JP2021143575A - Damage prediction apparatus, learning model, and method of generating learning model - Google Patents

Abstract

To provide a damage prediction apparatus which predicts damage when external force is applied to a target structure without performing structure analysis every time.SOLUTION: A damage prediction apparatus comprises input means and output means. The input means accepts input including a map of degradation level distribution or soundness determination distribution in a beam structure obtained from inspection results of a target structure, a planar shape of the beam structure in the target structure, and unassumed external force applied to the target structure, as input date for prediction means utilizing a learnt model provided with teaching data to perform machine learning, the teaching data including input data and output data, the input data taking the map of degradation level distribution or soundness determination distribution in the beam structure of a pier or a bridge, the planar shape of the beam structure, and the external force applied to the beam structure, and the output data taking the information indicating damage condition of the beam structure obtained from a structure analysis result when the external force is applied to the beam structure, and the output means outputs information indicating damage condition predicted when unassumed external force is applied to the target structure.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for predicting damage when an external force is applied to a pier or a beam structure of a bridge. In addition, in this specification, a bridge means a fictitious structure for a vehicle defined in Article 2 of the Road Traffic Act to pass on the water surface or on land.

Techniques for applying machine learning to the construction or maintenance of structures are known. For example, in Patent Document 1, the strength of the ground, weathering alteration, crack interval, crack state, strike dip of crack, presence / absence of spring water, amount of spring water, deterioration status, etc. are monitored, and the support pattern is determined by inputting these. However, it is described that this support pattern is output.

The technique described in Patent Document 1 is for performing construction management of a structure, and cannot predict damage to a pier or a bridge. Generally, in order to predict or evaluate damage when an external force is applied to the beam structure of a pier or a bridge, a detailed periodic inspection and diagnosis described in Non-Patent Document 1 is carried out, and a structural analysis based on the result is performed. Is done. Since detailed periodic inspection diagnosis and structural analysis based on the results are costly, a method of relatively easily evaluating the residual proof stress from the deterioration degree judgment result obtained from the general inspection diagnosis has been proposed (Non-Patent Document 2). ).

JP-A-2017-117147

Supervised by the Port Bureau of the Ministry of Land, Infrastructure, Transport and Tourism, "Maintenance and Management Technical Manual for Port Facilities (Revised Edition)", Coastal Technology Research Center, July 2018, p. 296, p. 317 Shuhiko Uno and Mitsuyasu Iwanami "Application of Residual Yield Strength Evaluation Method Utilizing Deterioration Degree Judgment Results to Actual Piers" JSCE Proceedings B3 (Ocean Development), Vol. 74, No. 2, pp. I_55-I_60, 2018.

In the technique described in Non-Patent Document 2, in order to predict the damage of the pier while considering the influence of the degree of deterioration, the structural analysis of the target pier had to be performed each time.

On the other hand, the present invention provides a technique for predicting damage when an external force is applied to a target structure without performing structural analysis each time.

In the present invention, a map of deterioration degree distribution or soundness judgment distribution in a beam structure of a pier or a bridge, a planar shape of the beam structure, and an external force applied to the beam structure are used as input data, and the external force is applied to the beam structure. It is a target as input data for a prediction means using a trained model that has been machine-learned by giving teacher data whose output data is information indicating the damaged state of the beam structure obtained from the structural analysis result at the time of Map of deterioration degree distribution or soundness judgment distribution in the beam structure obtained from the inspection result of the structure, plane shape of the beam structure in the target structure, and input of unexpected force applied to the target structure. Damage degree prediction having an input means for receiving the above and an output means for outputting information indicating a damage state predicted when the unexpected force is applied to the target structure obtained from the trained model. Provide the device.

The output means may output at least one or more information indicating the damage state, the range of the damage, the degree of the damage, and the rate of the damage.

The output means may output a map of the damage degree distribution based on the damage classification in the beam structure as information indicating the damage state, and the map of the damage degree distribution may be mapped on the diagram showing the beam structure. ..

The damage degree predicting device has a predicting means for predicting the deterioration degree distribution or the soundness judgment distribution at a future time point according to a probability model using the deterioration degree distribution or the soundness judgment distribution with respect to the beam structure, and has the output. The means may output the information indicating the damage state predicted when the unexpected force is applied, which is obtained by inputting the estimated deterioration degree distribution or the soundness judgment distribution into the trained model. good.

The map of the degree of deterioration distribution or the soundness determination distribution may have a common spatial size for a plurality of beam structures.

