JP7558738B2 - Information processing device, information processing method, and program - Google Patents

Description

The present invention relates to an information processing device, a control method for an information processing device, and a program, and in particular to a technology for determining the soundness of a structure.

When inspecting bridges, tunnels, and other structures, their soundness is judged using information on abnormalities such as cracks. Soundness is an index that shows the degree of soundness of a structure. Soundness can also be said to be an index that shows the degree of deterioration or damage to a structure. In conventional soundness judgments, inspection engineers would determine the soundness of a structure based on information such as the number and width of cracks, and the presence or absence of water leakage, after understanding the specified judgment criteria and judgment rules.

Generally, the soundness of a structure is judged for each part. For example, in the case of a bridge, the soundness is judged for each deck and pier. In the case of a tunnel, the soundness is judged for each construction span. Alternatively, any area, such as part of a structural member, may be used as the evaluation area. In this way, the soundness is judged for each predetermined evaluation area. In Patent Document 1, the soundness of each span of such a tunnel is automatically judged by an information processing device.

Patent No. 4279159

R. Agrawal and R. Srikant: Fast algorithms for mining association rules, In Proceedings of 20th Int. Conf. Very Large Data Bases, VLDB, pp. 487-499, 1994. Guozhu Dong, Jinyan Li, “Efficient Mining of Emerging Patterns: Discovering Trends and Differences”, KDD 1999: pp. 43-52 Guozhu Dong, Xiuzhen Zhang and Limsoon Wong, “CAEP: Classification by Aggregating Emerging Patterns,” Proceedings of The 2nd International Conference on Discovery Science, 30-42, 1999

When judging the soundness of a given evaluation area, deformations related to the evaluation area are identified, and the soundness of the evaluation area is judged based on the associated deformations. However, for deformations that cross the boundaries of the evaluation area, for example, there are multiple association methods for the evaluation area, and the result of the soundness judgment may change depending on the association method. For example, when judging the soundness based on the length of a crack, the longer the length of the crack included in the evaluation area, the lower the soundness is judged to be (not sound). In this case, if there is a long crack that crosses the boundary of the evaluation area, the result of the soundness judgment may change depending on whether the soundness is judged based on the length of the crack within the evaluation area or whether the soundness is judged including the crack length beyond the evaluation area.

In conventional technology, when automatically assessing soundness, no consideration was given to setting an association with an evaluation area for such deformations that cross the boundaries of the evaluation area. Therefore, in conventional technology, it was not possible to make an appropriate assessment of soundness after comparing and examining the soundness that may change depending on the setting of the association with the evaluation area.

The present invention was made in consideration of the above problems, and aims to provide a technology that makes it possible to set the deformations to be associated with evaluation areas when assessing the soundness of a structure, thereby obtaining more appropriate assessment results of the soundness.

In order to solve the above problems, the information processing device of the present invention comprises:
A setting means for selecting an association rule for associating a deformation used for judging the soundness of a structure with the evaluation area to be judged for the soundness from among a plurality of association rules, and setting the selected association rule as the association rule to be used for the judgment;
A determination means for determining a soundness of the evaluation area based on a defect associated with the evaluation area according to the association rule set by the setting means;
The present invention is characterized by having the following.

According to the present invention, when assessing the soundness of a structure, a more appropriate assessment result of the soundness can be obtained by setting the deformation to be associated with the evaluation area.

FIG. 1 is a diagram illustrating a hardware configuration of an information processing device according to an embodiment of the present invention. FIG. 2 is a block diagram showing a functional configuration of the information processing device. 2 is a process flowchart of the present embodiment. 2 is a diagram illustrating data handled by the information processing device according to the embodiment. FIG. FIG. 13 is a diagram for explaining learning of a soundness determination method. FIG. 13 is a diagram illustrating a method for determining the soundness. 4 is a diagram for explaining the basis of judgment output by a healthiness judging unit; FIG. FIG. 4 is a diagram illustrating an example of display information according to the first embodiment. 10A to 10C are diagrams illustrating another example of display information according to the first embodiment. FIG. 11 is a diagram illustrating an example of an association criterion change according to the first embodiment. 13A and 13B are diagrams illustrating another example of the association criterion change according to the first embodiment. 13A and 13B are diagrams illustrating another example of the association criterion change according to the first embodiment. FIG. 11 is a diagram illustrating a determination rule according to the second embodiment. 13A and 13B are diagrams for explaining learning of a soundness determination method according to a third embodiment. FIG. 13 is a diagram illustrating a change in association criteria according to the fourth embodiment. FIG. 13 is a diagram illustrating a change in association criteria according to the fifth embodiment. FIG. 13 is a diagram illustrating a configuration of an information processing apparatus according to a sixth embodiment.

<Embodiment 1>
Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, a processing operation related to an information processing method by an information processing device 100 for determining the soundness of a structure will be described.

First, the data used in the soundness determination by the information processing device 100 of this embodiment will be described with reference to FIG. 4. FIG. 4 shows data showing a portion of an unfolded view of a tunnel to be inspected. FIG. 4 shows an example of a tunnel as the inspection object, but the inspection object of this embodiment is not limited to a tunnel and may be other structures such as a bridge, a dam, or a building.

4A is a display example of a tunnel development diagram by the information processing device 100 according to the present embodiment. The tunnel development diagram is data in which a wall image of a tunnel wall is superimposed on a drawing of the tunnel development. That is, in this embodiment, the tunnel development diagram is data in which at least a part of a drawing of the inner wall part (curved surface) of the tunnel, which is a structure, and/or a photographed image is cut out and arranged on a two-dimensional coordinate system. In addition, the tunnel development diagram of this embodiment is data in which a superimposed image in which a design drawing of the tunnel wall surface and a photographed image of the tunnel inner wall surface are superimposed is developed on a plane with the vertical axis in the circumferential direction of the tunnel and the horizontal axis in the extension direction of the tunnel. An example of a method for creating a tunnel development diagram will be briefly described. First, multiple photographing devices are installed on a vehicle, and the tunnel wall surface is comprehensively photographed while the vehicle is moving. The images thus photographed are arranged on a two-dimensional coordinate system, and a wall image is created by linking them based on matching of the features. Then, the wall image and the design drawing of the developed tunnel wall surface are aligned and superimposed to create a tunnel development diagram. The alignment in this case may be performed based on input from the user, or may be performed using a matching process based on image processing. Note that, although the data used in the soundness assessment by the information processing device 100 is data of a tunnel development diagram in which a drawing of the structure and a photographed image of the structure are superimposed, this is not limiting, and data showing only either the drawing or the photographed image may be used.

Using the example of a tunnel, where the target surface of the structure to be assessed for soundness is a curved surface, the use of a development diagram of the curved surface has been described as data to be used in the soundness assessment. However, if the target surface is a flat surface in three-dimensional space, it is not necessary to use a development diagram. In this case, a drawing of the target surface (flat surface) of the structure to be assessed for soundness and/or an image of the target surface may be used as data to be used in the soundness assessment. Also, if it is not possible to capture an image so that the target surface (flat surface) of the structure and the imaging surface of the imaging device are parallel, a tilt-corrected image may be used as data to be used in the soundness assessment.

Furthermore, the tunnel development diagram shown in FIG. 4 is recorded in association with deformation information showing deformation of the tunnel. For example, FIG. 4(A) shows a state in which objects drawn based on the deformation information recorded in association with the tunnel development diagram, such as cracks 401, 402, 403, water leakage 411, and efflorescence 412, 413, are also displayed. In other words, in FIG. 4(A), cracks 401, 402, 403, water leakage 411, and efflorescence 412, 413, which are objects for identifying the deformation based on the deformation information, are further superimposed and displayed on the tunnel development diagram.

For the crack 402, an object drawn as a thick line is superimposed on the position where the crack was detected on the tunnel development diagram. The object for identifying the deformation is drawn according to the attributes of the detected deformation. For example, the object showing the crack may be drawn with a different color or line width according to the width of the crack. In FIG. 4, the crack 402 is drawn with a thicker line than the crack 401, which indicates that the crack width of the crack 402 is wider than that of the crack 401. In this embodiment, it is assumed that information showing the crack width such as 0.2 mm width or 0.5 mm width is recorded in the deformation information showing each crack. The crack width is recorded as the maximum crack width value for one crack. The crack width stored as the deformation information is not limited to the maximum value, but may be an average value, a minimum value, or multiple values. The type of deformation is not limited to these examples and may include exposed rebar, junk, rust, etc.

The information processing device 100 creates this abnormality information based on input from the user regarding abnormalities observed by the user at the site of the structure to be inspected. Alternatively, the user may visually check an image of the wall surface of the structure superimposed on the drawing as shown in Figure 4, and input the location and range of each abnormality. Furthermore, the information processing device 100 may automatically detect abnormalities based on the images, and use the results as abnormality information. Automatic detection of abnormalities can be achieved, for example, by applying the detection target to a detection model that has been machine-learned in advance using existing images of abnormalities.

In this embodiment, the information processing device 100 stores the position and range of deformation information indicating cracks as linear vector data. In addition, for deformation information represented by an area such as a water leak, the area is stored as polygon data. Hereinafter, these vector data and polygon data will be collectively referred to as vector data. This vector data is data that represents the coordinates of each point in a coordinate system based on the drawing.

Table 420 in Figure 4 (B) shows an example of deformation information. For example, deformation information indicating a crack with ID C401 includes vector data in the drawing coordinate system in the position field, and further includes crack width information in the width field. Also, the size field includes crack length information. Note that for deformations in areas other than cracks, the size field indicates the area of the deformation.

The deformation information may also include information indicating deformations that are composed of a combination of multiple deformations. For example, the dotted frame 422 in FIG. 4 is an object drawn based on deformation information indicating a closed crack composed of multiple cracks. H422 in the deformation information table 420 indicates this closed crack, and the position item indicates coordinate information indicating the dotted frame 422. Deformation information composed of multiple cracks includes, in addition to closed cracks, tortoiseshell cracks and lattice cracks.

