KR20200033822A - Apparatus and Method for Detecting/Analyzing Defect of Windturbine Blade - Google Patents

Abstract

The present invention provides an apparatus and a method for recognizing/analyzing a defect of a blade of a wind power generator. According to an embodiment of the present invention, the method for recognizing and analyzing a defect of a wind power generator from a photographed image of an unmanned aerial vehicle by using deep learning comprises: a receiving process of receiving an image of the wind power generator photographed from the unmanned aerial vehicle; a blade detection process of detecting an image including a blade area of the wind power generator from the image of the wind power generator; a defect detection process of determining whether a defect exists in pixel units from the detected blade area image; and a defect analysis process of analyzing a property of a defect of a detection image in pixel units with a recognized defect from the blade area image.

Description

Apparatus and Method for Detecting / Analyzing Defect of Windturbine Blade

The present invention relates to a defect recognition / analysis apparatus and method, and more particularly, to an apparatus and method for detecting and analyzing a defect by analyzing an image of a wind power generator blade.

The contents described in this section merely provide background information for this embodiment, and do not constitute a prior art.

A wind power generator is a device that converts wind energy into electrical energy, and rotates the wings of the wind power generator to produce electricity with the rotational force of the wings. Typically, wind power generators have a height of several tens of meters, some of which exceed 100 meters (m). Also, its blades are usually several tens of meters (m) in size.

As such, due to the huge size of the wind turbine blades, it was a lot of risk and inconvenience for facility managers to climb the wind generators or blades and inspect the defects directly, especially the various types of defects in each blade of the wind generator. Since the wind power generator is installed in an environment in which strong wind is constantly blowing, the wind power generator manager must perform the work while facing the strong wind at a high altitude, so there is a lot of danger to the manager. In addition, even if the blade image of the wind power generator is secured, it is disadvantageous in that its result is significantly different depending on the physical / mental situation of the skilled person, since the determination of the defect is made only based on the expert's long experience.

Therefore, there is a need for a method of safely and quickly analyzing the presence or absence of blade defects in a wind power generator.

One embodiment of the present invention, using a deep learning (deep learning) from the image photographed through the unmanned flight apparatus recognizes the presence or absence of defects in the blade of the wind power generator, and furthermore a defect analysis device for analyzing the properties of the recognized defects And to provide a method.

Defect recognition and analysis method of a wind generator using deep learning from a photographed image of an unmanned aerial vehicle according to an embodiment of the present invention, receiving an image of the wind generator photographed from the unmanned aerial vehicle Receiving process; A blade detection process of detecting an image including a blade region of the wind generator from the image of the wind generator; A defect detection process of determining whether a defect exists from the detected blade region image in units of pixels and detecting the defect; And a defect analysis process of analyzing a defect property of a pixel-based detection image in which a defect is recognized from the blade region image.

According to an embodiment of the present invention, it is characterized in that it further comprises the step of storing the detected image in the pixel unit where the defect is recognized and the detected image in the pixel unit where the defect is not recognized, respectively, from the blade region image. Is done.

According to an embodiment of the present invention, the analyzed pixel-based detection image is classified according to the defect type and the defect level according to the defect property, and the classified image is further comprised of storing in a separate folder. Is done.

According to an embodiment of the present invention, the pixel unit image data stored in the folder is also characterized in that location data of the corresponding pixel is also included based on the received image of the wind generator.

According to an embodiment of the present invention, the defect detection process, the pixel containing the detected blade area in units of pixels, a non-blade area pixel, a blade area or a defect-free pixel and a blade area and a defective pixel It is characterized by distinguishing and displaying in different colors.

According to one embodiment of the present invention, the blade detection process, the defect detection process and the defect analysis process are characterized by using a convolutional neural network algorithm of the deep learning technique, but using different techniques in each process.