Further, the present invention comprises a map of deterioration degree distribution or soundness judgment distribution in a beam structure of a pier or a bridge, a planar shape of the beam structure, an input layer into which an external force applied to the beam structure is input, and the beam structure. An output layer that outputs information indicating the damaged state of the beam structure obtained from the structural analysis result when the external force is applied to the beam, a map of the deterioration degree distribution or the soundness judgment distribution in the beam structure, and the beam structure. Teacher data in which the planar shape and the external force applied to the beam structure are input data, and the information indicating the damage state of the beam structure obtained from the structural analysis result when the external force is applied to the beam structure is output data. A map of the degree of deterioration distribution or soundness judgment distribution in the beam structure obtained from the inspection results of the target structure, the plan shape of the beam structure, and the relevant Data showing the unexpected force applied to the beam structure is acquired, and the map of the deterioration degree distribution or the soundness judgment distribution in the beam structure of the target structure, the planar shape of the beam structure, and the assumption given to the beam structure. Data indicating the external force is input to the input layer, the calculation is performed in the intermediate layer, and information indicating the damaged state of the beam structure when the expected external force is applied to the beam structure is output from the output layer. Provides a trained model for the computer to work.

Further, the present invention is a model generation method for generating a trained model to be applied to a damage degree prediction device for predicting the damage degree of a pier or a bridge, and is a deterioration degree distribution or a soundness determination distribution in the beam structure. Map, the planar shape of the beam structure, the external force applied to the beam structure, and the teacher data including information indicating the damage state of the beam structure obtained from the structural analysis result when the external force is applied to the beam structure. The step of acquiring the above and the map of the deterioration degree distribution or the soundness judgment distribution in the beam structure, the planar shape of the beam structure, and the external force applied to the beam structure are input using the teacher data, and the beam structure is input. Provided is a method for generating a trained model having a step of generating a trained model that outputs information indicating a damaged state of the beam structure obtained from the structural analysis result when the external force is applied to the beam.

The information indicating the damage state of the beam structure may include an image showing the distribution of the degree of damage based on the damage category on the diagram showing the beam structure, which corresponds to the map of the degree of deterioration distribution or the soundness determination distribution.

The map of the degree of deterioration distribution or the soundness determination distribution may have a common spatial size for a plurality of beam structures.

According to the present invention, damage when an external force is applied to a target structure can be predicted in a shorter time than in the case of performing structural analysis each time.

The figure which illustrates the functional structure of the damage degree prediction apparatus 1 which concerns on one Embodiment. The figure which illustrates the functional structure of the learning apparatus 2. The figure which illustrates the hardware composition of the damage degree prediction apparatus 1. FIG. The figure which illustrates the structure of the machine learning model 9 which concerns on one Embodiment. A flowchart illustrating the process of preparing teacher data. The figure which exemplifies the judgment criteria of the degree of deterioration. The figure which exemplifies the judgment criteria of soundness. The figure which exemplifies the map of the degree of deterioration distribution. The figure which illustrates the information which shows the planar shape of a beam structure. The figure which illustrates the external force condition. The figure which illustrates the result of the structural analysis. The flowchart which illustrates the learning process of the machine learning model 9. A sequence chart illustrating the damage degree prediction process. The figure which illustrates the information which shows the predicted damage state. The figure which illustrates the deterioration degree distribution and plane shape data which concerns on a modification.

1. 1. Configuration FIG. 1 is a diagram illustrating the functional configuration of the damage degree prediction device 1 according to the embodiment. The damage degree prediction device 1 is a system that predicts damage when an external force is applied to a target structure. The target structure is a beam structure of a pier or a bridge. The damage degree prediction device 1 predicts the damage of the target structure by using machine learning without relying on the structural analysis of the structure.

The damage degree prediction device 1 is a client device or user terminal that receives an instruction or data input from a user and outputs information according to the input instruction or data. The damage degree prediction device 1 includes a storage means 11, an input means 12, a prediction means 13, and an output means 14. The storage means 11 stores various data and programs. In this example, the data stored in the storage means 11 includes the trained model M. The trained model M is a machine learning model in which teacher data is given to perform machine learning. The trained model M is, for example, a model using a neural network, a random forest, a decision tree, a support vector machine, or deep learning. This teacher data can be used as input data (or explanatory variables) (1) Map of deterioration degree distribution or soundness judgment distribution in the beam structure of a pier or bridge.
It includes at least the planar shape of the beam structure and (3) the external force applied to the beam structure. The teacher data also includes at least information indicating the damaged state of the beam structure obtained from the structural analysis result when the external force is applied to the beam structure as output data (or objective variable). This teacher data is data about at least one pier or bridge. The trained model M is generated by the learning device 2.

