JP7323694B2 - Structure deterioration diagnosis method - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for diagnosing the state of deterioration of a structure, and more particularly to a method for diagnosing the state of deterioration of a structure from the results of detection by an image analysis program and artificial intelligence incorporated in a computer.

For example, in structures such as bridges, dams, and high-rise buildings built during the period of high economic growth, deterioration has become a problem. It is important to grasp the state of deterioration in advance. For this reason, various devices and methods have been developed for inspecting damaged portions due to cracks or the like in structures (see, for example, Patent Documents 1 and 2).

On the other hand, in recent years, with the aim of improving productivity at construction sites, the Ministry of Land, Infrastructure, Transport and Tourism has been promoting the effective use of 3D data in a series of construction work processes from surveying, design/construction planning, construction, and inspection. In addition to announcing the introduction of i-Construction, we are actively introducing ICT (Information and Communication Technology) such as drones and CIM (Construction Information Modeling/Management).

Preferably, by utilizing drones, for example, it would take about a week to survey thousands of points using conventional surveying methods, but it would be possible to complete surveys of millions of points in about 15 minutes. In addition, it is possible to take an overall image of the structure from various directions with a high pixel count (ultra-high pixel count) of, for example, 10000×10000=100 million pixels or more. Therefore, development of technology for diagnosing the state of deterioration of structures, preferably from high-pixel (ultra-high-pixel) images obtained by imaging structures using drones, using known image recognition technology based on artificial intelligence. is desired. An image of a structure captured with a high pixel count (ultra-high pixel count) can, for example, combine multiple images into a single image. is possible.

Japanese Unexamined Patent Application Publication No. 2011-231602 JP 2018-90981 A

However, when diagnosing the deterioration state of the structure from a high-pixel image of the structure by a known image recognition technology based on artificial intelligence, for example, a trained image recognition model for the damage state of the structure created in advance by artificial intelligence In addition, it is thought that it will be possible to diagnose the deterioration state of fine structures by inputting captured images. This image requires a lot of time to analyze the damage state due to the excessive number of pixels, and depending on the part of the structure, deformation that cannot occur in that part An erroneous diagnosis such as diagnosing "deterioration" may occur.

The present invention aims to efficiently diagnose the deterioration state of a structure by reducing the time required to analyze the damage state when diagnosing the deterioration state of the structure from a high-pixel image of the structure by image recognition technology using artificial intelligence. To provide a method for diagnosing a deterioration state of a structure capable of effectively avoiding misdiagnosis such as diagnosing a deformation that cannot occur in a specific part as ``deterioration''. for the purpose.

The present invention uses ultra-high pixel images of 100 million pixels or more and low pixel images obtained by imaging structures, from the detection results by an image analysis program and artificial intelligence incorporated in a computer, structures A deterioration state diagnosis method for diagnosing the deterioration state of a structure, comprising: an image acquisition step of acquiring an ultra-high pixel image of a structure; and a computer acquiring the ultra-high pixel image of the structure by an image analysis program. A site identification step of specifying a predetermined site and type in a low pixel image of the structure obtained by reducing the pixels by artificial intelligence, and a computer prepares in advance regarding the damage state of the specific site and type The super-high pixel image of the structure acquired in the image acquisition step of the part and type specified in the part identification step is input to the trained image recognition model for deterioration detection, and artificial intelligence recognizes the structure and a site-by-site damage detection step for outputting a damage state of a predetermined site of the site-based damage detection step, followed by an erroneous detection confirmation step, wherein the erroneous detection confirmation step includes , determining whether or not the damage state for each predetermined site and type in the structure detected in the site-based damage detection step is damage that cannot occur in the predetermined site and type; And when damage that cannot occur in the type is detected, the detected damage that cannot occur is removed to prevent erroneous detection, so that only the correct damage state is detected, The above object is achieved by providing a method for diagnosing the state of deterioration of a structure in which a diagnostician diagnoses the state of deterioration of the structure from the displayed state of damage of a predetermined portion.

