JP6984265B2 - Information processing equipment, information processing methods, and programs - Google Patents

Description

The present invention relates to an information processing apparatus or the like using a model by statistical analysis.

For bridges, tunnels or road structures, inspectors perform visual or tapping inspection work on a regular basis, and the structures are based on the experience and knowledge of each inspector according to the damage status of the structure. The status is determined.

It is difficult to repair all the damages existing in many structures at the same time due to cost and human resource constraints, and the structure manager repairs the structure based on the condition judgment obtained by the inspection work. I'm making a plan. Therefore, the condition determination of the structure is very important information in the repair plan.

In the state judgment, since the organization / person in charge of the inspection work is assigned to each place, there are cases where the judgment result is different even if the same degree of damage occurs in the structure. As a result, there is a possibility that the target to be repaired with priority should be overlooked.

In Patent Document 1, it is assumed that the priority index indicating the priority of maintenance for each predetermined section of the civil engineering structure follows the gamma distribution, and a model by statistical analysis is generated for the result of evaluating the inspection items. Further, an example showing the possibility that the evaluation result may be observed as having a defect is disclosed.

Japanese Unexamined Patent Publication No. 2016-122391

Even in the case of constructing a model by statistical analysis as in Patent Document 1, the judgment result (evaluation result) of the inspector includes some wrong judgment result. Therefore, when improving the accuracy of the model by statistical analysis, it is necessary to review the judgment result used for constructing the model.

On the other hand, when reviewing the judgment result based on the inspection record of the structure, the number of skilled specialists (inspectors, etc.) for re-judgment is limited for a huge number of structures. Therefore, the time required for reviewing the determination result is enormous.

An object of the present invention is to provide an information processing apparatus or the like capable of efficiently reviewing a determination result based on an inspection record of a structure.

The information processing apparatus according to one aspect of the present invention has a determination model generation unit that generates a determination model showing the relationship between an inspection record for each structure and the determination result, the determination model, the inspection record, and the determination result. Based on the above, a certainty calculation unit that calculates the certainty indicating the ambiguity of the determination result for each of the structures, and a re-judgment extraction unit that extracts the structure to be re-determined based on the certainty. , Equipped with.

The information processing method according to one aspect of the present invention generates a determination model showing the relationship between the inspection record for each structure and the determination result, and based on the determination model, the inspection record and the determination result, the structure. For each, the degree of certainty indicating the ambiguity of the determination result is calculated, and the structure to be re-determined is extracted based on the degree of certainty.

The program according to one aspect of the present invention generates a determination model representing the relationship between the inspection record for each structure and the determination result on a computer, and the structure is based on the determination model, the inspection record and the determination result. For each object, the degree of certainty indicating the ambiguity of the determination result is calculated, and the structure to be re-determined is extracted based on the degree of certainty.

INDUSTRIAL APPLICABILITY According to the present invention, the determination result based on the inspection record of the structure can be efficiently reviewed.

It is a block diagram which shows the structure of the determination support apparatus which concerns on 1st Embodiment. It is a data sheet which shows an example of the learning data in 1st Embodiment. It is a data sheet which shows the correspondence of the structure ID and the degree of certainty in 1st Embodiment. It is a flowchart which shows the operation of the determination support apparatus which concerns on 1st Embodiment. It is a block diagram which shows the characteristic structure of the determination support apparatus which concerns on 1st Embodiment. It is a block diagram which shows the structure of the determination support apparatus which concerns on 2nd Embodiment. It is a data sheet which shows an example of the learning data with a reliability in 2nd Embodiment. It is a flowchart which shows the operation of the determination support device which concerns on 2nd Embodiment. It is a block diagram which shows the characteristic structure of the determination support apparatus which concerns on 2nd Embodiment. It is a block diagram which shows the hardware composition which realized the determination support apparatus in 1st and 2nd Embodiment by a computer.

An information processing device according to an aspect of the present invention will be described with reference to an example of a determination support device.

(First Embodiment)
First, the configuration of the determination support device 110 according to the first embodiment will be described.

FIG. 1 is a block diagram showing a configuration of a determination support device 110 according to the first embodiment. The determination support device 110 shown in FIG. 1 includes a determination model generation unit 111, a certainty calculation unit 112, a re-determination extraction unit 113, and a learning data storage unit 121.