FIG. 2 is a diagram illustrating the functional configuration of the learning device 2. The learning device 2 gives teacher data to the model to perform machine learning, and obtains a trained model M. The learning device 2 has a storage means 21 and a learning means 22. The storage means 21 stores the teacher data and the trained model (or the model before learning). The learning means 22 gives teacher data to the model and causes machine learning to obtain a trained model. The learning device 2 outputs the trained model M to the damage degree prediction device 1.

See FIG. 1 again. The input means 12 accepts the input of explanatory variables for input to the trained model M. The prediction means 13 gives an explanatory variable input via the input means 12 to the trained model M, and obtains a corresponding objective variable. The explanatory variables input to the trained model M are
(A) Map of deterioration degree distribution or soundness judgment distribution in the beam structure obtained from the inspection result of the target structure,
It includes at least the planar shape of the beam structure in the target structure and (c) the unexpected force applied to the target structure. For these explanatory variables, the objective variable (ie, the result of the prediction) obtained from the trained model M contains at least information indicating the damage state of the target structure. The output means 14 outputs the information indicating the damaged state of the structure obtained by the predictor means 13.

FIG. 3 is a diagram illustrating a hardware configuration of the damage degree prediction device 1. The damage degree prediction device 1 is a computer device having a CPU (Central Processing Unit) 101, a memory 102, a storage 103, an input device 104, and a display device 105, for example, a personal computer. The CPU 101 is a control device that executes a program, performs various calculations, and controls other hardware elements in the damage degree prediction device 1. The memory 102 is a main storage device that functions as a work area when the CPU 101 executes a program, and includes, for example, a RAM. The storage 103 is a non-volatile storage device that records various programs and data, and includes, for example, an SSD (Solid State Drive) or an HDD (Hard Disk Drive). The input device 104 is a device for a user to input an instruction or data to the damage degree prediction device 1, and includes, for example, at least one of a touch screen, a keyboard, and a microphone. The display device 105 is a device that presents information to the user, and includes, for example, an OLED (Organic Light Emitting Diode) display or an LCD (Liquid Crystal Display).

In this example, in the program stored in the storage 103, a program for causing the computer device to function as the damage degree prediction device 1 (hereinafter referred to as “client program”) is stored. When the CPU 101 executes the client program, the function as the damage degree prediction device 1 is implemented in the computer device. At least one of the memory 102 and the storage 103 is an example of the storage means 11 in a state where the CPU 101 is executing the client program. The CPU 101 is an example of the prediction means 13. The input device 104 is an example of the input means 12. The display device 105 is an example of the output means 14.

Although the hardware configuration of the learning device 2 is not shown, it is a computer device having the same hardware configuration as the damage degree prediction device 1. The learning device 2 may be mounted on the same computer device as the damage degree prediction device 1, or may be mounted on a computer device different from the damage degree prediction device 1, for example, a server device on a network.

2. An outline of the machine learning model used in the present embodiment will be described prior to the description of the operation of the motion damage degree prediction device 1.

FIG. 4 is a diagram illustrating the structure of the machine learning model 9 according to the embodiment. In this example, the machine learning model has an input layer 91, an intermediate layer 92, and an output layer 93. In the input layer 91, a map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure of the pier or the bridge, the planar shape of the beam structure, and the external force applied to the beam structure are input. The output layer 93 outputs information indicating a damaged state of the beam structure obtained from the structural analysis result when the external force is applied to the beam structure. Information indicating the damage status includes at least one or more information on the extent of damage, the degree of damage, and the rate of damage. More specifically, the information indicating the damage state includes a map of the damage degree distribution. The intermediate layer 92 is a layer located between the input layer 91 and the output layer 93, and includes parameters for obtaining data output via the output layer 93 from data input via the input layer 91. In the intermediate layer 92, the map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure, the planar shape of the beam structure, and the external force applied to the beam structure are used as input data, and the external force is applied to the beam structure. The parameters are learned using the teacher data whose output data is the information indicating the damaged state of the beam structure obtained from the structural analysis result at that time.

Hereinafter, the operations of the damage degree prediction device 1 and the learning device 2 will be described. The operation of the damage degree prediction device 1 and the learning device 2 is roughly divided into three stages: preparation of teacher data, generation of a trained model M (that is, learning), and prediction using the trained model M. Hereinafter, each process of teacher data preparation, learning, and prediction will be described. In the following, functional elements such as the learning means 22 may be described as the main body of the process, which means that the hardware elements such as the CPU 101 execute the process in cooperation with other hardware elements according to the program. means. Further, in the following, a pier will be described as an example of the target structure.

