JP2013096741A - Infrared survey method of structure and infrared survey arithmetic device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a material for determining whether or not an abnormal part extracted from a thermal image of an infrared camera includes a defect.SOLUTION: An infrared survey method of structures includes: identifying factors affecting a thermal image of a structure and a relational expression of a multivariable analysis for obtaining a probability that a defect is included in the abnormal part extracted from the thermal image of the structure, using information of the factors; capturing the structure with the infrared camera to obtain the thermal image; extracting the abnormal part a temperature of which is different from that of the periphery, from the thermal image; determining the information of the factors in the abnormal part; and digitizing the determined factor information, and applying the numerical value to the relational expression of the multivariable analysis to obtain the probability that the defect is included in the extracted abnormal part.

Description

  The present invention relates to an infrared survey of a structure for investigating a defect in the structure using a thermal image taken by an infrared camera, and particularly includes an abnormality in an abnormal portion extracted from the thermal image. Probability is calculated.

  In concrete structures such as bridges, overpasses, and buildings (hereinafter simply referred to as “structures”), there are defects such as damage caused by aging and initial defects existing from the beginning of construction. The defect itself may be a cavity (including a crack) or may be the starting point of a cavity (including a crack). When the defect is accompanied by the cavity as described above, the surface layer is likely to be peeled off, and it is dangerous if it comes off. For this reason, it is required to find a defect with a cavity and prevent peeling.

  A defect with a cavity may be manifested as a crack on the surface of the structure. Therefore, in the past, peeling was prevented by visually inspecting the appearance of the structure and checking for cracks on the surface of the structure. However, visual inspection can only find defects with cracks. In other words, you can't find a defect that doesn't crack. Although some cracks have a low risk of not coming off immediately, visual inspection cannot determine the risk of cracking. For these reasons, visual inspection is considered insufficient as a means for preventing peeling.

  In recent years, an infrared survey using an infrared camera has been proposed instead of a visual inspection. The infrared camera detects energy in the infrared band emitted from the subject, converts the detected energy into temperature, and generates temperature distribution image data. This two-dimensional image of the temperature distribution is called a thermal image or a thermographic image. An example of an infrared survey is disclosed in Patent Documents 1 and 2 below, for example.

  Under the influence of outside air and sunlight, the structure repeatedly absorbs heat from the outside to the inside of the structure and releases heat from the inside of the structure to the outside. Since the defect accompanied by the cavity during heat absorption and heat dissipation functions as a heat insulating layer, heat transfer is blocked by the defect. As a result, a temperature difference is generated between a portion having a defect and a portion having no defect, resulting in a difference in infrared emissivity. When a thermal image of a structure is taken with an infrared camera in such a state, a defective portion and a non-defective portion are displayed in different colors. At this time, the part with the defect is locally displayed in the part without the defect. In this way, a portion having a different color from the surroundings is referred to as an abnormal portion, and the other portions are referred to as healthy portions. By observing the thermal image and determining whether or not there is an abnormal part, it is possible to determine the presence or absence of a defect in the structure, and further to determine the position of the defect. Infrared surveys have the advantage that damage can be prevented by determining the presence or absence of defects inside the structure that cannot be determined by visual observation.

JP 2005-140622 A JP 2006-329760 A

  However, in the infrared survey, it may be determined as an abnormal part even though it is actually a healthy part. This is because the cause of the abnormal portion in the thermal image is not only the malfunction inside the structure. For example, when a structure is repaired, the infrared emissivity in the repaired trace changes depending on the difference in thermal conductivity of the repair material used. Further, the infrared emissivity is different between a structure having a deposit such as free lime on the surface and a structure having no deposit. That is, repairs and deposits cause abnormal portions in the thermal image. There are other factors that cause abnormal portions in the thermal image.

  In the thermal image, it is impossible to determine whether the cause of the abnormal part is a defect or a repair or a deposit. For this reason, even if an abnormal part is extracted from a thermal image, there is actually no defect in the abnormal part, and there is no risk of peeling and there is no need for repair. Thus, the determination of the presence or absence of a defect using a thermal image includes an erroneous determination.