Defect recognition and analysis device of a wind power generator using deep learning from a photographed image of an unmanned aerial vehicle according to an embodiment of the present invention, receiving an image of the wind generator photographed from the unmanned aerial vehicle Communication department; A blade detector configured to detect an image including a blade region of the wind generator from the image of the wind generator; A defect detection unit for determining whether a defect exists from the detected blade region image in units of pixels and detecting the defect; And a defect analysis unit analyzing a defect property of a pixel-based detection image in which a defect is recognized from the blade region image.

According to an embodiment of the present invention, it is characterized in that the detected image in the pixel unit where the defect is recognized from the blade region image and the detected image in the pixel unit where the defect is not recognized further include a database that is stored in a separate folder. do.

According to an embodiment of the present invention, the database is characterized in that the detected pixel-by-pixel detection images are classified according to defect types and defect levels according to defect properties, and the classified images are stored in separate folders. .

According to an embodiment of the present invention, the database is characterized in that it includes pixel-based image data stored in the folder, and also includes location data of a corresponding pixel based on the image of the received wind power generator.

According to an embodiment of the present invention, the defect detection unit divides an image including the detected blade region into pixels, not a blade region, a blade region or a defect-free pixel, and a blade region and a defective pixel. And, it characterized in that the display in different colors.

According to an embodiment of the present invention, the blade detection unit, the defect detection unit and the defect analysis unit are characterized by using a convolutional neural network algorithm of the deep learning technique, but using different techniques in each process.

As described above, according to an aspect of the present invention, deep learning is used to recognize the presence or absence of a defect in a wind turbine blade from an image photographed through an unmanned aerial vehicle, and furthermore, the property of the recognized defect. It has the advantage of being able to analyze safely and quickly.

1 is a configuration diagram of a wind turbine defect analysis system according to an embodiment of the present invention.
2 is a block diagram of an apparatus for analyzing a defect in a wind power generator according to an embodiment of the present invention.
3 is a configuration diagram of an unmanned aerial vehicle according to an embodiment of the present invention.
Figure 4 is a flow chart showing a defect recognition / analysis method of a wind power generator according to an embodiment of the present invention.
5 is a flowchart illustrating a detailed defect recognition / analysis method of a wind power generator blade according to an embodiment of the present invention.
6 is a flowchart illustrating an image stitching technique in connection with an embodiment of the present invention.

The present invention can be applied to various changes and can have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each drawing, similar reference numerals are used for similar components.

Terms such as first, second, A, and B can be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from other components. For example, the first component may be referred to as a second component without departing from the scope of the present invention, and similarly, the second component may be referred to as a first component. The term and / or includes a combination of a plurality of related described items or any one of a plurality of related described items.

When an element is said to be "connected" or "connected" to another component, it is understood that other components may be directly connected to or connected to the other component, but may exist in the middle. It should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that no other component exists in the middle.

Terms used in the present 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 indicates otherwise. It should be understood that terms such as “include” or “have” in the present application do not preclude the existence or addition possibility of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification. .

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains.

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

In this specification, a defect determination of the object is described from image data of an object obtained by photographing an object or a part of the object through an unmanned aerial vehicle. Hereinafter, the object is described as an example of a wind generator as a facility and a part of the object as an example of a blade (blade) of the wind generator, but the scope of the present invention is not necessarily limited thereto. Meanwhile, the defect may be defined in various terms such as damage or destruction, and the defect may also be variously defined according to the subject's recognition of a defect in the corresponding technical field. Normally, a defect refers to all matters that may be a problem in relation to a functional element of an object, but is not limited thereto, and a non-functional element may also be included. In other words, the defect can be used in a sense that encompasses all states that are not complete without defects.

On the other hand, even though it is not specifically classified in this specification and is simply named and described as a wind power generator, the meaning may indicate all or part of the wind power generator according to the meaning of the context. In particular, a part of the wind power generator may be a blade of the wind power generator. have.

1 is a configuration diagram of a wind turbine defect analysis system 100 according to an embodiment of the present invention.

Referring to FIG. 1, a wind turbine defect analysis system 100 according to an embodiment of the present invention includes a wind turbine defect analysis device 110 and an unmanned aerial vehicle 120.

