Internal defect inspection method and internal defect inspection system

Abstract

This invention provides an internal defect inspection method and inspection system that can detect internal defects in concrete slabs with high accuracy by analyzing thermal images taken of the surface of asphalt pavement. [Solution] The thermal image NG38 includes a temperature fluctuation calculation step S11 in which, for each pixel G constituting the thermal image NG38, the low-frequency drift component Tav(kx,ky) contained in the temperature T of Nx pixels G arranged in the X-axis direction is calculated, and the high-frequency temperature fluctuation ΔT(kx,ky) is calculated by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky). The defective pixel extraction step S12 extracts pixels G(kx,ky) whose temperature fluctuation ΔT(kx,ky) exceeds a reference value Tth as defective pixels KG(kx,ky). Based on the position information of the defective pixels KG(kx,ky) extracted in the defective pixel extraction step S12, the defective region identification step S13 identifies a defective region KKR in the concrete slab 26 where an internal defect NK has occurred when the structure 24 to be inspected is viewed from above.

Description

Technical Field

[0001] The present invention relates to an internal defect inspection method and an internal defect inspection system that detect internal defects in a structure to be inspected by analyzing a thermal image obtained by imaging the surface of the structure to be inspected. Background Art

[0002] Many bridges, viaducts and the like have roads formed by laying asphalt pavement on the upper surface of concrete decks. In this type of structure, internal defects such as delamination and voids may occur inside the concrete deck or at the boundary between the concrete deck and the asphalt pavement due to aging or poor construction, and it is necessary to detect and address such internal defects at an early stage to prevent the occurrence of serious accidents.

[0003] As the most direct inspection method, there is a method of removing the asphalt pavement and inspecting the condition of the concrete deck by visual inspection or hammering. However, this method has problems such as prolonged traffic control, increased burden on workers and increased construction costs. A method of performing hammering inspection from above the asphalt pavement is also conceivable, but this requires inspectors with special skills, making it difficult to inspect a large number of structures located all over the country using this method.

[0004] In recent years, in order to efficiently inspect a large number of structures, inspection methods have been proposed that detect internal defects by imaging the surface of the structure with a thermal camera mounted on an unmanned aerial vehicle (such as a drone) and analyzing the captured thermal image. Generally, the surface temperature of a structure has the property that a temperature difference occurs between a region where an internal defect exists inside (defective region) and a region where no internal defect exists (sound region). Therefore, by analyzing a thermal image, the presence or absence of internal defects in the structure, as well as the position and size of the internal defects, can be estimated in a non-contact manner.

[0005] For example, the deformation detection method disclosed in Patent Document 1 involves photographing the surface of the concrete to be investigated using an infrared thermography device, and identifying the deformation in the concrete surface based on the thermal image of the surface to be investigated. The thermal image consists of multiple pixels aligned horizontally and vertically, and each pixel has temperature data.

[0006] To briefly explain the processing flow of Figure 1 in Patent Document 1, first, the horizontal temperature gradient of each pixel is calculated [S4, S5], and it is determined whether the calculated temperature gradient is greater than a threshold [S6], and pixels greater than the threshold are set to 1 and pixels smaller than the threshold are set to 0 to create a horizontal binarized image [S7]. Furthermore, the vertical temperature gradient of each pixel is calculated [S8, S9], and it is determined whether the temperature gradient is greater than a threshold [S10], and pixels greater than the threshold are set to 1 and pixels smaller than the threshold are set to 0 to create a vertical binarized image [S11]. Then, the horizontal and vertical binarized images are superimposed [S12], and the area of ​​the surface under investigation corresponding to the area enclosed by pixels that are 1 in the superimposed binarized image is determined to be the deformed area [S13, S14]. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2011-99687 [Overview of the project] [Problems that the invention aims to solve]

[0008] The deformation detection method described in Patent Document 1 detects deformation in the surface layer of concrete based on thermal images taken directly from the surface of the concrete being investigated. However, as mentioned above, many bridges and elevated bridges have asphalt pavement laid on the top surface of the concrete deck, so it is unclear whether this method is effective in cases where asphalt pavement is laid.

[0009] Furthermore, when imaging with a thermal camera, the optical characteristics of the thermal camera cause a problem in which a low-frequency drift component is generated in the characteristics of the temperature data of the thermal image. For example, even if the temperature of the actual surface being inspected is uniform, in a thermal image captured by a thermal camera, the center will be the hottest, and the temperature will gradually decrease as you move away from the center. This temperature shift, which occurs regardless of the presence or absence of internal defects, becomes a low-frequency drift component in the temperature characteristic curve of the thermal image data. In addition, when inspecting road surfaces such as bridges, there may be center lines drawn on the road surface or tire marks left by automobiles, and these also become sources of error when imaging with a thermal camera. Therefore, in order to perform high-precision inspections based on thermal images, it is necessary to consider the above-mentioned error factors (drift components, etc.). However, the deformation detection method in Patent Document 1 does not particularly consider this point.

[0010] The present invention has been made in view of the above-mentioned background art, and aims to provide an internal defect inspection method and an internal defect inspection system that can detect internal defects in concrete slabs with high accuracy by analyzing thermal images taken of the surface of asphalt pavement. [Means for solving the problem]

[0011] The present invention is an internal defect inspection method performed by a computer system, which involves analyzing thermal images taken of the upper surface of a structure to be inspected, where asphalt pavement is laid on the upper surface of a concrete slab to form a walkway, in order to detect internal defects in the structure to be inspected. In this invention, when the axis parallel to the length direction of the passage is defined as the X-axis and the axis parallel to the width direction of the passage is defined as the Y-axis, the plurality of pixels G constituting the thermal image are arranged sequentially in the X-axis direction from a reference point, with Nx pixels arranged sequentially in the Y-axis direction, and each pixel G is represented as pixel G(kx,ky) [1≦kx≦Nx, 1≦ky≦Ny], and the temperature data of pixel G(kx,ky) is represented as temperature T(kx,ky), A temperature variation calculation step is performed to calculate the low-frequency drift component Tav(kx,ky) contained in the temperature T of Nx pixels G arranged in the X-axis direction for each pixel G, and to calculate the high-frequency temperature variation ΔT(kx,ky) by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky), and A defective pixel extraction step is performed in which pixels G(kx,ky) whose temperature fluctuation ΔT(kx,ky) calculated in the above temperature fluctuation calculation step exceeds the reference value Tth are extracted as defective pixels KG(kx,ky), The system includes a defect region identification step, which identifies a defect region in the concrete slab where an internal defect has occurred, based on the positional information of the defect pixels KG(kx,ky) extracted in the defect pixel extraction step, when the structure to be inspected is viewed from above.

[0012] In the temperature fluctuation calculation step, the average value of the temperatures T of a specified number of pixels G adjacent to pixel G(kx,ky) and arranged sequentially in the X-axis direction, or the average value of the temperatures T of a specified number of pixels G including pixel G(kx,ky) and arranged in the X-axis direction, can be calculated as the drift component Tav(kx,ky). Alternatively, in the temperature fluctuation calculation step, the drift component Tav(kx,ky) may be calculated by performing polynomial fitting on the temperatures T of Nx image Gs arranged in the X-axis direction.