The learning data storage unit 121 stores the learning data. The learning data includes each inspection record of one or more structures to be learned, and a determination result of the structure based on the inspection record. FIG. 2 is a data sheet showing an example of learning data in the first embodiment. In the example of FIG. 2, as learning data, the inspection record and the determination result of the structure are associated with each structure ID (Identifier). As the items of the inspection record, as shown in FIG. 2, the type and magnitude of the damage of the structure and the age of the structure are set. The damage may include different types such as water leakage and exposure of reinforcing bars in addition to the same type.

The item indicating the determination result indicates the state of the structure determined by an expert such as an inspector based on the inspection record. In the example of FIG. 2, it is assumed that the smaller the value of the state, the more preferable the structure. An example of a condition is soundness. It is assumed that the learning data is stored in the learning data storage unit 121 in advance.

The determination model generation unit 111 reads the learning data of the learning data storage unit 121 and generates a determination model for determining the state of the structure. The judgment model shows the relationship between the judgment result items and the inspection record items by, for example, a multiple regression analysis method in which each item of the inspection record is used as an explanatory variable and the state of the structure which is the judgment result is used as the objective variable. Expressed as a formula.

For example, in the example of FIG. 2, Y is the state of the structure, X ₁ is the damage A, X ₂ is the damage B, and X ₃ is the variable of the age of the building. The explanatory variables can be changed as appropriate, such as the number of items in the inspection record, regardless of the example of FIG.

By substituting the training data (m set of Y, X ₁ , X ₂ , X ₃ ) into the above formula, the combination of _{coefficients b 0} , b ₁ , b ₂ , and b ₃ that minimizes the error is actually measured. The multiple regression equation is obtained by calculating the residual of the prediction by the method of least squares.

The determination model generated by the determination model generation unit 111 is sent to the certainty calculation unit 112.

The determination model using the above multiple regression analysis method is an example, and may be realized by another statistical analysis method. For example, it can be expressed as a mathematical formula showing the relationship between the items of the judgment result and each item of the inspection record by the multi-class logistic regression analysis method in which each item of the inspection record is used as an explanatory variable and the item of the judgment result is used as the objective variable. Alternatively, according to C4.5, the relationship between each item of the inspection record and the state may be expressed by the relationship of the if-then rule, and it may be generated as a judgment model used for judging the state.

The certainty calculation unit 112 takes the judgment model generated by the judgment model generation unit 111 as an input, reads the learning data from the learning data storage unit 121, and calculates the certainty for the state of each structure. FIG. 3 is a data sheet showing an example of the structure ID and the degree of certainty.

Here, the certainty degree represents the ambiguity of the determination result, and is calculated after using the determination model for the structure and the learning data. The certainty can be calculated from the judgment model, for example, by using a multi-class logistic regression analysis method, for each structure, the probability of which state it is in.

Specifically, the state of a certain structure in the training data is divided into five classes of 1, 2, 3, 4, and 5, and the probability of occurrence for each class indicating the state of the structure is calculated.

And. Based on this probability of occurrence

The degree of certainty (ambiguity) can be calculated by calculating the entropy represented by (the bottom is 2). The larger the entropy value for the state of a certain structure, the more ambiguous the judgment result of that state. Incidentally, when the state of the determination result of the five classes, the possible range of the entropy becomes 0 ≦ entropy ≦ log 2 _5.

Then, the certainty calculation unit 112 outputs the certainty of each structure to the redetermination extraction unit 113.

The re-judgment extraction unit 113 takes the certainty of each structure calculated by the certainty calculation unit 112 as an input, and extracts the structure (re-judgment structure) to be re-determined. Specifically, the re-determination extraction unit 113 extracts the structure (structure ID) associated with the highest certainty from the certainty of each structure.

The number of structures extracted by the redetermination extraction unit 113 may be one or more. In the case of a plurality of structures, the redetermination extraction unit 113 may extract the structure according to the magnitude of the certainty as a criterion for extracting the structure. The criteria for extraction may be other than the above, and for example, a structure having a certainty higher than a certain threshold value may be extracted. Furthermore, the number of structures to be extracted may be set in the redetermination extraction unit 113 via an input device (not shown).