The defect analysis device 110 acquires an image of the wind power generator 130 from the unmanned aerial vehicle 120, recognizes whether there is a defect on the blade of the wind power generator or the wind power generator from the obtained image, and if the defect is recognized, the corresponding fault Analyze the properties of. Here, the property of the defect is meant to include at least one of, for example, the type or type of defect, the degree of defect, and the like. The defect analysis device 110 may store an image of a wind generator having a predetermined number or more, and learn the stored image using, for example, a deep learning technique. Image of various types of wind generators, such as images with wind generators, images without wind generators, images with flaws in the blades of the wind generators, images with no defects in the blades of the wind generators, etc. It can be stored in (110) and used for learning through the deep learning. Meanwhile, the storage may be performed not only inside the defect analysis device 110 but also through an external storage medium connected to the defect analysis device 110. According to the related art, even if the wind power generator manager directly inspects and analyzes the defects in the wind power generator, or analyzes the defects using the image of the wind power generator, the wind power generator manager has analyzed the presence of the defects in numerous photographed images. However, the defect analysis apparatus 110 according to the present invention can inspect and analyze defects of the wind power generator 130 easily and easily in a short time by itself from an image of the wind power generator 130 received using a deep learning technique. Defect recognition and analysis through the deep learning will be described later and detailed description thereof will be omitted.

The unmanned aerial vehicle 120 is a device that moves according to a steering input signal and collects data for fault recognition and analysis of the wind power generator 130 by photographing the wind power generator 130. The unmanned aerial vehicle 120 flies around the wind power generator 130 according to the control input signal, photographs the wind power generator 130, and transmits the captured image data to the defect analysis device 110. As described above, since the wind power generator 130 has a huge size, all defects in the wind power generator 130 cannot be analyzed with only one image. Therefore, the unmanned aerial vehicle 120 photographs the wind power generator 130 with a large number of images while maintaining a certain distance from the wind power generator 130, so that all defects in the wind power generator 130 can be accurately analyzed. In particular, the unmanned aerial vehicle 120 acquires image data for the blades of the wind power generator 130 while flying in response to the movement of the blades of the wind power generator 130, that is, static or dynamic motions, to the defect analysis device 110. Can transmit.

2 is a configuration diagram of a defect generator 110 for a wind power generator according to an embodiment of the present invention.

Referring to FIG. 2, the wind power generator defect analysis device 110 according to an embodiment of the present invention includes a communication unit 210, a blade detection unit 220, a defect detection unit 230, a defect analysis unit 240, and a control unit 250 ) And the database 260.

The communication unit 210 may exchange and transmit various signals through data communication with the unmanned flying device 120, and in particular, may transmit the image data of the wind power generator 130 photographed by the unmanned flying device 120 to the unmanned flying device ( 120).

The blade detection unit 220 detects the target object of defect recognition / analysis, that is, the image data of the blade region of the wind power generator, from the received image data of the wind power generator 130. This is because the defects in the wind power generator 130 mainly occur in the continuously moving blade. Accordingly, the blade detection unit 220 extracts image data including the blade region from the received image data, and detects only the image data of the blade region from the extracted image data. As described above, when targeting an image for a large number of wind power generators 130, it takes a lot of computation time and wastes a lot of resources, so it is preceded for efficient defect recognition / analysis. In other words, the blade detection unit 220 first determines an image including a blade region in each image received, and detects only the blade region from the image in which the blade region is present. The blade detection unit 220 performs deep learning on the blade using images of a number of wind power generators stored in the database 260. Using the image in which the blade stored in the database 260 exists and the image in which the blade does not exist, the blade detection unit 220 performs deep learning on its own to determine the presence or absence of the blade in the image. In this case, the blade detection unit 220 may use a single-shot multibox detection (SSD) of a convolutional neural network (CNN) as a method of deep learning. The blade detector 220 classifies the image of the wind power generator 130 into regions, and detects a region of interest (ROI) in which the blade is present. As described above, the blade detection unit 220 may quickly detect the region of interest where the blade is present, by classifying each region using an SSD method.