  Normally, when an abnormal part is extracted, an operation for judging the necessity of repair, for example, a hammering test is performed. However, if the abnormal part is obtained by erroneous determination, this work itself is useless. In particular, when there are a large number of abnormal parts and there are many misjudgments, useless work becomes enormous. In order to reduce such useless work, it is desired to be able to determine the validity of the determination of the presence or absence of defects.

  The present invention has been made in view of such a situation, and an object of the present invention is to provide a material for determining whether or not a defect is included in an abnormal portion extracted from a thermal image of an infrared camera.

The first invention is
In an infrared survey method for a structure that finds defects in the structure by extracting abnormal parts from the thermal image of the structure photographed with an infrared camera,
Identify the factors that affect the thermal image and the relational expression for multivariate analysis that calculates the probability that the abnormal part extracted from the thermal image using the information of the factor contains a defect,
Take a thermal image by photographing the structure with an infrared camera,
Using the acquired thermal image, extract the abnormal part where the temperature is different from the surroundings,
Determine the information of the factor in the extracted abnormal part,
The information of the determined factor is digitized, the numeric value is applied to the relational expression of the multivariate analysis, and the probability that a defect is included in the extracted abnormal part is obtained.

  In the first invention, a logistic regression equation can be used as the relational expression for the multivariate analysis.

  In the first invention, it is possible to use the presence or absence of cracks on the surface of the structure, the surface state of the structure, and the service life of the structure as the factors.

  In the first invention, it is possible to make a determination as to the failure using the obtained probability. Specifically, it is possible to obtain the probabilities in a plurality of abnormal portions, respectively, and determine to deal with the defect in order from the abnormal portion with the highest probability. In addition, it is possible to determine that the defect is dealt with with respect to an abnormal part having the probability equal to or higher than a threshold value.

Also, the second invention is
An infrared investigation computing device used in infrared investigation of a structure to find a defect in the structure by extracting an abnormal part from a thermal image of the structure taken with an infrared camera,
A storage unit for storing a factor that affects the thermal image, and a relational expression of multivariate analysis for obtaining a probability that the abnormal part extracted from the thermal image using the information of the factor includes a defect;
An input unit for inputting information of the factor;
Obtaining information of the factor from the input unit, obtaining the relational expression of the multivariate analysis from the storage unit, and calculating a probability that the abnormal part extracted from the thermal image includes a defect,
And a notifying unit for notifying the probability obtained by the calculating unit.

  In the second invention, a logistic regression equation can be used as the relational expression for the multivariate analysis.

  In the second invention, as the factors, it is possible to use the presence or absence of cracks on the surface of the structure, the surface state of the structure, and the service life of the structure.

  In the second aspect of the invention, it is possible to identify an abnormal part that needs to be dealt with using the obtained probability.

  According to the present invention, an abnormal part is extracted from a thermal image taken by an infrared camera, and the probability that the abnormal part includes a defect is obtained. It can be determined whether or not a defect is included in the abnormal portion extracted based on the probability level. Therefore, it is possible to greatly reduce the case where there is no defect when the extracted abnormal part is actually inspected.

  Moreover, since it is determined by the numerical value called the probability whether or not a defect is included in the abnormal part, the determination is objective. In this way, a determination result that is not influenced by the skill of the user can be obtained.

  Moreover, since the urgency of repairing a defect can be determined based on the obtained probability level, it is possible to preferentially repair a defect having a high risk of peeling / peeling.

FIG. 1 shows the configuration of an infrared survey system according to the present invention. FIG. 2 shows the result of decision tree analysis. FIG. 3 shows the average values of the defect probabilities of the abnormal part that does not require repair and the abnormal part that requires repair. FIG. 4 shows the frequency for each defect probability. FIG. 5 shows the processing procedure of the infrared survey according to the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[1. Overview of infrared survey system configuration]
FIG. 1 shows the configuration of an infrared survey system according to the present invention. The infrared survey system 1 detects infrared rays emitted from the structure 3 and generates thermal image data, and outputs a thermal image of the structure 3 using the thermal image data generated by the infrared camera 10. A thermal image output device 20 that captures the structure 3 and generates appearance image data, and an appearance image output device that outputs an appearance image of the structure 3 using the appearance image data generated by the camera 30 40, and an arithmetic device 50 for obtaining a probability that the abnormal part extracted from the thermal image includes a defect (hereinafter referred to as “probability probability”) based on information on factors input from the outside.