Next, the operation of the determination support device 110 according to the first embodiment will be described.

FIG. 4 is a flowchart showing the operation of the determination support device 110 according to the first embodiment.

The determination model generation unit 111 reads out the learning data stored in the learning data storage unit 121 (step S11).

The determination model generation unit 111 uses the read learning data to generate a determination model showing the relationship between the inspection record for each structure and the determination result (step S12).

The certainty calculation unit 112 calculates the certainty for the state of each structure by using the determination model generated by the determination model generation unit 111 and the learning data of the learning data storage unit 121 (step S13).

The re-determination extraction unit 113 takes the certainty of each structure calculated by the certainty calculation unit 112 as an input, and extracts the structure to be re-determined from the learning data (step S14).

Next, the characteristic configuration of the first embodiment will be described.

FIG. 5 is a block diagram showing a characteristic configuration of the first embodiment. Referring to FIG. 5, the determination support device 110 (also referred to as an information processing device) includes a determination model generation unit 111, a certainty calculation unit 112, and a re-determination extraction unit 113. The determination model generation unit 111 generates a determination model showing the relationship between the inspection record for each structure and the determination result. The certainty calculation unit 112 calculates the certainty indicating the ambiguity of the judgment result for each structure based on the judgment model, the inspection record and the judgment result. The re-judgment extraction unit 113 extracts a structure to be re-determined from the training data based on the certainty.

Next, the effect of the first embodiment will be described.

According to the determination support device of the first embodiment, it is possible to efficiently review the determination result based on the inspection record of the structure.

The reason is that the judgment model generation unit 111 generates a judgment model showing the relationship between the inspection record for each structure included in the training data and the judgment result, and the certainty calculation unit 112 generates the judgment model. Based on the inspection record and the judgment result, the degree of certainty indicating the ambiguity of the judgment result is calculated for each structure in the training data. The re-judgment extraction unit 113 extracts a structure to be re-determined from the learning data based on the calculated certainty.

As a result, the determination support device 110 can efficiently identify the structure in which the determination result is ambiguous and it is preferable to re-determine from the inspection result for each structure and the learning data including the determination result. It is possible to shorten the time required for reviewing the results.

(Second embodiment)
Next, the configuration of the determination support device 210 according to the second embodiment will be described.

FIG. 6 is a block diagram showing a configuration of the determination support device 210 according to the second embodiment. The determination support device 210 shown in FIG. 6 includes a determination model generation unit 211, a certainty calculation unit 212, a re-judgment extraction unit 213, a reliability calculation unit 214, a learning data storage unit 221 and a feedback unit 231.

In the learning data storage unit 221 shown in FIG. 6, reliability is added to the learning data stored in the learning data storage unit 121 shown in FIG. Specifically, the learning data storage unit 221 stores, as learning data, the inspection record for each structure and the determination result, as well as the reliability of the determination result. FIG. 7 is a data sheet showing an example of learning data in the second embodiment.
The range of reliability is 0 to 1, and the initial value of each reliability is all 1.

The determination model generation unit 211, the certainty calculation unit 212, the re-judgment extraction unit 213, and the learning data storage unit 221 of the second embodiment shown in FIG. 6 each generate the determination model of the first embodiment shown in FIG. It has functions corresponding to the unit 111, the certainty calculation unit 112, the re-judgment extraction unit 113, and the learning data storage unit 121.

The determination model generation unit 211 inputs the learning data including the reliability stored in the learning data storage unit 221 and learns the determination model. For example, by multiple regression analysis by the weighted least squares method, a model that expresses the relationship between the inspection record, the judgment result, and the reliability by a mathematical formula is generated.

For example, in the example of FIG. 6, Y is the state of the structure, X ₁ is the damaged A, X ₂ is the damaged B, X ₃ is the age, X ₄ is the reliability variable, and w _j is every 1 to m sets. It is expressed by a multiple regression equation weighted by. The explanatory variables can be changed as appropriate, such as the number of items in the inspection record, regardless of the example of FIG.