The defect detection unit 230 detects whether a portion determined as a defect exists in the region of interest detected by the blade detection unit 220. The defect detection unit 230 performs deep learning with respect to a blade, particularly a surface of a blade, using a blade image among images of a number of wind power generators stored in the database 260. The defect detection unit 230 performs deep learning on its own by using an image in which the defect exists in the blade stored in the database 260 and an image in which the defect does not exist in the blade. In this case, the defect detector 230 may use a semantic segmentation method of a convolutional neural network as a deep learning method. The defect detection unit 230 determines whether the portion can be determined as a defect by determining the pixel unit by using semantic division. Since semantic segmentation is determined on a pixel-by-pixel basis, accuracy is relatively high compared to methods such as SSDs that are detected for each area. The defect detection unit 230 detects whether a portion that can be determined as a defect exists in each pixel unit by using semantic division of the detected regions of interest, and only the regions of interest where a portion that can be determined as a defect exists To detect.

The defect detection unit 230 may detect a region of interest by dynamically changing a region to be detected as a defect. The defect detection unit 230 narrows a region of interest determined to exist as a region of interest within the region of interest (the entire blade) detected by the blade detection unit 220, and then performs deep learning within the narrowed region of interest. It is possible to detect whether a portion judged to be defective exists. Depending on the size of the facility, the defect detection unit 230 may perform a step of narrowing the range of the region of interest from the region of interest (the entire blade) detected by the blade detection unit 220, and gradually narrowing the region of interest By winding, it is possible to more accurately detect a portion determined to be a defect.

The defect analysis unit 240 analyzes the regions of interest detected by the defect detection unit 230 in detail to analyze the type of defect or how much the defect is. The defect analysis unit 240 performs deep learning by itself using the image in which the defect exists in the blade stored in the database 260 and the image in which the defect does not exist in the blade. At this time, the defect analysis unit 240 may use a multi-label classification method of a convolutional neural network as a deep learning method. Accordingly, the defect analysis unit 240 determines whether the portion (which can be determined as a defect) detected by the defect detection unit 230 is an actual defect. In addition, when it is determined that the defect is a defect, the defect analysis unit 240 analyzes the property of the defect, such as what kind of defect it is and how much it is damaged due to the defect. Finally, the defect analysis unit 240 analyzes whether there is a defect in the blade, if so, what kind of defect, and to what extent.

The control unit 250 controls the operation of each component. When the communication unit 210 receives the images of the wind power generator 130, the control unit 250 controls each configuration 220 to 240 to analyze whether a defect exists in the blade.

In addition, when the defect analysis unit 240 finally analyzes the defect, the control unit 250 displays an image of the wind generator 130 in which the defect exists, the blade detection unit 220, the defect detection unit 230, or the defect analysis unit ( 240) is stored in the database 260 in correspondence with the analyzed result. Accordingly, the blade detection unit 220, the defect detection unit 230 and the defect analysis unit 240 can use the corresponding data stored in the database 260 to more accurately detect or analyze the blade or defect. That is, the control unit 250 controls to continuously store the received images and inspection results in the database 260 so that the database 260 can form big data, and deep learning of each component 220 to 240 is performed. Improve learning and reasoning performance.

The database 260 stores images of the wind generator 230 that has been previously received or stored by the control unit 250. The database 260 stores a number of images for wind generators, especially images with blades, images without blades, images with blade defects and images without blade defects. . The database 260 stores each type of image at a preset rate. For example, each type of image (images with different blades and blades with different defects) may be provided in equal proportions, and the image with the blades is relatively larger than the image without the blades. In addition, when a blade is present, an image in which the defect is present in the blade may be stored at a relatively higher rate than an image in which the defect is not present. As described above, the database 260 stores each type of images at a predetermined rate, thereby significantly improving deep learning and inference performance of each component 220 to 240.

3 is a configuration diagram of an unmanned aerial vehicle according to an embodiment of the present invention.