  The infrared camera 10 detects the energy of the infrared band emitted from the structure 3 with an infrared detection element, converts the detected energy into temperature, and generates thermal image data for imaging the converted temperature distribution. To do. As the infrared detection element, an element that detects mid-infrared rays such as InSb, QWIP, and μ-bolometer is used. The infrared camera 10 transmits the generated thermal image data as a thermal image signal S 1 directly to the thermal image output device 20 or stores it in the storage medium 12.

  The thermal image output device 20 acquires thermal image data by the thermal image signal S1 output from the infrared camera 10 or the storage medium 12, and displays a thermal image based on the thermal image data on the display unit 22. A thermal image can be printed instead of being displayed on the display unit 22.

  The camera 30 generates appearance image data of the structure 3 according to the photographing of the structure 3. The camera 30 corresponds to a general still camera or video camera. Any camera that can capture the appearance of the structure 3 regardless of whether it is a still image or a moving image can be used. The camera 30 transmits the generated appearance image data as an appearance image signal S 2 directly to the appearance image output device 40 or stores it in the storage medium 32.

  The appearance image output device 40 acquires appearance image data from the appearance image signal S2 output from the camera 30 or the storage medium 32, and displays an appearance image based on the appearance image data on the display unit. It is also possible to print the appearance image instead of displaying it on the display unit 42.

  The appearance image output device 40 has an image analysis unit (not shown). The image analysis unit analyzes the appearance image, determines information on factors used when obtaining the defect probability, and outputs the determination result to the outside. As will be described later, in this embodiment, the factors used when determining the probability of failure are `` crack presence / absence '', `` surface condition '', and `` service life '', and the image analysis section includes `` crack presence / absence ''. And “surface condition”.

  On the other hand, the appearance image output device 40 may not have the image analysis unit. In this case, it is necessary for the user to visually check the appearance image or directly look at the structure to determine information on factors to be used when determining the defect probability. The thermal image output device 20 can also be used as the appearance image output device 40.

  The computing device 50 is a device for determining a failure probability, and includes an input unit 52, a storage unit 54, a computing unit 56, and a notification unit 58. As the arithmetic unit 50, a personal computer can be used.

  The input unit 52 inputs information on factors used in obtaining the defect probability to the calculation unit 56 in accordance with the input operation of the user and / or the determination result output from the appearance image output device 40.

  The storage unit 54 stores software and various numerical values used when determining the failure probability. [3. As described in [Example of calculation of failure probability], in this embodiment, the failure probability is obtained using a logistic regression equation. For this reason, the memory | storage part 54 memorize | stores the various numerical values used with the software for performing the calculation using a logistic regression equation, and a logistic regression equation.

  The calculation unit 56 obtains the failure probability using the factor information input from the input unit 52, the software stored in the storage unit 54, and various numerical values. In this embodiment, the calculation unit 56 uses the information on “presence / absence of cracks”, “surface state”, “service life” input from the input unit 52 and various numerical values stored in the storage unit 54 as variables of the logistic regression equation. Replace it to find the failure probability. The calculation will be described in [3. An example of calculation of failure probability] will be described.

  The notification unit 58 notifies the failure probability obtained by the calculation unit 56. Examples of notification forms include display display, printing, and voice output.

[2. Identification of factors]
The inventors of the present invention specified the factors to be used when determining the defect probability as follows.

  The inventors have conducted an extensive infrared survey. In an infrared survey, a thermal image is obtained by photographing a structure with an infrared camera, a plurality of abnormal parts are extracted from the obtained thermal image, each extracted abnormal part is observed, and a hammering test is performed. The relationship with other factors was organized. And in order to specify the factor used by a present Example, the decision tree analysis was performed using the arranged result. FIG. 2 shows the result of decision tree analysis. From this result, it is estimated that the factors affecting the thermal image and controlling the failure probability are the “surface presence / absence of cracks” on the surface of the structure, the “surface condition” of the structure, and the “service life” of the structure. The

Here, the breakdown of the result of the decision tree analysis shown in FIG. 2 will be described.
The total number of abnormal parts extracted from the thermal image was 663 (node 0). As a result of the hammering test, there were 490 abnormal parts determined to be unnecessary for repair, and 173 abnormal parts were determined to be required for repair. Node 0 is set as a root node, and the nodes are sequentially classified into lower nodes, and each node is further classified according to the necessity of repair.