Furthermore, information on the degree of damage to a specific part, such as a structural member of a structure, may also be included in the abnormality information. The degree of damage is an index that indicates the degree of damage to an object. For example, 431 in FIG. 4 indicates a rock bolt of a tunnel in a wall image, and the result of evaluating the degree of damage to this rock bolt may also be included in the abnormality information. L431 in table 420 is abnormality information indicating damage to a rock bolt, and the coordinates of range 432 indicating rock bolt 431 are recorded in the position field. Note that range 432 in FIG. 4(A) is an object drawn based on the abnormality information indicating damage to a rock bolt. In the damage degree field of L431 in table 420, the damage degree of rock bolt 431 is recorded as C. The damage degree is expressed, for example, in three stages from A to C, with A being the most severe damage.

The damage level may be evaluated by a human being by checking the object on-site and judging the damage level, or by a human being by visually checking and judging the image of the object. The information processing device 100 may automatically judge the damage level from the image of the object. In this case, a judgment model for automatically judging the damage level from the image may be generated by machine learning, and the information processing device 100 may judge the damage level of the judgment object using this judgment model. The damage level judgment model can be constructed by preparing a large amount of data of a set of an image of the damage level judgment object and a teacher label using existing data in advance, and learning this as learning data. The damage level judgment object image is, for example, an image of a rock bolt cut out from a wall image with a size indicated by the range 432 indicating the rock bolt in FIG. 4. The result of a human judging the damage level for such an image is used as the teacher label. Learning can be performed using any machine learning algorithm such as Deep Learning.

As described above, the deformation information includes information on crack width, deformation size, and damage level. In what follows, all information other than the position information will be collectively referred to as attribute information.

As described above, the abnormality information in this embodiment is information such as that shown in table 420 in FIG. 4(b). As will be described later, this abnormality information is stored in the information storage unit 206.

Next, the soundness level determined in this embodiment will be explained. The soundness level is determined by using the deformation information described above to determine the soundness of the structure. Note that, as described below, the soundness level may also be determined using the structure's specification information. In this embodiment, the soundness level is determined for each predetermined range of the structure. For example, in the tunnel example of FIG. 4, the soundness level is determined for each construction span (hereinafter referred to as span) of the tunnel. Specifically, the soundness level is determined for each of spans A, B, and C. In the following, such an area for determining the soundness level is called an evaluation area. Note that the evaluation area is not limited to the construction span, and a predetermined unit area such as every 10 m may be used as the evaluation area. In addition, the soundness level is determined on a five-level scale from 1 to 5. The soundness level is set to 5 as the soundest and 1 as the most severely damaged state of the structure. Note that the soundness level is not limited to 5 levels.

The following describes the configuration and processing of the information processing device 100 according to this embodiment, taking as an example a case where a soundness determination is performed using the abnormality information in Figure 4.

The configuration of the information processing device 100 in this embodiment will be described using Figures 1 and 2.

(Hardware configuration)
FIG. 1 is a hardware configuration diagram of an information processing device 100 according to this embodiment. As shown in FIG. 1, the information processing device 100 has a CPU 101, a ROM 102, a RAM 103, a HDD 104, a display unit 105, an operation unit 106, and a communication unit 107. The CPU 101 is a central processing unit (Central Processing Unit) that performs calculations and logical judgments for various processes and controls each component connected to a system bus 108. The ROM (Read-Only Memory) 102 is a program memory that stores programs for control by the CPU 101, including various processing procedures described later. The RAM (Random Access Memory) 103 is used as a temporary storage area such as the main memory and work area of the CPU 101. The program memory may be realized by loading a program into the RAM 103 from an external storage device connected to the information processing device 100.

HDD 104 is a hard disk for storing electronic data and programs according to this embodiment. An external storage device may also be used to fulfill a similar role. Here, the external storage device can be realized, for example, by a medium (recording medium) and an external storage drive for realizing access to the medium. Known examples of such media include flexible disks (FDs), CD-ROMs, DVDs, USB memories, MOs, and flash memories. The external storage device may also be a server device connected via a network.

The display unit 105 is, for example, a CRT display or a liquid crystal display, and is a device that outputs images to a display screen. The display unit 105 outputs video according to display control by the CPU 101. The display unit 105 may be an external device connected to the information processing device 100 by wire or wirelessly. The operation unit 106 includes a keyboard, a mouse, etc., and accepts various operations by the user. The communication unit 107 performs two-way communication by wire or wirelessly with other information processing devices, communication devices, external storage devices, etc., using known communication technology.

Note that the information processing device 100 is described as a single device, but is not limited to this and may be multiple devices. Furthermore, the information processing device 100 may be a logical device virtualized by multiple devices.

(Functional configuration)
FIG. 2 is an example of a block diagram showing the functional configuration of the information processing device 100 according to this embodiment. The information processing device 100 has an evaluation area management unit 201, a soundness determination unit 202, a deformation information analysis unit 203, a display information processing unit 204, a setting unit 205, a deformation information storage unit 206, a specification information storage unit 207, and a learning unit 208. Each of these functional units is realized by the CPU 101 expanding a program stored in the ROM 102 into the RAM 103 and executing processing according to each flowchart described later. Then, the execution results of each processing are stored in the RAM 103 or the HDD 104. In addition, for example, when configuring hardware as an alternative to software processing using the CPU 101, a calculation unit or a circuit corresponding to the processing of each functional unit described here may be configured. Each functional unit may be configured, for example, with an ASIC or FPGA.

Next, an overview of each functional unit in FIG. 2 will be described. The deformation information storage unit 206 stores the deformation information described in FIG. 4. The specification information storage unit 207 stores the specification information of the structure to be inspected. The specification information is various information about the structure, and includes, for example, the number of years since the start of service (years of service), the construction method, the type of concrete, geographical conditions such as coastal or cold region, and environmental conditions for use such as traffic volume. The evaluation area management unit 201 associates deformation information with the evaluation area of soundness based on a predetermined association criterion (association rule). The soundness determination unit 202 uses the deformation information and specification information associated by the evaluation area management unit 201 to determine the soundness of the evaluation area. The soundness determination unit 202 also outputs the basis for the soundness determination. The deformation information analysis unit 203 analyzes the deformation included in the basis for the soundness determination. The display information processing unit 204 displays information on abnormalities related to the soundness assessment on the display unit 105 based on the analysis results of the abnormality information analysis unit 203. The setting unit 205 accepts user input via the operation unit 106 and controls the criteria for associating abnormalities with evaluation areas. Furthermore, the learning unit 208 learns the soundness assessment method of the soundness assessment unit 202. Note that it is not necessary for all of these functional units to be included in the information processing device 100, and at least some of the functions may be implemented by an external device that can be connected to the information processing device 100. Furthermore, each functional configuration may be realized by distributed processing by multiple devices.

(process)
Next, the process according to this embodiment will be described. The process according to this embodiment will be described below with reference to the flowchart shown in FIG. 3. The flowchart shown in FIG. 3 is a process in a soundness determination application started by the information processing device 100. When a structure to be determined as a target for determining soundness is selected based on a user instruction in the soundness determination application, the flowchart shown in FIG. 3 is started. When a target structure is selected, the information processing device 100 reads a superimposed image of the drawing of the target structure and the photographed image and the abnormality information in the soundness determination application. In the following, the tunnel shown in FIG. 4 is selected as a target for determining soundness, and the tunnel development diagram and the abnormality information shown in FIG. 4 are read in the soundness determination application.

(Processing related to the evaluation area management unit)
In FIG. 3, first, steps S301 to S303 are processes executed by the evaluation area management unit 201. Step S301 is a step of setting an association criterion. The association criterion is information indicating a method of setting the abnormality information to be used for the evaluation area to determine the soundness. Specifically, the association criterion is a judgment criterion such as whether or not the abnormality existing at the boundary of the evaluation area is used for the soundness judgment. The association criterion can be changed by the setting unit 205, and the details will be described later. Here, in step S301, the initial association criterion is set as the association criterion for performing the subsequent association process. The initial setting of the association criterion in this embodiment is set to "use the abnormality information within the evaluation area range of the evaluation target for the soundness judgment".

In this embodiment, a soundness assessment is performed sequentially for each evaluation area of the inspected structure. Step S302 is a step for selecting the evaluation area to be the current target for soundness assessment. In the following, the explanation will be given assuming that span A shown in FIG. 4(A) is selected as the evaluation area for soundness assessment in step S302.

In the next step S303, the evaluation area management unit 201 associates the abnormality information with the evaluation area according to the association criteria. That is, in step S303, the information processing device 100 selects the abnormality information to be used when determining the soundness of the evaluation area according to the association criteria. This process will be described with reference to FIG. 4.

As mentioned above, the explanation will be given assuming that the initial association criterion of "deformation information within the evaluation area range of the evaluation target is used for soundness assessment" was set in step S301. The evaluation area management unit 201 selects the deformation information included in the coordinate range of span A from the deformation information storage unit 206 according to the set association criterion. For example, the range of span A in Figure 4 includes deformations such as cracks 401 and 402, water leakage 411, and efflorescence 412. In step S303, the deformation information of these deformations is set as the deformation information to be used for soundness assessment of span A.

Here, crack 403 in Figure 4 is a crack that spans span A and span C. For this crack 403, according to the current association criteria, the range 421 of crack 403 that is included in span A is set as the abnormality information to be used in determining the soundness of span A. Therefore, the evaluation area management unit 201 changes the abnormality information indicating crack 403 to abnormality information indicating a crack up to the boundary of span A.

Specifically, the evaluation area management unit 201 acquires the deformation information of the crack 403 from the deformation information storage unit 206, and changes the vector data to the range 421 included in span A only. Furthermore, the size of the attribute information (crack length) is also changed to the length of the range 421. The deformation information thus changed according to the association criteria for the evaluation area is added to the deformation information for determining the soundness of span A. Note that the evaluation area management unit 201 may convert the deformation information indicating the crack 403 into deformation information indicating three cracks that exist in span A, span B, and span C, respectively.

By the above processing of the evaluation area management unit 201, the abnormality information for determining the soundness of span A is set.

(Processing related to the learning unit)
In the next step S304 of Figure 3, the information processing device 100 determines the soundness of the evaluation area using the deformation information associated with the evaluation area in the evaluation area management unit 201 and the specification information stored in the specification information storage unit 207.

First, a method for determining the health level will be described. In this embodiment, the health level is determined using a trained model that has been machine-learned. This learning is performed in advance by the learning unit 208.