Further, in the method for diagnosing the state of deterioration of a structure of the present invention, it is preferable that the super-high-pixel image of the structure is an image obtained by imaging the structure with a drone.

Further, in the method for diagnosing the state of deterioration of a structure of the present invention, it is preferable that the part and type identified in the part identifying step are the part and type identified by semantic segmentation by artificial intelligence.

Furthermore, in the method for diagnosing the state of deterioration of a structure according to the present invention, the creation of the learned image recognition model for identifying deterioration and the output of the damage state of a predetermined portion of the structure by artificial intelligence are performed using a convolutional neural network as an algorithm. It is preferably performed by deep learning.

Further, in the structure deterioration state diagnosis method of the present invention, the damage state of the predetermined portion of the structure output in the portion-based damage detection step is crack, corrosion, deterioration of anticorrosion function, water leakage / free lime, or Peeling and exposure of reinforcing bars are preferred.

Further, in the method for diagnosing a state of deterioration of a structure of the present invention, in the part identifying step, the learned image recognition model for identifying a part region, which has been created in advance for a specific part and type of the structure, is added to the image reduction step It is preferable that an image of a structure with reduced pixels is input and a predetermined portion and type of the structure are specified by artificial intelligence.

According to the method for diagnosing the state of deterioration of a structure of the present invention, the time for analyzing the state of damage is reduced when diagnosing the state of deterioration of a structure from a high-pixel image of the structure by image recognition technology using artificial intelligence. Therefore, it is possible to efficiently diagnose the state of deterioration, and to effectively avoid misdiagnosis such as diagnosing a deformation that cannot occur in a specific part as "deterioration". .

FIG. 1 is a schematic diagram illustrating the system configuration of a diagnostic system in which a method for diagnosing the state of deterioration of a structure according to a preferred embodiment of the present invention is implemented. FIG. 2 is a block diagram for explaining the system configuration of the diagnostic system. FIG. 3 is a flow chart for explaining the processing steps of the method for diagnosing the state of deterioration of a structure according to a preferred embodiment of the present invention. FIG. 4 is an explanatory diagram of a high-pixel image of a bridge structure captured using a drone; FIG. 10 is an explanatory diagram of a predetermined portion in the image of the structure with reduced pixels, which is identified in the portion identifying step; FIG. 10 is an explanatory diagram of a high-pixel image of a part identified in the part identification step, which is input to the trained deterioration detection image recognition model in the part-based damage detection step; FIG. 10 is an explanatory diagram illustrating a damage state of a predetermined portion of a structure detected in the portion-based damage detection step;

A method for diagnosing the state of deterioration of a structure according to a preferred embodiment of the present invention is adopted as a method for diagnosing the state of deterioration of a structure such as a bridge structure 20 made of reinforced concrete. That is, the bridge structure 20 shown in FIG. 1 was built during the period of high economic growth, and for example, 40 years or more have passed since its construction, so damage such as cracks due to aged deterioration has occurred. By adopting the method for diagnosing the state of deterioration of structures according to the present embodiment, it is possible to accurately diagnose the state of deterioration due to damage such as cracks and to grasp the cause thereof, preferably by an appropriate repair method. It allows the damaged part to be repaired.

Further, the method for diagnosing the state of deterioration of a structure according to the present embodiment is implemented using a diagnosis system 10 having the system configuration shown in FIG. 1, for example. The diagnostic system 10 includes a drone 12 having a function capable of imaging a bridge structure 20 with a high pixel count (ultra-high pixel count), which are connected to each other via a wired or wireless communication network 11, for example. A personal computer 13 installed in an office and a well-known machine learning tool capable of implementing a neural network preferably using a convolutional neural network (CNN) as an AI (Artificial Intelligence) algorithm. (software) is installed in the server 14.