  When classifying the abnormal part of node 0 by the presence or absence of cracks, there were 617 abnormal parts determined to have no cracks (node 1), and 46 abnormal parts determined to have cracks (node 2). Among the abnormal parts of node 1, there were 483 abnormal parts that were determined to be unnecessary for repair, and 134 abnormal parts that were determined to be required for repair. In addition, among the abnormal parts of node 2, there were seven abnormal parts that were determined to be unnecessary for repair, and 39 abnormal parts that were determined to be required for repair.

  If the abnormal part of node 1 is classified according to the surface state, there are 435 abnormal parts (node 3) that are determined as healthy, uneven color (no water effect), non-land surface (including steps), or repair marks, There were 182 abnormalities (node 4) that were judged as foreign matter adhesion (spacer, wood pieces), rust (including free lime), free lime, joints, rebar exposure, and discoloration (with water effect). Among the abnormal parts of node 3, there were 381 abnormal parts determined to be unnecessary for repair, and 54 abnormal parts were determined to be required for repair. In addition, among the abnormal parts of node 4, there were 102 abnormal parts determined to be unnecessary for repair, and 80 abnormal parts were determined to be required for repair.

  When the abnormal part of node 3 is classified by the service years, 131 abnormal parts are determined to be within 12 years (node 5), and 252 abnormal parts are determined to be between 12 years old and 19 years old. 52 nodes (node 7) were (node 6), and the number of abnormal parts determined to be over 19 years old. Of the abnormal parts of node 5, 127 abnormal parts were determined to be unnecessary for repair, and four abnormal parts were determined to be required for repair. Of the abnormal parts of node 6, 222 abnormal parts were determined to be unnecessary for repair, and 30 abnormal parts were determined to be required for repair. In addition, among the abnormal parts of node 7, there were 32 abnormal parts determined to be unnecessary for repair, and 20 abnormal parts were determined to be required for repair.

  When classifying the abnormal part of node 4 by the number of years of service, there are 48 abnormal parts determined to be less than 12 years (node 8), and 131 abnormal parts determined to be over 12 years old (node 9). there were. Of the abnormal parts of node 8, there were 48 abnormal parts determined to be unnecessary for repair, and 3 abnormal parts were determined to be required for repair. Of the abnormal parts of node 9, 54 abnormal parts were determined to be unnecessary for repair, and 77 abnormal parts were determined to be required for repair.

  As described above, the present inventors have inferred that the decision tree analysis dominates the three factors “crack presence / absence”, “surface condition”, and “service life”, but the decision tree analysis conditions can be changed. For example, other factors may be inferred. Other factors that can be inferred include, for example, “structure type (bridge type in the case of a bridge)”, “structure part (bridge part in the case of a bridge)”, “abnormal part shape”, “abnormality” Part size ”,“ shooting time ”, and the like.

[3. Example of failure probability calculation]
The present invention obtains the failure probability by multivariate analysis. The inventors decided to use logistic regression analysis as multivariate analysis. Then, logistic regression analysis is performed on the three factors “presence / absence of cracks”, “surface condition”, and “service life” specified in the decision tree analysis described above, and a logistic regression equation expressed by the following equation (1) is created. The numerical values shown in Tables 1 and 2 below were obtained.

Prob (event): failure probability α: constant (1.911777049)
β1X1: A term determined according to information on “presence / absence of crack” β2X2: A term determined according to information on “surface state” β3X3: A term determined according to information on “service life”.

  Β1X1, β2X2, and β3X3 in the formula (1) are determined using the numerical values in Tables 1 and 2 below. Table 1 shows the relationship between each item of each factor and a dummy variable, and Table 2 shows the regression coefficient of each item.

  As shown in Table 1, in the present embodiment, information on three factors such as “presence / absence of crack”, “surface condition”, and “service life” is classified into a plurality of items and specified. Specifically, the information on “presence / absence of cracks” is classified into two items [none, yes]. Also, the information on the “surface condition” is 10 items [sound, uneven color (no water effect), uneven surface (including step), discoloration (with water effect), foreign matter adhesion (spacer, wood chip), free lime, rust (free Lime), joints, rebar exposure, repair marks]. In addition, the “service category” information is classified into 3 items [within 12 years, over 12 years to 19 years, over 19 years]. Each information is associated with a dummy variable and the number of regressions in order to quantify each information in the form of βX.

