JP7442059B2 - Structural information display device, structural design support device, and structural design support model learning device - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act On November 17, 2017, http://www. takenaka. co. jp/news/2017/11/04/index. html? Published on the website on 2017-11-17

Application of Article 30, Paragraph 2 of the Patent Act Published on the 14th page of the Nihon Keizai Shimbun daily on November 17, 2017 on November 17, 2017

Application of Article 30, Paragraph 2 of the Patent Act Published on the third page of the morning edition of Kensetsu Tsushin Shimbun on November 20, 2017, November 20, 2017

Application of Article 30, Paragraph 2 of the Patent Act Published on the second page of the morning edition of the Nikkan Kensetsu Sangyo Shimbun on November 27, 2017

Application of Article 30, Paragraph 2 of the Patent Act Published on page 11 of the morning edition of the Denki Shimbun on November 30, 2017, on November 30, 2017

The present invention relates to a structural information display device, a structural design support device, and a structural design support model learning device.

BACKGROUND ART Conventionally, a building structural design support system that supports structural design in construction work is known (for example, Patent Document 1).

Japanese Patent Application Publication No. 2009-26303

The structural design support system described in Patent Document 1 supports structural design by determining the strength of a building. However, the technique described in Patent Document 1 does not take structural design performance information into consideration, and therefore may not be appropriate.

In consideration of the above facts, the present invention aims to appropriately support structural design based on performance information regarding structural design of construction work.

In order to achieve the above object, the structural information display device of the present invention includes a reception unit that receives input information that is information about a building and display target information that is information that specifies track record information about the structure of the building; The information corresponding to the input information and the display target information received by the reception unit from a database storing structural design data representing information regarding the building and track record information regarding the structure of the building for each of the buildings. It is configured to include an information acquisition unit that acquires performance information of a building, and a display unit that displays performance information regarding the structure of the building acquired by the information acquisition unit. This allows information on past construction work to be displayed appropriately, making it possible to support structural design.

The information regarding the building of the structural information display device of the present invention includes at least one of structure type, building use, floor type, foundation type, and number of floors, and the track record information regarding the building structure includes information regarding the quantity of materials, It may include at least one of a verification ratio, a live load, an earthquake load, a column dominated area, and a column span. This allows information on past construction work to be displayed appropriately, making it possible to support structural design.

The structural design support device of the present invention includes an acquisition unit that acquires information regarding the building to be designed and information regarding the quantity of materials of the building to be designed, and information regarding the building to be designed acquired by the acquisition unit; Based on information regarding the quantity of materials of the building to be designed, information regarding the learning building, and a learned model machine-learned in advance from learning data representing information regarding the quantity of materials of the learning building, and an estimation unit that outputs information regarding whether the structural design of the building to be designed is appropriate. This makes it possible to appropriately support structural design based on a trained model that reflects past construction work performance information.

The information regarding the quantity of materials of the learning building in the learning data of the present invention is information regarding the quantity of materials corresponding to each location of the learning building, and the estimating unit is configured to include the acquiring unit Based on the information regarding the building to be designed and the learned model obtained by the above, it is possible to further output information regarding whether the structural design of each part of the building to be designed is appropriate. As a result, information on whether the structural design of each part of the building is appropriate is output based on the trained model that reflects past construction work performance information, so it can appropriately support structural design. I can do it.

Further, the structural design support device of the present invention includes an acquisition unit that acquires an initial condition of a building to be designed, and an acquisition unit that acquires an initial condition of the building to be designed, and a a simulation unit that performs at least one of a simulation related to structural calculation of the building and a simulation related to the quantity of building materials for the structural design of the building; and an evaluation of the structural design of the building according to the simulation results obtained by the simulation unit. a learning unit that performs reinforcement learning on a model for outputting the structural design of the building from the initial conditions of the building based on the value to obtain a trained model that outputs the structural design of the building from the initial conditions of the building; and a design output section that outputs a structural design of the building to be designed based on the learned model obtained by the learning section. Structural design can be appropriately supported by outputting multiple structural designs of a building to be designed based on evaluation values for structural designs according to simulation results.

The learning unit of the structural design support device of the present invention generates a model for outputting the next structural design of the building to be designed based on the previously obtained structural design of the building to be designed. Reinforcement learning is repeated to update the trained model, and the design output unit calculates the structural design of the building to be designed based on the previously obtained structural design of the building to be designed based on the trained model updated by the learning unit. It is possible to output the next structural design of the building to be designed. By using a learned model that is updated sequentially, structural design candidates can be obtained appropriately and efficiently.

The design output unit of the structural design support device of the present invention reads the performance information from a database storing performance information regarding building structures, and reads the performance information from each of the structural designs output based on the learned model. The structural design having a degree of similarity with the information above a threshold value can be output. By selecting structural design candidates that are similar to the track record information, it is possible to obtain structural design candidates that are similar to the structural designs of past construction works.

The structural design support model learning device of the present invention extracts information about the quantity of building materials from information about the building based on learning data representing information about the learning building and information about the quantity of materials for the learning building. It is configured to include a learning unit that performs machine learning on a model to be output and obtains a trained model that outputs information regarding the quantity of materials of the building from information regarding the building. Thereby, it is possible to obtain a trained model that outputs information regarding the quantity of building materials in accordance with information regarding the building and information regarding the quantity of building materials.

According to the present invention, it is possible to appropriately support structural design based on performance information regarding structural design of construction work.