Here, in order to show the user the basis for the soundness determination and to show information on abnormalities related to the soundness determination, it is preferable that the soundness determination method is composed of judgment criteria that can be understood by humans, and for example, the judgment criteria are preferably binary judgments. Among machine learning algorithms, a decision tree, which is a collection of binary judgments, is an algorithm suitable for this embodiment. Below, an embodiment in which a decision tree is used in the soundness determination method is described. Note that the soundness determination method is not limited to a decision tree, and soundness determination may be performed using a trained model that uses other machine learning algorithms.

Regarding the method of learning a judgment model for soundness judgment, first, learning data will be described. Learning data for soundness judgment is created from abnormality information such as existing inspection data and data with known soundness. A certain piece of learning data _Di is expressed as follows:

D _i is data related to one evaluation area of a certain structure. X _i is deformation information associated with a certain evaluation area and specification information of a certain structure. y _i is soundness data determined by a human using the information of X _i , and indicates a soundness of 1 to 5 in this embodiment. In the learning data, this soundness y _i indicates the teacher class label. A large amount of such learning data D _i is collected to create a learning data set D = {D _i }.

Next, a method for learning a decision tree for determining soundness will be explained with reference to FIG. 5(A). A decision tree is composed of branching nodes 501, 502 and a terminal node (leaf node) 503. Branching nodes 501, 502 determine the branching direction based on a predetermined determination criterion. In FIG. 5(A), if the determination is True, proceed to the node on the right, and if the determination is False, proceed to the node on the left.

Next, the judgment criteria of the branch node will be described. The judgment criteria are a binary judgment using the deformation information and the specification information. Simply, the presence or absence of deformation can be the judgment criteria, such as "Are there cracks in the evaluation area?", "Are there water leaks in the evaluation area?", and "Are there closed cracks?", using the deformation information. In addition, a criterion combining the deformation type and the attribute information can be the judgment criterion. For example, in the judgment criterion using the crack width of the attribute information, judgment criteria such as "Are there cracks with a crack width of 0.5 mm or more present?" and "Are there cracks with a crack width of 1.0 mm or more present?" can be created. In other words, the presence or absence of deformation, the size or width of the deformation can be the judgment criterion of the soundness. At this time, by setting a threshold value such as a width of 0.5 mm or a width of 1.0 mm for the attribute information, multiple different judgment criteria can be created. In addition, in the judgment criterion using the deformation size of the attribute information, judgment criteria such as "Are there cracks with a length of 5 m or more present?" and "Are there water leaks of 1 ^m2 or more present?" can be created. In other words, the length or area of the deformation can be the judgment criterion of the soundness. Furthermore, judgment criteria using the damage level of the attribute information can be created such as "Is bolt damage level A present?" or "Is bolt damage level B present?" In other words, the damage level can be the judgment criterion for soundness. The number of abnormalities may also be the judgment criterion for soundness. Note that in this embodiment, the judgment criteria are set so that a structure is judged to be more unsound when there is an abnormality that satisfies the judgment criterion compared to when there is no abnormality that satisfies the judgment criterion.

In this way, judgment criteria using deformation information can be created based on the presence or absence of a type of deformation, the type of deformation, and attribute information (such as the size of the deformation). Judgment criteria can also be created by combining these. For example, judgment criteria such as "Are there cracks with a width of 0.5 mm or more and efflorescence present?" can be created.

Judgment criteria using specification information can be created in the same way as judgment criteria using deformation information. For example, judgment criteria can be created such as "Has it been in service for more than 30 years?", "Has it been in service for more than 50 years?", "Located in a coastal area?", "Located in a cold region?", "Is it precast concrete?", etc. Judgment criteria can also be created that combine deformation information and specification information. For example, judgment criteria can be created such as "Has it been in service for more than 30 years and does it have cracks wider than 0.5 mm?".

As described above, various patterns of judgment criteria can be created. In decision tree learning, from these various judgment criteria, a judgment criterion that is effective for separating the training data is selected and set for each branching node. Below, as an example of a decision tree learning method, we will explain a learning method in which judgment criteria for each branching node are set from judgment criterion candidates prepared in advance by a human.

Table 500 in FIG. 5(A) shows n candidate criterion values C1 to Cn that have been prepared in advance by a human. First, the process of learning the criterion value for the first branch node 501 will be described. In the first step of learning branch node 501, a binary decision is made for learning data set D using n candidate criterion values C1 to Cn to classify the data. In the second step, the candidate criterion value that most favorably classified the learning data is selected from the data classification results using the n candidate criterion values, and this is set as the criterion value for node 501. Here, "classifying the learning data most favorably" refers to a state in which the soundness of the learning data set after division is most suitable for classification. A specific example of this will be described using FIG. 5(B).

In FIG. 5B, B1 is a state where the branch node 501 is set to the criterion candidate C1 "Are there any cracks with a width of 0.5 mm or more?", and B2 is a state where the branch node 501 is set to the criterion candidate C4 "Are there any leaks of 1 ^m2 or more?". The histogram 510 is a histogram of the soundness included in the learning data set D. In this example, it is assumed that the same number of learning data for soundness levels 1 to 5 are prepared. In B1, the learning data determined to be False by the criterion candidate C1 is classified into the left node, and the learning data determined to be True is classified into the right node. Furthermore, histograms 511 and 512 of the soundness of the data set classified into each node are shown. B2 also shows a state where the learning data set is classified by the criterion candidate C4, and histograms 513 and 514 of the soundness of each node are shown. Here, the histograms 513 and 514 of the nodes in B2 show that the soundness is not sufficiently classified.

On the other hand, in histograms 511 and 512 of B1, the health levels are classified with high accuracy. Specifically, histogram 511 shows that the learning data classified in the left node contains only data with health levels of 4 and 5, while the learning data classified in the right node contains only data with health levels of 1, 2, and 3. In such a case, candidate criterion C1 is a better criterion for classifying the learning data than candidate criterion C4. By making such a comparison for the learning data classification results using each candidate criterion, the criterion that provides the best classification is set as the criterion for the node.

The basis for this comparison is information gain. Information gain IG is expressed by the following formula:

Here, D is a data set, the subscript p is a parent node, the subscripts left and right indicate the left and right child nodes, respectively, _IG is Gini impurity, and N is the number of data in the data set. Since a larger value of the information gain IG indicates that the data is better classified, the information gain IG is calculated for the classification results using each criterion candidate, and the criterion candidate with the highest information gain IG is selected as the criterion for the node.

In addition, since the information gain IG indicates the judgment performance of each node, it can be used to calculate the importance of the judgment criteria, which will be described later. Therefore, the information gain IG of each node calculated during learning of the decision tree is recorded as information for each node, together with the judgment criteria.

In this way, the criterion to be set in node 501 can be determined. In the learning of the next node, the criterion that best classifies the data is selected for the learning data set classified by the higher-level parent node, in the same manner as described above. By repeating this process, the learning of the decision tree can proceed.

As a method for terminating the learning of a decision tree, there is a method of repeating classification until the node contains only one type of health level. However, since a decision tree learned in this way generally has a tendency to overlearn, it is preferable to perform node learning up to a preset depth (number of stages). The example of FIG. 5(A) shows a situation in which learning is terminated at depth 3. As a result, a terminal node (leaf node) is created, for example, as shown in node 503 in FIG. 5(A). FIG. 5(A) shows a health level histogram 505 of the learning data set contained in terminal node 503. Since histogram 505 contains multiple health levels, the health level that is most frequently included is selected here and is set as the health level indicated by the terminal node. In FIG. 5(A), the health level indicated by terminal node 503 is health level 5.

(Processing related to soundness determination unit)
In step S304 in Fig. 3, the soundness of the data to be judged is judged using the soundness judgment model learned as described above. Fig. 6 is a diagram for explaining the soundness judgment process. First, the deformation information and specification information of the evaluation area are input as input. Then, branching to child nodes is performed using the judgment criteria of each node. As a result, it is assumed that the terminal node 601 has been reached in Fig. 6. Also, it is assumed that the soundness indicated by the terminal node 601 is soundness level 3. As a result, the soundness of the input data is judged to be soundness level 3.

In the above description, a method of learning a decision tree has been described in which criterion candidates are prepared in advance and the criterion for each branch node is set from the criterion candidates. The method of learning the criterion for a node is not limited to this and may be other methods. For example, when learning each node, multiple criterion candidates may be randomly generated and the criterion with the best classification performance may be selected from these criterion candidates. As a specific example of a method of randomly creating criterion candidates, first, the type of deformation and the type of specification information are randomly selected. In addition, the threshold value of the attribute information is also randomly selected, and a criterion candidate is created by combining the type of deformation and the type of specification information with the attribute information threshold value. By repeating the creation of criterion candidates by such random selection multiple times, multiple criterion candidates are created. In addition, these criterion candidates may be combined to create a criterion candidate consisting of a combination of multiple deformation information and specification information, such as C7 and C8 in FIG. 5.

Furthermore, the soundness determination unit 202 extracts a determination criterion related to the soundness determination from the determination criteria of the soundness determination method, and outputs it as the determination basis. This process corresponds to step S305 in FIG. 3. In this embodiment, the determination criterion of the node in the decision tree where the determination is True is used as the determination basis. For example, in FIG. 6, the determination is True at node 502. At this time, it is assumed that the determination criterion of node 502 is "Total crack length 10 m or more?" In this case, this "Total crack length 10 m or more?" becomes the determination basis. In the simple example of FIG. 6, there is one determination basis, but if multiple nodes in the decision tree are determined to be True, the soundness determination unit 202 outputs multiple determination reasons. Note that an example has been shown in which the determination criterion set to True at a node in the decision tree is used as the determination basis, but this is not limited to this. For example, the determination criterion set to False at a node in the decision tree may be used as the determination basis.

(Processing related to the Deformation Information Analysis Unit)
Next, the process of step S306 in Fig. 3 will be described. Step S306 is a process executed by the deformation information analysis unit 203, which analyzes deformations related to the soundness determination based on the determination basis output by the soundness determination unit 202. One specific process of analyzing deformations is to identify the type of deformation included in the determination basis. In the example of Fig. 6, the soundness determination unit 202 outputs "total crack length 10m or more?" as the determination basis. In this case, the deformation information analysis unit 203 identifies the deformation that was used as the basis for the soundness determination of the target evaluation area.

