to analyze images using these characteristics, it is advantageous to use the clustering technique, which is an unsupervised learning method among recent machine learning methods.

Representative cluster analysis methods include k-means clustering, agglomerative clustering, and density-based spatial clustering of application with noise (DBSCAN).

In the present invention, the degree of deterioration was classified using the merged clustering method with the area and circumference of the particle as variables.

The k-means clustering technique is not appropriate because different groups are changed each time an analysis is performed. In the case of DBSCAN, the coefficients were combined several times, but the groups could not be classified smoothly. It is judged that the characteristic distribution of image data is not suitable for classification.

The surface replication image inspection method has a disadvantage in that it is a method of inspecting a partial area of the entire facility area.

Since the tube is located in the flow of combustion air in the boiler, the metal temperature distribution varies greatly not only in the longitudinal direction but also in the circumferential direction.

Therefore, it is necessary to increase the accuracy of diagnosing the condition of the entire facility by increasing the inspection location and area.

As mentioned above, after connecting each newly graded image through merge clustering in N×N, checking the grade distribution and comparing it with other equipment and parts with a histogram, etc., it is possible to evaluate the degree of deterioration more reliably.

The deterioration degree evaluation apparatus of the present invention can acquire a surface replica image (ie, a plurality of image data) for each facility location.

In power plants, non-destructive inspections are conducted if necessary along with visual inspections at every maintenance for vulnerable locations of high-temperature and high-pressure main facilities to determine operation status and maintenance intervals.

At this time, among the traditional non-destructive methods, classification using surface replication images is being carried out, and as the operating time increases, the degree of equipment damage increases, so there is a need to inspect more locations and wider areas to increase the reliability of the evaluation technique.

Although images of similar facilities in power plants are acquired and retained over a long period of time in various locations, perfect results cannot be expected because the same expert cannot accurately perform all image deterioration grade determinations.

When tens of thousands of images of various facilities are classified through a classification algorithm through machine learning rather than classification by an expert, the degree of deterioration can be accurately compared for each facility in each power plant, and the change in deterioration degree according to the operating period can be accurately analyzed. can

In the case of the acquired surface replication image, the acquired facility, location, and date and time information are included in the file name as shown in FIG.

7 is a diagram illustrating an example of image acquisition and file name standardization for each facility.

The control unit 180 may binarize the acquired image (a plurality of image data) (ie, generate a plurality of binary images).

The acquired image is distributed in various gray levels between white and black.

For image analysis, it is advantageous to separate into two colors, white and black, and this is called a binary image.

Determining the boundary point separating white and black when performing binary imaging is called thresholding, and there are several techniques.

8 is a diagram showing a resulting image according to the image binarization technique.

There is a global thresholding method that divides the entire image into one threshold value, and there is an Otzu evolution algorithm that improves this. However, this has a disadvantage in that the boundary cannot be clearly distinguished when the brightness of the entire image is not constant.

To improve this, there is an adaptive threshold technique for dividing an image into several areas of the present invention and obtaining a boundary value using only pixel values around the area, and as shown in FIG. Compared to the algorithm, it is possible to distinguish particles uniformly without deviation according to location.

The controller 180 may analyze image features based on a contour.

Contour means a line that connects places with the same value, such as contour lines and isobars, and image contour is a line that connects the edge boundaries of parts with the same color or color intensity, and is used to identify the shape and area of a target.

For a more accurate analysis, it is necessary to use a binary image, and the area and circumferential length of each black particle can be analyzed for the image obtained by the binaryization method above.

10 is a view showing the analysis of Contour-based particle area and circumferential length.

Referring to FIG. 10 , when the analysis was performed for each image, it was confirmed that more than 10,000 particles were distributed in one image, and it could be confirmed that each particle was sequentially and accurately counted.

11 is a table showing the analysis results of the area and circumferential length of each particle in the image.

FIG. 11 shows values obtained by calculating the area and circumference of each particle for the first image analyzed in FIG. 10 .

In order to represent one image, the controller 180 calculates the sum of the area and circumference of all particles and the average particle area and circumference in consideration of the number of particles to obtain characteristic values for each image. have.

The controller 180 may perform image classification using an agglomerative clustering algorithm.

The control unit 180 is a step of classifying and sequencing the entire image into an arbitrary number of groups by performing image analysis using the merged clustering technique, which is unsupervised learning, from the characteristics of the number of particles, area, and circumference of each image.

Clustering algorithms, which are a classification of unsupervised learning within machine learning, largely include k-means clustering, merged clustering, and DBSCAN. In the present invention, image analysis was performed using each algorithm and the optimal algorithm was selected.

12 is a diagram showing k-means clustering analysis results.

In the present invention, k-mean clustering analysis was performed on 130 sample images based on the area and circumference of each particle, and as a result, three clusters and each central value were obtained as shown in FIG. 12.

k-means is the most popular clustering algorithm because it is relatively easy to understand, easy to implement, and relatively fast, but the biggest drawback is that the center point is initially randomly selected and calculated, so the shape of the cluster changes every time depending on the initial value set as a random number. .