2-1. Preparation of teacher data FIG. 5 is a flowchart illustrating a process of preparing teacher data. Here, structural analysis is performed on piers of various sizes by giving each of them a plurality of analysis conditions (for example, conditions relating to deterioration degree distribution and external force). Known structural analysis software (for example, "Engineer's Studio" manufactured by FORUM8 Co., Ltd.) is used for structural analysis. As a result of the structural analysis, data on the extent of damage, the degree of damage, and the rate of damage are obtained. These data are used as teacher data.

In step S101, the learning device 2 (learning means 22) acquires information indicating the deterioration degree distribution of the pier. Here, the data format of the information indicating the degree of deterioration may be any. In one example, the information indicating the degree of deterioration includes information for identifying a partial beam in the beam structure of the target pier (for example, information indicating an identifier of the partial beam or information indicating the relative position of the partial beam) and the degree of deterioration of the beam. It is a combination with the information shown. In one example, the deterioration degree distribution information is character string data indicating this combination. The degree of deterioration is information indicating the degree of deterioration of the beam, and in this example, it is information obtained by an inspector inspecting an actual pier according to a predetermined inspection standard or by an automatic determination by an inspection device.

FIG. 6 is a diagram illustrating a criterion for determining the degree of deterioration of the pier (cited from Non-Patent Document 1). In this example, a reference is provided for each of the side surface portion and the lower surface portion of the beam. For both the side surface portion and the lower surface portion, four categories of a, b, c, and d are defined as the degree of deterioration. The degree of deterioration a is the state where the degree of deterioration is the heaviest (or the degree of deterioration is high), and the degree of deterioration d is the state where the degree of deterioration is the lightest or no deterioration (or the degree of deterioration is low). The beam structure is divided into a plurality of partial beams, and the degree of deterioration is determined for each partial beam. The inspector observes the lower surface portion and the side surface portion of the target partial beam in the beam structure, and judges the degree of deterioration of the partial beam by comparing with the judgment criteria. In this example, the inspector determines the degree of deterioration of the lower surface portion and the degree of deterioration of the side surface portion of the partial beam, respectively, and adopts a heavier degree of deterioration as the degree of deterioration of the partial beam. For example, when the degree of deterioration of the lower surface is c, the degree of deterioration of one of the two side surfaces is b, and the degree of deterioration of the other is d, the degree of deterioration of the partial beam is b.

If the target structure is a bridge (that is, a road bridge) instead of a pier, the index of soundness is used instead of the degree of deterioration.

FIG. 7 is a diagram illustrating the criteria for determining soundness (quoted from the road bridge periodic inspection procedure “Road Bureau, Ministry of Land, Infrastructure, Transport and Tourism 2019.2”). In this example, the soundness of the road bridge is defined as four categories I, II, III, and IV. Category I is the most healthy (or lightly deteriorated) state, and Category IV is the most severely deteriorated state. The inspector observes the target partial beam in the beam structure and judges the soundness of the partial beam by comparing it with the judgment criteria. The criteria for determining the degree of deterioration and soundness illustrated in FIGS. 6 and 7 are merely examples, and other criteria may be used. The degree of deterioration and soundness are not limited to four, and may be classified into five or more.

See FIG. 5 again. In step S102, the learning means 22 converts the deterioration degree distribution acquired in step S101 into a map (that is, an image). This map contains an image showing a standardized beam arrangement and information showing the degree of deterioration of the corresponding beam on the image.

FIG. 8 is a diagram illustrating a map of the degree of deterioration distribution. In this example, the map has a beam structure that combines a plurality of beams in the vertical and horizontal directions. For example, the horizontal direction indicates a direction parallel to the shore, and the vertical direction indicates a direction perpendicular to the shore. Hereinafter, the rightward direction is defined as the x direction, and the upward direction is defined as the y direction. In this figure, the gray part indicates a beam joint (that is, a part where two beams intersect) or a pile head (that is, a part supported by a pile from below). The beam joint, the part sandwiched between the pile heads, or the end of the beam (not the beam joint) (that is, the part sandwiched between the two gray-painted parts, or the end of the beam (not the beam joint)) Each is a partial beam. In this map, all partial beams have a common size. That is, this map does not contain information that indicates the actual size of the beam structure. This is the purpose of "showing a standardized beam arrangement".

In this map, the degree of deterioration of each partial beam is represented by color (color cannot be expressed in this figure, so it is distinguished by hatching instead). For example, the portion of the degree of deterioration a is represented by red, the portion of the degree of deterioration b is represented by yellow, the portion of the degree of deterioration c is represented by green, and the portion of the degree of deterioration d is represented by blue.