Figure 7 shows information output by the soundness determination unit 202 as the basis for the soundness determination of tunnel span A in Figure 4. This example will be used to further explain an embodiment for identifying anomalies related to the soundness determination, and also to explain the calculation of importance.

Figure 7 shows the result when span A is judged to have a soundness level of 3, and three judgment criteria are extracted as the judgment basis. Figure 7 shows a case where, when the deformation information and judgment basis of span A are input to the decision tree, the input data is judged to be True at nodes N502, N511, and N533 of the decision tree. In other words, span A is judged to have a soundness level of 3 based on the judgment criteria of nodes N502, N511, and N533. The soundness judgment unit 202 uses these judgment criteria as the judgment basis. Figure 7 shows the judgment basis obtained in this way and information related to the judgment basis. The deformation information analysis unit 203 regards the deformation type included in the judgment basis as a deformation related to the soundness judgment. From the judgment basis in Figure 7, cracks and water leakage are identified as deformations related to the soundness judgment, as listed in the related deformation type item.

Next, the process of deriving the importance of deformations related to the soundness assessment will be described. In the example of Figure 7, as described above, the information gain at the time of learning is recorded in each node of the decision tree. In Figure 7, the information gains of nodes N502, N511, and N533 of the decision tree are shown. This information gain is tallied for each type of deformation related to the soundness assessment to calculate the importance of each type of deformation. Note that the importance is an index that indicates the degree of contribution of the target deformation to the soundness assessment result. In the example of Figure 7, the node related to cracks is N502, so the importance of cracks is 0.32. Furthermore, the nodes related to water leakage are N511 and N533, and the sum of these information gains gives the importance of water leakage of 0.26.

The above describes an embodiment in which the importance of each type of deformation is calculated based on the information gain recorded in the nodes of the decision tree. The method of calculating the importance of each type of deformation is not limited to this, and other methods may be used. For example, the number of times a type of deformation appears in the judgment basis may be used as the importance. If the number of times a type of deformation appears in the judgment basis is used as the importance, in the example of Figure 7, cracks only appear in the judgment basis of node N502, so the importance of cracks is 1. Water leakage appears in nodes N511 and N533, so the importance is 2.

The deformation information analysis unit 203 also acquires the position information of each deformation related to the soundness judgment. Specifically, the deformation used as the judgment basis is identified from the deformation in the evaluation area, and its position is acquired from the deformation information storage unit 206. For example, in FIG. 7, the judgment basis of node N502 is "Are there any cracks with a crack width of 0.5 mm or more?", so cracks with a crack width of 0.5 mm or more are extracted from the deformation in the evaluation area. Then, position information (vector data) of cracks with a crack width of 0.5 mm or more is acquired from the deformation information storage unit 206. In addition, the judgment basis of node N511 is "Is there any leakage?", so the leakage is extracted from the deformation in the evaluation area, and the position information of the leakage is acquired from the deformation information storage unit 206. Note that the judgment basis of N533 is "Has the number of years elapsed been 30 years or more and has a leakage of 1 ^m2 or more?", and since the position information of the leakage has already been acquired, the process of acquiring the position information for this judgment basis is omitted.

As described above, in step S306, the deformation information analysis unit 203 identifies deformations related to the basis for the judgment, acquires their location information, and calculates the importance of each deformation. The deformation information analysis unit 203 also identifies deformations that satisfy the judgment criteria that served as the basis for the judgment. In the following, the deformation information analysis unit 203 uses this information created by the deformation information analysis unit 203 to perform processing to visualize deformations related to the soundness judgment.

In the next step S307 in FIG. 3, if the soundness assessment process for all evaluation areas of the inspected structure has been completed, the process proceeds to step S308. If the soundness assessment process for all evaluation areas has not been completed, the process returns to step S302, another evaluation area is selected, and the process of steps S303 to S306 described above is executed.

(Processing related to the display information processing unit)
Step S308 in Fig. 3 is a process executed by the display information processing unit 204. The display information processing unit 204 creates and outputs display information indicating the health assessment result and information on abnormalities related to the health assessment, and causes the display information to be displayed on the display unit 105. That is, the display information processing unit 204 generates and outputs information on abnormalities that satisfy the assessment criteria that served as the basis for the assessment. Fig. 8 is an example of a display screen displayed in the health assessment application. Fig. 8 is an example of information created by the display information processing unit 204 and displayed on the display unit 105. Below, the display information processing unit 204 will be explained using this image.

The display window 800 in FIG. 8 shows a GUI application window. The display window 800 also has an image display area 801 and an importance display area 802. First, the image display area 801 displays the information described in FIG. 4. Specifically, the structure drawing, the image of the structure superimposed on the drawing, and the abnormality information are displayed. The abnormality information is superimposed on the image as polylines and polygons based on the vector data of the position information. In addition, the soundness assessment results are displayed for each evaluation area; for example, 811 displays the soundness assessment results for span A.

In the image display area 801, based on the information created by the deformation information analysis unit 203, deformations related to the soundness judgment are highlighted. For example, cracks 822 and 823 are cracks with a crack width of 0.5 mm or more, and are deformations related to the soundness judgment, so they are highlighted. On the other hand, crack 821 is a crack with a crack width of less than 0.5 mm, and is a deformation that was not selected as the basis for the soundness judgment, so it is not highlighted. In FIG. 8, as an example of highlighting, cracks 822 and 823 are displayed with thicker polylines than crack 821 that was not selected as the basis for the soundness judgment. The highlighting method is not limited to this, and other display methods may be used. For example, the deformation may be displayed in a noticeable color such as red, or may be displayed in a flashing manner. As another example of highlighting, the deformation that was the basis for the soundness judgment may be surrounded by a circle or a specified object.

Similarly, water leakage 824 is also displayed with a thick outline as an abnormality related to the soundness assessment. On the other hand, efflorescence 825 is an abnormality that was not selected as a basis for the soundness assessment, so the outline is not highlighted. Note that an abnormality related to the soundness assessment may be extracted and further displayed in a separately provided display area that makes it possible to identify that it was the basis for the assessment.

In addition, if the abnormality related to the soundness assessment is a combination of multiple abnormalities or the degree of damage to a specific part of a structural member of a structure, the area may be highlighted. For example, span B in Figure 8 has been judged to have a soundness level of 2, and the basis for this judgment is a closed crack. In Figure 8, the area 826 surrounding the closed crack is highlighted with a thick line. Although not shown in Figure 8, if the degree of damage to a rock bolt is related to the soundness assessment, for example, the rock bolt part may be highlighted.

In addition, information on the basis of the determination may be displayed for the abnormality related to the soundness determination in the image display area 801. For example, speech bubble 830 indicates that crack 822 is related to the soundness determination because it satisfies the determination criterion "Are there any cracks with a crack width of 0.5 mm or more?". The display position of speech bubble 830 can be determined based on the position information of the abnormality obtained by the abnormality information analysis unit 203.

Span C has a soundness level of 5, which means that no unhealthy abnormalities are present and no abnormalities related to the soundness assessment exist. Therefore, no abnormalities related to the soundness assessment are highlighted in the assessment area for span C.

Next, the display contents of the importance display area 802 will be described. In this embodiment, the importance of the deformation type is displayed for each evaluation area. The importance display area 802 in FIG. 8 displays the importance of the deformation type with respect to the soundness assessment of span A. The user can switch the evaluation area for viewing the importance by, for example, operating the pull-down menu 804. The importance is calculated by the deformation information analysis unit 203 as described above, and is shown in FIG. 8 by an importance bar graph 803. This display shows that the deformation types involved in the soundness assessment of span A are cracks and water leaks, and that cracks in particular are strongly related to the assessment.

In addition, the importance information may be used to control the degree of emphasis when highlighting abnormalities in the image display area 801. For example, the more important the abnormality, the more emphasized it may be, for example, by making the displayed lines thicker depending on the importance of the abnormality. Specifically, in the example of FIG. 8, cracks are more important than water leaks, so the lines of cracks 822 and 823 are made thicker than the outline of water leak 824, resulting in a relatively emphasized display.

As described above, by displaying the soundness assessment result and information on abnormalities related to the soundness assessment as shown in FIG. 8, the user can easily understand the abnormalities related to the soundness assessment. This makes it easier for the user to check the soundness automatically assessed by the information processing device 100. In particular, by showing the location and importance of the abnormalities related to the soundness assessment rather than simply showing the basis for the assessment (assessment criteria), the relationship between the assessment result and the abnormalities can be easily understood.

Next, another example of a method for displaying abnormalities related to soundness assessment will be explained with reference to FIG. 9. In FIG. 9, the abnormalities that are the basis for the soundness assessment are displayed in the form of an inspection report based on the information created by the abnormality information analysis unit 203. In an inspection report, images of important abnormalities and their information are generally recorded. An abnormality related to soundness assessment is an important abnormality in terms of inspection, and by automatically organizing this in the form of an inspection report, the user's document creation work can be simplified.

In the display window 900 in Figure 9, the health of span A is 3, and information on abnormalities related to this health assessment is displayed. In Figure 9, images of two abnormalities, damage numbers 1 and 2, and information on the abnormalities are displayed. Note that in Figure 9, a page forwarding function 901 is provided, and multiple abnormalities are organized in the form of an inspection report. Here, when the abnormalities are arranged in order, such as damage numbers 1, 2, the display order may be determined based on importance. For example, the abnormalities may be arranged in order of importance, so that the more important the abnormality is for the health assessment, the earlier it is displayed.

An image of the area of deformation used as the basis for the soundness assessment, such as image 902, is created by cutting out from an image superimposed on a drawing, based on the position information of the deformation acquired by the deformation information analysis unit 203. Specifically, the rectangular coordinates of the range surrounding the deformation are determined from the position information of the deformation in the drawing coordinate system. Since the image is superimposed on the drawing, it has a coordinate system that is common to the drawing or can be converted. Therefore, an image such as image 902 can be created by cutting out the image range indicated by these rectangular coordinates. Specifically, image 902 is an image obtained by cutting out the image surrounding crack 823, based on the coordinates of crack 823 in Figure 8. Similarly, image 903 is an image obtained by cutting out the image surrounding crack 822.