DBSCAN's density-based clustering analysis technique is an analysis technique that assumes that the data in the same cluster will be in similar locations, that is, the density will be high.

This algorithm can find even complex shapes, and can distinguish points that do not belong to any class.

However, it has a disadvantage that it is slightly slower than merged clustering or k-means. Using this, the image was analyzed to obtain the result shown in FIG. 13 . 13 is a diagram showing the results of density-based clustering analysis.

Although the EPS and the minimum cluster coefficient, which are the main coefficients for forming clusters, were adjusted, it was difficult to obtain two or more clusters as shown in FIG. 13.

This is because the image analysis results were difficult to separate into several clusters using the density-based clustering technique.

On the other hand, the merged clustering algorithm applied in the present invention is not randomly selected at the beginning like k-means clustering, but designates a feature value for each image as one cluster and merges neighboring points in order until the condition is satisfied. get it done

The images were classified using the merged clustering technique, and the results shown in FIG. 14 were obtained. 14 is a diagram showing the results of image analysis using the merged clustering technique.

It can be seen that the 130 images are divided into 5 clusters based on the average area and average circumferential length of the particles.

In the present invention, a dendrogram can be used as another method of expressing merged clusters.

15 is a diagram showing merged cluster analysis results expressed as a dendrogram.

The apparatus for evaluating the degree of deterioration of the present invention initially treats each data as one cluster, goes through a process of sequentially merging with neighboring clusters, and can show the result as a dendrogram as shown in FIG. 15 .

The dendrogram represents the number of each data value in turn, and it is possible to check the process of clustering through which path each data goes through, so that the degree of similarity between each data can be visually checked, and it can be easily checked visually when changing the number of clusters.

The y-axis of the dendrogram does not simply indicate when two clusters are merged in the merging algorithm, it shows how far apart the merged clusters are.

The deterioration evaluation apparatus of the present invention can classify through a dendrogram when classifying deterioration grades into an arbitrary number of groups.

In addition, the deterioration evaluation apparatus may use the elbow technique shown in FIG. 16 .

16 is a diagram showing a method for determining the number of clusters using the elbow technique.

The elbow technique has the advantage of being able to check as an indicator how much the distortion on the y-axis decreases as the number of groups (K-value) on the x-axis increases.

In the case of these 130 images, the deformation value rapidly decreases until the number of groups is 3, and then gradually decreases.

Therefore, it is intuitively determined that it is appropriate to determine the number of groups from 3 to 5, and in order to more quantitatively determine the number of groups, it is possible to use Equation 1 below.

[Equation 1]

The k value when the k-1th and kth deformation value decreases to less than 0.4 of the k-th deformation value can be determined as the number of groups. In the case of FIG. 16, when the k value is 5, it is 0.43 and when it is 6, it is 0.36.

The controller 180 may classify (rank) the final deterioration degree (deterioration grade) in consideration of the operation history and maintenance information (maintenance history) of the facility.

The controller 180 may determine the final deterioration grade in consideration of facility maintenance and operation history.

The control unit 180 may check whether there is a maintenance history such as parts replacement and heat treatment in the corresponding part by searching the existing maintenance history for the deterioration grade obtained through image analysis.

In addition, the controller 180 may convert the temperature at which the actual facility is exposed based on temperature information for each location input in real time.

Through this process, the controller 180 can finally sort the deterioration levels in order within any one facility and classify them.

17 and 18 are diagrams showing results of merged cluster analysis expressed as a dendrogram after classification of the final degradation grade.

Referring to FIGS. 17 and 18, FIG. 17 shows the first highest (1) graded photo, the 65th middle grade (3), and the most deteriorated among 130 images as a result after the final degradation grade classification. It shows the last 130th image of the advanced 5th grade in order.

In the first photo, the shape of the lath, the microstructure of the earliest tempered martensite, is best preserved, and as the deterioration progresses, the interval of the lath gradually increases and the average area and circumference decrease.

In addition, the present invention has an effect of reducing errors generated from the results performed by experts in the past through such a classification technique to almost zero.

The controller 180 may create (draw) a damage map for a wider area than before by connecting the images classified for each group.

19 is a diagram illustrating a damage map.

19 shows a method of scanning a peripheral area with an image acquisition device. As shown in the upper figure of FIG. 19, when images are acquired in the direction of the arrow, classified by the unsupervised learning method from 1 to 5, and then connected in a 7 × 5 neighborhood, it can be visualized as a damage map as shown in the figure below.

Through this, the present invention has the advantage of being able to recognize the entire damaged area and shape within the part.

20 is a diagram comparing degrees of deterioration between similar facilities using a histogram.

Referring to FIG. 20, the present invention can quantitatively compare the same parts in the boiler by using a histogram as shown in FIG. 20 when quantitatively comparing the same parts.