[0013] In the defect region identification step, it is preferable to cluster the multiple defective pixels KG(kx,ky) extracted in the defect pixel extraction step by applying a predetermined clustering algorithm, and to define the resulting cluster or the region surrounding the cluster as the defect region.

[0014] It is preferable to configure the system to include a defect area display step in which the area of ​​the upper surface of the structure to be inspected, excluding the defect area, is defined as a healthy area, and a binarized image is displayed on the display device in which the healthy area and the defect area are represented in different colors. In this case, in the defect pixel extraction step, a plurality of reference values ​​Tth are set and a defect pixel KG(kx,ky) is extracted for each reference value Tth, in the defect area identification step, the defect area is identified for each reference value Tth, and in the defect area display step, the binarized images for each reference value Tth are displayed on the display device, with the healthy area represented in the same color and the defect area in different colors.

[0015] In the defective pixel extraction step, it is preferable to set a first reference value Tth1 and a second reference value Tth2, which are reference values, and to set the first reference value Tth1 to a value of h1 times the standard deviation σ when the distribution of Nx temperature fluctuations ΔT(kx,ky) constituting each row is considered to be a normal distribution (h1 is a constant h1>0), and to set the second reference value Tth2 to an arbitrary constant greater than zero, and to extract pixels G(kx,ky) in which the temperature fluctuation ΔT(kx,ky) calculated in the temperature fluctuation calculation step exceeds both reference values ​​Tth1 and Tth2 as defective pixels KG(kx,ky).

[0016] Furthermore, the system can be configured to include a defect depth estimation step in which the area of ​​the upper surface of the structure under inspection, excluding the defective area, is defined as a sound area, and the depth location of the internal defect occurring in the structure under inspection is estimated based on the rate of change Rp of the temperature fluctuation ΔT(kx,ky) occurring at the boundary between the defective area and the sound area, or based on the average value Tmean of the temperature T(kx,ky) and the rate of change Rp.

[0017] In this case, the system includes a database preparation step of preparing a database in which model structure-related information, which is numerical information about a virtual model structure that simulates the structure to be inspected, and model structure temperature information, which is information calculated by performing a heat conduction analysis using the model structure-related information and which shows the temperature distribution of each part of the model structure, including the defective area and the sound area, are linked together and registered. The aforementioned model structure-related information includes, at a minimum, structural information indicating the structural characteristics of the model structure, ambient temperature information indicating the characteristics of the ambient temperature at the installation site of the model structure, site information indicating the characteristics of the amount of solar radiation on the upper surface of the model structure, and defect information indicating the presence or absence of internal defects in the model structure and the characteristics of those internal defects. In the defect depth estimation step, it is preferable to prepare a defect depth estimation model created by machine learning using the model structure-related information and the model structure temperature distribution information registered in the database as training data, and to input the inspection target structure-related information, which is numerical information corresponding to the information other than the defect information in the model structure-related information, and the temperature T(kx,ky) information of the inspection target structure, or the temperature fluctuation ΔT(kx,ky) information calculated in the temperature fluctuation calculation step, or both, into the defect depth estimation model, thereby estimating the depth location of the internal defect occurring in the inspection target structure.

[0018] Furthermore, the present invention is an internal defect inspection system comprising a computer system that detects internal defects in a structure to be inspected by analyzing thermal images taken of the upper surface of the structure to be inspected, in which asphalt pavement is laid on the upper surface of a concrete slab to form a passageway.

[0019] When an axis parallel to the length direction of said passage is defined as the X-axis and an axis parallel to the width direction of said passage is defined as the Y-axis, the plurality of pixels G constituting said thermal image are arranged in order of Nx pixels in the X-axis direction and Ny pixels in the Y-axis direction from a reference point, where an individual pixel G is represented as pixel G(kx,ky) [1≦kx≦Nx, 1≦ky≦Ny], and the temperature data at the position of pixel G(kx,ky) is represented as temperature T(kx,ky), a temperature fluctuation calculation unit that calculates, for each pixel G, a low-frequency drift component Tav(kx,ky) included in the temperatures T of Nx pixels G arranged in the X-axis direction, and calculates a high-frequency temperature fluctuation ΔT(kx,ky) by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky); a defective pixel extraction unit that extracts, as a defective pixel KG(kx,ky), the pixel G(kx,ky) for which the temperature fluctuation ΔT(kx,ky) calculated by said temperature fluctuation calculation unit exceeds a reference value Tth; a defect region specifying unit that specifies, based on positional information of the defective pixels KG(kx,ky) extracted by said defective pixel extraction unit, a defect region where an internal defect occurs in said concrete floor slab when the inspection target structure is viewed from the top.

[0020] The temperature fluctuation calculation unit may be configured to calculate, as said drift component Tav(kx,ky), an average value of the temperatures T of a prescribed number of pixels G sequentially arranged in the X-axis direction adjacent to the pixel G(kx,ky), or an average value of the temperatures T of a prescribed number of pixels G arranged in the X-axis direction including the pixel G(kx,ky). Alternatively, the temperature fluctuation calculation unit may be configured to calculate said drift component Tav(kx,ky) by performing polynomial fitting on the temperatures T of Nx pixels G arranged in the X-axis direction.

[0021] It is preferable that the defect region specifying unit is configured to cluster the plurality of defective pixels KG(kx,ky) extracted by said defective pixel extraction unit by applying a predetermined clustering algorithm, and set the obtained cluster or a region surrounding said cluster as said defect region.

[0022] a display device; and it is preferable to employ a configuration comprising a defect region display unit that defines a region on the upper surface of the structure to be inspected other than the defect region as a sound region, and causes the display device to display a binarized image in which the sound region and the defect region are represented in mutually different colors. In this case, the defective pixel extraction unit sets a plurality of reference values Tth and extracts defective pixels KG(kx,ky) for each reference value Tth; the defect region specifying unit specifies the defect region for each reference value Tth; and the defect region display unit can be configured such that the binarized image for each reference value Tth is represented with the sound regions in the same color as each other and the defect regions in mutually different colors, respectively, and these are superimposed and displayed on the display device.

[0023] Furthermore, it is preferable to employ a configuration comprising a defect depth estimation unit that defines a region on the upper surface of the structure to be inspected other than the defect region as a sound region, and estimates the depth position of the internal defect occurring in the structure to be inspected based on the change rate Rp of temperature fluctuation ΔT(kx,ky) occurring at the boundary between the defect region and the sound region, or based on the average value Tmean of temperature T(kx,ky) and the change rate Rp.