The abnormality information display area 904 also displays coordinate information of the abnormality and the name of the type of abnormality. The basis for the judgment may also be converted into a general sentence and displayed as the observation information. For example, the abnormality information display area 904 displays the sentence "There is a crack with a width of 0.5 mm or more" as the observation information created from the information on the basis for the judgment. Furthermore, the abnormality information display area 904 may also be provided with a comment field as a function for creating an inspection report.

As described above, FIG. 9 describes an embodiment in which anomalies related to soundness assessment are displayed in the form of an inspection report. The example shown in FIG. 9 may further include a function for outputting inspection report data in the format of an arbitrary text file.

The above describes a method for displaying abnormalities related to the soundness assessment. The user checks these results, and if the automatic soundness assessment result differs from the user's judgment, adjusts the soundness assessment result. One method for adjusting the results is to correct the abnormality information itself. Specifically, for example, if it is determined that the crack record is incorrect, the crack information can be deleted or, conversely, a crack can be added. In this way, the soundness result can be adjusted by correcting the abnormality record and then reassessing the soundness.

Another method for adjusting the soundness assessment results is to change the criteria for associating the anomaly with the evaluation area. For example, in the above embodiment, the soundness was assessed by associating the anomaly with the evaluation area using the association criteria that determines the soundness using only the anomaly within the evaluation area range. If the association criteria for the evaluation area is changed here, for example, to "for an anomaly that exists on the evaluation area boundary, the entire anomaly is associated with the evaluation area to determine the soundness," the soundness assessment result may change.

For example, for crack 403 in Figure 4, the above embodiment describes an embodiment in which only crack portion 421, which is within the range of span A, is used to determine the soundness of span A. On the other hand, if the entire crack 403 is used to determine the soundness of span A, the total length of the crack used to determine the soundness of span A will be longer, and a determination may be made that the soundness is worse than the soundness determination result in the above embodiment. In this way, the result of the soundness determination changes depending on how the abnormality is associated with the evaluation area of the soundness determination, so the method of associating the abnormality with the evaluation area is one method of adjusting the soundness determination result.

Below, we will explain in detail how to change the association criteria of the evaluation area management unit 201 via the setting unit 205 and thereby adjust the healthiness.

(Processing related to the setting unit)
The setting unit 205 sets the association criteria by displaying information on the display unit 105 (described below) and accepting user input from the operation unit 106 .

In the following, an embodiment will be described in which, after determining the soundness by the process described so far, the criteria for associating the abnormality to the evaluation area are changed to adjust the soundness determination. FIG. 10 shows the display unit 105 when adjusting the soundness by changing the association criteria. FIG. 10 shows a display window 1000 in a state in which the association criteria for span A are being changed, and the evaluation area for which the association criteria are being changed can be changed, for example, by a pull-down menu 1001.

In the image display area 1002 in the display window 1000, an image of the tunnel wall surface superimposed on the tunnel drawing and information on the abnormality are displayed, similar to the image display area 801 in FIG. 8. In addition, in the soundness display area 1004, the processing result of the soundness judgment according to the embodiment described above is displayed as "soundness 3". In addition, the image display area 1002 shows the association of the abnormality to the evaluation area span A. Here, in the above embodiment, the abnormality information is associated with the evaluation area according to the initial association criterion. This initial association criterion was "deformation in the evaluation area is used for the soundness judgment". Therefore, in the image display area 801, abnormalities existing within the range of span A, such as cracks 1020 and water leakage 1021, are highlighted as being associated with span A. Specifically, in FIG. 10, the lines and contours of the abnormalities existing within the range of span A are displayed as solid and thick. On the other hand, abnormalities not associated with span A (deformations in spans B and C) are displayed as dotted lines.

Crack 1010 is a deformation that spans spans A to C, but according to this association criterion, crack portion 1011 within the range of span A is associated with span A. Therefore, only portion 1011 of crack 1010 is highlighted. In this way, the image display area 1002 in FIG. 10 highlights the deformation associated with a specified evaluation area. In this way, in this embodiment, the deformations associated according to the set association criterion are displayed in a manner that is identifiable to the user. This makes it easier for the user to understand which deformations are associated with an evaluation area in order to determine the soundness of that evaluation area.

In the display window 1000 of FIG. 10, the association of the deformation to the evaluation area can be confirmed by pressing the setting button 1006. FIG. 11 (A) shows an association criteria setting screen displayed by pressing the setting button 1006. In this screen, the association method for the deformation existing at the evaluation area boundary to the evaluation area is displayed for each deformation type such as cracks and efflorescence. In this embodiment, three types of association criteria, association criteria 1 to 3, are prepared for the deformation existing at the evaluation area boundary, and one of the association criteria is set for each deformation type. Association criteria 1 (pattern 1) is "For deformations existing across the evaluation area boundary, the entire deformation is used to determine the soundness". Association criteria 2 (pattern 2) is "For deformations existing across the evaluation area boundary, only the deformations within the range of the evaluation area are used to determine the soundness". Note that pattern 2 is an association criterion set as the initial setting in the above description. Association criterion 3 (pattern 3) is "Exclude any abnormalities that exist across the boundaries of the evaluation area from the abnormalities used to determine the soundness of the evaluation area." Figure 11 (A) shows the setting state of the association pattern for abnormalities at the boundaries of these evaluation areas for each type of abnormality. In Figure 11 (A), the association criteria for all types of abnormalities are set as association pattern 2 as the initial association criteria.

Next, the image display area 1003 and the like present at the bottom of the display window 1000 in Figure 10 will be described. These are displays for adjusting the settings of the association criteria for associating anomalies with evaluation areas. Using this display and the operations described below, the user can set association criteria different from the association criteria at the top of the display window 1000 (hereinafter referred to as setting 1), reassess the healthiness, and adjust the healthiness assessment results.

The procedure for adjusting the soundness assessment by changing the association criteria will be described below. First, when the user presses the setting button 1007, the association criteria setting screen in FIG. 11(B) is displayed. This screen is the same as that in FIG. 11(A). The user can select any association criteria for each type of deformation by clicking a radio button. Here, an example is shown in which only cracks are changed to association pattern 1 from setting 1 in FIG. 11(A). Hereinafter, this setting will be referred to as setting 2. The image display area 1003 in FIG. 10 shows the deformation associated with span A by this setting 2. Unlike the image display area 1002 of setting 1, in the image display area 1003, the entire crack 1012 existing at the boundary of the evaluation area (span A) is associated with the evaluation area, and the entire crack 1012 is highlighted.

In step S301 of FIG. 3, the association criteria can be set based on the user's instructions. For example, when the association criteria of setting 2 are set, the user presses the judge button 1008 to judge the soundness based on this association criteria. This executes the processes of steps S302 to S306 in FIG. 3, and judges the soundness of span A (span A is selected as the evaluation area in step S302). This judgment result is displayed in the soundness display section 1005. In FIG. 10, the soundness judgment results before and after changing the association criteria are displayed in the soundness display sections 1004 and 1005, respectively. By displaying the soundness before and after the change on the same screen in this way, the user can compare the results and decide which soundness judgment result to adopt. Note that, in the above, an embodiment in which the soundness judgment process is executed by the user pressing the judge button 1008 has been described, but the execution of the soundness judgment process may be automatically executed at the timing when the association criteria is set.

The method described above makes it possible to change the association criteria for the anomaly to the evaluation area and adjust the health assessment results. Note that in the above explanation, an example was given in which only the association criteria for span A are changed, but it is also possible to change the association criteria for all evaluation areas at once and rerun the health assessment process.

In the above embodiment, the evaluation area is the span of the tunnel, but the evaluation area is not limited to this, and any range may be used as the evaluation area as long as it is a predetermined range for determining the soundness.

Figure 12 explains another example of an evaluation area, where a part of a bridge deck is set as the evaluation area. Like Figure 10, Figure 12 shows a display window 1200 for changing the association criteria. Image display areas 1201, 1202, and 1203 display wall images and abnormality information superimposed on drawings of the bridge deck 1210 and pier 1211. Furthermore, frames 1221, 1222, and 1223 shown on the deck indicate the evaluation area for soundness. Here, an example is shown in which the area near the deck pier is set as the evaluation area for determining the soundness of the deck, rather than the entire deck.

In Figure 12, image display areas 1201 and 1202 show the association status of anomalies to evaluation areas when the association criteria are changed for each type of anomaly. Specifically, image display area 1201 shows the state in which only anomalies within the evaluation area are associated, and image display area 1202 shows the state in which, for anomalies that exist across an evaluation area boundary, the entire anomaly is associated with the evaluation area. As with Figure 10, these are highlighted so that the range of anomalies associated with the evaluation area can be seen.

As a variation of the method for setting the criteria for associating deformations with evaluation areas, the range of the evaluation area may be changed. Evaluation area 1223 of image evaluation area 1203 is set as an evaluation area with a larger range than evaluation areas 1221 and 1222. In this way, the setting unit 205 may allow the evaluation range near the pier of the deck to be arbitrarily set based on user instructions.

Below, we will explain a variation of the first embodiment.

First, in the first embodiment, an embodiment in which a soundness determination method is learned using a machine learning algorithm is described, and a decision tree is described as an example. The machine learning algorithm used in the first embodiment is not limited to a decision tree, and other methods may be used. In particular, an embodiment using a randomized tree that is configured with a tree structure of binary determinations, similar to a decision tree, is a preferred modification.

Also, as shown in FIG. 8, when the health assessment results of multiple evaluation areas and the abnormalities related to the assessments are highlighted, the display method of the abnormalities may be changed depending on the value of the health level. For example, first, a predetermined color is set for each of health levels 1 to 5. Then, the color of the abnormalities related to the health assessments to be highlighted in each evaluation area may be displayed in the color of the health level of that evaluation area. In this way, it becomes easier to check the health assessment results of multiple evaluation areas and the abnormalities related to each health assessment at the same time. For the same purpose, the degree of highlighting of the abnormalities related to the health assessment may be changed depending on the health level. For example, the abnormalities related to the health assessment of an evaluation area judged to have a health level of 1 are highlighted more than the abnormalities related to the health assessment of an evaluation area judged to have a health level of 4. To this end, the lines of the abnormalities related to the assessment of health level 1 and the lines of the area contours may be displayed thicker or in a more noticeable color.