  Β1X1 in the equation (1) is determined as follows. If it is determined that “crack presence / absence” is “none”, the dummy variable (1) of “crack presence / absence” becomes code “1” as shown in Table 1, and as a result, the dummy variable (1) corresponds to the dummy variable (1). The regression coefficient (regression coefficient of crack presence / absence (1) in Table 2) is used, and β1X1 in the equation (1) becomes “−2.665773762 × 1”. When “presence / absence of crack” is determined as “present”, the dummy variable (1) of “presence / absence of crack” is code “0” as shown in Table 1, and β1X1 in the above equation (1) is “0”. "

  Β2X2 in the equation (1) is determined as follows. When the “surface state” is determined to be “sound”, the dummy variable (1) of the “surface state” is code “1” and the dummy variables (2) to (9) are code “0” as shown in Table 1. As a result, the dummy variable (1) (regression coefficient of the surface condition (1) in the regression coefficient table 2 corresponding to) is used, and β2X2 in the equation (1) is “0.16584942 × 1”. When the “surface condition” is determined as “color unevenness (no water effect)”, as shown in Table 1, the dummy variable (2) of the “surface condition” is code “1”, the dummy variables (1), ( 3) to (9) become the code “0”, and as a result, the regression coefficient corresponding to the dummy variable (2) (regression coefficient of the surface state (2) in Table 2) is used, and β2X2 in the equation (1) is used. Becomes “0.005544404 × 1”. Since β2X2 is determined in the same manner for other items of “surface state” (“non-land surface (including steps)” to “rebar exposure”), the description thereof is omitted here. If the “surface state” is determined as “repair trace”, the dummy variables (1) to (9) of the “surface state” are all code “0” as shown in Table 1, and the equation (1) β2X2 is “0”.

  Β3X3 in the equation (1) is determined as follows. When the “service category” is determined as “˜12 years”, the dummy variable (1) of the “service category” is code “1” and the dummy variable (2) is code “0” as shown in Table 1, As a result, the regression coefficient corresponding to the dummy variable (1) (the regression coefficient of the service category (1) in Table 2) is used, and β3X3 in the equation (1) is “−3.35658201 × 1”. If the “service category” is determined as “13-19 years”, the dummy variable (2) of the “service category” is code “1” and the dummy variable (1) is code “0” as shown in Table 1. As a result, the regression coefficient corresponding to the dummy variable (2) (the regression coefficient of the service category (2) in Table 2) is used, and β3X3 in the equation (1) is “−1.0163386375 × 1”. Become. If it is determined that “service category” is “20 years or more”, dummy variables (1) and (2) of “service category” are both code “0” as shown in Table 1, and the above equation (1) Β3X3 of this becomes “0”.

  Here, an example of the calculation using the equation (1) is shown assuming a specific situation of each factor.

・ (Situation 1)
Check for cracks ... No,
Surface condition ... sound,
In-service category: up to 12 years α = 1.9117777049
β1X1 = −2.665737762
β2X2 = 0.16584942
β3X3 = −3.35658201
α + β1X1 + β2X2 + β3X3 = −3.993693304
Prob (event) = 0.019139175
As a result of the calculation, the failure probability is 0.0191139175.

・ (Situation 2)
Check for cracks ... Yes,
Surface condition: repair marks,
In-service category: 20 years or more α = 1.91177049
β1X1 = 0
β2X2 = 0
β3X3 = 0
α + β1X1 + β2X2 + β3X3 = 1.91117777049
Prob (event) = 0.871218658
As a result of the calculation, the failure probability is 0.871218658.

  As described above, in this embodiment, the information on the three factors “presence / absence of crack”, “surface condition”, and “service life” is discriminated, and the information on each factor is digitized and applied to the logistic regression equation. Find the failure probability.

  In addition to using logistic regression analysis, a neural network or multiple regression analysis may be used as multivariate analysis for determining the failure probability.