[0024] In this case, a database is provided in which model structure-related information, which is numerical information for a virtual model structure simulating the structure to be inspected, and model structure temperature information, which is information calculated by heat conduction analysis calculation using the model structure-related information and is information indicating the temperature distribution of each part of the model structure including the defect region and the sound region, are registered in association with each other, the model structure-related information includes at least: structure information indicating structural features of the model structure; environmental temperature information indicating features of the environmental temperature at the installation location of the model structure; on-site information indicating features related to the amount of solar radiation on the upper surface of the model structure; and defect information indicating the presence or absence of an internal defect in the model structure and features of the internal defect, Preferably, the defect depth estimation unit prepares a defect depth estimation model created by machine learning using the model structure-related information and the model structure temperature distribution information registered in the database as training data, and estimates the depth location of the internal defect occurring in the structure under inspection by inputting the inspection target structure-related information, which is numerical information corresponding to information other than the defect information in the model structure-related information, and the temperature T(kx,ky) information of the inspection target structure, or the temperature fluctuation ΔT(kx,ky) information calculated by the temperature fluctuation calculation unit, or both, into the defect depth estimation model. [Effects of the Invention]

[0025] The internal defect inspection method and system of the present invention analyze temperature data from a thermal image of the upper surface of a structure to be inspected (a structure in which asphalt pavement is laid on the upper surface of a concrete slab to form a walkway) by canceling error factors such as low-frequency drift components included in the characteristics of the temperature data using a unique method. As a result, the presence or absence of internal defects and the location of defect regions containing internal defects can be easily and accurately detected. [Brief explanation of the drawing]

[0026] [Figure 1] This is a block diagram showing the first embodiment of the internal defect inspection system of the present invention. [Figure 2] This is an overall flowchart (a) showing the first embodiment of the internal defect inspection method of the present invention (inspection method performed by the internal defect inspection system in Figure 1), and a flowchart (b) showing the contents of the temperature fluctuation calculation step and the defect pixel extraction step. [Figure 3] (a) is a perspective view showing how a thermal image of the inspection area on the upper surface of the structure under inspection is captured, and (b) is a graph of the temperature T of pixels G(1,ky) to G(Nx,ky) within the pixels G that make up the thermal image. [Figure 4](a) is a graph of the temperature fluctuation ΔT calculated for each pixel G(1,ky) to G(Nx,ky), and (b) is a figure comparing a thermal image representing the defect region detected by impact testing with a binarized image representing the defect region identified in the defect region identification step. [Figure 5] Graphs (a) and (b) show the results of verification tests of the internal defect inspection method of the first embodiment. [Figure 6] Figures (a) and (b) show two examples of how to set the reference value Tth in the defective pixel extraction step. [Figure 7] This is a block diagram showing a second embodiment of the internal defect inspection system of the present invention. [Figure 8] Figure 7 shows a second embodiment of the internal defect inspection method of the present invention (the inspection method performed by the internal defect inspection system in Figure 7), and Figure 8 shows a complete flowchart (a) and a block diagram (b) showing the flow of information when creating a database in the database preparation step. [Figure 9] (a) is a graph showing the relationship between the rate of change of temperature Rp at the boundary between a defective region and a healthy region on the surface of a model structure, and the depth and location Zf of the internal defect, as well as a schematic diagram (b) showing an overview of the model structure. [Figure 10] Block diagram (a) shows the flow of information in the defect depth estimation step, and block diagram (b) shows how to create the defect depth estimation model. [Figure 11] This is a diagram illustrating an example of model structure-related information. [Figure 12] This is a diagram showing an example of temperature change information for a model structure. [Figure 13] This is a diagram illustrating an example of general information about a structure to be inspected. [Figure 14] Figure (a) shows an example of information related to the structure being inspected, and Figure (b) shows an example of information on the temperature T and temperature fluctuation ΔT of pixel G. [Figure 15] Figure (a) shows an example of boundary temperature information derived during the defect depth estimation step, and Figure (b) shows an example of internal defect depth and size information output as the estimation result of the defect depth estimation step. [Modes for carrying out the invention]

[0027] <<Internal defect inspection system 10 / internal defect inspection method of the first embodiment>> The first embodiment of the internal defect inspection system and internal defect inspection method of the present invention will be described below with reference to Figures 1 to 6. The internal defect inspection system 10 of this embodiment is a computer system that inspects internal defects of a target structure by analyzing thermal images taken of the upper surface of the target structure, which has asphalt pavement laid on the upper surface of a concrete floor slab to form a passageway. As shown in Figure 1, the internal defect inspection system 10 includes functional blocks such as a temperature fluctuation calculation unit 12, a defect pixel extraction unit 14, a defect area identification unit 16, and a defect area display unit 18, and further includes a display device 20 such as a display and an information input device 22 such as a keyboard or mouse for user operation.

[0028] The first embodiment of the temperature measurement and prediction method of the present invention is an inspection method performed by an internal defect inspection system 10, and as shown in Figure 2(a), it consists of a temperature fluctuation calculation step S11 performed by a temperature fluctuation calculation unit 12, a defect pixel extraction step S12 performed by a defect pixel extraction unit 14, a defect region identification step S13 performed by a defect region identification unit 16, and a defect region display step S14 performed by a defect region display unit 18. The temperature measurement result prediction method (and temperature measurement result prediction system 10) of this embodiment will be explained in accordance with the overall flowchart in Figure 2(a), but before that, the structure to be inspected and the thermal image will be explained.

[0029] The structures to be inspected are, for example, bridge girders, etc., where asphalt pavement 28 is laid on the upper surface of a concrete deck slab 26 to form a road 30, as shown in the inspected structure 24 in Figure 3(a). A white or yellow center line 32 is drawn in the center of the width direction of the upper surface of the road 30. Note that auxiliary members such as railings are omitted in this figure.

[0030] The thermal image was captured using a thermal camera 36 mounted on an unmanned aerial vehicle 34, capturing the inspection area 38 on the upper surface of the structure 24 (the upper surface of the road 30). The thermal image NG38 shown in the upper part of Figure 3(b) is composed of a large number of pixels G arranged in an aligned manner, with temperature data and coordinate data assigned to each pixel G.

[0031] When the axis parallel to the length of road 30 is defined as the X-axis and the axis parallel to the width of road 30 is defined as the Y-axis, pixels G are arranged sequentially in the X-axis direction from the reference point, with Nx pixels arranged sequentially in the Y-axis direction. Here, each pixel G is assigned an XY coordinate, and each pixel G is represented as pixel G(kx,ky)[1≦kx≦Nx, 1≦ky≦Ny]. The temperature data of each pixel G is represented as temperature T(kx,ky). When analyzing the thermal image NG38, it will be treated not so much as an image that provides visual information, but rather as image data (a collection of temperature data and coordinate data for each pixel).

[0032] The thermal image NG38 has Ny rows of Nx pixels G arranged sequentially along the X-axis. Focusing on the row of pixels G(1,ky) to G(Nx,ky), the distribution of temperature T(1,ky) to T(Nx,ky) is as shown in the lower graph of Figure 3(b). Looking at this graph, the central part of the temperature characteristic curve bulges upward gently, which is the low-frequency drift component of the temperature characteristic curve explained in the background technology section. The high-frequency fluctuation component superimposed on the low-frequency drift component shown in the graph of Figure 3(b) is important data for detecting the state of internal defects.

[0033] <Temperature fluctuation calculation step S11 / Temperature fluctuation calculation unit 12> In the temperature fluctuation calculation step S11, a thermal image NG38 is acquired, and the low-frequency drift component is canceled from the temperature T(kx,ky) to extract the high-frequency fluctuation component. Specifically, steps S111 and S112 shown in Figure 2(b) are performed in order. First, for each pixel G, the average value (moving average) of the temperature T of a specified number of pixels G adjacent to pixel G(kx,ky) and arranged sequentially in the X-axis direction is calculated as the drift component Tav(kx,ky) [Step S111]. Alternatively, for each pixel G, the average value (moving average) of the temperature T of a specified number of pixels G including pixel G(kx,ky) and arranged in the X-axis direction is calculated as the drift component Tav(kx,ky) [Step S111]. The choice of which moving average to use is optional, but empirically, almost the same results can be obtained. The drift components Tav(1,ky)~Tav(Nx,ky) for the row of pixels G(1,ky)~G(Nx,ky) correspond to the "low-frequency drift component" in the lower graph of Figure 3(b) explained earlier.