Furthermore, in the above embodiment, an embodiment in which the soundness of each evaluation area is determined has been described, but there are also cases in which the representative soundness of the entire structure is viewed. In such a case, the most unsound soundness (soundness closest to 1) among the soundness of each evaluation area of the structure is set as the representative soundness of the entire structure. Therefore, in a state in which the representative soundness of the entire structure is displayed, as described above, an abnormality related to the soundness determination result may be displayed for the evaluation area in which the soundness of the representative soundness has been determined. Note that the soundness of the entire structure may be configured to use the average value of the soundness of multiple evaluation target areas. In this case, an abnormality related to the soundness determination result may be displayed for the evaluation area in which the soundness of each evaluation area of the structure has been determined to be close to the average value.

<Embodiment 2>
In the first embodiment, an embodiment in which a judgment model learned by machine learning is used to judge the health level has been described. The method of judging the health level of this embodiment is not limited to this, and other methods may be used. In the second embodiment, a method of judging the health level using a preset rule as a judgment criterion will be described. Note that the hardware configuration and functional configuration of the second embodiment are the same as those of the first embodiment, and therefore description thereof will be omitted.

In the second embodiment, the rules for judging each health level are designed and set in advance by a human being. The information processing device 100 (specifically, the health level judgment unit 202) automatically judges the health level by using these judgment rules. Each judgment rule corresponds to the judgment criterion in the first embodiment. The health level judgment process of the health level judgment unit 202 in the second embodiment will be described below.

Figure 13 shows examples of judgment rules for health levels 2 and 3, and examples of health judgment results for a certain evaluation area. First, each judgment rule performs a binary judgment based on the deformation information associated with the evaluation area and the structure's specification information. For example, rule R2-1 for judging health level 2 in Figure 13 is "Are there any cracks with a crack width of 1.0 mm or more?" According to this rule, if a crack with a crack width of 1.0 mm or more exists in the evaluation area, the health level of the evaluation area will be 2. In this way, if the deformation and specification information of the evaluation area are True for any of the judgment rules for health level 2, the health level will be 2. In addition, in the health judgment method using the judgment rules, the lowest health level for which the judgment rule is True will be the health judgment result. For example, in Figure 13, the judgment result based on the deformation information and specification information of a certain evaluation area is shown in the judgment item 1301. In this example, all of the judgment rules for health level 1 are False.

As shown in FIG. 13, for the judgment rule for health level 2, the abnormality information and specification information for the evaluation area are all False. On the other hand, for the judgment rule for health level 3, two judgment rules are True. As a result, regardless of the results of the judgment rules for health levels 4 and 5, the health level of the target evaluation area is judged to be 3.

The health level determination unit 202 determines the health level as described above, and outputs the basis for determining the health level, as in the first embodiment. The basis for the determination in the second embodiment is the determination rule (determination criterion) that was determined to be true in the health level determination. In the example of FIG. 13, the basis for determining the health level as 3 is the determination rules R3-1 and R3-3.

The deformation information analysis unit 203 identifies deformation information related to the soundness judgment based on the judgment basis output by the soundness judgment unit 202. For example, based on the judgment rule R3-1 in the example of FIG. 13, it is assumed that cracks in the evaluation area with a crack width of 0.5 mm or more are related to the soundness judgment. Then, the position information of cracks in the evaluation area with a crack width of 0.5 mm or more is obtained from the deformation information storage unit 206. Similarly, based on the judgment rule R3-3, the position information of lattice-shaped cracks (lattice spacing 50 cm) is obtained from the deformation information storage unit 206. The display information processing unit 204 uses the position information of these deformations to highlight deformations related to the soundness judgment and cut out images of the deformation range, as described in the first embodiment.

In addition, even in the soundness determination method using the determination rule, the importance of the deformation type related to the soundness determination may be calculated. In the first embodiment, the importance was calculated based on the information gain obtained when learning the decision tree, but since there is no information gain in the determination rule, the importance is calculated by a different method. As a method of calculating the importance in the second embodiment, for example, the number of times the determination rule became True for each deformation type may be counted. For example, in the example of FIG. 13, the determination rule related to cracks becomes True for both the determination rules R3-1 and R3-3, and the crack counter is set to 2. As a result, the importance of cracks with respect to the soundness determination result becomes 2. Such processing is performed for each deformation type, and the importance for each deformation type is determined. In this way, the display information processing unit 204 can also display the importance for each deformation type, as in the first embodiment. In addition, the determination basis output by the soundness determination unit 202 may be displayed in the same manner as in the first embodiment.

As described above, even when a judgment rule is used as the judgment method for the health judgment, it is possible to display information on abnormalities related to the health judgment, and the user can understand the basis for the health judgment of the information processing device 100.

<Embodiment 3>
In the second embodiment, an embodiment in which the soundness is judged based on an artificially created judgment rule has been described. The method of artificially creating a judgment rule has the advantage that the judgment criteria for the soundness judgment are easy to understand and the judgment criteria can be controlled based on human knowledge. On the other hand, when it is desired to prepare complex judgment criteria in order to correctly judge the soundness, there is a problem that it is difficult for a human to manually design all the judgment criteria. For example, when the judgment criteria are set including not only the presence or absence and size of deformation such as cracks, but also information such as the number of years in service of the structure and the geographical location, the number of combinations of conditions becomes enormous, making it difficult for a human to determine the judgment criteria considering all the conditions. Therefore, there are cases in which it is preferable to statistically learn the judgment criteria for the soundness judgment by machine learning. The learning of the judgment criteria using the decision tree of the first embodiment is an example of a method for solving this problem.

In this embodiment, an embodiment in which judgment rules are learned by frequent pattern mining will be described. Learning judgment rules by frequent pattern mining is one form of learning binary judgment criteria that humans can understand. This is one embodiment in which judgment criteria are acquired by learning, and is a modified embodiment of embodiment 1. Note that the hardware configuration and functional configuration of embodiment 3 are similar to those of embodiment 1, so a description thereof will be omitted.

In frequent pattern mining, items or combinations of items that frequently occur under specified conditions are extracted. Below, an embodiment in which this technology is applied to learning judgment rules for health assessment will be explained using FIG. 14. FIG. 14 is a diagram showing an overall picture of judgment rule learning using frequent pattern mining.

First, in the context of frequent pattern mining, items and item sets will be described in order to explain the judgment rule learning of this embodiment. In this embodiment, a single judgment criterion for judging soundness is defined as an item. Table 1401 in FIG. 14 shows an example of item definition. For example, item I1 is a judgment criterion of "Are there cracks in the evaluation area?" In this way, in this embodiment, a single judgment criterion is a single item. Note that an item of a judgment criterion including a threshold value for attribute information such as the size of the deformation may be created. For example, item I2 in FIG. 14 is "Are there cracks with a crack width of 0.5 mm or more?", which is a judgment criterion composed of deformation information and attribute information threshold value. Similarly, judgment criteria related to specification information are also defined as items. For example, item IM is defined as an item with a judgment criterion of "Has the number of years since the start of service been 50 or more?". Hereinafter, the entire set of items is represented by I _all = {I1, I2, ... IM}, and a subset of I _all is called an item set. An item set includes a single item or a combination of items. In frequent pattern mining, the objective is to extract a set of items that can be used to judge soundness from the entire set of items I _all .

In the above item definition, an example was shown in which an item is defined using a threshold value for the deformation size, but it is also possible to define items according to discretized numerical values. Specifically, for example, the classification of crack widths may be discretized, and items such as "Are there any cracks with a crack width of 0.2 mm to 0.5 mm?" and "Are there any cracks with a crack width of 0.5 mm to 1.0 mm?" may be defined.

Next, in order to perform frequent pattern mining, transaction data is created from the learning data D. As in the first embodiment, the learning data D is D={D _i }, where D _i is data related to one evaluation area of a certain structure. D _i is defined as follows:

_Xi is the deformation information associated with a certain evaluation area and the specification information of a certain structure. _Yi is the correct answer data of the soundness determined by a human based on _Xi , and in this embodiment, indicates the soundness of 1 to 5.

One transaction T _i is created from this one learning data D _i . The created transaction T _i is, for example, T _i = {I1, I2, I9, I20; y _i }. In this case, it is shown that X _i (deformation information and specification information of the evaluation area) of data D _i matches the conditions of item definitions I1, I2, I9, and I20. As another specific example, if X _i contains a crack with a crack width of 0.5 mm, items I1 and I2 match the conditions. If the conditions of other items do not match X _i , T _i = {I1, I2; y _i }.

In this way, transaction data _Ti is created for each D _i of the learning data D. If the set of items included in the transaction data _Ti is I _i , the transaction data _Ti is expressed as _Ti = {I _i ; y _i }. Also, the data set of all transaction data is T, where T = { _Ti }.

Below, frequent pattern mining is performed using this transaction dataset T. In frequent pattern mining, a set of items that frequently appear in the transaction dataset T is extracted and used as a judgment rule. The process of extracting frequent patterns is also performed for each soundness level, and judgment rules for each soundness level are learned. Below, the process of extracting frequent patterns is explained using the learning of judgment rules for soundness level 3 as an example.

In this embodiment, Apriori (Non-Patent Document 1) is used to extract frequent patterns. In Apriori, the frequency of occurrence of a certain item set in a transaction dataset is called support. Support is expressed by the following formula.

N is the number of data in the transaction dataset T, and σ _T (I) is the number of transaction data T _i that include the itemset I. In Apriori, a threshold called minimum support is given in advance, and itemsets that show support equal to or greater than the minimum support are extracted as frequent patterns. Since Apriori is a known algorithm, a detailed explanation will not be given, but by specifying the minimum support in advance when searching for frequent itemsets, itemsets that have equal to or greater than the minimum support can be efficiently discovered.

In learning the judgment rule for soundness level 3, attention is focused only on transaction data with soundness level 3. If the transaction data set consisting only of transaction data with soundness level 3, i.e., transaction data with y _i =3, is denoted as T ⁽³⁾ , the support thereof is given by

N ⁽³⁾ is the number of data in T ⁽³⁾ . By executing Apriori using this support, it is possible to extract a set of event sets that frequently appear in the transaction data set with health level 3. This set of event sets is represented as I ⁽³⁾ = {Ia, Ib, ..., In}, where Ia, Ib, ... are event sets such as Ia = {I1, I4, I10}, Ib = {I8}, etc.