[4. Verification of defect probability 1]
[3. The verification result of the failure probability obtained in [Example of failure probability calculation] will be described. FIG. 3 shows the average values of the defect probabilities of the abnormal part that does not require repair and the abnormal part that requires repair.

  The present inventors have described [2. The 663 abnormal parts described in “Specification of Factors”, that is, 663 abnormal parts extracted from the thermal image in the actual infrared survey, are categorized according to whether or not they need repair, and each abnormal part has a defect. The probability was calculated. Then, an average value of defect probabilities of 490 abnormal parts determined to be unnecessary for repair and an average value of defect probabilities of 173 abnormal parts determined to be repaired were obtained. The result is shown in FIG.

  As shown in FIG. 3, the average value (0.65) of the defect probability of the group that needs repair is higher than the average value (0.12) of the defect probability of the group that does not need repair. From this, [3. It is considered that the defect probability obtained in the example of calculation of defect probability functions as an index indicating whether or not a defect is included in the abnormal part extracted from the thermal image.

[5. Verification of defect probability 2]
FIG. 4 shows the frequency for each defect probability.
The present inventors have described [2. The 663 abnormal parts described in “Specification of Factors”, that is, 663 abnormal parts extracted from the thermal image in the actual infrared survey, are categorized according to whether or not they need repair, and each abnormal part has a defect. The probability was calculated and the frequency of failure probability was examined. The result is shown in FIG. Furthermore, binary values of 0.3 and 0.65 were assumed as threshold values for determining whether or not a defect is included in the abnormal part. That is, two cases were assumed: a case where it is determined that a defect is included in an abnormal part with a defect probability of 0.3 or higher, and a case where it is determined that a defect is included in an abnormal part where the defect probability is 0.65 or higher.

  Since the graph of FIG. 4A shows the defect probabilities of 490 abnormal parts that do not require repair, it is ideal that this graph has all the defect probabilities below the threshold. However, in FIG. 4A, the defect probability of 92 abnormal parts is 0.3 or more. Assuming that the threshold value is 0.3, the “abnormal determination” that the 92 abnormal parts include a defect although the defect is not included occurs. Similarly, in FIG. 4A, the defect probability of 10 abnormal parts is 0.65 or more. About these 10 abnormal parts, although it does not include a defect, an “incorrect determination” that includes a defect occurs.

  Since the graph of FIG. 4B shows the defect probabilities of 173 abnormal parts that need repair, it is ideal that this graph has all the defect probabilities equal to or higher than the threshold value. In FIG.4 (b), the malfunction probability of 135 abnormal parts is 0.3 or more. Assuming that the threshold value is 0.3, these 135 abnormal parts actually contain defects, so the defect probability is considered “medium”. Similarly, in FIG. 4B, the defect probability of 52 abnormal parts is 0.65 or more. Assuming that the threshold value is 0.65, these 52 abnormal parts actually contain defects, so the defect probability is considered “medium”.

  The inventors determined the coverage ratio and the hit ratio by using the number of “false determinations” in FIG. 4A and the “target” in FIG. The results are shown in Table 3 below. Table 3 shows the hit, misjudgment, cover rate, and hit rate for the specific failure probability (0.3, 0.65) in FIG.

  Coverage ratio indicates the ratio of abnormal parts that need repair by judgment using the probability of defects, that is, the ratio of the number of abnormal parts that have a "probable" defect probability out of the total number of abnormal parts that need repair It is obtained by (number of abnormal parts whose defect probability is “successful”) / (total number of abnormal parts requiring repair). In the case of FIG.4 (b), the coverage of the defect probability 0.3 becomes 135/173 = 0.7803 ..., and is calculated | required with 78.0%. Further, the coverage rate of the defect probability 0.65 is 52/173 = 0.3005... 30.1%.

  The hit rate indicates the ratio of the number of abnormal parts whose failure probability is “target” out of the total number of abnormal parts whose failure probability is equal to or higher than a threshold. Number) / (number of abnormal parts whose failure probability is “target” + number of abnormal parts whose failure probability is “false determination”). In the case of FIGS. 4A and 4B, the hit rate of the defect probability 0.3 is 135 / (135 + 92) = 0.5947. Further, the hit ratio of the defect probability 0.65 is 52 / (52 + 10) = 0.8387.