[0034] Then, for each pixel G, the temperature fluctuation ΔT(kx,ky) is calculated by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky) [Step S112]. The graph in Figure 4(a) shows the distribution of temperature fluctuations ΔT(1,ky) to ΔT(Nx,ky) for the row of pixels G(1,ky) to G(Nx,ky). The temperature fluctuations ΔT(1,ky) to ΔT(Nx,ky) correspond to the "high-frequency fluctuation component" explained earlier, and by performing steps S111 and S112, this important data can be accurately extracted.

[0035] Furthermore, this temperature fluctuation calculation step S11 is characterized in that, when calculating the low-frequency drift component, it uses a "moving average in the X-axis direction (length direction of the road 30)" rather than a "moving average in the Y-axis direction (width direction of the road 30)". Roads 30 often have a center line 32 or tire marks from automobiles in the length direction (X-axis direction), so temperature unevenness may occur in the thermal image NG38 due to the difference in solar radiation absorptive coefficient between the parts with and without these. However, by using a "moving average in the X-axis direction (length direction of the road 30)" when calculating the low-frequency drift component, the effect of temperature unevenness can be minimized.

[0036] <Defective pixel extraction step S12 / Defective pixel extraction unit 14> In the defective pixel extraction step S12, a pixel G(kx,ky) for which the temperature fluctuation ΔT(kx,ky) calculated in the temperature fluctuation calculation step S11 exceeds a reference value Tth is extracted as a defective pixel KG(kx,ky). As described above, when there is an internal defect in a structure, there is a property that the temperature difference between a region (defective region KKR) where the internal defect exists inside on the surface of the structure and a region (sound region KZR) where no internal defect exists becomes large. Therefore, a pixel G(kx,ky) for which the temperature fluctuation ΔT(kx,ky) exceeds the reference value Tth is determined as a defective pixel KG(kx,ky) that has a high possibility of belonging to the defective region KKR.

[0037] In this embodiment, the reference value Tth is set as a value of h1 times the standard deviation σ (h1 is a constant satisfying h1>0) when the distribution of Nx temperature fluctuations ΔT(kx,ky) constituting each row is regarded as a normal distribution. For example, in the graph of Fig. 4(a), comparing the case where the reference value Tth is set to σ [h1=1] and the case where it is set to 2σ [h1=2], the number of defective pixels KG(kx,ky) is larger when the reference value Tth is set to σ. This means that the smaller the reference value Tth is, the higher the defect detection sensitivity becomes.

[0038] Note that other methods for determining the reference value Tth are also conceivable. For example, there is a method of setting the reference value Tth to the h2-th value (1<h2<Nx) counted from the largest value among the Nx temperature fluctuations ΔT(kx,ky) constituting each row, and there is also a method of setting the reference value Tth to any arbitrary constant larger than zero.

[0039] <Defective region identification step S13 / Defective region identification unit 16><Defective region display step S14 / Defective region display unit 18> In the defective region identification step S13, based on the position information of the defective pixels KG(kx,ky) extracted in the defective pixel extraction step S12, the defective region KKR in the thermal image NG38 is identified, and other regions are set as the sound region KZR. For example, one method is to use the position of each defective pixel KG(kx,ky) extracted in the defective pixel extraction step S12 as the defective region KKR, or to use the region surrounding all defective pixels KG(kx,ky) as the defective region KKR. In addition, to remove noise data, one method is to cluster the multiple defective pixels KG(kx,ky) extracted in the defective pixel extraction step S12 by applying a predetermined clustering algorithm, and use the resulting cluster or the region surrounding the cluster as the defective region KKR.

[0040] Then, in the defective area display step S14, a binarized image 40 is displayed on a display device 20 such as a display, in which the defective area KKR and the healthy area KZR are represented in different colors. The binarized image 40 shown in the lower part of Figure 4(b) is a binarized image when the reference value Tth=σ[h1=1], in which the defective area KKR is represented in black and the healthy area KZR is represented in white.

[0041] In the upper thermal image NG38 of Figure 4(b), the major defect regions KKR detected by a skilled inspector during impact testing are enclosed in rectangular frames. The defect regions KKR shown in the binarized image 40 are generally in similar locations to the actual defect regions (defect regions KKR detected by impact testing).

[0042] In addition, in the defect area display step S14, the binarized image 40 when the reference value Tth = σ and the binarized image 40 when the reference value Tth = 2σ may be displayed overlaid on each other. In this case, the defect area KKR is represented by different colors (for example, black and red), and the healthy area KZR is represented by the same color (for example, white), allowing the user to visually recognize the difference in the defect area KKR due to the difference in the reference value Tth.

[0043] <Supplementary information regarding steps S11-S13 mentioned above> Up to this point, in order to make the explanation easier to understand, the contents of steps S11 to S13 have been explained as "analyzing the thermal image NG38 of a specific inspection area 38 and identifying the defect area KKR within the inspection area 38." However, since the general structure to be inspected 24 is very large, in many cases the surface of the structure to be inspected 24 is divided into multiple inspection areas 38, and multiple thermal images NG38 taken for each inspection area 38 are analyzed to inspect the entire structure to be inspected 24.

[0044] Therefore, the process for identifying the KKR defect area of ​​the entire structure 24 under inspection could be, for example, "combining all thermal images NG38 before step S11 to create data for the overall thermal image, and then sequentially executing steps S11 to S13 on the overall thermal image data to identify the KKR defect area of ​​the entire structure 24 under inspection." Alternatively, "executing each step S11 to S13 individually for each thermal image NG38 to identify the KKR defect area for each inspection area 38, and then combining the results to obtain the KKR defect area for the entire structure 24 under inspection" could also be considered. In either case, almost the same results can be obtained. The stage at which information obtained from multiple thermal images NG38 is combined is flexible, and ultimately, the goal is to identify the KKR (Known Key Point Relief) area of ​​the entire structure 24 being inspected.

[0045] <Results of verification tests for the first embodiment> Next, a verification test conducted to verify the defect detection accuracy of the internal defect inspection method (and internal defect inspection system 10) of this embodiment will be described. The verification test compared the defect area KKR detected using the internal defect inspection method of this embodiment with the defect area KKR detected by a skilled inspector using tapping inspection on a specific structure that is actually in operation, and evaluated how well the former defect area KKR matched the latter defect area KKR.

[0046] Here, we decided to evaluate using two indicators: area-based conformance and count-based conformance. The area-based conformance is calculated using the total surface area of ​​the structure (the sum of the area of ​​defective regions KKR [R] and the area of ​​sound regions KZR [S]) as the denominator, and the area of ​​the portion where the detection results by the internal defect inspection method of this embodiment and the detection results by the impact sound inspection match (the sum of the area of ​​defective regions KKR [A] and the area of ​​sound regions KZR [D]) as the numerator. The count-based conformance is calculated using the number of defective regions KKR detected by the impact sound inspection [R] as the denominator, and the number of internal defects KKR detected by the internal defect inspection method of this embodiment [A] within that number [R] as the numerator.