Substituting the training data (m set of Y, X ₁ , X ₂ , X ₃ , X ₄ ) into the above equation, the coefficients b ₀ , b ₁ , b ₂ , b ₃ , b so as to minimize the error. _The multiple regression equation is obtained by calculating the residuals of the actual measurement and the prediction by the least squares method for the combination of 4.

The determination model generated by the determination model generation unit 211 is sent to the certainty calculation unit 212.

In the determination support device 210, the re-judgment extraction unit 213 extracts the structure to be re-determined and sends the extracted structure ID to the feedback unit 231.

The feedback unit 231 has an input unit (not shown) and an output unit (not shown). The feedback unit 231 controls the output of the inspection record corresponding to the extracted structure ID in a manner that can be displayed to the user of the determination support device 210. The output unit is, for example, a display device. An example of a user is an expert in inspection work and the like. The redetermination result of the state of the structure output by the output unit is input by the input unit (not shown) of the feedback unit 231. The input unit is, for example, a keyboard, a touch panel, or the like.

The feedback unit 231 may be an interface for outputting information about the re-judgment structure or inputting a re-judgment result.

The reliability calculation unit 214 inputs the re-judgment result of the re-judgment structure and the learning data with reliability of the learning data storage unit 221, and updates the reliability of the determination result according to the re-judgment result. Specifically, when the determination result for the re-determination structure is changed, the reliability calculation unit 214 calculates the reliability of the changed determination result. The calculated reliability Rc is expressed by the following formula.
Rc ＝ Rp / Rn
Here, Rp represents the reliability of the judgment result before the change for a certain structure, and Rn represents the number of different judgment results.

Specifically, when the state of a certain structure is changed from "3" to "4" in the training data, the judgment of the state of a certain structure is divided into "3" and "4". It will be. Rp has a reliability of 1 (initial value) when the state of a certain structure is "3", and Rn has a reliability Rc = 1/2 = calculated because there are two different judgments. It becomes 0.5.

The calculation of reliability is not limited to the above formula. The reliability calculation unit 214 may calculate the reliability by another formula. For example, the reliability Rca may be calculated using the following formula.
Rca = Rp / (Cnt + 1)
Rp represents the reliability of the judgment result before the change for a certain structure, and Cnt represents the number of changes of the judgment result for a certain structure. Specifically, when the state of the structure changes from "3" to "4" for the first time by re-judgment and the judgment result of the structure is changed for the first time, the reliability before the change of Rp is 1 (initial). As a value), Rca is 1 / (1 + 1) = 0.5.

If the re-judgment extraction unit 213 extracts the re-judgment structure as a re-judgment structure, but the judgment result from the feedback unit 241 does not change, the reliability of the judgment result by the reliability calculation unit 214 is not calculated.

The reliability calculation unit 214 stores the reliability calculated by changing the determination result in the learning data storage unit 221 and updates the learning data with reliability. FIG. 7 is a data sheet showing an example of learning data with reliability in the second embodiment. In the example shown in FIG. 7, the state of the structure for the structure ID “100” is changed from “3” to “4”, and the reliability is also updated from “1” to “0.5”.

When the re-judgment extraction unit 213 extracts a plurality of re-judgment structures, the operation from reading the learning data with reliability to calculating the reliability is repeated a plurality of times according to the number of the extracted re-judgment structures. The calculated reliability is output to the learning data storage unit 221. Alternatively, an embodiment may be adopted in which the accuracy of the reliability of the determination model is improved by repeating any number of times determined by the user.

Next, the operation of the determination support device 210 according to the second embodiment will be described.

FIG. 8 is a flowchart showing the operation of the determination support device 210 according to the second embodiment.

The determination model generation unit 211 reads out the learning data with reliability stored in the learning data storage unit 221 (step S21).

The determination model generation unit 211 generates a determination model showing the relationship between the inspection record for each structure, the determination result, and the reliability by using the learning data with reliability (step S22).

The certainty calculation unit 212 calculates the certainty for the state of each structure by using the judgment model generated by the judgment model generation unit 211 and the learning data with reliability of the learning data storage unit 221 (step S23). ).

The re-judgment extraction unit 213 inputs the certainty of each structure calculated by the certainty calculation unit 212, and extracts the structure to be re-determined from the learning data (step S24).