As described above, the deterioration evaluation apparatus of the present invention includes the steps of acquiring a large number of surface replication images from vulnerable parts for each facility in order to classify the deterioration grade of any facility (eg, power plant facility); converting the acquired image into a binary image using an adaptive threshold technique; Analyzing the area and circumference length (interface length) of each particle based on the contour of the binary image; Classifying into an arbitrary number of groups by using Agglomerative clustering, which is unsupervised learning, from the characteristics of the number, area, and circumference of particles for each analyzed image; And a step of determining the final number of groups and the degree of deterioration based on the power plant facility maintenance and operation history can be performed.

Steps of a method or algorithm described in connection with an embodiment of the present invention may be implemented directly in hardware, implemented in a software module executed by hardware, or implemented by a combination thereof. A software module may include random access memory (RAM), read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or It may reside in any form of computer readable recording medium well known in the art to which the present invention pertains.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art to which the present invention pertains can be implemented in other specific forms without changing the technical spirit or essential features of the present invention. you will be able to understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

Claims (20)

generating a plurality of image data by photographing a plurality of sections of a facility;
generating a plurality of binary images by binarizing the plurality of image data;
extracting a plurality of pipe information included in the plurality of binary images;
inputting the plurality of pipe information into a pre-learned algorithm, grouping the plurality of binary images, and ranking deterioration degrees of each group; and
Deterioration evaluation method comprising the step of determining the final number of groups and the degree of degradation of each group based on at least one of the operation history and maintenance information of the facility. According to claim 1,
Generating the plurality of binary images,
Deterioration evaluation method, characterized in that for generating the plurality of binary images from the plurality of image data by utilizing an adaptive threshold technique. According to claim 2,
Generating the plurality of binary images,
For the plurality of image data, the image is divided into several regions, a threshold value is set, and a contrast ratio of the threshold value is set as a boundary to divide the image into two regions to generate the plurality of binary images. Way. According to claim 1,
The plurality of piping information includes at least one of the size, number and circumference of the particles. According to claim 1,
The step of extracting the plurality of pipe information,
Deterioration evaluation method, characterized in that for extracting the plurality of pipe information using a contour technique indicating the boundary of the particle. According to claim 5,
Wherein the contour technique is a technique for identifying a boundary of a particle by connecting an edge boundary of a part having the same color or the same color intensity. According to claim 1,
The pre-learned algorithm includes a clustering technique, which is one of the unsupervised learning methods,
The step of sequencing the degree of deterioration of each group,
Deterioration evaluation method characterized in that the plurality of binary images are grouped using a clustering method, which is one of the unsupervised learning methods. According to claim 7,
The clustering method is a deterioration evaluation method, characterized in that the merged clustering method. According to claim 1,
The step of sequencing the degree of deterioration of each group,
A deterioration degree evaluation method characterized by grouping a plurality of binary images into an arbitrary number of groups using a dendrogram and classifying the degree of deterioration of each group. According to claim 1,
Deterioration evaluation method further comprising the step of generating a damage map by connecting the images classified for each group. an image acquisition module for generating a plurality of image data by photographing a plurality of sections of a certain facility; and
Binaryizing the plurality of image data to generate a plurality of binary images, extracting a plurality of pipe information included in the plurality of binary images, and inputting the plurality of pipe information to a pre-learned algorithm, Deterioration degree evaluation device comprising a control unit that groups binary images, ranks the degree of deterioration of each group, and determines the final number of groups and the degree of deterioration of each group based on at least one of operation history and maintenance information of the facility . According to claim 11,
The control unit,
Deterioration evaluation apparatus, characterized in that for generating the plurality of binary images from the plurality of image data by utilizing an adaptive threshold technique. According to claim 12,
The control unit,
For the plurality of image data, the image is divided into several regions, a threshold value is set, and a contrast ratio of the threshold value is set as a boundary to divide the image into two regions to generate the plurality of binary images. Device. According to claim 11,
The plurality of piping information includes at least one of the size, number and circumferential length of particles. According to claim 11,
The control unit,
Deterioration evaluation device, characterized in that for extracting the plurality of pipe information using a contour technique representing the boundary of the particle. According to claim 15,
The contour technique is a technique for identifying a boundary of a particle by connecting an edge boundary of a part having the same color or the same color intensity. According to claim 11,
The pre-learned algorithm includes a clustering technique, which is one of the unsupervised learning methods,
The control unit,
Deterioration evaluation device, characterized in that for grouping the plurality of binary images using a clustering method, which is one of the unsupervised learning methods. 18. The method of claim 17,
The clustering method is a deterioration evaluation device, characterized in that the merged clustering method. According to claim 11,
The control unit,
A deterioration degree evaluation device characterized by grouping a plurality of binary images into an arbitrary number of groups using a dendrogram and classifying the degree of deterioration of each group. According to claim 11,
The control unit,
Deterioration evaluation device, characterized in that for generating a damage map by connecting the images classified for each group.

Images