  It is desirable that both the coverage ratio and the hit ratio are high. However, according to the results in Table 3, there is a tendency that the higher the coverage ratio, the lower the hit ratio, and conversely, the higher the hit ratio, the lower the cover ratio. Therefore, when it is determined how to treat the abnormal part according to the obtained defect probability, it is preferable to set the defect probability threshold value appropriately according to the situation at that time. For example, when priority is given to the coverage rate, the failure probability may be set low, and when priority is given to the hit rate, the failure probability may be set higher.

[6. Processing procedure]
FIG. 5 shows the processing procedure of the infrared survey according to the present invention. Here, a processing procedure using the infrared survey system shown in FIG. 1 will be described.

  Before taking a picture of a structure on site, specify the factors that affect the thermal image and control the failure probability, and the relational expression for multivariate analysis that uses this factor information to determine the failure probability. (Step S51). In the present embodiment, three factors of “presence / absence of crack”, “surface condition”, and “age of service” and a logistic regression equation represented by the above equation (1) are specified. Then, software for performing calculations using the information in Tables 1 and 2 and the logistic regression equation is stored in the storage unit 54 of the calculation device 50.

  Next, the structure is actually photographed on site. At this time, the infrared camera 10 captures the structure 3 to obtain a thermal image (step S52). Further, a portion photographed by the infrared camera 10 is photographed by a normal camera 30 to obtain an appearance image thereof.

  When photographing by the infrared camera 10 and the normal camera 30 is completed, the acquired thermal image is displayed on the display unit 22 of the thermal image output unit 20, and an abnormal part having a temperature different from that of the surrounding is extracted from the displayed thermal image ( Step S53).

  If no abnormal part is extracted in any thermal image, it is determined that there is no defect in the structure. On the other hand, when an abnormal part is extracted from any one of the thermal images, the appearance image acquired by the normal camera 30 is displayed on the display unit 42 of the appearance image output unit 40, and the abnormal part of the thermal image among the displayed appearance images. The information of the factor corresponding to is determined (step S54). Specifically, from the appearance image, information on “presence / absence of cracks” of the part corresponding to the abnormal part is determined from the two items shown in Table 1, and information on the “surface state” of the part corresponding to the abnormal part is obtained. Is determined from the 10 items shown in Table 1. Further, the information on the “service life” of the structure is determined from the three items shown in Table 1.

  When the information on each factor is determined, the determined information on each factor is input to the calculation unit 56 using the input unit 52 of the calculation device 50. Then, the calculating part 56 digitizes the inputted factor information, and obtains the defect probability in the abnormal part extracted using the numerical value and the logistic regression equation (step S55). That is, the calculation unit 56 is configured as described in [3. As described in “Problem probability calculation example”, β1X1 corresponding to “crack presence / absence” information, β2X2 corresponding to “surface condition” information, and β3X3 corresponding to “service years” state are stored. The calculation is performed by using Table 1 and Table 2 stored in the unit 54, and applying β1X1, β2X2, and β3X3 to the logistic regression equation represented by the equation (1). When the calculation is completed, the notification unit 58 outputs the calculation result to the outside in a form such as display or printing.

  In this way, in this embodiment, the abnormal part is extracted from the thermal image acquired using the infrared camera, and the defect probability Prob (event) at the abnormal part is obtained.

  Furthermore, it is possible to improve the accuracy of the logistic regression equation by performing the logistic regression analysis again using the results obtained by the series of processes as teacher data.

  Note that the processing shown in FIG. 5 can be modified as follows.

  In the process shown in FIG. 5, the imaging of the structure using the infrared camera 10 and the imaging of the structure using the normal camera 30 are performed simultaneously (step S52). However, the imaging of the structure using the normal camera 30 may be performed only on the extracted abnormal portion after extracting the abnormal portion from the thermal image, that is, after step S53.

  In the processing shown in FIG. 5, factor information is determined using an appearance image acquired by a normal camera 30 (step S54). However, instead of using an appearance image, the structure information may be discriminated by directly viewing the structure.

  In the process shown in FIG. 5, the relational expression between the factor and the multivariate analysis is specified before photographing the structure with the infrared camera (step S51). However, the identification of the relational expression between the factor and the multivariate analysis may be performed at any time before actually using the relational expression for the factor or the multivariate analysis, that is, before step S54.