[0047] The graph in Figure 5(a) shows the verification results when the reference value Tth in the defective pixel extraction step S12 is set to σ[h1=1], where the area-based precision was approximately 85% or higher, and the number-based precision was 85%. The graph in Figure 5(b) shows the verification results when the reference value Tth in the defective pixel extraction step S12 is set to 2σ[h1=2], where the area-based precision was approximately 85% or higher, and the number-based precision was 62%.

[0048] Focusing on the area-based conformance rate, good results of approximately 85% or higher were obtained, both when the reference value Tth was σ and when it was 2σ. The primary objective of the present invention is to analyze thermal images to easily identify defective areas and efficiently narrow down structures that require rigorous inspection, such as impact testing. Therefore, if the area-based conformance rate is 85% or higher, the internal defect inspection method of this embodiment can be said to be sufficiently practical.

[0049] Incidentally, looking at the precision on a case-by-case basis, setting the reference value Tth to σ[h1=1] yielded better results (85%) than setting it to 2σ[h2=2]. However, we did not consider making the reference value Tth even smaller than σ[h1=1]. This is because when the reference value Tth=σ[h1=1], the defect detection sensitivity is higher than when it is 2σ[h1=2], so the number of cases [C] that were incorrectly identified as KKR defects also increased, and it is undesirable to further increase the defect detection sensitivity. Currently, the precision on a case-by-case basis remains at 85%, which is thought to be partly due to the influence of noise data contained in the thermal image NG38. Therefore, if the noise data contained in the thermal image NG38 can be reduced, for example by using a more high-performance thermal camera 36, ​​the precision on a case-by-case basis is expected to improve further.

[0050] Furthermore, there is room for improvement in the method of setting the reference value Tth. In the verification test described above, the method chosen was to set the reference value Tth to h1 times the standard deviation σ (h1>0). This is to enable statistical and objective evaluation, but on the other hand, as shown in the graph on the right side of Figure 6(a), there is a possibility that pixels G(kx,ky) located in areas with many healthy regions KZR are being judged as defective pixels KG(kx,ky) too strictly.

[0051] Therefore, it is considered appropriate to establish a first criterion value Tth1 and a second criterion value Tth2, setting the first criterion value Tth1 to a value h1 times the standard deviation σ (where h1 is a constant h1>0), and the second criterion value Tth2 to an arbitrary constant greater than zero, and extracting pixels G(kx,ky) where the temperature fluctuation ΔT(kx,ky) exceeds both criterion values ​​Tth1 and Tth2 as defective pixels KG(kx,ky). As a result, as shown in the graph on the right side of Figure 6(b), pixels G(kx,ky) in areas with a large amount of healthy regions KZR can be judged as defective pixels KG(kx,ky) with the same strictness as pixels G(kx,ky) in areas with a large amount of defective regions KKR, and it is thought that the precision rate on a number basis will further improve.

[0052] <Summary of the first embodiment> As described above, the internal defect inspection method and internal defect inspection system 10 of this embodiment analyze the temperature data of a thermal image NG38 taken from the upper surface of a structure to be inspected 24 (a structure in which asphalt pavement 28 is laid on the upper surface of a concrete slab 26 to form a road 30). By canceling error factors such as low-frequency drift components contained in the temperature data using a unique method, it is possible to easily and accurately detect the presence or absence of internal defects and the location of the defect region KKR in which internal defects exist.

[0053] <<Second Embodiment of Internal Defect Inspection System 42 / Internal Defect Inspection Method>> Next, a second embodiment of the internal defect inspection system and internal defect inspection method of the present invention will be described with reference to Figures 7 to 15. Here, components identical to those in the first embodiment are denoted by the same reference numerals and their description is omitted. As shown in Figure 7, the internal defect inspection system 42 of this embodiment has the same configuration as the internal defect inspection system 10 described above, and is a computer system that further adds a defect depth estimation unit 44 and a database 46 to provide new functions. The new function is to estimate the depth location of internal defects NK occurring in the structure 24 under inspection based on the thermal image NG38.

[0054] A second embodiment of the temperature measurement and prediction method of the present invention is an inspection method performed by an internal defect inspection system 42, and as shown in Figure 8(a), in addition to the temperature fluctuation calculation step S11, defect pixel extraction step S12, defect region identification step S13, and defect region display step S14 described above, a database preparation step S21 for preparing a database 46 and a defect depth estimation step S22 performed by a defect depth estimation unit 44 are added.

[0055] Next, we will explain the temperature measurement result prediction method (and temperature measurement result prediction system 42) of this embodiment, focusing on the two newly added steps S21 and S22. Before that, however, we will explain the characteristics that form the technical basis for estimating the depth location of internal defects.

[0056] The graph in Figure 9(a) is a simulation result of a heat conduction analysis showing the relationship between the rate of change Rp of the temperature at the boundary portion KYB between the defective region KKR and the sound region KZR on the surface of the model structure X, which is a simulation model, and the depth position Zf of the internal defect NK. Figure 9(b) is a schematic diagram showing an overview of the model structure X. The inventors of the present invention have performed simulations as shown in Figures 9(a) and (b) using various model structures X and have found that there is a correlation between the rate of change Rp (=temperature gradient) of the surface temperature at the boundary portion KYB and the depth position Zf of the internal defect NK.

[0057] There are several ways to quantify the rate of change of surface temperature Rp, but here, as shown in the graph on the right side of Figure 9(a), the rate of change of surface temperature Rp is quantified in the form Rp = Ta / Tb. Ta is the difference between the "main temperature of the defective region KKR" and the "temperature at the boundary point between the defective region KKR and the healthy region KZR," and Tb is the difference between the "main temperature of the defective region KKR" and the "main temperature of the healthy region KZR." Therefore, the rate of change of surface temperature Rp = Ta / Tb becomes larger when the temperature gradient is large and smaller when the temperature gradient is small.

[0058] The graph on the left in Figure 9(a) shows the length of the internal defect NK in the X-axis direction (2 × Lx) on the horizontal axis and the rate of change of surface temperature Rp on the vertical axis. The solid line represents the characteristics when the internal defect NK is located at a depth Zf = 45 mm, that is, the characteristics when the internal defect NK (void) is located at the boundary between the asphalt pavement 28 and the concrete slab 26 in this embodiment. The dashed line in the same graph represents the characteristics when the internal defect NK is located at a depth Zf = 85 mm, that is, the characteristics when the internal defect NK (void) is located inside the concrete slab 26, which is deeper than the asphalt pavement 28. From this graph, it can be seen that if the length of the internal defect NK (2 × Lx) is greater than a certain value, a strong correlation occurs between the rate of change of surface temperature Rp and the depth position Zf of the internal defect NK.

[0059] Therefore, if the correlation between the rate of change of surface temperature Rp and the depth position Zf is known in advance, the depth position Zf can be easily calculated by substituting the rate of change of temperature Rp derived from the thermal image NG38 into the formula representing that correlation, or by substituting the rate of change of temperature Rp and the mean value Tmean derived from the thermal image NG38, and the size of the internal defect NK can also be easily calculated. Alternatively, a conversion table can be created based on the correlation, and the calculation can be performed using that conversion table. Incidentally, the mean value Tmean is the average value of the temperature T of each pixel G in the thermal image NG38. Furthermore, the rate of change of surface temperature Rp can be any numerical representation of the magnitude of the temperature gradient, and an index other than Ta / Tb may be used.