This map has a common spatial size for multiple beam structures so that different piers of different sizes can be processed with a common image size. For example, this map has the largest spatial size in the world to record the degree of deterioration of a pier. As an example, the world's largest pier in the direction perpendicular to the shore has 1,000 beams in the direction parallel to the shore, and the world's largest pier in the direction parallel to the shore has 500 beams in the direction perpendicular to the shore. With beams, this map has an image corresponding to a beam structure of 1,000 rows x 500 columns.

For example, when the target is a pier having a beam structure of 5 rows × 7 columns, in the learning means 22, the reference portion (for example, the upper left partial beam) of the beam structure is the reference position (for example, the upper left partial beam) on the map. Color the partial beam so that it corresponds to the upper left partial beam). The part outside the beam structure of interest in the map remains uncolored and colorless. It should be noted that instead of the map of the degree of deterioration having a shared spatial size for a plurality of beam structures, each beam structure may have an individual spatial size.

See FIG. 5 again. In step S103, the learning means 22 acquires information indicating the planar shape of the target beam structure. Here, any data format of information indicating the planar shape of the beam structure may be used. In one example, the information indicating the planar shape is a combination of information indicating a partial beam in the beam structure of the target pier and information indicating the size of the beam.

FIG. 9 is a diagram illustrating information indicating the planar shape of the beam structure. Here, FIG. 9A shows the identification number of the partial beam. Identification numbers are assigned to the vertical and horizontal partial beams, respectively. The table of FIG. 9B includes information indicating the size of the target beam structure (the number of partial beams in each of the vertical and horizontal directions) and information indicating the dimensions of each partial beam. This beam structure is composed of 80 partial beams in the x direction and 60 partial beams in the y direction. For example, the beam of beam number 101 has a length of 0.95 m in the x direction and a length of 0.50 m in the y direction.

See FIG. 5 again. In step S104, the learning means 22 acquires the external force condition applied to the target beam structure.

FIG. 10 is a diagram illustrating external force conditions at the pier. The external force condition in the bridge may be defined separately (exemplification is omitted). The rows of this table show the external force conditions, and the columns show the explanatory variables. In this example, five types of explanatory variables are used: static (sea ← land), static (sea → land), L1, L2-1, and L2-2. "Static (sea ← land)" represents the traction force of a given ship. "Static (sea → land)" indicates the berthing force of a given vessel. “L1” indicates a level 1 seismic motion, that is, an earthquake with a high probability of occurring during the service period. “L2” indicates a level 2 earthquake motion, that is, an earthquake that is unlikely to occur during the service period but emits a large amount of energy, and L2-1 is a type I: plate boundary type level 2 earthquake ground motion (for example, off the Pacific coast of the Tohoku region). (Earthquake, etc.). L2-2 indicates a type II: inland type level 2 seismic motion (for example, the Hyogo-ken Nanbu Earthquake). The static (sea ← land) and static (sea → land) described as external force conditions are sound as they act not only as the traction force and berthing force of the specified ship but also as the design load. It is also used to see if there is any damage that should not occur at the pier.

In this table, the value of each cell is either "0" or "1". A value "0" indicates that the external force is not applied, and a value "1" indicates that the external force is applied. For example, the external force condition "static (sea ← land)" indicates that only the traction force of the ship is given to the symmetrical beam structure, and no other external force is given. In this example, only an external force condition in which a single type of external force is applied is shown, but an external force condition in which a plurality of types of external forces are applied may be used. Specifically, for example, an external force condition in which the traction force of the ship and the level 1 seismic motion are given at the same time may be used.

See FIG. 5 again. In step S105, the learning means 22 applies the external force indicated by the external force condition acquired in step S104 to the beam structure having the planar shape acquired in step S103 and the deterioration degree distribution acquired in step S101. Obtain the structural analysis result when assuming. The structural analysis may be performed by the learning means 22 itself, or may be performed by a device other than the learning device 2.

FIG. 11 is a diagram illustrating the results of structural analysis. In this example, the damage degree distribution and the damaged area ratio data are obtained as a result of the structural analysis. In one example, the results of the structural analysis are mapped onto an image of a standardized beam structure. In this map, all partial beams have a common size. That is, this map, like the example in FIG. 8, does not contain information indicating the actual size of the beam structure.

In this map, the damage to each partial beam is represented by a color (the color cannot be represented in this figure, so it is distinguished by hatching instead). For example, crack damage is shown in yellow, yield damage is shown in red, and ultimate damage is shown in black. Since these damages are displayed on the map of the beam structure, it can be said that this figure shows the extent and degree of damages.