Referring to FIG. 3, the unmanned aerial vehicle 120 according to an embodiment of the present invention includes a communication unit 310, a control unit 320, a wing unit 330, and an image photographing unit 340.

The communication unit 310 receives a control input signal for the unmanned flight device 120 from a control device (not shown) and transmits it to the flight control unit 320. In addition, the communication unit 310 transmits the image data of the wind generator taken by the image photographing unit 340 to the defect analysis device 110.

The control unit 320 controls the wing unit 330 to fly according to the received steering input signal. In addition, the control unit 320 controls the image capturing unit 340 so as to photograph the wind power generator.

The wing part 330 includes a rotor that generates rotational force and a propeller that converts the rotational force to propulsion, so that the unmanned aerial vehicle 120 can fly. The wing part 330 changes the flight direction of the unmanned flying device 120 or the altitude of the unmanned flying device 120 under the control of the control unit 320.

The image photographing unit 340 photographs the wind power generator 130 under the control of the control unit 320.

Figure 4 is a flow chart showing a defect recognition / analysis method of a wind power generator according to an embodiment of the present invention.

The defect analysis device 110 receives image data of the wind power generator 130 from the unmanned flight device 120 (S410).

The defect analysis device 110 extracts only the blade region of the wind power generator 130 from the received image data of the wind power generator 130 (S420).

The defect analysis device 110 detects a portion determined to be a defect in the blade area (S430).

The defect analysis device 110 analyzes whether the detected portion is an actual defect and the nature of the defect (S440).

Hereinafter, the defect recognition / analysis method of the wind power generator in the defect analysis apparatus 110 using a deep learning technique will be described in more detail.

5 is a flowchart illustrating a detailed defect recognition / analysis method of a wind turbine blade according to an embodiment of the present invention, and FIG. 6 is a diagram illustrating an image stitching technique in connection with an embodiment of the present invention It is a flowchart.

The blade defect recognition / analysis method of the wind power generator according to the present invention is largely performed in three stages based on a deep learning technique, first extracting a region of interest (ROI) region image corresponding to the blade region image from the input image. , Recognizing and storing the defective area in units of pixels from the extracted ROI area image, and finally storing and analyzing the properties of the recognized defective area, and storing the segmented area. (cycle). On the other hand, deep learning techniques for performing the above steps may be different.

2 and 5, the blade detection unit 220 receives an image of the wind power generator 130 from the unmanned aerial vehicle 120 through the communication unit 210 (S510).

The blade detection unit 220 recognizes the blade area from the received wind generator 130 image (S520). In this case, a deep learning technique that recognizes a blade region as a region of interest (ROI) from the received image may be, for example, an SSD technique of a Convolution Neural Network (CNN) algorithm. At this time, the CNN algorithm or SSD technique is referred to in the context of the present invention, and the detailed description is omitted, and only the necessary contents are described in the present specification.

In other words, the blade detector 220 separates the background region and the blade region from the received wind generator image (eg, one image). At this time, the size of the separated blade area may be changed each time because it is not based on a general fixed slide window. In relation to this, the SSD technique mainly enables a high-speed (eg, 15 Hz) execution in an embedded environment by modifying, learning, and applying a known model as a model for object detection. Accordingly, only the blade region of the wind generator can be detected as an ROI as a result of step S520 from the image of the wind generator received according to the SSD technique of the deep learning CNN according to an embodiment of the present invention.

Next, the defect detection unit 230 receives the ROI, that is, the blade area image data, detected by the blade detection unit 220 to recognize defects on the blade surface, that is, normal / abnormality (S530, S540). Here, the deep learning technique for determining whether the blade surface is defective from the received blade region image data is not an SSD technique specialized for object detection described above, but, for example, a semantic segmentation (Segnet) technique of the CNN algorithm may be used. You can. At this time, the previously published contents of the semantic segmentation technique are also referred to in connection with the present invention, and a detailed description is omitted, and only necessary contents are described in this specification. In other words, the defect detection unit 230 may determine the presence or absence of defects in units of pixels using the ROI, that is, the blade region image data, and display the defects. At this time, in relation to the display according to the presence or absence of the defect, for example, in units of pixels of the blade area image data, pixels other than the blade area correspond to black, blade areas, but pixels without defects correspond to green and blade areas. Defective pixels can be marked as red. The display color for the pixel is only an example, and is not limited thereto. As a result, as a result of the determination of step S540, the image data of whether the defect is recognized in the pixel unit from the ROI area becomes the location data of the pixel. The latter location data of the corresponding pixel may be based on, for example, the output of step S520 or the input image of step S530.