1 is a block diagram showing a schematic configuration of an yield structure information display device according to a first embodiment; FIG. It is a figure showing an example of information stored in a structural design database. FIG. 3 is a diagram showing an example of the allowance corresponding to each location of a building. FIG. 3 is an explanatory diagram for explaining input information and display target information. FIG. 3 is a diagram illustrating an example of a structure information display processing routine in the first embodiment. FIG. 2 is a block diagram showing a schematic configuration of a structural design support system according to a second embodiment. FIG. 2 is a block diagram showing a schematic configuration of a client terminal according to a second embodiment. FIG. 2 is a block diagram showing a schematic configuration of a structural design device according to a second embodiment. FIG. 2 is a block diagram showing a schematic configuration of a structural design support device according to a second embodiment. FIG. 7 is a diagram showing an example of a learned model according to the second embodiment. FIG. 7 is an explanatory diagram for explaining processing in the second embodiment. It is a figure which shows an example of the learning process routine in 2nd Embodiment. It is a figure showing an example of an estimation processing routine in a 2nd embodiment. FIG. 2 is a block diagram showing a schematic configuration of a structural design support system according to a third embodiment. FIG. 3 is a block diagram showing a schematic configuration of a structural design support device according to a third embodiment. It is a figure showing an example of a structural design support processing routine in a 3rd embodiment. FIG. 7 is a diagram illustrating another configuration example of a trained model.

Embodiments of the present invention will be described in detail below.

<System configuration of the structural information display device according to the first embodiment>

FIG. 1 is a block diagram showing an example of the configuration of a structure information display device according to the first embodiment. Functionally, the structural information display device 10 can be represented by a configuration including an operating device 12, a computer 20, and a display device 30, as shown in FIG.

In the first embodiment, a structural information display device that displays track record information regarding past structural designs will be described. By displaying performance information regarding structural design, it is possible to support the structural designer's design work.

The operating device 12 receives operating information input by the user of the structural information display device 10.

The computer 20 includes a CPU (Central Processing Unit), a ROM (Read Only Memory) that stores programs for implementing each processing routine, a RAM (Random Access Memory) that temporarily stores data, and a memory that serves as a storage means. , network interface, etc. Functionally, the computer 20 includes a structural design database 22, a reception section 24, an information acquisition section 26, and a display section 28.

The structural design database 22 stores structural design data representing information regarding the building and track record information regarding the structure of the building for each of a plurality of buildings.

FIG. 2 shows an example of information stored in the structural design database 22. As shown in FIG. 2A, the building table showing information about the building includes structure type, building purpose, floor type, foundation type, and number of floors.

Further, as shown in FIG. 2(B), the yield table showing performance information regarding the structure of the building includes the yield of each material. Yield is the quantity of material per unit area or unit volume. For example, the yield table includes concrete yield, formwork yield, reinforcing bar (R/A) yield, and reinforcing bar (R/C) yield. Note that the yield is an example of information regarding the quantity of materials of the present invention.

Further, as shown in FIG. 2C, the structure information table showing performance information regarding the structure of the building includes verification ratio, live load, earthquake load, column dominated area, and column span. Note that the verification ratio is the ratio (N/F) between the member strength (F) of the building and the acting force (N) acting on the building.

The yield for each material in the yield table shown in FIG. 2(B) includes the yield corresponding to each location of the building. For example, as shown in Figure 3, the concrete allowances include the concrete allowance for columns, the concrete allowance for major beams, the concrete allowance for small beams, the concrete allowance for walls, and the concrete allowance for slabs. It is being

The receiving unit 24 receives input information that is information regarding a building and display target information that is information specifying performance information regarding the structure of the building. For example, the user inputs information about a building similar to the building he/she is designing. Further, the user inputs desired information about the building set as input information as display target information.

FIG. 4 shows an explanatory diagram for explaining input information and display target information accepted by the reception unit 24. For example, the user appropriately selects locations C and D on the display screen X shown in FIG. 4 to input input information. In the example shown in FIG. 4, the structure type, building purpose, floor type, foundation type, number of floors, etc. are input as input information.

Further, the user appropriately selects locations A, A1, B, and B1 on the display screen X shown in FIG. 4 to input display target information. In the example shown in FIG. 4, which one of the yield, verification ratio, live load, earthquake load, column controlled area, and column span is selected is input as input information.

For example, when "Yukake" is selected at locations A and B, the user can select concrete, formwork, reinforcing bars (R/A), reinforcing bars (R/C), and steel frame etc. can be further selected. Furthermore, the user can further select columns, girders, beams, braces, walls, and slabs at locations A1 and B1.

The information acquisition unit 26 acquires the input information and display target information received by the reception unit 24. Then, the information acquisition unit 26 acquires building performance information corresponding to the input information and display target information from the structural design database 22.

The display unit 28 displays performance information regarding the structure of the building acquired by the information acquisition unit 26. For example, as shown in Y in FIG. 4, the display unit 28 selects "Hyukake" as the display target information A and B, and "Concrete" and "Tally" as the display target information A1, In addition, when "reinforcing bars (R/A)" and "totalization" are selected as the display target information B1, a display screen Y as shown in FIG. 4 is displayed.

Thereby, the user can grasp the approximate cost of building similar to the building he/she is designing. For example, if the building being designed is S construction, the concrete yield is approximately between 0.4 and 0.8, and the steel reinforcement (R/A) yield is approximately 0.4 to 0.8, as shown in Figure 4 above. It can be seen that it is approximately between 0.1 and 0.8.