[0060] It should be noted that the surface temperature of a structure is affected by changes in ambient temperature over time. For example, the surface temperature of a structure at 12:00 (ambient temperature 20°C) will differ slightly depending on whether the ambient temperature at 6:00 that day was 0°C or 10°C. Therefore, in order to determine the depth position Zf with higher accuracy, it is preferable to analyze the thermal image NG38 while also considering the history of ambient temperature. The database preparation step S21 and defect depth estimation step S22, described below, also take into account changes in ambient temperature over time.

[0061] <Database preparation step S21 / Database 28> In the database preparation step S21, as shown in Figure 8(b), a database 46 is prepared in which model structure-related information and model structure temperature change information are linked and registered together. Model structure X is a virtual structure that simulates the structure to be inspected 24, and is a simulation model for heat conduction analysis.

[0062] First, let's explain the model structure-related information. Model structure-related information is numerical information used to perform heat conduction analysis simulations for model structure X (model structures 1, 2, ...), and includes at least structural information that shows the structural characteristics of model structure X, ambient temperature information that shows the characteristics of the ambient temperature at the installation site of model structure X, site information that shows the characteristics of the amount of solar radiation on the upper surface of model structure X, and defect information that shows the presence or absence of internal defects NK in model structure X and the characteristics of internal defects NK.

[0063] In the example shown in Figure 11, the structural information items include the thickness of the asphalt and concrete, and the thermal conductivity values ​​of the asphalt and concrete, while the environmental temperature information items include data on the outside temperature for each month and hour. For the outside temperature data, it is advisable to use data for appropriate locations published by, for example, the Japan Meteorological Agency.

[0064] Furthermore, the field information items include the value of the solar radiation absorptivity of asphalt and data on global solar radiation for each month and hour. For the global solar radiation data, it is advisable to use data for appropriate locations published by the Japan Meteorological Agency, for example. The values ​​of the solar radiation absorptivity of asphalt and global solar radiation are used when calculating the equivalent ambient temperature necessary for heat conduction analysis. In addition, if it is assumed that internal defects NK (voids) exist, the defect information items include values ​​such as the depth, location, thickness, length, and thermal conductivity of the air layer.

[0065] Next, we will explain the model structure temperature change information. The model structure temperature change information is data on the temperature distribution of each part calculated by performing a heat conduction analysis simulation based on model structure-related information. As shown in Figure 12, it is the temperature of each part of model structure X (model structure 1, 2, ...) calculated for each month and time.

[0066] The calculated temperature distribution data includes at least temperature data for the defective region KKR and the sound region KZR on the surface of the model structure X. If internal defects NK, such as voids, exist inside the model structure X, the temperature difference between the defective region KKR and the sound region KZR on the structure surface increases, and the isotherms 48 at the boundary between the two regions tend to become denser. Therefore, the degree of density of the isotherms 48 at the boundary becomes important information.

[0067] Thus, the database 46 contains model structure-related information as shown in Figure 11 and model structure temperature change information as shown in Figure 12, linked together. This information serves as training data when creating the defect depth estimation model 44a (machine learning model), which will be described later. Incidentally, "preparing database 46" means creating database 46 and storing it in the system, making database 46 already stored in the system usable, and making information from database 46 stored on other devices or storage media retrievalable.

[0068] <Defect depth estimation step S22 / Defect depth estimation unit 26> The defect depth estimation step S22 is performed after the defect area identification step S13 and the database preparation step S21, as shown in Figure 8(a). In the defect depth estimation step S22, as shown in Figure 10(a), a predetermined defect depth estimation model 44a (machine learning model) is prepared, and the depth location Zf of the internal defect NK occurring in the structure under inspection is estimated by inputting the structure under inspection-related information and the temperature T(kx,ky) information of the thermal image NG38 of the structure under inspection 24, or the temperature fluctuation ΔT(kx,ky) information calculated in the temperature fluctuation calculation step S12, or both, into the defect depth estimation model 44a.

[0069] First, let's explain the defect depth estimation model 44a. As shown in Figure 10(b), the defect depth estimation model 44a is a machine learning model created using model structure-related information and model structure temperature change information stored in the database 46 as training data. Therefore, the defect depth estimation model 44a inevitably performs estimations that take into account the "correlation between the rate of change of surface temperature Rp and the depth location Zf" and the "time-dependent change in ambient temperature." The machine learning algorithm is not particularly limited; for example, a model equation created using multiple regression analysis may be used, or methods such as random forests and neural networks can be used.

[0070] Incidentally, "preparing the defect depth estimation model 44a" means creating the defect depth estimation model 44a and storing it in the system, making the defect depth estimation model 44a already stored in the system usable, and making the defect depth estimation model 44a stored in other devices or storage media usable, etc.

[0071] Next, we will explain the information related to the structure under inspection that is input into the defect depth estimation model 44a. Figure 14(a) shows an example of the information related to the structure under inspection, and each item of the information related to the structure under inspection is numerical information corresponding to the information other than defect information (structural information, ambient temperature information, and site information) in the model structure-related information shown in Figure 11.

[0072] Information related to the structure to be inspected [=numerical information] is basically provided by the user, but some items are difficult for the user to identify unless they have specialized knowledge. Therefore, in order to reduce the burden on the user, in this embodiment the user provides text information [=character information or numerical information] of the structure to be inspected, and a process is performed to derive the information related to the structure to be inspected from the provided structure overview information.

[0073] To explain this point by comparing the example of the structure inspection overview information shown in Figure 13 with the example of the structure inspection-related information shown in Figure 14(a), for example, the "asphalt thickness" item in the structure inspection-related information can be identified relatively easily even by users without specialized knowledge, so the structure inspection overview information [=numerical information] of "25" is obtained from the user and used as is in the structure inspection-related information [=numerical information].

[0074] On the other hand, for example, the "thermal conductivity of asphalt" item in the information related to the structure being inspected is difficult for users to determine without specialized knowledge. Therefore, the system obtains summary information about the structure being inspected [=text information] such as "standard (asphalt density)" and "low moisture content (asphalt moisture content)" from the user, and automatically derives the information related to the structure being inspected [=numerical information] such as "1.45 (thermal conductivity of asphalt)" based on this information. The derivation method is not particularly limited, but one possible method is to prepare a conversion table in advance that converts specific text information into specific numerical information.

[0075] Furthermore, for example, the "outside temperature by time" item in the information related to the inspected structure can be determined by the user by researching past data from the Japan Meteorological Agency, but researching it each time is troublesome. Therefore, the system obtains summary information of the inspected structure [=text information] such as "April 2nd (inspection date)" and "Toyama City XX district (installation location)" from the user, and automatically derives the information related to the inspected structure [=numerical information], which is the outside temperature data by time, based on this. The method of deriving is not particularly limited, but for example, one possible method is to store temperature data from all over the country published by the Japan Meteorological Agency in advance in a database, automatically search the database based on the text information, and extract the corresponding numerical information.