The method for diagnosing the state of deterioration of a structure according to the present embodiment preferably uses a super-high-pixel image of the bridge structure 20 captured by the drone 12 to detect fine deterioration occurring in various parts of the bridge structure 20. The state can be properly diagnosed by image recognition technology using artificial intelligence, and the time to analyze the damage state can be reduced to efficiently diagnose the deterioration state. It has a function to avoid misdiagnosis, such as diagnosing a deformation that does not exist as "deterioration".

Then, as shown in FIGS. 1 to 3, the method for diagnosing the state of deterioration of a structure according to the present embodiment has a high pixel count of 100 million pixels or more obtained by imaging, for example, a bridge structure 20 made of reinforced concrete as a structure. A deterioration state diagnosis method for diagnosing the deterioration state of a bridge structure 20 from detection results by an image analysis program and artificial intelligence incorporated in a computer 14 using an image (ultra-high pixel), wherein the computer 14 is a bridge structure An image acquisition step S1 for acquiring a high-pixel image of the structure 20; an image low-pixel reduction step S2 for reducing the pixels of the high-pixel image of the bridge structure 20 by an image analysis program; , a part identification step S3 in which artificial intelligence identifies a predetermined part and type in a low-pixel image of a structure, and a computer 14 prepares a pre-learned degradation detection model related to the damage state of the specific part and type and a site-specific damage detection step S4 for inputting a high-pixel image of the site and type specified in the site specifying step S3 and outputting the damage state of a predetermined site of the structure by artificial intelligence. , the diagnostician diagnoses the deterioration state of the bridge structure 20 from the displayed damage state of the predetermined portion.

In the present embodiment, the drone 12, which constitutes the diagnostic system 10 and has a function capable of imaging with high pixels, is used at the construction site of the bridge structure 20 to be diagnosed. Preferably, the entire bridge structure 20 is imaged. to shoot. When imaging the bridge structure 20 with the drone 12, it may be difficult to capture the entire bridge structure 20 in a single image. It is also possible to diagnose the state of deterioration by photographing only the area that includes the site to be investigated. The image of the structure photographed with a high number of pixels can be combined into one image by synthesizing a plurality of images in the personal computer 13, for example.

The drone 12 has a function capable of capturing, preferably, the entire image from various directions of the bridge structure 20 with a high pixel count (ultra-high pixel count) of, for example, 10000×10000=100 million pixels or more. In addition, since the image of the bridge structure 20 captured with a high number of pixels has high definition, it is possible to diagnose even minute deterioration states. The drone 12 can fly around the bridge structure 20 under the control of a photographing/transmitting/display unit 13a of a personal computer 13 installed in, for example, a management office, which constitutes the diagnostic system 10. 13b, an area including part or all of the bridge structure 20 can be photographed.

A personal computer 13 that constitutes the diagnostic system 10, for example, installed in a management office, functions as a photographing/transmitting/displaying unit 13a, controls the flight of the drone 12, and receives commands from the photographing unit 13b. allows the drone 12 to capture an area including part or all of the bridge structure 20, import the captured image into the capturing/transmitting/displaying unit 13a, and store it in, for example, a storage unit (not shown). It's like The photographing/transmitting/displaying unit 13a transmits the captured image of the bridge structure 20 as data for detecting the damage state of a predetermined portion of the bridge structure 20 via the transmitting unit 13c. In addition to transmitting to the server 14, the detection result of the damage state of the predetermined portion of the bridge structure 20 transmitted from the server 14 can be displayed on, for example, the display 13e via the display unit 13d.

Here, the personal computer 13 includes a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), I/F (Interface), HDD (Hard Disk Drive), and the like. The CPU can control various processes while using the RAM as a work area according to various programs incorporated in the ROM. In addition, the CPU functions storage means, input means, display means, output means, etc., and various information is displayed on the display 13e and output from the printer, as various computer programs are incorporated in the ROM. It is designed to be able to

Further, in this embodiment, the server 14 that constitutes the diagnostic system 10 is a computer, in which a known image analysis program and a known artificial intelligence are incorporated, thereby forming a damage detector 14a. In the damage detection unit 14a of the server (computer) 14, the image acquisition step S1, the image pixel reduction step S2, the site identification step S3, and the site-by-site damage detection step S4 of the method for diagnosing the state of deterioration of the structure of the present embodiment are performed. It is designed to be