[7. Usage example of defect probability]
Aged structures and extensive structures often have multiple defects. When an infrared survey is performed on such a structure, a plurality of abnormal portions are extracted from the acquired thermal image. When a plurality of abnormal parts are extracted in this way, for example, the defect probability of each abnormal part may be obtained, and a hammering inspection or repair may be performed in order from the abnormal part having the highest defect probability. At this time, it is also possible to identify an abnormal part having a high defect probability with the arithmetic device 50 shown in FIG. 1 or an apparatus associated with the arithmetic device 50, and to display position information, survey date and time, etc. in order from the abnormal part having the highest defect probability. good.

  Further, a threshold value of the defect probability may be determined, and a hammering inspection or repair may be performed on an abnormal part whose calculated defect probability is equal to or higher than the threshold value or exceeds the threshold value. At this time, it is possible to identify an abnormal part having a defect probability equal to or higher than a threshold value or exceeding the threshold value by the arithmetic device 50 shown in FIG. Anyway.

  The two forms may be combined. In other words, when a plurality of abnormal parts are extracted by setting a threshold value of a defect probability, a hammering test is performed in order from an abnormal part having a higher defect probability among abnormal parts having a calculated defect probability equal to or higher than the threshold value or exceeding the threshold value. Or repairs may be carried out.

DESCRIPTION OF SYMBOLS 1 ... Infrared investigation system 10 ... Infrared camera 20 ... Thermal image output apparatus 30 ... Camera 40 ... Appearance image output apparatus 50 ... Arithmetic apparatus 52 ... Input part 54 ... Memory | storage part 56 ... Calculation part 58 ... Output part

Claims (10)

In an infrared survey method for a structure that finds defects in the structure by extracting abnormal parts from the thermal image of the structure photographed with an infrared camera,
Identify the factors that affect the thermal image and the relational expression for multivariate analysis that calculates the probability that the abnormal part extracted from the thermal image using the information of the factor contains a defect,
Take a thermal image by photographing the structure with an infrared camera,
Using the acquired thermal image, extract the abnormal part where the temperature is different from the surroundings,
Determine the information of the factor in the extracted abnormal part,
An infrared investigation method for a structure characterized in that information on the determined factor is digitized, and the numerical value is applied to the relational expression of the multivariate analysis to obtain a probability that the extracted abnormal part includes a defect.   The infrared survey method for a structure according to claim 1, wherein a logistic regression equation is used as a relational expression for the multivariate analysis.   The infrared survey method for a structure according to claim 1 or 2, wherein the presence or absence of cracks on the surface of the structure, the surface state of the structure, and the service life of the structure are used as the factors.   The infrared inspection method for a structure according to any one of claims 1 to 3, wherein a failure countermeasure is determined using the obtained probability.   The infrared inspection method for a structure according to claim 4, wherein the probabilities in a plurality of abnormal portions are respectively determined, and it is determined that the defects are dealt with in order from the abnormal portion having the highest probability.   The infrared inspection method for a structure according to claim 4 or 5, wherein it is determined that the defect is dealt with to an abnormal part having the probability equal to or higher than a threshold value. An arithmetic device for infrared survey used in an infrared survey of a structure to find a defect in the structure by extracting an abnormal part from a thermal image of the structure photographed by an infrared camera,
A storage unit for storing a factor that affects the thermal image, and a relational expression of multivariate analysis for obtaining a probability that the abnormal part extracted from the thermal image using the information of the factor includes a defect;
An input unit for inputting information of the factor;
Obtaining information of the factor from the input unit, obtaining the relational expression of the multivariate analysis from the storage unit, and calculating a probability that the abnormal part extracted from the thermal image includes a defect,
An infra-red survey computing device comprising: a notifying unit for notifying the probability obtained by the computing unit.   The infrared survey method for a structure according to claim 7, wherein a logistic regression equation is used as a relational expression for the multivariate analysis.   9. The arithmetic unit for infrared survey according to claim 7, wherein presence / absence of cracks on the surface of the structure, surface state of the structure, and service life of the structure are used as the factors.   The calculation device for infrared investigation of a structure according to any one of claims 7 to 9, wherein an abnormal part that needs to be dealt with is identified using the obtained probability.

Images