Further, for example, as shown in FIG. 4, performance information is extracted from the structural design database 22, and the performance information is drawn on the display screen Y in an appropriate graph. Further, when a plot point E on the display screen Y is specified, details of the construction work corresponding to the plot point are displayed. This allows the structural designer to refer to past performance information regarding the structural design of buildings, thereby providing appropriate support for structural design.

The display device 30 displays a display screen according to the control of the display unit 28. The display device 30 is realized by, for example, a display. For example, the display device 30 displays a display screen X, a display screen Y, etc., as shown in FIG. 4 above.

<Function of structure information display device>

Next, the operation of the structure information display device 10 will be explained. When structural design data is stored in the structural design database 22 and input information and display target information are input by the user via the operating device 12, the structural information display device 10 executes the structural information display processing routine shown in FIG. Execute.

In step S100, the receiving unit 24 receives input information and display target information.

In step S102, the information acquisition unit 26 acquires the input information and display target information received in step S100. Then, the information acquisition unit 26 acquires building performance information corresponding to the input information and display target information from the structural design database 22.

In step S104, the display unit 28 displays the performance information regarding the structure of the building acquired in step S102.

The display device 30 displays performance information regarding the structure of the building.

As described in detail above, in the first embodiment, building track record information corresponding to input information and display target information is displayed from a database storing structural design data for each of a plurality of buildings. This allows the performance information of past construction work to be appropriately displayed and supports structural design.

<Configuration of structural design support system according to second embodiment>

In the second embodiment, past construction work track record information is set as learning data, and a learned model is generated by machine learning based on the learning data. This embodiment differs from the first embodiment in that structural design is supported using the generated trained model.

FIG. 6 is a block diagram showing an example of the configuration of a structural design support system according to the second embodiment. Functionally, the structural design support system 210 can be represented by a configuration including a client terminal 211, a structural design device 212, and a structural design support device 213, as shown in FIG. The client terminal 211, the structural design device 212, and the structural design support device 213 are connected through a predetermined communication means 14.

[Client terminal]

The client terminal 211 includes an operating device 220, a computer 230, and a display device 240, as shown in FIG.

The operating device 220 receives operating information input by the user. For example, design information regarding a predetermined building input by a user who is a structural designer is accepted.

The computer 230 includes a CPU, a ROM that stores programs for implementing each processing routine, a RAM that temporarily stores data, a memory as a storage means, a network interface, and the like. Functionally, the computer 230 includes a reception section 232, a design section 234, a control section 236, and a transmission/reception section 238.

The reception unit 232 receives operation information input from the operation device 220.

The design section 234 designs a predetermined building according to the design information received by the reception section 232. For example, if a predetermined CAD (computer-aided design) system is installed in the computer 230, the design unit 234 generates CAD data representing a predetermined building based on design information input by the user. As a result, the user performs a building design work, and information regarding the building to be designed (hereinafter simply referred to as "designed building information") is generated. The designed building information includes, for example, the building scale, structure type, purpose, and various design data of the building to be designed.

The control unit 236 performs control processing according to the operation information received by the reception unit 232. For example, the control unit 236 controls the transmitting/receiving unit 238 to transmit designed building information to the structural design apparatus 212 and the structural design apparatus 212, which will be described later, according to the operation information received by the receiving unit 232.

The transmitter/receiver 238 transmits and receives information to and from the structural design device 212 under the control of the controller 236 . Further, the transmitting/receiving unit 238 transmits and receives information to and from the structural design support device 213 under the control of the control unit 236 .

The display device 240 displays a predetermined screen under the control of the control unit 236.

[Structural design equipment]

The structural design device 212 is constituted by a computer including a CPU, a ROM that stores programs for implementing each processing routine, a RAM that temporarily stores data, a memory as a storage means, a network interface, and the like. Functionally, the structural design device 212 includes a transmitting/receiving section 250 and a yield calculation section 252, as shown in FIG.

The transmitter/receiver 250 transmits and receives information to and from the client terminal 211 .

The yield calculation section 252 calculates the yield of the building to be designed according to the designed building information acquired by the transmitting/receiving section 250. For example, if the building to be designed is an RC structure, the construction unit 252 calculates the size of each pillar, the size of each beam, and the size of each floor of the building to be designed, which are included in the design building information. Calculate the concrete yield accordingly.

The transmitting/receiving unit 250 transmits the rate calculated by the rate calculation unit 252 to the client terminal 211 .

[Structural design support device]

The structural design support device 213 is composed of a computer including a CPU, a ROM that stores programs for implementing each processing routine, a RAM that temporarily stores data, a memory as a storage means, a network interface, etc. . As shown in FIG. 9, the structural design support device 213 functionally includes a structural design database 260, a learning section 262, a learned model storage section 264, a transmitting/receiving section 266, and an estimating section 268. ing. The transmitter/receiver 266 is an example of an acquisition unit of the present invention.

The structural design database 260 stores a building table that shows information about the learning building and a yield table that shows information about the yield of the learning building. The information stored in the structural design database 260 is used as learning data by a learning section 262, which will be described later. For example, for one building, the building structure type, building use, floor type, foundation type, and number of floors, the building's concrete yield, formwork yield, reinforcing bar (R/A) yield, and The yield of reinforcing bars (R/C) is used as one learning data.

The learning unit 262 performs machine learning on the model using learning data for each of the plurality of buildings stored in the structural design database 260 to generate a learned model. Specifically, the learning unit 262 performs machine learning on a model for outputting information regarding the yield from the designed building information.