Here, the computer serving as the server 14 includes a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), I/F (Interface), HDD (Hard Disk Drive), storage means, input means. , display means, output means, and the like. The CPU controls image processing by the image analysis program and machine learning by AI while using the RAM as a work area according to various programs incorporated in the ROM. In addition, since various computer programs are incorporated in the ROM, the CPU functions as storage means, input means, display means, output means, etc., and also receives data sent from the personal computer 13 and analysis results by AI. For example, it can be stored in a database unit, or can be displayed as predetermined information on a display 13e or output from a printer via a personal computer 13 or the like installed in a management office.

In the method for diagnosing the state of deterioration of a structure according to the present embodiment, the damage state of a predetermined portion of the structure is detected by artificial intelligence according to the treatment steps shown in the flowchart of FIG. It can be detected and used for diagnosis of the deterioration state of the bridge structure 20 by a diagnostician.

That is, preferably the entire image of the bridge structure 20 captured by the drone 12 in accordance with a command from the imaging unit 13b of the imaging/transmission/display unit 13a by the personal computer 13 is sent to the transmission unit of the imaging/transmission/display unit 13a. 13c, to the site identification unit 14b (see FIG. 2) of the damage detection unit 14a by the server (computer) 14, in the image acquisition step S1, as shown in FIG. 4, for example, 100 million pixels or more is acquired by the computer 14 as an image of the entire bridge structure 20 with a high number of pixels.

Next, the high-pixel image of the entire bridge structure 20 acquired in the image acquisition step S1 is preferably stored in a storage unit (not shown) in the damage detection unit 14a of the server (computer) 14. Then, in the image pixel reduction step S2, the image is reduced in pixels by an image analysis program, for example, to obtain an image of about 100,000 pixels or less, preferably 200×200=40,000 pixels. The entire low-pixel image of the bridge structure 20, which has undergone bottom-pixel reduction in the image-lowering step S2, is preferably stored in a storage unit (not shown), and the bottom pixel is reduced by performing the site specifying step S3. A predetermined portion and type of the bridge structure 20 in the pixelated image are specified by artificial intelligence.

In the present embodiment, the part identification step S3 preferably includes an existing trained image recognition model for identifying a specific part and type of the bridge structure 20, which is created in advance, and the image recognition step S2 to reduce the pixels. By inputting the converted image of the bridge structure 20, artificial intelligence can specify a predetermined part and type of the bridge structure 20. FIG. The existing trained image recognition model for identifying a part region uses a convolutional neural network as training data, for example, a data set in which the part or type is color-coded for each pixel, which is a large number of images of existing bridge structures. It can be created by deep learning. In addition, a predetermined portion and type of the bridge structure 20 can also be specified by inputting the low-pixel image of the bridge structure 20, which has been reduced in pixels in the image-lowering step S2, to the trained image recognition model for specifying a portion region. Moreover, it can be easily performed by deep learning using a convolutional neural network as an algorithm.

In the part identification step S3, the predetermined parts and types of the bridge structure 20 identified by deep learning using a convolutional neural network as an algorithm are preferably subjected to semantic segmentation by artificial intelligence, as shown in FIG. As shown in FIG. 6, the region images of the identified regions, for example, the main girder 20a and a plurality of images of the chest walls 20b, 20c, and 20d can be extracted with the site numbers attached. The region-based region image of each of the specified predetermined regions extracted is converted back to the high-pixel region image, and the region-based damage detection unit 14 c of the server (computer) 14 is converted as a region-based high-pixel region image. (see FIG. 2c), a site-specific damage detection step S4 will detect the damage status at each identified site.