The learning device 2 repeats the processes of steps S101 to S105 for each of the plurality of piers having various shapes and sizes, and accumulates data. The data accumulated in this way is used as teacher data in the learning process described below. In this teacher data, the map of the degree of deterioration distribution acquired in step S102, the planar shape acquired in step S103, and the external force condition acquired in step S104 are input data, and the structural analysis result obtained in step S105 is output data. Is. The learning device 2 stores these teacher data in the storage means 21.

2-2. Learning FIG. 12 is a flowchart illustrating the learning process of the machine learning model 9. The flow of FIG. 12 is started, for example, when a learning instruction is given from the administrator or the user of the damage degree prediction device 1. Alternatively, the flow of FIG. 12 may be started when a predetermined amount of teacher data is accumulated.

In step S201, the learning means 22 acquires teacher data used for learning. In this example, the learning means 22 acquires teacher data prepared in an external device different from the damage degree prediction device 1.

In step S202, the learning means 22 gives the machine learning model 9 teacher data to perform machine learning. This teacher data includes input data and output data. In one example, the input data is a map of the degree of deterioration distribution acquired in step S102, a planar shape acquired in step S103, and an external force condition acquired in step S104.

In step S203, the learning means 22 stores the machine learning model 9 in which the parameters of the intermediate layer 92 are learned by machine learning in the storage means 21 as the learned model M. The learning device 2 outputs the trained model M to the damage degree prediction device 1. The damage degree prediction device 1 stores the learned model M in the storage means 11. In this way, the damage degree prediction device 1 obtains the trained model M.

2-3. Prediction FIG. 13 is a sequence chart illustrating the damage degree prediction process. The degree of damage is predicted in response to a request from the user.

The user instructs the damage degree prediction device 1 to predict the damage degree. Specifically, the user inputs explanatory variables for the target structure. The input means 12 accepts the input of explanatory variables, specifically the following data (a) to (c) (step S301).
(A) Deterioration degree distribution or soundness judgment distribution in the beam structure obtained from the inspection results of the target structure.
(B) Planar shape of the beam structure in the target structure.
(C) Unexpected force applied to the target structure.

In one example, the damage degree predictor 1 itself provides a user interface for inputting the inspection result of the structure. The user inputs the inspection result into the damage degree prediction device 1. Alternatively, the damage degree prediction device 1 may receive the data of the deterioration degree distribution or the soundness determination distribution from the device that records the inspection result of the target structure. The data of the deterioration degree distribution or the soundness judgment distribution may have any data format as in the case of the teacher data.

The user inputs the planar shape data (for example, image data) of the target structure to the damage degree prediction device 1. Alternatively, the data of the planar shape of the structure is recorded in a database in an external device such as a server, and the damage degree prediction device 1 uses the database to indicate the planar shape of the target structure according to the user's instruction. Data may be acquired.

For the unexpected force, for example, the same conditions used in the teacher data are used. Alternatively, the damage degree prediction device 1 may use only the conditions specified by the user in the table of FIG. 10 as explanatory variables.

In step S302, the input means 12 generates a request for predicting the degree of damage. This request includes an explanatory variable input, acquired, or specified in step S301. In step S303, the input means 12 requests the predicting means 13 to predict the degree of damage.

In step S304, the prediction means 13 receives a request for prediction. In this example, the prediction means 13 converts the deterioration degree distribution data included in the request into a map (that is, an image). This map contains an image showing a standardized beam arrangement and information showing the degree of deterioration of the corresponding beam on the image. The process of converting the deterioration degree distribution data into a map may be performed by an external device such as a server. When the conversion to the map is performed in the external device, the damage degree prediction device 1 transmits the map of the deterioration degree distribution to the external device.

In this example, the information input to the damage degree prediction device 1 is not exactly the same as the explanatory variables input to the trained model M (because the deterioration degree distribution data is converted into a map), but the deterioration Considering that a map of the deterioration degree distribution can be obtained from the degree distribution data, it can be said that the input means 12 substantially inputs the following three types of explanatory variables into the trained model M.
(A) Map of deterioration degree distribution or soundness judgment distribution in the beam structure obtained from the inspection result of the target structure,
(B) The planar shape of the beam structure in the target structure, and (c) the unexpected force applied to the target structure.

In step S305, the prediction means 13 inputs the map of the degree of deterioration of the beam, the planar shape of the beam structure, and the unexpected force as explanatory variables into the trained model M. The trained model M outputs an objective variable, that is, information indicating a damage state according to the input explanatory variable. The information indicating the damage state is, in one example, data in the same format as the damage distribution map used as the teacher data illustrated in FIG. In step S306, the prediction means 13 acquires the objective variable output from the trained model M. The prediction means 13 transmits the acquired objective variable to the device that sends the request (step S307). In step S308, the output means 14 outputs information indicating a predicted damage state according to the objective variable output from the trained model M. The output means 14 may output the information output from the learned model M to the user as it is, or may output the information output from the learned model M after some processing has been performed on the information.