When designing the structure of a building, it is necessary to perform the design work appropriately according to information about the building (building structure type, building purpose, floor type, foundation type, and number of floors). However, since structural design work is done manually, there may be omissions or omissions in the design content. In order to suppress this, studies are conducted by multiple structural designers. At this time, information regarding the yield, which is an example of information regarding the quantity of building materials, may be useful in considering the content of the structural design.

For example, based on performance information of past construction work, if the building size of a building is approximately "X", it is possible to associate in advance from the performance information a relationship such that the building size of the building is approximately "Y". Therefore, if the building size of the building to be designed is "X", and the construction rate is significantly larger than "Y" or significantly smaller than "Y", the design content is inappropriate. It can be determined that the possibility is high.

Therefore, in the second embodiment, a learned model is generated using learning data in which information about buildings and information about commissions are associated with each other. As a result, for example, information that the construction cost for a building of this size is about this amount is reflected in the trained model. Therefore, using the trained model, it is possible to support the determination of whether the structural design content of the building to be designed is appropriate.

Regarding the learned model of the second embodiment, for example, a neural network as shown in FIG. 10 can be used as an example of the model, and deep learning can be used as an example of the learning algorithm. The trained model of the second embodiment is a trained model that outputs information regarding the yield from the designed building information.

For example, based on the learning data stored in the structural design database 260, the learning unit 262 stores the designed building information for the learning data in which the designed building information is "X" and the walk rate is "A". A neural network is trained so that when "X" is input, the probability of receiving "A" is high.

The learning unit 262 then stores the obtained trained model in the trained model storage unit 264.

The trained model storage unit 264 stores the trained model generated by the learning unit 262.

The transmitting/receiving unit 266 receives the designed building information and the walk rate transmitted from the client terminal 211.

The estimating unit 268 determines whether the structural design of the building to be designed is appropriate, based on the designed building information and yield received by the transmitting/receiving unit 266 and the trained model stored in the trained model storage unit 264. (hereinafter simply referred to as "inspection information").

For example, first, the estimating unit 268 inputs the designed building information received by the transmitting/receiving unit 266 into the trained model. By inputting the designed building information to the learned model, the probabilities of each walk rate for each inputted designed building information are output as shown in FIG. 10 above.

Then, as shown in FIG. 11, the estimating unit 268 compares the walk rate “A” with the highest probability with the walk rate “X” received by the transmitting/receiving unit 266, and determines the structural design of the building to be designed. Generate inspection information indicating whether the inspection is appropriate or not.

For example, if the difference between the walk rate "A" and the walk rate "X" is greater than or equal to a predetermined threshold, the estimation unit 268 generates inspection information indicating that the structural design of the building to be designed is inappropriate. generate. Furthermore, if the difference between the walk rate "A" and the walk rate "X" is less than a predetermined threshold, the estimation unit 268 generates inspection information indicating that the structural design of the building to be designed is appropriate. generate.

The transmitting/receiving unit 266 transmits the test information estimated by the estimating unit 268 to the client terminal 211.

The transmitting/receiving unit 238 of the client terminal 211 receives the inspection information transmitted from the structural design support device 213 and outputs the inspection information to the control unit 236. The control unit 236 causes the test information to be displayed on the display device 240.

A user operating the client terminal 211 checks inspection information indicating whether the designed building information of the building he/she is designing is appropriate. This allows the user to know whether the structural design of the building he or she is designing is appropriate.

<Function of structure information display device>

Next, the operation of the structural design support system 210 will be explained. First, in the structural design support device 213, a learning processing routine shown in FIG. 12 is executed.

<Learning processing routine>

In step S200, the learning unit 262 reads each piece of learning data stored in the structural design database 260.

In step S202, the learning unit 262 causes a neural network, which is an example of a model, to learn by deep learning based on the learning data read out in step S200, thereby obtaining a learned model.

In step S204, the learning unit 262 stores the trained model obtained in step S202 above in the trained model storage unit 264.

Next, the user operating the client terminal 211 of the structural design support system 210 operates the client terminal 211 to create designed building information for the building to be designed. The user then transmits the created designed building information to the structural design device 212.

The transmitting/receiving unit 250 of the structural design device 212 receives the designed building information transmitted from the client terminal 211. Then, the yield calculation section 252 of the structural design device 212 calculates the yield of the designed building according to the designed building information received by the transmitting/receiving section 250. The transmitting/receiving unit 250 of the structural design device 212 transmits the walk rate of the building to be designed, calculated by the walk rate calculation unit 252, to the client terminal 211.

The transmitter/receiver 238 of the client terminal 211 receives the yield transmitted from the transmitter/receiver 250 of the structural design device 212 . Then, the user operates the operating device 220 of the client terminal 211 to transmit the designed building information obtained by the design unit 234 and the yield transmitted from the structural design device 212 to the structural design support device 213.

When the transmitting/receiving unit 266 of the structural design support device 213 receives the designed building information and the walk rate transmitted from the client terminal 211, the structural design support device 213 executes the estimation processing routine shown in FIG. 13.

In step S250, the estimating unit 268 acquires the designed building information and the walk rate received by the transmitting/receiving unit 266.

In step S252, the estimation unit 268 reads out the trained model stored in the trained model storage unit 264.

In step S254, the estimation unit 268 inputs the designed building information acquired in step S250 above to the learned model read out in step S252. As a result, the probability of each walk rate corresponding to the designed building information is output from the learned model.

In step S256, the estimating unit 268 identifies the rate with the highest probability among the probabilities of each rate obtained in step S254.

In step S258, the estimation unit 268 compares the yield obtained in step S250 and the yield specified in step S256. For example, the estimation unit 268 calculates the difference between the turn rate acquired in step S250 and the turn rate identified in step S256.