As a result of determining whether the defect is recognized in the pixel unit in step S540, the display data in the pixel unit without defects is stored in the normal image folder (S550), and the display data in the defective pixel unit is stored in the image abnormal folder ( S560). As described above, in step S550 and step S560, images recognized for pixel defects from the blade area images are classified and stored in folders, and the image data in pixel units stored in each folder is a corresponding folder based on the pixel location data. It can also be stored separately within.

Finally, the defect analysis unit 240 determines and classifies the properties of the defects, that is, the type and degree of the defects, from the image data of the pixel units in which the defects are recognized, and stores the classified pixel-level images in a state-damaged defect folder. (S570, S580). At this time, the image data in units of pixels recognized as the defects may be received from the abnormal state folder. Also, the status defect folder may exist in the status abnormal folder. On the other hand, with regard to the determination and classification of the defect property, a multi-label classification method of the CNN algorithm may be used as a deep learning technique. At this time, in addition, reference is made to the multi-label classification technique, and a detailed description is omitted, but description of necessary contents in connection with the present invention is as follows. In other words, the defect analysis unit 240 cuts an image of a pixel unit into an image of a certain size, and recognizes the type and degree of the defect using the image itself. This is called object recognition, and the location of the recognized result is displayed and classified based on the pixel location on the initially received wind power generator image, and stored in the status defect folder. The type and degree of defects classified in the above may be changed according to predefined settings. For example, the type of defects may be set to 15, and the degree of defects to be classified as 5.

In summary, according to the present invention, a deep learning technology is used to first recognize and recognize a blade area that is a target of defect determination from an image of a facility (wind power generator) obtained from an unmanned aerial vehicle 120, and the blade area image thus recognized is recognized. The defects are again determined in units of pixels, and each is separately stored for each folder.

Meanwhile, referring to FIG. 6, a stitching-based subject, that is, a photographing result of an object may be corrected.

In other words, an object photographing result (continuous picture) is received (S610), and image stiching is performed (S620).

The stitching result is corrected based on the state variable with respect to the subject image on which the image stitching is performed (S630). When correcting the stitching result, GPS position and posture at the time of continuous image shooting may be referred to as one of the state variables.

Meanwhile, the entire subject photographing result image corrected as a result of stitching based on the state variable may be used as photographed image data of the wind turbine input in step S510 of FIG. 5.

In the above, reference is made to the published information related to the image stitching technique, and detailed description is omitted.

The technique of the deep learning technique performed in each step of FIG. 5 described above is used, for example, as an embodiment of the present invention, but is not limited thereto. In other words, with respect to the present invention, techniques applied at this stage may be replaced or referred to in accordance with future technological developments.

4 to 6, it is described that each process is executed sequentially, but this is merely illustrative of the technical idea of an embodiment of the present invention. In other words, a person having ordinary skill in the art to which one embodiment of the present invention pertains may execute or change one or more of the processes described in FIGS. 4 to 6 without departing from the essential characteristics of one embodiment of the present invention. 4 to 6 are not limited to time-series order, as various modifications and variations can be applied to the process in parallel.