[0076] Furthermore, to reduce the burden on the user, items such as "solar radiation absorption rate of asphalt" and "total solar radiation amount per hour" within the information related to the inspected structure are automatically derived from the user's textual summary information of the inspected structure, based on this information. If the user can specify the numerical values ​​for all items, the system may be configured to allow the user to provide all the information related to the inspected structure [=numerical information].

[0077] Next, regarding the temperature information of the structure under inspection 24 input to the defect depth estimation model 44a, the temperature information is, as shown in Figure 14(b), the temperature T(kx,ky) information of each pixel G in the thermal image NG38, or the temperature fluctuation ΔT(kx,ky) information of each pixel G, or both. As described above, this defect depth estimation step S22 estimates the depth position Zf of the internal defect NK based on the "correlation between the rate of change of surface temperature Rp and the depth position Zf," so information is needed to enable the estimation (or calculation) of the "rate of change of surface temperature Rp" of the structure under inspection 24. Therefore, at least the temperature T or the temperature fluctuation ΔT information is sufficient, and if both information is available, the "rate of change of surface temperature Rp" can be estimated (or calculated) even more efficiently.

[0078] Figure 15(a) shows an example of time-dependent temperature information derived during the process of the defect depth estimation model 44a outputting its estimation results, specifically, an example of time-dependent temperature information for the defect region KKR (defect region 1, 2, ...) on the surface of the structure under inspection 24. Based on this temperature information, the defect depth estimation model 44a estimates the depth position Zf for each internal defect NK (internal defect 1, 2, ...) and outputs it in a user-friendly format, as shown in Figure 15(b). The size of the internal defect NK can also be easily calculated, so the size of the internal defect NK may also be output.

[0079] <Summary of the second embodiment> As described above, the internal defect inspection method and internal defect inspection system 42 of this embodiment can achieve the same effects as the first embodiment, and furthermore, it can accurately estimate the depth position Zf of the internal defect NK occurring in the structure 24 under inspection.

[0080] <<Other embodiments and variations>> It should be noted that the internal defect inspection method and internal defect inspection system of the present invention are not limited to the embodiments described above. For example, in the description of the above embodiments, it was explained that a thermal image NG38 of the structure to be inspected 24 is captured by a thermal camera 36 mounted on an unmanned aircraft, but the method of capturing the thermal image is flexible, and the size of the thermal image and the number of pixels (Nx × Ny) are not particularly limited.

[0081] In the temperature fluctuation calculation step (temperature fluctuation calculation unit) of the above embodiment, the moving average method is used to calculate the low-frequency drift component Tav(kx,ky) contained in the temperature T of Nx pixels G arranged in the X-axis direction. However, other methods may be used. For example, a method can be considered in which polynomial fitting is performed on the temperature T of Nx images G arranged in the X-axis direction to calculate the drift component Tav(kx,ky). Although the moving average method is a very simple and easy-to-use method, if there are many pixels G that contain temperature information other than the drift component among the specified number of pixels G when calculating the moving average, the drift component cannot be calculated appropriately. In contrast, if a polynomial fitting method is used, the processing becomes somewhat more complex, but the drift component can be calculated more accurately. It is preferable to use polynomials for fitting, but if the conditions are met, various functions such as trigonometric functions, exponential functions, and sigmoid functions may be used, or fitting processing using a low-pass filter may be performed.

[0082] Furthermore, the specific details (items and numerical values) of the model structure-related information, the overview information of the structure to be inspected, and the information related to the structure to be inspected shown in Figures 11, 13, and 14(a) are merely examples and should be modified as appropriate according to the method of heat conduction analysis simulation and the characteristics of the structure. [Explanation of symbols]

[0083] 10,42 Internal Defect Inspection System 12 Temperature fluctuation calculation unit 14. Defective pixel extraction unit 16. Defect area identification unit 18. Defect area display section 20 Display device 22 Information Input Device 24 Structures subject to inspection 26 Concrete slab 28 Asphalt pavement 30 Road (passage) 38 Examination Areas 40. Binarized image 44 Defect depth estimation unit 44a Defect depth estimation model 46 Databases S11 Temperature fluctuation calculation step S12 Defective pixel extraction step S13 Defect area identification step S14 Defect area display step S21 Database Preparation Steps S22 Defect depth estimation step G(kx,ky) pixels KG(kx,ky) Defective Pixel KKR defect area KYB boundary part KZR healthy area NG38 Thermal image of the examination area NK internal defect Rp surface temperature change rate T(kx,ky) Temperature of each pixel Tav(kx,ky): Temperature moving average of each pixel Tmean: The average temperature of each pixel in a thermal image. ΔT(kx,ky) Temperature fluctuation

Claims

1. An internal defect inspection method, performed by a computer system, which involves analyzing thermal images taken of the upper surface of a structure to be inspected, in which asphalt pavement is laid on the upper surface of a concrete slab to form a walkway, to detect internal defects in the structure to be inspected, When the axis parallel to the length direction of the passage is defined as the X-axis and the axis parallel to the width direction of the passage is defined as the Y-axis, the plurality of pixels G constituting the thermal image are arranged sequentially in the X-axis direction from the reference point, with Nx pixels arranged sequentially in the Y-axis direction, and each pixel G is represented as pixel G(kx, ky) [1≦kx≦Nx, 1≦ky≦Ny], and the temperature data of pixel G(kx, ky) is represented as temperature T(kx, ky), A temperature variation calculation step is performed to calculate the low-frequency drift component Tav(kx,ky) contained in the temperature T of Nx pixels G arranged in the X-axis direction for each pixel G, and to calculate the high-frequency temperature variation ΔT(kx,ky) by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky), and A defective pixel extraction step is performed to extract pixels G(kx, ky) in which the temperature fluctuation ΔT(kx, ky) calculated in the above temperature fluctuation calculation step exceeds the reference value Tth, as defective pixels KG(kx, ky), An internal defect inspection method characterized by comprising: a defect area identification step, which identifies a defect area in the concrete slab where an internal defect has occurred, based on the positional information of the defect pixels KG(kx, ky) extracted in the defect pixel extraction step, when the structure to be inspected is viewed from above.

2. The internal defect inspection method according to claim 1, wherein in the temperature fluctuation calculation step, the average value of the temperature T of a specified number of pixels G adjacent to pixel G(kx,ky) and arranged sequentially in the X-axis direction, or the average value of the temperature T of a specified number of pixels G including pixel G(kx,ky) and arranged in the X-axis direction, is calculated as the drift component Tav(kx,ky).

3. The internal defect inspection method according to claim 1, wherein the drift component Tav(kx, ky) is calculated by performing polynomial fitting on the temperature T of Nx images G arranged in the X-axis direction in the temperature fluctuation calculation step.

4. The internal defect inspection method according to claim 1, wherein in the defect region identification step, a plurality of defect pixels KG(kx, ky) extracted in the defect pixel extraction step are clustered by applying a predetermined clustering algorithm, and the resulting cluster or the region surrounding the cluster is defined as the defect region.

5. The internal defect inspection method according to claim 1, further comprising a defect area display step, in which the area of ​​the upper surface of the structure to be inspected, other than the defect area, is defined as a healthy area, and a display device displays a binarized image in which the healthy area and the defect area are represented by different colors.