FIG. 14 is a diagram illustrating information indicating a predicted damage state output from the output means 14. In this example, the degree of damage (ie, the degree of damage) of each partial beam is divided into a plurality of divisions (that is, damage divisions), and each division is visually represented (for example, in different colors). The categories used here are common to those illustrated in FIG. 11, for example, crack damage is represented in yellow, yield damage is represented in red, and ultimate damage is represented in black.

Further, the information output from the output means 14 includes the damage rate. Here, the damage ratio means the ratio of the area of the partial beam having a predetermined damage to the partial beam of the target beam structure. For example, when the total area of the partial beam in the target beam structure is 5,000 m2 and the total area of the partial beam with crack damage at least partially is 3,000 m2, the crack area ratio is 60%. Is. Alternatively, the percentage of damage may be expressed as a percentage of the number of damaged partial beams rather than the area percentage.

The format of the information indicating the predicted damage state is not limited to the example of FIG. The information indicating the damage state may include at least one or more information on the extent of damage, the degree of damage, and the rate of damage. For example, the output means 14 does not distinguish between damage categories, and the presence or absence of damage may be represented by a binary value (that is, with or without damage). Alternatively, the output means 14 may output only the damage rate without outputting the damage state on the map.

As described above, according to the damage degree prediction device 1, it is possible to obtain information indicating the damage state without performing the structural analysis of the target structure each time.

3. 3. Modifications The present invention is not limited to the above-described embodiment, and various modifications can be performed. Hereinafter, some modification examples will be described. Two or more of the items described in the following modifications may be used in combination.

3-1. Degradation degree distribution data format In the embodiment, an example in which the deterioration degree distribution data is character string data including a plurality of pairs of a partial beam identifier and the deterioration degree of the partial beam has been described. However, the data format of the deterioration degree distribution is not limited to this. For example, the map (that is, image) data of the deterioration degree distribution as illustrated in FIG. 8 may be input to the learning device 2 as the deterioration degree distribution data. This applies to both learning and prediction processing. In this case, the process of converting the deterioration degree distribution data into a map may be omitted.

Alternatively, when the deterioration degree distribution data is input to the learning device 2 as a map, the arrangement of the beams in the map does not have to be standardized. That is, the map of the degree of deterioration distribution may include information on both the degree of deterioration and the planar shape of the beam structure.

FIG. 15 is a diagram illustrating data on the degree of deterioration and the planar shape according to this modified example. In this figure, unlike the example of FIG. 8, the arrangement of the beams is not standardized. That is, the size of the partial beam in this figure reflects the actual size of the partial beam. For example, the ratio of the lengths of the two partial beams in this figure corresponds to the ratio of the actual lengths of the partial beams. As shown in FIG. 15, this data contains information indicating the scale.

When the deterioration degree distribution data is input to the learning device 2 as a map, the learning means 22 corresponds to FIG. 15 from the deterioration degree distribution map and the plan shape of the beam structure (actual partial beam size). You may generate a map (reflecting) and input it into the machine learning model 9 or the trained model M. Further, the damage degree distribution map output by the output means 14 may reflect or scale the size of the partial beam in the same manner as the deterioration degree distribution map.

3-2. The probabilistic model predicting means 13 may use a predetermined probabilistic model to predict the degree of deterioration or soundness in the future. This probabilistic model is a model that estimates future state transitions based on probabilities based on past state transitions, and an example is a Markov chain model. In this example, the prediction means 13 makes a prediction by combining the trained model M and the probability model. More specifically, the prediction means 13 inputs the output from the trained model M into the probabilistic model. The prediction means 13 further inputs the current time and the future time to be predicted into the probability model. These times are input by the user, for example, via the input means 12.

3-3. Others When the target beam structure has a multi-layer beam structure, the target beam structure is not limited to any single layer. Learning and prediction processing may be applied to a plurality of layers.

The order of processing in the preparation, learning, and prediction of teacher data is not limited to that illustrated in FIGS. 5, 12, and 13. For example, the input order of the explanatory variables may be any.

The hardware configuration of the damage degree prediction device 1 is not limited to that exemplified in the embodiment. The damage degree prediction device 1 may have any hardware configuration as long as it can implement the required functions. For example, a part of the functions of the damage degree prediction device 1, for example, the prediction means 13 and the trained model M may be mounted on an external device such as a server device. In this case, the computer device having the input means 12 and the output means 14 cooperates with the external device and functions as the damage degree prediction device 1 according to the embodiment as a whole.