In the site-by-site damage detection step S4, an existing image recognition model for deterioration detection that has been learned in advance regarding the damage state of the specific site and type is subjected to high-pixel resolution of each site and type identified in the site identification step S3. By inputting an image, artificial intelligence outputs and detects the damage state of a predetermined portion of the bridge structure 20 . The existing trained image recognition model for deterioration detection uses a large number of images of existing bridge structures captured for each predetermined part and type, including damage locations. Training data can be created by deep learning using a convolutional neural network as an algorithm. The damage state of the predetermined portion of the bridge structure 20, which is color-coded for each pixel, can preferably be cracks, corrosion, deterioration of anti-corrosion function, water leakage/free lime, or exfoliation/reinforcing bar exposure. The detection of the damage state in each part by inputting the high-pixel image of each part and type identified in the part identification step S3 into the trained image recognition model for deterioration detection is also performed using a convolutional neural network as an algorithm. It can be easily done by deep learning.

In the site-by-site damage detection step S4, the site-by-site high-pixel area images of the specified sites converted into high-pixel images in the site-specifying step S3 are preferably selected in accordance with the order of the site numbers attached thereto. The image recognition models for deterioration detection which have been trained in advance and which are related to the damage state of each part and type are sequentially inputted. By sequentially inputting high-pixel images of the specified parts and types, deep learning using a convolutional neural network based on artificial intelligence as an algorithm is performed, preferably for cracks, corrosion, and anticorrosion functions of predetermined parts of the bridge structure 20. deterioration, water leakage/free lime, or exfoliation/reinforcing bar exposure can be sequentially output, and the damage state at each site can be detected.

Further, in the present embodiment, following the site-based damage detection step S4, the damage state for each predetermined site and type in the bridge structure 20 detected in the site-based damage detection step S4 is displayed for the predetermined site and type. includes an erroneous detection confirmation step S5 for determining whether or not the damage is improbable so that erroneous detection does not occur. In the erroneous detection confirmation step S5, the damage state such as cracks, corrosion, deterioration of anti-corrosion function, water leakage / free lime, or exfoliation / rebar exposure of the predetermined part detected in the part-by-part damage detection step S4 is specified. When damage that cannot occur in the predetermined part and type is detected, for example, when concrete cracks or exposed reinforcing bars are detected in the steel part of the bridge structure 20, such an error Detection is removed to allow only correct damage states to be detected more accurately.

The data regarding the damage state of the specified part and type of the bridge structure 20 detected by the damage detection unit 14a is sent to the photographing/transmission/display unit 13a by the personal computer 13, and is transmitted via the display unit 13d. can be displayed on the display 13e, for example. For example, as shown in FIGS. 7(a) and 7(b), the state of damage of a predetermined part and type of the bridge structure 20 shown in the photograph on the left can be displayed as shown in the image on the right. It's becoming As a result, the diagnostician can observe the state of damage of the specified predetermined portion and type of the bridge structure 20 displayed on the display 13e, for example, while observing the state of deterioration of the bridge structure 20 without going to the construction site. can be easily diagnosed.

Then, according to the deterioration state diagnosis method of the present embodiment having the above-described configuration, when diagnosing the deterioration state of the bridge structure 20 from a high-pixel image of the bridge structure 20 by image recognition technology using artificial intelligence, It is possible to reduce the time required to analyze the damage state and efficiently diagnose the deterioration state. can be effectively avoided from occurring.

That is, the method for diagnosing the state of deterioration of a structure according to the present embodiment includes an image acquisition step S1 for acquiring a high-pixel image of the bridge structure 20, and a site for specifying a predetermined portion and type in the low-pixel image by artificial intelligence. In step S3, a super-high-pixel image of the specified part and type is input to a learned deterioration detection model regarding the damage state of the specified part and type, and damage to the specified part of the structure is performed by artificial intelligence. and a damage detection step S4 for each part for outputting the state, and an erroneous detection confirmation step S5 is further included following the damage detection step S4 for each part, so that the bridge structure 20 is imaged with a high number of pixels. By reducing the amount of data to be processed compared to the case where the image is input as it is to the trained image recognition model for deterioration detection and the damage state of the bridge structure 20 is detected by artificial intelligence, When diagnosing the state of deterioration of a structure using image recognition technology, it is possible to reduce the time required to analyze the state of damage and diagnose the state of deterioration efficiently. is limited to images of specified predetermined parts and types of the bridge structure 20, misdiagnosis occurs, such as diagnosing deformation that cannot occur in the specific part as "deterioration". can be effectively avoided.