Next, an event set specific to transactions with a soundness of 3 is selected from I ⁽³⁾ . The event set set I ⁽³⁾ is an event set that frequently appears in the learning data with a soundness of 3, but may also include event sets that frequently appear in the learning data with other soundness levels. In other words, it may include items or combinations of items that are likely to appear regardless of the soundness level. Therefore, a process of selecting an event set specific to the learning data set with a soundness of 3 is performed using a learning data set other than the soundness level 3. Specifically, the ratio of the support (occurrence frequency) of each event set in I ⁽³⁾ in the transaction data set T ⁽³⁾ to the support in the transaction data set other than the soundness level 3 is calculated, and the event set with the larger ratio is determined as the event set specific to the soundness level 3. This support ratio is called the Growth Rate and is expressed by the following formula.

Here, the transaction dataset created from the learning dataset other than the health level 3 is defined as T ^(not3) . I is the event set for which the Growth Rate is calculated, and is the event set included in I ⁽³⁾ . Hereinafter, Growth Rate will be abbreviated as GR. Note that patterns that appear disproportionately in one dataset are called Emerging Patterns (Non-Patent Document 2).

In the above example, the higher the GR, the more specific the item set is to the health level 3. Therefore, by selecting an item set having a GR equal to or greater than a certain threshold from the item set set I ⁽³⁾ , a set of item sets specific to health level 3 can be created. This item set R ⁽³⁾ = {R3-1, R3-2, R-3N} specific to health level 3 can be used as a judgment rule for judging health level 3. Specifically, as shown in table 1402 in FIG. 14, for example, R3-1 = {I2}, and item I2 "Are there any cracks with a crack width of 0.5 mm or more?" is learned as one of the judgment rules. In addition, judgment rules for combinations of multiple items, such as R3-2 = {I5, I10}, are also learned. Note that the selection of an event pattern based on GR may be performed by selecting event patterns with the top n GRs instead of threshold processing.

In this way, frequent pattern mining can learn the judgment rule for health level 3. Similar processing is performed for each health level to learn judgment rules for health levels 1 to 5. Using the learned judgment rule, the process of judging the health level of the target evaluation area and the process of displaying information on abnormalities related to the subsequent health level judgment can be performed in the same manner as in embodiment 2. In other words, a judgment rule is determined for which the input abnormality information and specification information are True, and this judgment rule is used as the judgment basis. Then, based on the abnormality information included in the judgment basis, the abnormality is highlighted or displayed.

In addition, the rule for determining health level 5 (healthiest) may not be learned, and if the health level is not determined to be 1 to 4, the health level may be determined to be 5.

GR may also be used to calculate the importance of the anomaly type. In this case, during learning, the GR of each judgment rule, for example, R3-1, R3-2, etc., is saved together with the judgment rule. The higher the GR, the more likely the event set is specific to that health level. Therefore, by using GR in the same way as the information gain of each node of the decision tree in embodiment 1, it becomes possible to calculate the importance of each anomaly type with respect to the health level judgment result.

Furthermore, the learned judgment rules may be displayed to the user so that the user can adjust the judgment rules. In this case, the judgment rules learned by the above method are first displayed so that the user can understand the contents of the items. Specifically, the items in the item set in table 1402 in FIG. 14 are displayed as the contents of the specific items. For example, the content of the judgment rule R3-1 is displayed as the judgment criterion "Are there any cracks with a crack width of 0.5 mm or more?" This allows the user to confirm what judgment rules have been learned by frequent pattern mining. Based on this display, the user can adjust the judgment rules. Specifically, a judgment rule that is judged to be inappropriate as a judgment rule is removed, or a judgment rule that has not been learned but is judged to be added as a judgment rule is added. As a result, it becomes possible to create judgment rules based on human knowledge and experience based on the mechanically acquired judgment rules.

In addition, in general, it takes time to accumulate data in order to prepare learning data. Therefore, in an operation in which soundness determination is performed by an information processing device, first, as described in embodiment 2, soundness determination is performed based on a judgment rule created by a human in the early stages of operation. The input data (deformation information and specification information of the evaluation area) at this time may be accumulated, and when a predetermined amount of data is accumulated, the data may be used as learning data to learn the judgment rule. The labels of soundness 1 to 5 to be learned are data in which a human has confirmed the soundness determination result of the information processing device, or the result of a human having corrected the soundness determination result of the information processing device. This makes it possible to operate automatic soundness determination by the information processing device from an early stage when no learning data exists. In addition, learning of the judgment rule associated with data accumulation may be performed not only once, but also multiple times in stages depending on the accumulation status of the learning data. Note that the embodiment in which learning data is accumulated and learning is performed at the timing when a predetermined amount of learning data is accumulated can also be implemented in learning the soundness determination method by the machine learning algorithm of embodiment 1.

In the above explanation, the two-step judgment rule learning has been described: extracting frequent patterns using Apriori and selecting item sets specific to each soundness level by calculating GR. However, the judgment rule learning method of this embodiment is not limited to this. In other words, other methods may be used as long as they are methods for extracting specific item sets using training data. For example, in addition to Apriori, a method called FP-Growth is also well known as a method for extracting frequent patterns. Alternatively, frequent patterns may be extracted using other computationally efficient algorithms.

In the above, the learned judgment rule is used in the same manner as in the second embodiment. That is, when the input data corresponds to one of the judgment rules of a certain soundness, the soundness is obtained as the soundness judgment result. However, the judgment rules obtained by the above learning have different reliability as indicated by GR. Therefore, a score may be calculated from the judgment rule using the support and GR, and the soundness may be judged based on the score. As described above, the judgment rule is a set of items that appear in a transaction dataset of a specific soundness, and this is called Emerging Patterns. An ensemble identification method using these Emerging Patterns is described in Non-Patent Document 3. In the following embodiment, an embodiment in which the soundness is judged from the learned judgment rule using this method will be described. Note that, in order to implement this method, the support and GR of each judgment rule are recorded when the judgment rule is learned.

First, the judgment score S _H of a certain health level H is expressed as follows based on Non-Patent Document 3.

Here, IN is an item included in the abnormality information and specification information of the evaluation area to be judged for soundness, and I ^(H) indicates the judgment rule for the soundness H. That is, in the judgment rule in which the input data of the abnormality information and specification information of the evaluation area is True, the sum of the values calculated based on GT and support becomes the score S _H for judging the soundness H. According to this formula, the higher the judgment rule in which GR and support indicate high values, the higher the score S _H obtained. This score S _H is calculated for each soundness, and the soundness that outputs the highest score S _H is set as the soundness H of the evaluation area.

In the above, in the third embodiment, a method for learning soundness judgment rules using frequent pattern mining has been described. By learning judgment rules using frequent pattern mining, it is possible to create judgment rules that are more complex and have higher judgment performance than judgment rules set by humans. In addition, since the learning results are judgment rules that can perform binary judgments, it is easy for humans to understand the learning results and it is also possible to adjust the judgment rules. Furthermore, by using the learned judgment rules for soundness judgment in the same manner as in the second embodiment, it is possible to display information on abnormalities related to the soundness judgment, and therefore it is possible to display the basis of the soundness judgment results in an easy-to-understand manner.

<Embodiment 4>
In each of the above-described embodiments, the setting of the association criteria for the deformation to the evaluation area is described for each deformation type. The association criteria may be set not only for each deformation type, but also for each individual deformation. In the fourth embodiment, an embodiment in which an association criterion for the evaluation area is set for each individual deformation is described. Note that the hardware configuration and functional configuration of the fourth embodiment are the same as those of the first embodiment, and therefore the description is omitted.

Figure 15 shows a display window 1500 displayed by the setting unit 205 on the display unit 105 in order to set an association with an evaluation area for each individual deformation. In Figure 15, a soundness evaluation area is set for each span of the tunnel, and the state in which the association criteria for deformation with the evaluation area of span E are set is shown. Four cracks with crack IDs 001 to 004 are displayed in the image information display unit 1501. Here, crack IDs 003 and 004 are located in positions unrelated to span E, so they are not associated with span E and are shown with dotted lines.

To set the association criteria for an individual anomaly, the user selects the anomaly for which settings are to be adjusted. For example, the user uses the mouse cursor 1502 to select the anomaly for which settings are to be adjusted. Figure 15 shows that crack ID 002 has been selected. The dotted line 1504 indicates that crack ID 002 is focused. When crack ID 002 is selected, information for selecting the association criteria for crack ID 002 to span E is displayed in the criteria setting area 1503.

In FIG. 15, as in the first embodiment, three types of association criteria, association patterns 1 to 3, are prepared for anomalies that exist on the boundary of an evaluation area as association criteria for the evaluation area. Association pattern 1 is a pattern in which the entire anomaly that exists on the boundary of an evaluation area is used to determine the soundness of the anomaly. Association pattern 2 is a pattern in which only the anomaly within the range of the evaluation area is used to determine the soundness of anomalies that exist on the boundary of an evaluation area. Association pattern 3 is a pattern in which anomalies that exist on the boundary of an evaluation area are excluded from the anomalies used to determine the soundness of the evaluation area.

Figure 15 shows a situation where the user has set association pattern 1 as the association criterion for crack ID002. With this setting, the entire crack ID002 is displayed in a thick line in the image information display unit 1501, indicating that it is associated with span E. On the other hand, although the setting status of crack ID001 is not shown, it is assumed that the association criterion of association pattern 2 is set. As a result, crack ID001 is displayed in a thick line to indicate that the range of span E is associated with span E, and the part outside the range of span E is displayed in dotted lines to indicate that it is not associated with span E. In this way, in embodiment 4, different association criteria can be set for crack ID001 and ID002, which are the same deformation type.

In addition, the screen for setting the association criteria for individual anomalies may be configured to display how the health level changes depending on the association settings. In the criteria setting area 1503 of FIG. 15, the health level of span E is determined when each association pattern is set for crack ID 002, and the results are displayed. In this way, it becomes possible to set the association criteria for anomalies to evaluation areas while checking the change in health level when the association criteria for individual anomalies are changed.

As described above, in the fourth embodiment, an embodiment has been described in which an association criterion for an evaluation area is set for each individual anomaly. This allows the association of anomalies with evaluation areas to be set in detail, allowing the user to fine-tune the health assessment results.