Software elements such as programs and trained models may be provided as recorded on a computer-readable recording medium such as a CD-ROM (Compact Disc Read only memory), or may be provided via a communication network such as the Internet. It may be downloaded.

1 ... Damage degree predictor, 2 ... Learning device, 9 ... Machine learning model, 11 ... Storage means, 12 ... Input means, 13 ... Prediction means, 14 ... Output means, 21 ... Storage means, 22 ... Learning means, 91 ... Input layer, 92 ... Intermediate layer, 93 ... Output layer, 101 ... CPU, 102 ... Memory, 103 ... Storage, 104 ... Input device, 105 ... Display device

Claims (9)

When the external force is applied to the beam structure, the map of the deterioration degree distribution or the soundness judgment distribution in the beam structure of the pier or the bridge, the planar shape of the beam structure, and the external force applied to the beam structure are used as input data. Inspection of the target structure as input data for the prediction means using a trained model that has been machine-learned by giving teacher data that uses information indicating the damaged state of the beam structure as output data obtained from the structural analysis results. A map of the degree of deterioration distribution or soundness judgment distribution in the beam structure obtained from the results, the planar shape of the beam structure in the target structure, and the input means for receiving the input of the unexpected force applied to the target structure. When,
A damage degree prediction device having an output means for outputting information indicating a damage state predicted when the unexpected force is applied to the target structure obtained from the trained model. The damage degree prediction device according to claim 1, wherein the output means outputs at least one or more information on the range of the damage, the degree of the damage, and the rate of the damage as information indicating the damage state. The output means outputs a map of the damage degree distribution based on the damage classification in the beam structure as information indicating the damage state.
The damage degree prediction device according to claim 2, wherein a map of the damage degree distribution is mapped on a diagram showing the beam structure. With respect to the beam structure, there is a predicting means for predicting the deterioration degree distribution or the soundness judgment distribution at a future time point according to the probability model using the deterioration degree distribution or the soundness judgment distribution.
The output means outputs information indicating a damage state predicted when the unexpected force is applied, which is obtained by inputting the estimated deterioration degree distribution or the soundness judgment distribution into the trained model. The damage degree prediction device according to any one of claims 1 to 3. The damage degree prediction device according to any one of claims 1 to 4, wherein the map of the deterioration degree distribution or the soundness determination distribution has a common spatial size for a plurality of beam structures. A map of the degree of deterioration distribution or soundness judgment distribution in the beam structure of a pier or a bridge, the planar shape of the beam structure, and the input layer into which the external force applied to the beam structure is input.
An output layer that outputs information indicating a damaged state of the beam structure obtained from the structural analysis result when the external force is applied to the beam structure, and
The map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure, the planar shape of the beam structure, and the external force applied to the beam structure are used as input data, and the structural analysis result when the external force is applied to the beam structure. It has an intermediate layer whose parameters have been learned using teacher data whose output data is information indicating the damaged state of the beam structure obtained from the above.
Obtain a map of the degree of deterioration distribution or soundness judgment distribution in the beam structure obtained from the inspection results of the target structure, the planar shape of the beam structure, and data showing the unexpected force applied to the beam structure.
A map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure of the target structure, the planar shape of the beam structure, and data indicating the unexpected force applied to the beam structure are input to the input layer, and the intermediate A trained model for operating a computer to calculate from the layer and output information indicating the damage state of the beam structure when the unexpected force is applied to the beam structure from the output layer. A model generation method that generates a trained model for application to a damage degree predictor that predicts the degree of damage to a pier or bridge.
The map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure, the planar shape of the beam structure, the external force applied to the beam structure, and the structural analysis result obtained when the external force is applied to the beam structure. Steps to acquire teacher data containing information indicating the damaged state of the beam structure, and
Using the teacher data, a map of the degree of deterioration distribution or the soundness judgment distribution in the beam structure, the planar shape of the beam structure, and the external force applied to the beam structure are input, and the external force is given to the beam structure. A method of generating a trained model having a step of generating a trained model that outputs information indicating a damaged state of the beam structure obtained from the structural analysis result at that time. The information indicating the damage state of the beam structure includes an image showing the distribution of the damage degree based on the damage classification corresponding to the map of the deterioration degree distribution or the soundness judgment distribution on the diagram showing the beam structure. How to generate a trained model as described in. The method for generating a trained model according to any one of claims 6 to 8, wherein the map of the degree of deterioration distribution or the soundness determination distribution has a common spatial size for a plurality of beam structures.

Images