In step S260, the estimating unit 268 generates inspection information indicating whether or not the structural design of the building to be designed is appropriate, based on the difference between the walk rates obtained in step S258 and a preset threshold. generate.

The transmitting/receiving unit 266 of the structural design support device 213 transmits the inspection information obtained in step S260 above to the client terminal 211.

The transmitting/receiving unit 238 of the client terminal 211 receives the inspection information transmitted from the structural design support device 213 and causes the display device 240 to display the inspection information.

A user operating the client terminal 211 checks inspection information indicating whether the designed building information of the building he/she is designing is appropriate. This allows the user to know whether the structural design of the building he or she is designing is appropriate.

As described in detail above, in the second embodiment, information regarding the yield is output from information regarding the building based on information regarding the learning building and learning data representing the yield of the learning building. Perform machine learning on the model to obtain a trained model. Thereby, by using machine learning, it is possible to obtain a trained model that outputs information regarding the rate of sale in accordance with information regarding the building and information regarding the rate of sale.

In addition, in the second embodiment, the structural design of the building to be designed is appropriate based on the designed building information, information regarding the yield of the building to be designed, and a learned model that has been machine learned in advance from learning data. Outputs information regarding whether or not it is. Thereby, it is possible to appropriately support structural design based on a trained model that reflects past performance information.

<Configuration of structural design support system according to third embodiment>

In the third embodiment, simulation is performed according to initial conditions related to structural design. In the third embodiment, a trained model is generated by reinforcement learning based on the simulation results. This embodiment differs from the first and second embodiments in that structural design is supported using the obtained trained model.

FIG. 14 is a block diagram showing an example of the configuration of a structural design support system according to the third embodiment. Functionally, the structural design support system 310 can be represented by a configuration including a client terminal 211 and a structural design support device 213, as shown in FIG.

[Client terminal]

The user operates the operating device 220 of the client terminal 211 to set initial conditions for the building to be designed. The user then transmits the initial conditions of the building to be designed to the structural design support device 213 via the transmitter/receiver 238. As the initial conditions of the building to be designed, for example, CAD data in which the initial conditions of the building are set can be used.

[Structural design support device]

The structural design support device 213 is composed of a computer including a CPU, a ROM that stores programs for realizing each processing routine, a RAM that temporarily stores data, a memory as a storage means, a network interface, etc. . As shown in FIG. 15, the structural design support device 213 functionally includes a transmitting/receiving section 320, a design output section 321, a simulation section 322, a learning section 324, a learned model storage section 326, and a structure. and a design database 330. The transmitter/receiver 320 is an example of an acquisition unit of the present invention.

The transmitting/receiving unit 320 acquires the initial conditions of the building to be designed, which are transmitted from the client terminal 211.

The design output unit 321 changes the structure of the building to be designed corresponding to the initial conditions acquired by the transmitter/receiver 320, and generates structural design candidates for the building to be designed.

In the third embodiment, a trained model is generated by reinforcement learning. Therefore, the structure of the building to be designed that is changed by the design output unit 321 is set as the environment s in reinforcement learning. Furthermore, action a is to change the sizes of the structural members of the building to be designed, such as columns, beams, and floors. Further, the evaluation value corresponding to the state when the structural members of the building to be designed are changed is set as the reward r. Therefore, in this embodiment, the environment s is the structure of the building to be designed, the action a is to change the structure of the building, and the reward r is an evaluation based on the state when changing the structural members of the building. It is a value.

First, the design output unit 321 changes the CAD data indicating the building structure corresponding to the initial conditions or the building structural design candidate generated by the previous process.

For example, as a predetermined action _a1 , the design output unit 321 removes a predetermined pillar from a plurality of pillars included in the CAD data of the building to be designed. Alternatively, as a predetermined action _a2 , the design output unit 321 adds a pillar to a predetermined location in the CAD data of the building to be designed. Alternatively, as predetermined action _a3 , the design output unit 321 thickens the pillars included in the CAD data of the building to be designed. Alternatively, as predetermined action _a3 , the design output unit 321 thins the pillars included in the CAD data of the building to be designed. As a result, candidates for the structural design of the building to be designed are generated.

The simulation unit 322 evaluates each piece of CAD data obtained by changing the CAD data of the building to be designed. Specifically, the simulation unit 322 performs a simulation regarding structural calculation of a building on the structural design candidates generated by the design output unit 321, and evaluates the structural strength of the structural design candidates. Further, the simulation unit 322 performs a yield calculation, which is a simulation regarding the quantity of building materials, on the structural design candidates, and evaluates the cost of the structural design candidates. The simulation unit 322 then sets the reward r according to the simulation result.

For example, the simulation unit 322 evaluates each of the CAD data obtained by the predetermined actions a ₁ to a ₄ using a predetermined evaluation function related to structural design and yield. For example, with respect to CAD data obtained by a predetermined action _a1 , even if the yield is small, if the structural strength is low, the reward _r1 representing the evaluation will be small. Further, when the yield is small and the structural strength is high for the CAD data obtained by the predetermined action _a2 , the reward _r2 representing the evaluation becomes large.

The learning unit 324 performs reinforcement learning on a model for outputting the structural design of the building from the initial conditions of the building, based on the simulation results obtained by the simulation unit 322, to obtain a learned model. Specifically, the learning unit 324 performs reinforcement learning on a model for outputting the next structural design from the previously obtained structural design candidates based on the evaluation value for the structural design of the building among the simulation results. , iteratively updates the trained model.