The feedback unit 231 outputs the inspection record and the state of the structure related to the extracted re-judgment structure, and receives the re-judgment result for the re-judgment structure (step S25).

The reliability calculation unit 214 calculates a new reliability based on the re-determination result using the reliability before the change of the re-determination structure (step S26). The reliability calculation unit 214 updates the reliability corresponding to the redetermination structure stored in the learning data storage unit 221.

When the re-judgment extraction unit 213 extracts a plurality of re-judgment structures, steps S21 to S24 are repeated for each of the extracted re-judgment structures (step S27).

Next, the characteristic configuration of the second embodiment will be described.

FIG. 9 is a block diagram showing a characteristic configuration of the second embodiment. Referring to FIG. 9, the determination support device 210 (also referred to as an information processing device) includes a determination model generation unit 211, a certainty calculation unit 212, and a re-determination extraction unit 213. The determination model generation unit 211 generates a determination model showing the relationship between the inspection record for each structure and the determination result. The certainty calculation unit 212 calculates the certainty indicating the ambiguity of the judgment result for each structure based on the judgment model, the inspection record and the judgment result. The re-judgment extraction unit 213 extracts a structure to be re-determined from the training data based on the certainty. The re-judgment extraction unit 213 outputs the inspection record of the extracted re-judgment structure and the judgment result (state of the structure), and receives the re-judgment result for the re-judgment structure. The reliability calculation unit 214 calculates a new reliability based on the re-judgment result using the reliability before the change of the re-judgment structure.

Next, the effect of the second embodiment will be described.

According to the determination support device of the second embodiment, it is possible to efficiently review the determination result based on the inspection record of the structure as in the case of the determination support device of the first embodiment.

The reason is that each component of the determination support device functions as follows. That is, the determination model generation unit 211 generates a determination model showing the relationship between the inspection record for each structure included in the learning data with reliability and the determination result. The certainty calculation unit 212 calculates the certainty indicating the ambiguity of the judgment result for each structure in the learning data with reliability based on the generated judgment model, the inspection record and the judgment result. The re-judgment extraction unit 213 extracts a structure to be re-determined from the learning data based on the calculated certainty.

As a result, the determination support device can efficiently identify the structure for which the determination result is ambiguous and it is preferable to re-determine from the inspection result for each structure and the learning data including the determination result. It is possible to shorten the time required for reviewing.

Further, according to the determination support device of the second embodiment, it is possible to efficiently review the determination result based on the inspection record of the structure. The reason is that the information related to the re-judgment structure extracted by the re-judgment extraction unit 213 is displayed as the inspection record and the state of the structure related to the re-judgment structure extracted by the feedback unit 231, and the re-judgment structure is displayed. This is because the re-judgment result for is accepted.

Further, according to the determination support device of the second embodiment, the accuracy of the reliability of the determination result can be improved. The reason is that the reliability calculation unit 214 calculates a new reliability using the reliability before the change of the re-judgment structure based on the re-judgment result, and the re-judgment structure stored in the learning data storage unit 221. This is because the reliability corresponding to the object is updated.

Further, the determination support device of the second embodiment efficiently determines the accuracy of the reliability even when the person who performs the re-determination is divided into a plurality of persons or the reliability of one re-determination is low. Can be enhanced. The reason is that the determination support device of the second embodiment can repeatedly generate a determination model and redetermine the determination result for each extracted redetermination structure.

(Hardware configuration)
FIG. 10 is a diagram showing a hardware configuration in which the determination support device according to the first and second embodiments is realized by a computer.

In the first and second embodiments, each component of the determination support device indicates a block of functional units. A part or all of each component of the determination support device is realized by, for example, an arbitrary combination of the computer 600 and the program shown in FIG. As an example, the computer 600 includes the following configuration.
CPU (Central Processing Unit) 601
ROM (Read Only Memory) 602,
RAM (Random Access Memory) 603,
Program 604 loaded into RAM 603,
Storage device 605 for storing the program 604,
Drive device 607, which reads and writes the recording medium 606,
Communication interface 608 for connecting to the communication network 609,
Input / output interface 610 for inputting / outputting data,
Bus 611 connecting each component
Each component of the determination support device is realized by the CPU 601 acquiring and executing the program 604 that realizes these functions. The program 604 that realizes the functions of each component is stored in, for example, a storage device 605 or a RAM 603 in advance, and is read by the CPU 601 as needed. The program 604 may be supplied to the CPU 601 via the communication network 609, or may be stored in the recording medium 606 in advance, and the drive device 607 may read the program and supply the program to the CPU 601.