6. In the aforementioned defective pixel extraction step, multiple reference values ​​Tth are set and a defective pixel KG (kx, ky) is extracted for each reference value Tth. In the defect region identification step, the defect region is identified for each reference value Tth, The internal defect inspection method according to claim 5, wherein in the defect area display step, the binarized images for each reference value Tth are displayed such that the healthy areas are represented by the same color and the defect areas by different colors, and these are superimposed and displayed on the display device.

7. In the defective pixel extraction step, a first reference value Tth1 and a second reference value Tth2 are provided as reference values. The first reference value Tth1 is set to the value of h1 times the standard deviation σ (where h1 is a constant h1 > 0) when the distribution of the Nx temperature fluctuations ΔT(kx, ky) constituting each row is considered to be a normal distribution. Set the second reference value Tth2 to any constant greater than zero, The internal defect inspection method according to claim 1, wherein a pixel G(kx, ky) whose temperature fluctuation ΔT(kx, ky) calculated in the temperature fluctuation calculation step exceeds both reference values ​​Tth1 and Tth2 is extracted as a defective pixel KG(kx, ky).

8. An internal defect inspection method according to any one of claims 1 to 7, comprising a defect depth estimation step of estimating the depth location of the internal defect occurring in the structure to be inspected, based on the rate of change Rp of the temperature fluctuation ΔT(kx, ky) occurring at the boundary between the defective area and the healthy area, or based on the average value Tmean of the temperature T(kx, ky) and the rate of change Rp.

9. The system includes a database preparation step of preparing a database in which model structure-related information, which is numerical information about a virtual model structure that simulates the structure to be inspected, and model structure temperature information, which is information calculated by performing a heat conduction analysis using the model structure-related information, and which shows the temperature distribution of each part of the model structure, including the defective area and the sound area, are linked together and registered. The aforementioned model structure-related information includes, at a minimum, structural information indicating the structural characteristics of the model structure, ambient temperature information indicating the characteristics of the ambient temperature at the installation site of the model structure, site information indicating the characteristics of the amount of solar radiation on the upper surface of the model structure, and defect information indicating the presence or absence of internal defects in the model structure and the characteristics of those internal defects. The internal defect inspection method according to claim 8, wherein in the defect depth estimation step, a defect depth estimation model created by machine learning using the model structure-related information and the model structure temperature distribution information registered in the database as training data is prepared, and the depth location of the internal defect occurring in the structure to be inspected is estimated by inputting into the defect depth estimation model the inspection target structure-related information, which is numerical information corresponding to the information other than the defect information in the model structure-related information, and the temperature T(kx, ky) information of the structure to be inspected or the temperature fluctuation ΔT(kx, ky) information calculated in the temperature fluctuation calculation step or both.

10. An internal defect inspection system comprising a computer system detects internal defects in a structure under inspection by analyzing thermal images taken of the upper surface of the structure under inspection, in which asphalt pavement is laid on the upper surface of the concrete slab to form a walkway. When the axis parallel to the length direction of the passage is defined as the X-axis and the axis parallel to the width direction of the passage is defined as the Y-axis, the plurality of pixels G constituting the thermal image are arranged sequentially in the X-axis direction from the reference point, with Nx pixels arranged sequentially in the Y-axis direction, and each pixel G is represented as pixel G(kx, ky) [1≦kx≦Nx, 1≦ky≦Ny], and the temperature data at the position of pixel G(kx, ky) is represented as temperature T(kx, ky), A temperature fluctuation calculation unit calculates the low-frequency drift component Tav(kx,ky) contained in the temperature T of Nx pixels G arranged in the X-axis direction for each pixel G, and calculates the high-frequency temperature fluctuation ΔT(kx,ky) by subtracting the drift component Tav(kx,ky) from the temperature T(kx,ky). A defective pixel extraction unit extracts pixels G(kx, ky) where the temperature fluctuation ΔT(kx, ky) calculated by the temperature fluctuation calculation unit exceeds the reference value Tth, as defective pixels KG(kx, ky), An internal defect inspection system characterized by comprising: a defect area identification unit that identifies a defect area in the concrete slab where an internal defect has occurred, based on the positional information of the defect pixels KG(kx, ky) extracted by the defect pixel extraction unit, when the structure to be inspected is viewed from above.

11. The internal defect inspection system according to claim 10, wherein the temperature fluctuation calculation unit calculates the average value of the temperature T of a specified number of pixels G adjacent to pixel G(kx,ky) and arranged sequentially in the X-axis direction, or the average value of the temperature T of a specified number of pixels G including pixel G(kx,ky) and arranged in the X-axis direction, as the drift component Tav(kx,ky).

12. The internal defect inspection system according to claim 10, wherein the temperature fluctuation calculation unit calculates the drift component Tav(kx, ky) by performing polynomial fitting on the temperatures T of Nx images G arranged in the X-axis direction.

13. The internal defect inspection system according to claim 10, wherein the defect region identification unit clusters a plurality of defect pixels KG(kx, ky) extracted by the defect pixel extraction unit by applying a predetermined clustering algorithm, and defines the resulting cluster or the region surrounding the cluster as the defect region.

14. Display device and The internal defect inspection system according to claim 10, further comprising: a defect area display unit that defines the area of ​​the upper surface of the structure to be inspected, other than the defect area, as a sound area, and displays a binarized image on the display device in which the sound area and the defect area are represented by different colors.

15. The defective pixel extraction unit sets multiple reference values ​​Tth and extracts a defective pixel KG (kx, ky) for each reference value Tth. The defect region identification unit identifies the defect region for each reference value Tth, The internal defect inspection system according to claim 10, wherein the defect area display unit displays the binarized image for each reference value Tth, with the healthy areas being the same color and the defect areas being different colors, and these are superimposed and displayed on the display device.

16. An internal defect inspection system according to any one of claims 10 to 15, comprising a defect depth estimation unit that estimates the depth location of the internal defect occurring in the structure under inspection, based on the rate of change Rp of the temperature fluctuation ΔT(kx, ky) occurring at the boundary between the defective area and the healthy area, or based on the average value Tmean of the temperature T(kx, ky) and the rate of change Rp.

17. The system includes a database in which numerical information about a virtual model structure that simulates the structure to be inspected, which is model structure-related information, and model structure temperature information, which is information calculated by performing a heat conduction analysis using the model structure-related information and shows the temperature distribution of each part of the model structure, including the defective area and the sound area, are mutually linked and registered. The aforementioned model structure-related information includes, at a minimum, structural information indicating the structural characteristics of the model structure, ambient temperature information indicating the characteristics of the ambient temperature at the installation site of the model structure, site information indicating the characteristics of the amount of solar radiation on the upper surface of the model structure, and defect information indicating the presence or absence of internal defects in the model structure and the characteristics of those internal defects. The defect depth estimation unit prepares a defect depth estimation model created by machine learning using the model structure-related information and the model structure temperature distribution information registered in the database as training data, and estimates the depth location of the internal defect occurring in the structure to be inspected by inputting the inspection target structure-related information, which is numerical information corresponding to information other than the defect information in the model structure-related information, and the temperature T(kx, ky) information of the structure to be inspected or the temperature fluctuation ΔT(kx, ky) information calculated by the temperature fluctuation calculation unit or both, into the defect depth estimation model, as described in claim 16.

Images