For example, the learning unit 324 learns the action value function Q(s, a) according to the following equation (1) based on the simulation results (Internet (URL: http://www.sist.ac.jp/ ~kanakubo/research/reinforcement_learning.html).

(1)

Here, s _t represents the environment at time t. Further, a _t represents an action at time t. Due to the action a _t , the environment changes to s _t+1 . Further, r _t+1 is a reward given according to a change in the environment.

The term with max in the above equation (1) is the Q value multiplied by γ when the action a with the highest Q value (known at the time) is selected under the environment s _t+1 . Further, γ is a parameter satisfying 0<γ≦1 and is called a discount rate. α is a learning coefficient, and is in the range of 0<α≦1.

The behavior value function Q(s, a) as a model is learned according to the update formula shown in the above formula (1), and the behavior value function Q(s, a) as a learned model is generated. Note that the action value function Q(s, a) is sequentially updated according to the evaluation value regarding the structural strength of the structural design candidate and the evaluation value regarding the cost of the structural design candidate.

The learned model storage unit 326 stores the action value function Q(s, a) obtained by the learning unit 324. The action value function Q(s, a) is expressed, for example, by a Q table in which a Q value is stored for each combination of state s and action a. Therefore, the next action a (for example, an action that changes the size of columns, beams, floors, etc.) is determined according to the action value function Q(s, a) and the current building structure s. Ru.

When the design output unit 321 generates the next structural design candidate from the previously obtained structural design candidate, the design output unit 321 uses the action value function Q(s, a ), outputs structural design candidates for the building to be designed. By using the action value function Q(s, a), the design output unit 321 can generate candidates with high evaluation values regarding structural strength and high evaluation values regarding cost. Therefore, when generating the next candidate for structural design, it is easier to generate a more appropriate candidate than the previous one.

The generation of structural design candidates by the design output unit 321, the evaluation of structural design candidates by the simulation unit 322, and the learning process by the learning unit 324 are repeated a predetermined number of times. This generates multiple structural design candidates.

In this embodiment, each time a structural design candidate is generated, the action value function Q(s, a) is sequentially updated based on the reward that is the evaluation value of the structural design candidate obtained by the simulation unit 322. . Thereby, structural design candidates are generated whose evaluation value increases as the learning process by the learning unit 324 is repeated more often. Therefore, candidates suitable for structural design can be efficiently generated.

Next, after the process of the design output unit 321, the process of the simulation unit 322, and the process of the learning unit 324 are repeated a predetermined number of times, the design output unit 321 outputs the plurality of structural design candidates generated so far. Select. Specifically, the design output unit 321 selects a plurality of structural design candidates that have been generated so far, based on structural design performance information stored in the structural design database 330. For example, the structural design database 330 stores information as shown in FIG. 2 above as performance information.

By using the action value function Q(s, a), candidates with high evaluation values regarding structural strength and high evaluation values regarding cost are generated, while the structural design of a newly designed building is based on the past structural design. In some cases, it may be preferable to be similar to Therefore, in this embodiment, from a plurality of structural design candidates generated using the action value function Q(s, a), a candidate that is as close as possible to past structural design performance information is selected as a structural design candidate to be output. Sort so as to leave it as such.

Specifically, the design output unit 321 outputs the structural design candidates and the results stored in the structural design database 330 for each of the plurality of structural design candidates generated using the action value function Q(s, a). Calculate the degree of similarity between information. For example, the degree of similarity is calculated according to the structure type, building use, floor type, foundation type, number of floors, etc., and the verification ratio, live load, earthquake load, column dominated area, column span, etc. The more similar the structural design candidate is to the track record information of buildings constructed in the past, the greater the degree of similarity.

Next, the design output unit 321 selects structural design candidates whose degree of similarity is greater than a preset threshold value as structural design candidates to be output. Then, the design output unit 321 outputs a plurality of structural design candidates selected according to the track record information as a result.

The transmitting/receiving unit 320 transmits the plurality of structural design candidates output by the design output unit 321 to the client terminal 211.

The transmitting/receiving unit 238 of the client terminal 211 receives the plurality of structural design candidates transmitted from the structural design support device 213. Then, the control unit 236 of the client terminal 211 displays a plurality of structural design candidates on the display device 240.

The user selects the most appropriate structural design candidate as the final plan from among the plurality of structural design candidates displayed on the display device 240. This means that the structural design of the building to be designed can be obtained by setting only the initial conditions. The user then operates the operating device 220 of the client terminal 211 to transmit the structural design selected as the final plan to the structural design support device 213.

The transmitting/receiving unit 320 of the structural design support device 213 stores the final draft structural design transmitted from the client terminal 211 in the structural design database 330. Note that when storing the final plan structural design in the structural design database 330, weight information may be added to the final plan structural design and stored. Thereby, for example, structural design candidates can be selected according to the weight information.

<Function of the structural design support system>

Next, the operation of the structural design support system 310 will be explained. When the initial conditions of the building to be designed are transmitted from the client terminal 211 to the structural design support device 213, the structural design support device 213 executes the structural design support processing routine shown in FIG.

<Structural design support processing routine>

In step S300, the transmitting/receiving unit 320 acquires the initial conditions of the building to be designed, which are transmitted from the client terminal 211.

In step S302, the design output unit 321 changes the structure of the building to be designed corresponding to the initial conditions acquired in step S300 or the structural design candidate generated in the previous step S302 to Generate structural design candidates for the building. When generating structural design candidates by changing the structure of the building to be designed corresponding to the initial conditions, the design output unit 321 generates structural design candidates by, for example, randomly changing the structure of the building. . In addition, when the design output unit 321 generates the next structural design candidate from the structural design candidates obtained in the previous step S302, the design output unit 321 uses the trained model stored in the trained model storage unit 326. Based on the action value function Q(s, a), the next structural design candidate is output.