There are various modifications in the method of realizing the determination support device. For example, the determination support device may be realized by any combination of the computer 600 and the program, which are separate for each component. Further, a plurality of components included in the determination support device may be realized by any combination of one computer 600 and a program.

Further, a part or all of each component of the determination support device is realized by other general-purpose or dedicated circuits, processors, etc., or a combination thereof. These may be composed of a single chip or may be composed of a plurality of chips connected via a bus. Further, a programmable logic device such as FPGA (Field-Programmable Gate Array) may be used instead of the computer 600.

Further, a part or all of each component of the determination support device may be realized by a combination of the above-mentioned circuit or the like and a program.

Further, when a part or all of each component of the determination support device is realized by a plurality of computers, circuits, etc., the plurality of computers, circuits, etc. may be centrally arranged or distributed. good. For example, a computer, a circuit, or the like may be realized as a form in which each is connected via a communication network, such as a client and server or cloud computing.

Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above embodiments. The configurations of the embodiments can be modified in various ways as can be understood by those skilled in the art within the scope of the present invention.

110 Judgment support device 111 Judgment model generation unit 112 Confidence calculation unit 113 Rejudgment extraction unit 121 Learning data storage unit 210 Judgment support device 211 Judgment model generation unit 212 Confidence calculation unit 213 Rejudgment extraction unit 214 Reliability calculation unit 221 Learning Data storage unit 231 Feedback unit 600 Computer 601 CPU
602 ROM (Read Only Memory)
603 RAM
604 Program 605 Storage device 606 Recording medium 607 Drive device 608 Communication interface 609 Communication network 610 Input / output interface 611 Bus

Claims (8)

A judgment model generation means that generates a judgment model showing the relationship between the inspection record and the judgment result by using the inspection record for each structure and the judgment result as learning data.
A certainty calculation means for calculating the certainty indicating the ambiguity of the judgment result for each of the structures based on the judgment model, the inspection record, and the judgment result.
A re-judgment extraction means for extracting a structure to be re-determined from the learning data based on the certainty.
Information processing device equipped with. The certainty calculation means is
When the determination result of the structure is divided into n classes (n is a natural number), the occurrence probability of being determined to be the class.

On the basis of the,

Calculate the entropy of the formula represented by, and obtain the certainty for each structure.
The information processing apparatus according to claim 1. A learning data storage means to which the reliability of the determination result is attached for each structure of the learning data,
A reliability calculation means for calculating the reliability of the structure to be re-determined by inputting the re-judgment result for the structure to be re-determined is further provided.
The information processing apparatus according to claim 1 or 2. The re-judgment extraction means extracts the structure associated with the highest certainty from the certainty of each structure.
The information processing apparatus according to any one of claims 1 to 3. The re-judgment extraction means extracts a plurality of structures from the certainty of each structure according to the magnitude of the certainty.
The information processing apparatus according to any one of claims 1 to 3. The re-judgment extraction means extracts a structure associated with a certainty degree larger than a set threshold value from the certainty degree for each structure.
The information processing apparatus according to any one of claims 1 to 3. Information processing equipment
Using the inspection record for each structure and the judgment result as learning data, a judgment model showing the relationship between the inspection record and the judgment result is generated.
Based on the determination model, the inspection record, and the determination result, the degree of certainty indicating the ambiguity of the determination result is calculated for each of the structures.
Based on the certainty, a structure to be re-determined is extracted from the learning data.
An information processing method characterized by this.
Using the inspection record for each structure and the judgment result as learning data, a judgment model showing the relationship between the inspection record and the judgment result is generated.
Based on the determination model, the inspection record, and the determination result, the degree of certainty indicating the ambiguity of the determination result is calculated for each of the structures.
Based on the certainty, a structure to be re-determined is extracted from the learning data.
A program that lets a computer do things.

Images