It should be noted that the present invention is not limited to the above embodiments, and various modifications are possible. For example, the high-pixel image of the structure does not necessarily have to be an image obtained by imaging the structure with a drone, and it does not have to be an image that includes the entire structure. The high-pixel image of the structure may be an image containing only a predetermined portion of the structure to be investigated. The predetermined parts and types of the structure identified in the part identification step are not necessarily the parts and types identified by semantic segmentation, and may be the parts and types identified by the bounding box by artificial intelligence. good. Structures whose deterioration state is diagnosed may be bridge structures, dams, high-rise buildings, and various other structures made of reinforced concrete or steel. The image acquisition step, the image pixel reduction step, the region identification step, and the damage detection step for each region do not necessarily have to be performed in a server, which is a computer. and run on the computer of the administrative office.

10 diagnostic system 11 communication network 12 drone 13 personal computer 13a photographing/transmission/display unit 13b photographing unit 13c transmission unit 13d display unit 13e display 14 server (computer)
14a damage detection unit 14b site identification unit 14c site-based damage detection unit 20 bridge structure

Claims (6)

Using ultra-high-pixel images of over 100 million pixels and low-pixel images obtained by imaging structures, the state of deterioration of structures can be determined from the results of detection by an image analysis program and artificial intelligence built into a computer. A deterioration state diagnosis method for diagnosing,
an image acquisition step in which the computer acquires an ultra-high pixel image of the structure;
A site identification step in which a computer uses artificial intelligence to identify a predetermined site and type in a low-pixel image of the structure obtained by reducing the pixels of the super-high-pixel image of the structure using an image analysis program. and,
A computer stores a pre-trained image recognition model for deterioration detection related to the damage state of a specific part and type of the structure acquired in the image acquisition step of the part and type specified in the part specifying step. and a site-by-site damage detection step of inputting an ultra-high pixel image and outputting the damage state of a predetermined site of the structure by artificial intelligence,
Following the site-by-site damage detection step, the method further includes an erroneous detection confirmation step, in which the erroneous detection confirmation step determines the damage state for each predetermined site and type in the structure detected in the site-by-site damage detection step. , determine whether the damage is impossible for the predetermined part and type, and remove the detected impossible damage if the damage is detected. By preventing erroneous detection from occurring, only the correct damage state is detected.
A method for diagnosing the state of deterioration of a structure in which a diagnostician diagnoses the state of deterioration of the structure from the displayed damage state of a predetermined portion. 2. The method for diagnosing the state of deterioration of a structure according to claim 1 , wherein said ultra-high pixel image of said structure is an image obtained by imaging said structure with a drone. 3. The method for diagnosing a deterioration state of a structure according to claim 1, wherein the parts and types specified in said part specifying step are parts and types identified by semantic segmentation by artificial intelligence. 3. The structure according to claim 1 or 2, wherein the creation of the trained image recognition model for identifying deterioration and the output of the damage state of a predetermined portion of the structure by artificial intelligence are performed by deep learning using a convolutional neural network as an algorithm. deterioration state diagnosis method. 3. The structure according to claim 1 or 2, wherein the damage state of the predetermined portion of the structure output in the portion-based damage detection step is cracking, corrosion, deterioration of anticorrosion function, water leakage/free lime, or peeling/rebar exposure. A method for diagnosing the state of deterioration of an object. In the part identification step, an image of the structure reduced in pixels by the image analysis program is input to an image recognition model for identifying a part region that has been previously created for a specific part and type of the structure, and the artificial intelligence 3. The method for diagnosing the state of deterioration of a structure according to claim 1, wherein the predetermined portion and type of the structure are specified by