In step S304, the simulation unit 322 performs a simulation regarding the structural calculation of the building on the structural design candidates generated in step S302, and evaluates the structural strength of the structural design candidates. Further, the simulation unit 322 performs a yield calculation, which is a simulation regarding the cost of a building, on the structural design candidates generated in step S302, and evaluates the cost of the structural design candidates. The simulation unit 322 then sets the reward r according to the simulation result.

In step S306, the learning unit 324 performs reinforcement learning on the model for outputting the structural design of the building from the initial conditions of the building or the structural design candidates generated last time, based on the simulation results obtained in step S304. Then, the action value function Q(s, a) as a trained model is obtained.

In step S308, the learning unit 324 stores the action value function Q(s, a), which is the learned model obtained in step S306, in the learned model storage unit 326, and updates the learned model.

In step S310, it is determined whether the processes from step S302 to step S308 have been repeated a predetermined number of times. If the processes from step S302 to step S308 have been repeated a predetermined number of times, the process advances to step S312. On the other hand, if the processes from step S302 to step S308 have not been repeated a predetermined number of times, the process returns to step S302.

In step S312, the design output unit 321 reads performance information stored in the structural design database 330.

In step S314, the design output unit 321 calculates the degree of similarity between the plurality of structural design candidates generated in step S302 and the track record information read out in step S312. Then, the design output unit 321 selects structural design candidates whose degree of similarity is greater than a preset threshold value as structural design candidates to be output.

In step S316, the transmitting/receiving unit 320 transmits the plurality of structural design candidates obtained in step S314 above to the client terminal 211.

The transmitting/receiving unit 238 of the client terminal 211 receives the plurality of structural design candidates transmitted from the structural design support device 213. Then, the control unit 236 of the client terminal 211 displays a plurality of structural design candidates on the display device 240.

The user selects the most appropriate structural design candidate as the final plan from among the plurality of structural design candidates displayed on the display device 240. The user then operates the operating device 220 of the client terminal 211 to transmit the structural design selected as the final plan to the structural design support device 213.

The transmitting/receiving unit 320 of the structural design support device 213 stores the final draft structural design transmitted from the client terminal 211 in the structural design database 330.

As described in detail above, the third embodiment relates to structural calculation of a building for structural design of a building in which the structure of the building to be designed is changed according to the initial conditions of the building to be designed. Perform simulations and simulations regarding quantities of building materials. In the third embodiment, a model for outputting the structural design of the building from the initial conditions of the building is subjected to reinforcement learning based on the evaluation value for the structural design of the building according to the simulation result. In the third embodiment, a structural design of a building to be designed is output based on the obtained learned model. By using reinforcement learning, it is possible to obtain a trained model that outputs a building structural design from each of the building structural design candidates. In addition, by outputting multiple structural design candidates for a building based on the trained model obtained through reinforcement learning, structural design can be appropriately supported.

In addition, structural design candidates for the building to be designed can be obtained simply by inputting the initial conditions of the design target. In addition, a trained model is generated based on the evaluation value regarding structural calculation and the evaluation value regarding cost among the simulation results, and structural design candidates are output based on the trained model. Structural design candidates that are considered appropriate from the viewpoint of evaluation values can be obtained.

In addition, in the third embodiment, based on the simulation results, reinforcement learning is performed on a model for outputting the next structural design of the building to be designed from the structural design of the building to be designed that was obtained last time. Repeat to update the model. Therefore, by using a trained model that is updated sequentially, structural design candidates can be obtained appropriately and efficiently.

Furthermore, since the trained model obtained by reinforcement learning is also used in the structural design of other buildings to be designed, the trained model is updated every time processing is performed. Therefore, candidates for structural design can be obtained appropriately and efficiently.

Further, in the third embodiment, from each of the structural designs outputted based on the trained model obtained by reinforcement learning, a structural design whose similarity with track record information is equal to or higher than a threshold is outputted. Thereby, structural design candidates similar to the track record information are selected, so that structural design candidates similar to the structural designs of past construction work can be obtained.

Note that the present invention is not limited to the embodiments described above, and various modifications and applications can be made without departing from the gist of the present invention.

For example, in the above embodiment, the case where the yield is used as an example of information regarding the quantity of building materials has been described, but the information is not limited to this, and any information regarding the quantity of building materials may be used. information may be used.

In addition, in the second embodiment, an example is given in which information regarding whether the structural design of the building to be designed is appropriate is output using the yield, which is an example of information regarding the quantity of building materials. Although described above, the invention is not limited thereto. For example, information about whether the structural design of the building to be designed is appropriate may be output using information different from information about the quantity of building materials. In this case, for example, building scale, number of floors, floor area, structure type, structural calculation route, load table, span, floor height, wall amount, bracing ratio, material strength, material edge conditions, live load, finishing load, Additional loads, etc., earthquake load, wind load, soil water pressure load, building drawing recognition results, building use, column axial force, column pitch (spacing), building material, building designer, design year, construction site, cost estimate As a result, information regarding whether the structural design of the building to be designed is appropriate may be output using the verification ratio, floor type, roof type, presence or absence of braces, and the like.

Further, in the second embodiment, a case has been described in which a neural network model as an example of a learning model is trained by deep learning, but the present invention is not limited to this. For example, a model different from the neural network model may be trained using a learning method different from deep learning.