Meanwhile, the processes shown in FIGS. 4 to 6 can be implemented as computer readable codes on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data readable by a computer system is stored. That is, the computer readable recording medium includes magnetic storage media (eg, ROM, floppy disk, hard disk, etc.), optical reading media (eg, CD-ROM, DVD, etc.) and carrier waves (eg, the Internet). Storage). The computer-readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

The above description is merely illustrative of the technical idea of the present embodiment, and those skilled in the art to which this embodiment belongs may be capable of various modifications and variations without departing from the essential characteristics of the present embodiment. Therefore, the present embodiments are not intended to limit the technical spirit of the present embodiment, but to explain, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of the present embodiment should be interpreted by the claims below, and all technical spirits within the equivalent range should be interpreted as being included in the scope of the present embodiment.

100: defect analysis system
110: defect analysis device
120: unmanned aerial vehicle
130: wind power generator
210, 310: communication unit
220: blade detection unit
230: defect detection unit
240: defect analysis unit
250, 320: control unit
260: database
330: wing
340: video recording unit

Claims (12)

In a method for recognizing and analyzing a defect in a wind power generator using deep learning from a photographed image of an unmanned aerial vehicle,
A receiving process for receiving an image of the wind generator taken from the unmanned aerial vehicle;
A blade detection process of detecting an image including a blade region of the wind generator from the image of the wind generator;
A defect detection process of determining whether a defect exists from the detected blade region image in units of pixels and detecting the defect; And
A defect analysis process of analyzing a defect property of a pixel-based detection image in which a defect is recognized from the blade region image;
Wind turbine defect recognition and analysis method comprising a. According to claim 1,
A method for recognizing and analyzing a defect in a wind power generator, comprising the step of storing the detected pixel-based detected image and the detected defective pixel-based detected image in separate folders, respectively, from the blade area image. . According to claim 2,
The analyzed pixel unit detection image is classified according to the defect type and the defect level according to the defect property, and the classified image is further included in the step of storing in a separate folder. . The method of claim 2 or 3,
The pixel unit image data stored in the folder,
Method for recognizing and analyzing a defect in a wind generator, characterized in that location data of a corresponding pixel is also included based on the image of the received wind generator. According to claim 1,
The defect detection process,
Wind power generator characterized in that the image containing the detected blade area is divided into pixels, a non-blade area, a non-blade area or a defect-free pixel and a blade area and a defective pixel, and displayed in different colors. Defect recognition and analysis methods. According to claim 1,
The blade detection process, defect detection process and defect analysis process,
Wind turbine fault recognition and analysis method, characterized by using a convolutional neural network algorithm of the deep learning technology, but using a different technique in each process. In the apparatus for recognizing and analyzing a defect in a wind power generator using deep learning from a photographed image of an unmanned aerial vehicle,
A communication unit that receives an image of the wind power generator photographed from the unmanned aerial vehicle;
A blade detector configured to detect an image including a blade region of the wind generator from the image of the wind generator;
A defect detection unit for determining whether a defect exists from the detected blade region image in units of pixels and detecting the defect; And
A defect analysis unit analyzing a defect property of a pixel-based detection image in which a defect is recognized from the blade region image;
Wind power generator fault recognition and analysis device comprising a. The method of claim 7,
The apparatus for detecting and analyzing a defect in a wind power generator, further comprising a database for storing a detection unit of a pixel unit in which a defect is recognized from a blade region image and a detection unit of a pixel unit in which a defect is not recognized in a separate folder. The method of claim 8,
The database,
Wind power generator defect recognition and analysis device, characterized in that the detected pixel-by-pixel detection images are classified according to defect types and defect levels according to defect properties, and the classified images are stored in separate folders. The method of claim 8 or 9,
The database,
The pixel-by-pixel image data stored in the folder,
Wind power generator defect recognition and analysis device, characterized in that it also comprises the location data of the pixel based on the image of the received wind generator. According to claim 1,
The defect detection unit,
Wind power generator characterized in that the image containing the detected blade area is divided into pixels, a non-blade area, a non-blade area or a defect-free pixel and a blade area and a defective pixel, and displayed in different colors. Defect recognition and analysis device. According to claim 1,
The blade detection unit, the defect detection unit and the defect analysis unit,
A wind power generator fault recognition and analysis apparatus using the convolutional neural network algorithm of the deep learning technology, but using different techniques in each process.

Images