<Embodiment 5>
In each of the above-described embodiments, an embodiment has been described in which an anomaly existing within an evaluation area or on the boundary of an evaluation area is associated with the evaluation area to determine the soundness of the evaluation area. In this embodiment, the association of an anomaly with an evaluation area may be for an anomaly that does not directly overlap with the evaluation area. In the fifth embodiment, an embodiment will be described in which an anomaly that is not directly related to the evaluation area is associated as an anomaly to be used in determining the soundness of the evaluation area. That is, in the fifth embodiment, an association rule is set to associate an anomaly existing in an associated area that is different from the evaluation area but is related in determining the soundness as an anomaly to be used in determining the soundness of the evaluation area. Note that the hardware configuration and the functional configuration of the fifth embodiment are the same as those of the first embodiment, and therefore the description will be omitted.

Figure 16 shows a drawing and an image of a road bridge that is the subject of a soundness assessment. The road bridge in Figure 16 shows the deck 1602 of the span between piers 1603. The deck 1602 is the part seen from the ground, and Figure 16 also shows a drawing and an image of the road surface 1601. Here, in order to assess the soundness of the deck 1602, a process is carried out to associate the abnormalities used in the soundness assessment of the deck 1602. Note that the frame 1604 surrounding the deck 1602 indicates that the entire deck 1602 is the evaluation area to be processed.

In the example shown in FIG. 16, an embodiment is shown in which the deformation of the road surface 1601 on the back side of the deck 1602 is included in the deformation used to determine the soundness of the deck 1602. Specifically, the pothole 1611 is associated with the evaluation area 1604 and used to determine the soundness of the evaluation area 1604 of the deck 1602. In order to do this, the association criteria for the deformation of the evaluation area 1604 in FIG. 16 are set to "associate the pothole on the road surface on the back side with the evaluation area". In areas where the road surface is severely damaged, such as potholes, water such as rain can affect the deck slab, so it is preferable to judge the soundness of the deck including this. Note that a pothole is a round hole or dent that is formed when the surface layer of an asphalt road peels off. In this way, deformations that are not directly included in the evaluation area may also be associated with the evaluation area to determine the soundness. The change in the association criteria can be set for each type of deformation or for each individual deformation, as in the methods described in the first and fourth embodiments. For example, in FIG. 16, the crack 1612 in the road surface 1601 is not associated with the evaluation area 1604 in the deck 1602, and is not used to determine the soundness of the evaluation area 1604. The user can set whether or not to associate this association via the operation unit 105 as necessary.

When associating an abnormality that is not directly related to the evaluation area, the target area is set in advance as an area that is structurally or inspectably related to the evaluation area, such as the floor board and road surface in Figure 16. Then, the association criteria for whether or not to associate an abnormality in this related area with the evaluation area can be adjusted. As an alternative method, it is also possible to set an arbitrary abnormality in an arbitrary position to be individually associated with the evaluation area.

As described above, the method described in the fifth embodiment makes it possible to use anomalies that exist in a position that is not directly related to the evaluation area in the soundness assessment. This makes it possible to carry out a soundness assessment that takes into account anomalies outside the evaluation area that affect the soundness of the evaluation area. In addition, by adjusting the association criteria, it becomes possible to adjust the results of the soundness assessment.

<Embodiment 6>
In the above embodiments, various processes are performed by the information processing device 100 in Fig. 1. In the sixth embodiment, a process configuration for providing a service to a user using a system configuration of a server and a client will be described.

Figure 17 (A) is a configuration diagram of an information processing device 100 (information processing system) according to this embodiment. It is a diagram explaining one form of processing in a server and client configuration. In Figure 17 (A), a client terminal 1711 and a service providing system 1731 are information processing devices each having the same hardware configuration as the information processing device 100 in Figure 1. In Figure 17 (A), client terminals 1701 to 1703 operated by users 1701 to 1703 are connected to the service providing system 1731 via the Internet 1720.

In FIG. 17(A), only the processing related to the display unit 105 and operation unit 106 is executed on the client terminal, and other processing (e.g., processing of the health assessment unit 202) is executed on the service provider system 1731. With this configuration, only light processing is executed on the client terminal, and processing that requires a large amount of calculation, such as health assessment, can be executed on the high-performance service provider system 1731. This allows users 1701 to 1703 to receive services simply by preparing an inexpensive client terminal with low processing power.

The distribution of the process shown in FIG. 17(A) is not limited to the one shown in FIG. 17(A), and may be in another form. For example, FIG. 17(B) shows a configuration including an on-premise server 1742 as a processing configuration different from that shown in FIG. 17(A). In FIG. 17(B), the processing of this embodiment is distributed and executed by the client terminal 1741, the on-premise server 1742, and the service provider system 1743. For example, the service provider system 1743 learns a soundness determination method, creates a soundness determination model or a soundness determination rule as a result of the learning, and transmits it to the on-premise server 1742. The on-premise server performs processing such as soundness determination processing, and the client terminal 1741 displays the soundness determination results. In this way, even if the inspection image or soundness determination results are confidential, the on-premise server 1742 in-house performs soundness determination processing, so that the configuration is safe from the viewpoint of information security. On the other hand, the latest services can be received, such as the latest soundness determination model, from the service provider system 1743.

Furthermore, the configuration of this embodiment makes it possible to provide appropriate services for each user. For example, the method of determining the soundness may differ depending on the user. For example, users in this embodiment are assumed to be inspection companies, structure managers, infrastructure management departments of local governments, etc. These users may have their own inspection standards, and the soundness determination method and criteria differ for different structures to be managed. Therefore, by providing a soundness determination method (soundness determination model and determination rules) that is suitable for the user's requests and circumstances, it becomes possible to implement a soundness determination that is suitable for each user.

Similarly, appropriate association criteria for evaluation areas may be provided for each user. For example, the service provider system 1731 may store association criteria previously set by the user together with the user information. Then, when the user performs a new health assessment, the association criteria previously set by the user may be provided as the initial association criteria settings.

The association criteria may also be learned from the user's association setting criteria. For example, as described in the fourth embodiment, the association criteria can be set for each individual anomaly. In learning the association criteria, the user's tendency to set the association criteria is learned. Specifically, the size of the anomaly, its position relative to the evaluation area, the degree of overlap between the evaluation area and anomalies existing on the evaluation area boundary, etc. are used as input variables, and the presence or absence of an association with the evaluation area is used as the teacher label, and learning is performed using an arbitrary machine learning algorithm. The anomaly is associated with the evaluation area using the model learned as a result of this.

In this way, services can be provided that are suited to the trends in each user's health assessment.

[Other embodiments]
The present invention can also be realized by a process in which a program for implementing one or more of the functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program. The present invention can also be realized by a circuit (e.g., ASIC) that implements one or more of the functions.

201 Evaluation area management unit 202 Soundness determination unit 203 Deformation information analysis unit 204 Display information processing unit 205 Setting unit 206 Deformation information storage unit 207 Specifications information storage unit 208 Learning unit

Claims (18)

A setting means for selecting an association rule for associating a deformation used for judging the soundness of a structure with an evaluation area to be judged for the soundness from among a plurality of association rules, and setting the selected association rule as the association rule to be used for the judgment;
A determination means for determining a soundness of the evaluation area based on a defect associated with the evaluation area according to the association rule set by the setting means;
13. An information processing device comprising: The information processing device according to claim 1, characterized in that the setting means sets an association rule selected from the plurality of association rules based on a user instruction. The information processing device according to claim 1 or 2, further comprising a display control means for identifiably displaying the abnormality associated with the evaluation area according to the association rules set by the setting means. The information processing device according to claim 3, characterized in that the display control means highlights the abnormality associated with the evaluation area according to the association rules set by the setting means. The information processing device according to claim 1 or 2, further comprising a display control means for displaying a first judgment result by the judgment means based on an abnormality associated with the evaluation area according to a first association rule set by the setting means, and a second judgment result by the judgment means based on an abnormality associated with the evaluation area according to a second association rule set by the setting means that is different from the first association criterion. The information processing device according to any one of claims 1 to 5, characterized in that the setting means sets the association rules for each type of deformation. The information processing device according to any one of claims 1 to 5, characterized in that the setting means sets the association rule for each of a plurality of abnormalities. The information processing device according to any one of claims 1 to 7, characterized in that the setting means sets the association rules for determining the association of anomalies existing across the boundaries of the evaluation area to the evaluation area. The information processing device according to claim 8, characterized in that the plurality of association rules include a rule that, among the anomalies that exist across the boundaries of the evaluation area, an anomaly that exists within the evaluation area is associated with the evaluation area as an anomaly to be used in determining the soundness. The information processing device according to claim 8 or 9, characterized in that the plurality of association rules include a rule that an anomaly that exists across the boundary of the evaluation area is not associated with the evaluation area as an anomaly to be used in determining the soundness. The information processing device according to any one of claims 8 to 10, characterized in that the plurality of association rules include a rule that associates all of the anomalies present across the boundaries of the evaluation area with the evaluation area as an anomaly to be used in determining the soundness. The information processing device according to any one of claims 1 to 11, characterized in that the plurality of association rules include a rule that associates an abnormality that exists in an area related to the evaluation area, which is different from the evaluation area, with the evaluation area as an abnormality to be used in determining the soundness. The information processing device according to claim 12, characterized in that the relevant area is an area behind the evaluation area. The information processing device according to any one of claims 1 to 13, characterized in that the determination means determines the soundness using a trained model trained by machine learning. The information processing device according to claim 14, characterized in that the trained model is trained using a tree-structured machine learning algorithm. The information processing device according to any one of claims 1 to 15, characterized in that the deformation of the structure includes at least one of cracks, water leakage, potholes, exposed rebar, junk, rust, and efflorescence. An information processing method by an information processing device,
selecting an association rule from a plurality of association rules for associating the deformation used to judge the soundness of the structure with the evaluation area to be judged for the soundness;
The selected association rule is set as the association rule to be used for the judgment;
determining a health level of the evaluation area based on the abnormality associated with the evaluation area according to the set association rule;
23. An information processing method comprising: A program for causing a computer to function as the information processing device according to any one of claims 1 to 16.

Images