Furthermore, in the second embodiment, the case where the walk rate with the highest probability is compared with the walk rate calculated by the structural design device 212 is exemplified according to the probability of each walk rate output from the trained model. , but is not limited to this. For example, as shown in FIG. 17, the designed building information and the walk rate calculated by the structural design device 212 are input to the trained model, and the probability that the combination is appropriate (probability that it is OK) and the The trained model may be configured to output the probability that the combination is inappropriate (probability that the combination is NG).

Further, in the second embodiment, the case where information regarding whether the structure of the building to be designed is appropriate or not is simply outputted according to the designed building information and the walk rate has been explained as an example. It is not limited.

For example, information regarding the yield of the learning building in the learning data may be set as information regarding the yield corresponding to each location of the learning building. In this case, the learned model is configured to calculate the yield of each location of the building to be designed, and the estimation unit 268 determines whether the structural design of each location of the building to be designed is appropriate. It may also be possible to output further information regarding. As a result, it is automatically pointed out that, for example, the cross section of a beam or column at a specific location is designed to be small, so that support for structural design by a structural designer can be automatically provided. For example, in accordance with the design building information, by further outputting information indicating that for a building of this size, the beam yield on the You can understand whether the parts are inappropriate or not. Furthermore, the trained model may be configured to further provide advice regarding the building yield based on information regarding which locations are inappropriate. For example, advice may be given to the effect that the amount of concrete in a location determined to be inappropriate is insufficient.

Further, the trained model may be configured to provide advice regarding general structural design. In this case, for example, advice on analysis condition settings (for example, it will tell you which items have been changed from the default with a high probability when compared with similar cases), advice on trends in setting values (for example, when A If a change is made, B is also changed), advice on the validity of the amount of reinforcement (e.g. columns, girders, small beams, shear walls, etc.), advice on load settings (e.g. comparison with similar cases to reduce leakage). The model may be configured to output advice regarding the validity of the set cross section (beam size), and advice regarding the validity of the set cross section (column size). Further, as a past case example search, it may be configured such that past construction work cases, which are track record information, can be searched.

Regarding advice on setting analysis conditions and advice on trends in setting values, building scale, number of floors, floor area, structure type, structural calculation route, etc. are assumed to be factors, so advice is provided based on this information.

In addition, when it comes to advice on the validity of the amount of reinforcement, factors such as load tables, spans, story heights, wall amounts, bracing ratios, number of stories, material strength, and material edge conditions are assumed to be factors; Advice will be given based on this.

In addition, regarding load setting advice, factors such as live load, finishing load, additional load, earthquake load, wind load, earth and water pressure load, building drawing recognition results, building use, and building size are assumed to be factors. Advice is given based on this information.

In addition, as for advice on the validity of the set cross section (beam size), the live load, finishing load, span, number of floors, structure type, timber edge conditions, etc. are assumed to be factors, so advice is given based on this information. be exposed.

Further, regarding advice regarding the validity of the set cross section (column size), since the column axial force, number of floors, column pitch (spacing), structure type, etc. are assumed to be factors, advice is given based on this information.

Furthermore, the past case example search is configured to be performed in accordance with, for example, building scale, number of floors, floor area, structure type, and building drawing recognition results.

When giving advice on structural design, for example, multiple buildings are classified in advance by clustering multiple buildings corresponding to track record information, and then the structure Advice regarding design may also be provided. As an example of a model used when clustering a plurality of buildings, for example, the k-means method can be used.

In addition, although the above description has been made of a mode in which the program is stored (installed) in advance in the storage unit (not shown), the program can be recorded on any recording medium such as a CD-ROM, DVD-ROM, or micro SD card. It is also possible to provide it in the form shown below.

10 Structural information display device 12, 220 Operating device 20, 230 Computer 22, 260, 330 Structural design database 24, 232 Reception section 26 Information acquisition section 28 Display section 30, 240 Display device 210, 310 Structural design support system 211 Client terminal 212 Structural design device 213 Structural design support device 234 Design section 236 Control section 238, 250, 266, 320 Transmission/reception section 252 Yield calculation section 262, 324 Learning section 264, 326 Learned model storage section 268 Estimation section 321 Design output section 322 Simulation Department

Claims (3)

an acquisition unit that acquires initial conditions of the building to be designed;
At least one of a simulation regarding the structural calculation of the building and a simulation regarding the quantity of building materials for the structural design of the building when the structure of the building to be designed in the initial conditions acquired by the acquisition unit is changed. A simulation department that performs
Based on the evaluation value for the structural design of the building according to the simulation result obtained by the simulation unit, a model for outputting the structural design of the building from the initial conditions of the building is subjected to reinforcement learning to determine the initial conditions of the building. a learning unit that obtains a trained model that outputs the structural design of the building from;
a design output unit that outputs a structural design of the building to be designed based on the learned model obtained by the learning unit;
Structural design support equipment including The learning unit performs reinforcement learning on a model for outputting the next structural design of the building to be designed based on the previously obtained structural design of the building to be designed, based on the simulation result, and Iterate to update the model,
The design output unit outputs a next structural design of the building to be designed based on the previously obtained structural design of the building to be designed, based on the learned model updated by the learning unit.
The structural design support device according to claim 1. The design output unit reads the track record information from a database storing track record information regarding building structures, and calculates the degree of similarity between each of the structural designs output based on the learned model and the track record information. outputting the structural design that is greater than or equal to a threshold;
A structural design support device according to claim 1 or claim 2.